diff --git a/data/resources/shaders/reshade/Shaders/crt-royale.fx b/data/resources/shaders/reshade/Shaders/crt-royale.fx new file mode 100644 index 000000000..198b7e920 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale.fx @@ -0,0 +1,244 @@ +#include "ReShade.fxh" + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + +// Enable or disable the shader +#ifndef CONTENT_BOX_VISIBLE + #define CONTENT_BOX_VISIBLE 0 +#endif + +#include "crt-royale/shaders/content-box.fxh" + +#if !CONTENT_BOX_VISIBLE + #include "crt-royale/shaders/input-blurring.fxh" + #include "crt-royale/shaders/electron-beams.fxh" + #include "crt-royale/shaders/blurring.fxh" + #include "crt-royale/shaders/deinterlace.fxh" + #include "crt-royale/shaders/phosphor-mask.fxh" + #include "crt-royale/shaders/brightpass.fxh" + #include "crt-royale/shaders/bloom.fxh" + #include "crt-royale/shaders/geometry-aa-last-pass.fxh" +#endif + + +technique CRT_Royale +{ + // Toggle the content box to help users configure it + #if CONTENT_BOX_VISIBLE + pass contentBoxPass + { + // content-box.fxh + // Draw a box that displays the crop we'll perform. + VertexShader = PostProcessVS; + PixelShader = contentBoxPixelShader; + } + #else + #if ENABLE_PREBLUR + pass PreblurVert + { + // input-blurring.fxh + // Optionally blur the input buffer a little + VertexShader = contentCropVS; + PixelShader = preblurVertPS; + + RenderTarget = texPreblurVert; + + PrimitiveTopology = TRIANGLESTRIP; + VertexCount = 4; + } + pass PreblurHoriz + { + // input-blurring.fxh + VertexShader = PostProcessVS; + PixelShader = preblurHorizPS; + + RenderTarget = texPreblurHoriz; + } + #endif + pass beamDistPass + { + // electron-beams.fxh + // Simulate emission of the interlaced video as electron beams. + VertexShader = calculateBeamDistsVS; + PixelShader = calculateBeamDistsPS; + + RenderTarget = texBeamDist; + + // This lets us improve performance by only computing the mask every k frames + ClearRenderTargets = false; + } + pass electronBeamPass + { + // electron-beams.fxh + // Simulate emission of the interlaced video as electron beams. + VertexShader = simulateEletronBeamsVS; + PixelShader = simulateEletronBeamsPS; + + RenderTarget = texElectronBeams; + + // If the preblur passes are disabled, we have to crop in this pass + #if !ENABLE_PREBLUR + PrimitiveTopology = TRIANGLESTRIP; + VertexCount = 4; + #endif + } + pass beamConvergencePass + { + // electron-beams.fxh + // Simulate beam convergence miscalibration + // Not to be confused with beam purity + VertexShader = beamConvergenceVS; + PixelShader = beamConvergencePS; + + RenderTarget = texBeamConvergence; + } + pass bloomApproxPassVert + { + // bloom.fxh + VertexShader = PostProcessVS; + PixelShader = approximateBloomVertPS; + + RenderTarget = texBloomApproxVert; + } + pass bloomApproxPassHoriz + { + // bloom.fxh + VertexShader = PostProcessVS; + PixelShader = approximateBloomHorizPS; + + RenderTarget = texBloomApproxHoriz; + } + pass blurVerticalPass + { + // blurring.fxh + // Vertically blur the approx bloom + VertexShader = blurVerticalVS; + PixelShader = blurVerticalPS; + + RenderTarget = texBlurVertical; + } + pass blurHorizontalPass + { + // blurring.fxh + // Horizontally blur the approx bloom + VertexShader = blurHorizontalVS; + PixelShader = blurHorizontalPS; + + RenderTarget = texBlurHorizontal; + } + pass deinterlacePass + { + // deinterlace.fxh + // Optionally deinterlace the video if interlacing is enabled. + // Can help approximate the original crt-royale's appearance + // without some issues like image retention. + VertexShader = deinterlaceVS; + PixelShader = deinterlacePS; + + RenderTarget = texDeinterlace; + } + pass freezeFramePass + { + // deinterlace.fxh + // Capture the current frame, so we can use it in the next + // frame's deinterlacing pass. + VertexShader = freezeFrameVS; + PixelShader = freezeFramePS; + + RenderTarget = texFreezeFrame; + + // Explicitly disable clearing render targets + // scanlineBlendPass will not work properly if this ever defaults to true + ClearRenderTargets = false; + } + pass generatePhosphorMask + { + // phosphor-mask.fxh + VertexShader = generatePhosphorMaskVS; + PixelShader = generatePhosphorMaskPS; + + RenderTarget = texPhosphorMask; + + // This lets us improve performance by only computing the mask every k frames + ClearRenderTargets = false; + + PrimitiveTopology = TRIANGLESTRIP; + VertexCount = 4; + } + pass applyPhosphormask + { + // phosphor-mask.fxh + // Tile the scaled phosphor mask and apply it to + // the deinterlaced image. + VertexShader = PostProcessVS; + PixelShader = applyComputedPhosphorMaskPS; + + RenderTarget = texMaskedScanlines; + // RenderTarget = texGeometry; + } + pass brightpassPass + { + // brightpass.fxh + // Apply a brightpass filter for the bloom effect + VertexShader = brightpassVS; + PixelShader = brightpassPS; + + RenderTarget = texBrightpass; + } + pass bloomVerticalPass + { + // bloom.fxh + // Blur vertically for the bloom effect + VertexShader = bloomVerticalVS; + PixelShader = bloomVerticalPS; + + RenderTarget = texBloomVertical; + } + pass bloomHorizontalPass + { + // bloom.fxh + // Blur horizontally for the bloom effect. + // Also apply various color changes and effects. + VertexShader = bloomHorizontalVS; + PixelShader = bloomHorizontalPS; + + RenderTarget = texBloomHorizontal; + } + pass geometryPass + { + // geometry-aa-last-pass.fxh + // Apply screen geometry and anti-aliasing. + VertexShader = geometryVS; + PixelShader = geometryPS; + + RenderTarget = texGeometry; + } + pass uncropPass + { + // content-box.fxh + // Uncrop the video, so we draw the game's content + // in the same position it started in. + VertexShader = contentUncropVS; + PixelShader = uncropContentPixelShader; + + PrimitiveTopology = TRIANGLESTRIP; + VertexCount = 4; + } + #endif +} \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/bind-shader-params.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/bind-shader-params.fxh new file mode 100644 index 000000000..7bce0fff6 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/bind-shader-params.fxh @@ -0,0 +1,908 @@ +#ifndef _BIND_SHADER_PARAMS_H +#define _BIND_SHADER_PARAMS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +///////////////////////////// SETTINGS MANAGEMENT //////////////////////////// + +/////////////////////////////// BEGIN INCLUDES /////////////////////////////// +#include "helper-functions-and-macros.fxh" +#include "user-settings.fxh" +#include "derived-settings-and-constants.fxh" +#include "../version-number.fxh" + +//////////////////////////////// END INCLUDES //////////////////////////////// + +// Override some parameters for gamma-management.h and tex2Dantialias.h: +#ifndef _OVERRIDE_DEVICE_GAMMA + #define _OVERRIDE_DEVICE_GAMMA 1 +#endif + +#if __RENDERER__ != 0x9000 + #define _DX9_ACTIVE 0 +#else + #define _DX9_ACTIVE 1 +#endif + +// #ifndef ANTIALIAS_OVERRIDE_BASICS +// #define ANTIALIAS_OVERRIDE_BASICS 1 +// #endif + +// #ifndef ANTIALIAS_OVERRIDE_PARAMETERS +// #define ANTIALIAS_OVERRIDE_PARAMETERS 1 +// #endif + +#ifndef ADVANCED_SETTINGS + #define ADVANCED_SETTINGS 0 +#endif + +// The width of the game's content +#ifndef CONTENT_WIDTH + #define CONTENT_WIDTH BUFFER_WIDTH +#endif +// The height of the game's content +#ifndef CONTENT_HEIGHT + #define CONTENT_HEIGHT BUFFER_HEIGHT +#endif + +#if ADVANCED_SETTINGS == 1 + // Using vertex uncropping is marginally faster, but vulnerable to DX9 weirdness. + // Most users will likely prefer the slower algorithm. + #ifndef USE_VERTEX_UNCROPPING + #define USE_VERTEX_UNCROPPING 0 + #endif + + #ifndef NUM_BEAMDIST_COLOR_SAMPLES + #define NUM_BEAMDIST_COLOR_SAMPLES 1024 + #endif + + #ifndef NUM_BEAMDIST_DIST_SAMPLES + #define NUM_BEAMDIST_DIST_SAMPLES 120 + #endif + + #ifndef BLOOMAPPROX_DOWNSIZING_FACTOR + #define BLOOMAPPROX_DOWNSIZING_FACTOR 4.0 + #endif + + // Define this internal value, so ADVANCED_SETTINGS == 0 doesn't cause a redefinition error when + // NUM_BEAMDIST_COLOR_SAMPLES defined in the preset file. Also makes it easy to avoid bugs + // related to parentheses and order-of-operations when the user defines this arithmetically. + static const uint num_beamdist_color_samples = uint(NUM_BEAMDIST_COLOR_SAMPLES); + static const uint num_beamdist_dist_samples = uint(NUM_BEAMDIST_DIST_SAMPLES); + static const float bloomapprox_downsizing_factor = float(BLOOMAPPROX_DOWNSIZING_FACTOR); +#else + static const uint USE_VERTEX_CROPPING = 0; + static const uint num_beamdist_color_samples = 1024; + static const uint num_beamdist_dist_samples = 120; + static const float bloomapprox_downsizing_factor = 4.0; +#endif + +#ifndef HIDE_HELP_SECTIONS + #define HIDE_HELP_SECTIONS 0 +#endif + + +// Offset the center of the game's content (horizontal) +#ifndef CONTENT_CENTER_X + #define CONTENT_CENTER_X 0 +#endif +// Offset the center of the game's content (vertical) +#ifndef CONTENT_CENTER_Y + #define CONTENT_CENTER_Y 0 +#endif + +// Wrap the content size in parenthesis for internal use, so the user doesn't have to +static const float2 content_size = float2(int(CONTENT_WIDTH), int(CONTENT_HEIGHT)); + +#ifndef ENABLE_PREBLUR + #define ENABLE_PREBLUR 1 +#endif + + +static const float2 buffer_size = float2(BUFFER_WIDTH, BUFFER_HEIGHT); + + +// The normalized center is 0.5 plus the normalized offset +static const float2 content_center = float2(CONTENT_CENTER_X, CONTENT_CENTER_Y) / buffer_size + 0.5; +// The content's normalized diameter d is its size divided by the buffer's size. The radius is d/2. +static const float2 content_radius = content_size / (2.0 * buffer_size); +static const float2 content_scale = content_size / buffer_size; + +static const float content_left = content_center.x - content_radius.x; +static const float content_right = content_center.x + content_radius.x; +static const float content_upper = content_center.y - content_radius.y; +static const float content_lower = content_center.y + content_radius.y; + +// The xy-offset of the top-left pixel in the content box +static const float2 content_offset = float2(content_left, content_upper); +static const float2 content_offset_from_right = float2(content_right, content_lower); + +uniform uint frame_count < source = "framecount"; >; +uniform int overlay_active < source = "overlay_active"; >; + +static const float gba_gamma = 3.5; // Irrelevant but necessary to define. + + +// === HELP AND INFO === + +uniform int APPEND_VERSION_SUFFIX(version) < + ui_text = "Version: " DOT_VERSION_STR; + ui_label = " "; + ui_type = "radio"; +>; + +uniform int basic_setup_help < + ui_text = "1. Configure the Content Box if your game has letter-boxing.\n" + "2. Configure the Phosphor Mask.\n" + "3. Configure the Scanlines.\n" + "4. Configure the Colors and Effects.\n" + "5. Configure the Screen Geometry.\n" + "6. Configure or disable Preblur\n\n" + "- In Preprocessor Definitions, set ADVANCED_SETTINGS to 1 to access more settings.\n"; + ui_category = "Basic Setup Instructions"; + ui_category_closed = true; + ui_label = " "; + ui_type = "radio"; + hidden = HIDE_HELP_SECTIONS; +>; + +uniform int content_box_help < + ui_text = "1. Expand the Preprocessor Definitions section.\n" + "2. Set CONTENT_BOX_VISIBLE to 1.\n" + "3. Use the \"CONTENT_\" parameters to configure the Content Box.\n" + "4. Align the content box with the border of your game.\n" + "5. Set CONTENT_BOX_VISIBLE to 0 when you're done.\n\n" + "Parameters to focus on:\n" + "- CONTENT_HEIGHT and CONTENT_WIDTH\n" + "- CONTENT_CENTER_X and CONTENT_CENTER_Y\n" + "- CONTENT_BOX_INSCRIBED\n\n" + "Fancy Trick 1:\n" + "\tCONTENT_HEIGHT = BUFFER_HEIGHT\n" + "\tCONTENT_WIDTH = CONTENT_HEIGHT * 4.0 / 3.0\n" + "- Good if your game fills the screen vertically and has a 4:3 aspect ratio.\n" + "- Will also rescale automatically if you resize the window.\n\n" + "Fancy Trick 2:\n" + "\tCONTENT_HEIGHT = CONTENT_WIDTH * 9.0 / 16.0\n" + "\tCONTENT_WIDTH = 1500\n" + "- Good if your game is 1500 pixels wide with a 16:9 aspect ratio.\n" + "- Won't rescale automatically, but you'd only have to change the width.\n"; + ui_category = "Content Box Instructions"; + ui_category_closed = true; + ui_label = " "; + ui_type = "radio"; + hidden = HIDE_HELP_SECTIONS; +>; + + +// ==== PHOSPHOR MASK ==== +uniform int mask_type < + #if !HIDE_HELP_SECTIONS + ui_text = "Choose which kind of CRT you want.\n\n"; + #endif + ui_label = "Mask Type"; + ui_tooltip = "Selects the phosphor shape"; + ui_type = "combo"; + ui_items = "Grille\0" + "Slot\0" + "Shadow\0" + "LowRes Grille\0" + "LowRes Slot\0" + "LowRes Shadow\0"; + + ui_category = "Phosphor Mask"; + ui_category_closed = true; +> = mask_type_static; + +uniform uint mask_size_param < + ui_label = "Mask Size Param"; + ui_tooltip = "Switch between using Mask Triad Size or Mask Num Triads"; + ui_type = "combo"; + ui_items = "Triad Width\0" + "Num Triads Across\0"; + hidden = !ADVANCED_SETTINGS; + + ui_spacing = 2; + ui_category = "Phosphor Mask"; +> = mask_size_param_static; + +uniform float mask_triad_width < + ui_label = "Mask Triad Width"; + ui_tooltip = "The width of a triad in pixels"; + ui_type = "slider"; + ui_min = 1.0; + ui_max = 60.0; + ui_step = 0.1; + + ui_category = "Phosphor Mask"; +> = mask_triad_width_static; + +uniform float mask_num_triads_across < + ui_label = "Mask Num Triads Across"; + ui_tooltip = "The number of triads in the viewport (horizontally)"; + ui_type = "drag"; + ui_min = 1.0; + ui_max = 1280.0; + ui_step = 1.0; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Phosphor Mask"; +> = mask_num_triads_across_static; + +uniform float scale_triad_height< + ui_label = "Scale Triad Height"; + ui_tooltip = "Scales the height of a triad"; + ui_type = "drag"; + ui_min = 0.01; + ui_max = 10.0; + ui_step = 0.001; + + ui_spacing = 2; + ui_category = "Phosphor Mask"; +> = 1.0; + +uniform float2 phosphor_thickness < + ui_label = "Phosphor Thickness XY"; + ui_tooltip = "Makes the phosphors appear thicker in each direction"; + ui_type = "drag"; + ui_min = 0.01; + ui_max = 0.99; + ui_step = 0.01; + // hidden = !ADVANCED_SETTINGS; + + ui_category = "Phosphor Mask"; +> = 0.2; + +uniform float2 phosphor_sharpness < + ui_label = "Phosphor Sharpness XY"; + ui_tooltip = "Makes the phosphors appear more crisp in each direction"; + ui_type = "drag"; + ui_min = 1; + ui_max = 100; + ui_step = 1; + // hidden = !ADVANCED_SETTINGS; + + ui_category = "Phosphor Mask"; +> = 50; + +uniform float3 phosphor_offset_x < + ui_label = "Phosphor Offset RGB X"; + ui_tooltip = "Very slightly shifts the phosphor mask. Can help with subpixel alignment."; + ui_type = "drag"; + ui_min = -1; + ui_max = 1; + ui_step = 0.01; + // hidden = !ADVANCED_SETTINGS; + + ui_spacing = 2; + ui_category = "Phosphor Mask"; +> = 0; + +uniform float3 phosphor_offset_y < + ui_label = "Phosphor Offset RGB Y"; + ui_tooltip = "Very slightly shifts the phosphor mask. Can help with subpixel alignment."; + ui_type = "drag"; + ui_min = -1; + ui_max = 1; + ui_step = 0.01; + // hidden = !ADVANCED_SETTINGS; + + ui_category = "Phosphor Mask"; +> = 0; + +// static const uint pixel_grid_mode = 0; +// static const float2 pixel_size = 1; +/* +// ==== PIXELATION === +uniform uint pixel_grid_mode < + #if !HIDE_HELP_SECTIONS + ui_text = "- Fix issues displaying pixel art.\n" + "- Force high-res games to look low-res.\n\n"; + #endif + ui_label = "Pixel Grid Param"; + ui_tooltip = "Switch between using Pixel Size or Num Pixels"; + ui_type = "combo"; + ui_items = "Pixel Size\0" + "Content Resolution\0"; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Pixelation"; + ui_category_closed = true; +> = 0; + +uniform float2 pixel_size < + #if !HIDE_HELP_SECTIONS && !ADVANCED_SETTINGS + ui_text = "- Fix issues displaying pixel art.\n" + "- Force high-res games to look low-res.\n\n"; + #endif + ui_label = "Pixel Size"; + ui_tooltip = "The size of an in-game pixel on screen, in real-world pixels"; + ui_type = "slider"; + ui_min = 1.0; + ui_max = 30.0; + ui_step = 1.0; + + ui_category = "Pixelation"; + ui_category_closed = true; +> = float2(1, 1); + +uniform float2 pixel_grid_resolution < + ui_label = "Num Pixels"; + ui_tooltip = "The number of in-game pixels displayed on-screen in each direction"; + ui_type = "drag"; + ui_min = 1.0; + ui_max = 10000.0; + ui_step = 1.0; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Pixelation"; +> = content_size; +uniform float2 pixel_grid_offset < + ui_label = "Pixel Grid Offset"; + ui_tooltip = "Shifts the pixel-grid to help with alignment"; + ui_type = "slider"; + ui_min = -15.0; + ui_max = 15.0; + ui_step = 1.0; + + #if ADVANCED_SETTINGS + ui_spacing = 2; + #endif + ui_category = "Pixelation"; +> = float2(0, 0); +*/ + +// ==== SCANLINES ==== +uniform uint scanline_thickness < + #if !HIDE_HELP_SECTIONS + ui_text = "Configure the electron beams and interlacing.\n\n"; + #endif + ui_label = "Scanline Thickness"; + ui_tooltip = "Sets the height of each scanline"; + ui_type = "slider"; + ui_min = 1; + ui_max = 30; + ui_step = 1; + + ui_category = "Scanlines"; + ui_category_closed = true; +> = 2; + +uniform float scanline_offset < + ui_label = "Scanline Offset"; + ui_tooltip = "Vertically shifts the scanlines to help with alignment"; + ui_type = "slider"; + ui_min = -30; + ui_max = 30; + ui_step = 1; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Scanlines"; +> = 0; + +uniform uint beam_shape_mode < + ui_label = "Beam Shape Mode"; + ui_tooltip = "Select the kind of beam to use."; + ui_type = "combo"; + ui_items = "Digital (Fast)\0" + "Linear (Simple)\0" + "Gaussian (Realistic)\0" + "Multi-Source Gaussian (Expensive)\0"; + + ui_category = "Scanlines"; +> = 1; + +uniform bool enable_interlacing < + ui_label = "Enable Interlacing"; + + ui_spacing = 5; + ui_category = "Scanlines"; +> = false; + +uniform bool interlace_back_field_first < + ui_label = "Draw Back-Field First"; + ui_tooltip = "Draw odd-numbered scanlines first (often has no effect)"; + + ui_category = "Scanlines"; +> = interlace_back_field_first_static; + +uniform uint scanline_deinterlacing_mode < + ui_label = "Deinterlacing Mode"; + ui_tooltip = "Selects the deinterlacing algorithm, if any."; + ui_type = "combo"; + ui_items = "None\0" + "Fake-Progressive\0" + "Weaving\0" + "Blended Weaving\0"; + + ui_category = "Scanlines"; +> = 1; + +uniform float deinterlacing_blend_gamma < + ui_label = "Deinterlacing Blend Gamma"; + ui_tooltip = "Nudge this if deinterlacing changes your colors too much"; + ui_type = "slider"; + ui_min = 0.01; + ui_max = 5.0; + ui_step = 0.01; + + ui_category = "Scanlines"; +> = 1.0; + +uniform float linear_beam_thickness < + ui_label = "Linear Beam Thickness"; + ui_tooltip = "Linearly widens or narrows the beam"; + ui_type = "slider"; + ui_min = 0.01; + ui_max = 3.0; + ui_step = 0.01; + + ui_spacing = 5; + ui_category = "Scanlines"; +> = 1.0; + +uniform float gaussian_beam_min_sigma < + ui_label = "Gaussian Beam Min Sigma"; + ui_tooltip = "For Gaussian Beam Shape, sets thickness of dim pixels"; + ui_type = "drag"; + ui_min = 0.0; + ui_step = 0.01; + + ui_spacing = 5; + ui_category = "Scanlines"; +> = gaussian_beam_min_sigma_static; + +uniform float gaussian_beam_max_sigma < + ui_label = "Gaussian Beam Max Sigma"; + ui_tooltip = "For Gaussian Beam Shape, sets thickness of bright pixels"; + ui_type = "drag"; + ui_min = 0.0; + ui_step = 0.01; + + ui_category = "Scanlines"; +> = gaussian_beam_max_sigma_static; + +uniform float gaussian_beam_spot_power < + ui_label = "Gaussian Beam Spot Power"; + ui_tooltip = "For Gaussian Beam Shape, balances between Min and Max Sigma"; + ui_type = "drag"; + ui_min = 0.0; + ui_step = 0.01; + + ui_category = "Scanlines"; +> = gaussian_beam_spot_power_static; + +uniform float gaussian_beam_min_shape < + ui_label = "Gaussian Beam Min Shape"; + ui_tooltip = "For Gaussian Beam Shape, sets sharpness of dim pixels"; + ui_type = "drag"; + ui_min = 0.0; + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS; + + ui_spacing = 2; + ui_category = "Scanlines"; +> = gaussian_beam_min_shape_static; + +uniform float gaussian_beam_max_shape < + ui_label = "Gaussian Beam Max Shape"; + ui_tooltip = "For Gaussian Beam Shape, sets sharpness of bright pixels"; + ui_type = "drag"; + ui_min = 0.0; + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Scanlines"; +> = gaussian_beam_max_shape_static; + +uniform float gaussian_beam_shape_power < + ui_label = "Gaussian Beam Shape Power"; + ui_tooltip = "For Gaussian Beam Shape, balances between Min and Max Shape"; + ui_type = "drag"; + ui_min = 0.0; + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Scanlines"; +> = gaussian_beam_shape_power_static; + +uniform float3 convergence_offset_x < + ui_label = "Convergence Offset X RGB"; + ui_tooltip = "Shift the color channels horizontally"; + ui_type = "drag"; + ui_min = -10; + ui_max = 10; + ui_step = 0.05; + hidden = !ADVANCED_SETTINGS; + + ui_spacing = 5; + ui_category = "Scanlines"; +> = 0; +uniform float3 convergence_offset_y < + ui_label = "Convergence Offset Y RGB"; + ui_tooltip = "Shift the color channels vertically"; + ui_type = "drag"; + ui_min = -10; + ui_max = 10; + ui_step = 0.05; + hidden = !ADVANCED_SETTINGS; + ui_category = "Scanlines"; +> = 0; + +static uint beam_horiz_filter = beam_horiz_filter_static; +static float beam_horiz_sigma = beam_horiz_sigma_static; +static float beam_horiz_linear_rgb_weight = beam_horiz_linear_rgb_weight_static; + +// ==== IMAGE COLORIZATION ==== +uniform float crt_gamma < + #if !HIDE_HELP_SECTIONS + ui_text = "Apply gamma, contrast, and blurring.\n\n"; + #endif + ui_label = "CRT Gamma"; + ui_tooltip = "The gamma-level of the original content"; + ui_type = "slider"; + ui_min = 1.0; + ui_max = 5.0; + ui_step = 0.01; + + ui_category = "Colors and Effects"; + ui_category_closed = true; +> = crt_gamma_static; + +uniform float lcd_gamma < + ui_label = "LCD Gamma"; + ui_tooltip = "The gamma-level of your display"; + ui_type = "slider"; + ui_min = 1.0; + ui_max = 5.0; + ui_step = 0.01; + + ui_category = "Colors and Effects"; +> = lcd_gamma_static; + +uniform float levels_contrast < + ui_label = "Levels Contrast"; + ui_tooltip = "Sets the contrast of the CRT"; + ui_type = "slider"; + ui_min = 0.0; + ui_max = 4.0; + ui_step = 0.01; + + ui_spacing = 5; + ui_category = "Colors and Effects"; +> = levels_contrast_static; + +uniform float halation_weight < + ui_label = "Halation"; + ui_tooltip = "Desaturation due to eletrons exciting the wrong phosphors"; + ui_type = "slider"; + ui_min = 0.0; + ui_max = 1.0; + ui_step = 0.01; + + ui_spacing = 2; + ui_category = "Colors and Effects"; +> = halation_weight_static; + +uniform float diffusion_weight < + ui_label = "Diffusion"; + ui_tooltip = "Blurring due to refraction from the screen's glass"; + ui_type = "slider"; + ui_min = 0.0; + ui_max = 1.0; + ui_step = 0.01; + + ui_category = "Colors and Effects"; +> = diffusion_weight_static; + +uniform float blur_radius < + ui_label = "Blur Radius"; + ui_tooltip = "Scales the radius of the halation and diffusion effects"; + ui_type = "slider"; + ui_min = 0.01; + ui_max = 5.0; + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Colors and Effects"; +> = 1.0; + +uniform float bloom_underestimate_levels < + ui_label = "Bloom Underestimation"; + ui_tooltip = "Scale the bloom effect's intensity"; + ui_type = "drag"; + ui_min = FIX_ZERO(0.0); + ui_step = 0.01; + + ui_spacing = 2; + ui_category = "Colors and Effects"; +> = bloom_underestimate_levels_static; + +uniform float bloom_excess < + ui_label = "Bloom Excess"; + ui_tooltip = "Extra bloom applied to all colors"; + ui_type = "slider"; + ui_min = 0.0; + ui_max = 1.0; + ui_step = 0.01; + + ui_category = "Colors and Effects"; +> = bloom_excess_static; + +uniform float2 aa_subpixel_r_offset_runtime < + ui_label = "AA Subpixel R Offet XY"; + ui_type = "drag"; + ui_min = -0.5; + ui_max = 0.5; + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS || !_RUNTIME_ANTIALIAS_SUBPIXEL_OFFSETS; + + ui_category = "Colors and Effects"; +> = aa_subpixel_r_offset_static; + +static const float aa_cubic_c = aa_cubic_c_static; +static const float aa_gauss_sigma = aa_gauss_sigma_static; + + +// ==== GEOMETRY ==== +uniform uint geom_rotation_mode < + #if !HIDE_HELP_SECTIONS + ui_text = "Change the geometry of the screen's glass.\n\n"; + #endif + ui_label = "Rotate Screen"; + ui_type = "combo"; + ui_items = "0 degrees\0" + "90 degrees\0" + "180 degrees\0" + "270 degrees\0"; + + ui_category = "Screen Geometry"; + ui_category_closed = true; +> = 0; +uniform uint geom_mode_runtime < + ui_label = "Geometry Mode"; + ui_tooltip = "Select screen curvature type"; + ui_type = "combo"; + ui_items = "Flat\0" + "Spherical\0" + "Spherical (Alt)\0" + "Cylindrical (Trinitron)\0"; + + ui_category = "Screen Geometry"; +> = geom_mode_static; + +uniform float geom_radius < + ui_label = "Geometry Radius"; + ui_tooltip = "Select screen curvature radius"; + ui_type = "slider"; + ui_min = 1.0 / (2.0 * pi); + ui_max = 1024; + ui_step = 0.01; + + ui_category = "Screen Geometry"; +> = geom_radius_static; + +uniform float geom_view_dist < + ui_label = "View Distance"; + ui_type = "slider"; + ui_min = 0.5; + ui_max = 1024; + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS; + + ui_spacing = 2; + ui_category = "Screen Geometry"; +> = geom_view_dist_static; + +uniform float2 geom_tilt_angle < + ui_label = "Screen Tilt Angles"; + ui_type = "drag"; + ui_min = -pi; + ui_max = pi; + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Screen Geometry"; +> = geom_tilt_angle_static; + +uniform float2 geom_aspect_ratio < + ui_label = "Screen Aspect Ratios"; + ui_type = "drag"; + ui_min = 1.0; + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS; + + ui_category = "Screen Geometry"; +> = float2(geom_aspect_ratio_static, 1); +uniform float2 geom_overscan < + ui_label = "Geom Overscan"; + ui_type = "drag"; + ui_min = FIX_ZERO(0.0); + ui_step = 0.01; + hidden = !ADVANCED_SETTINGS; + + ui_spacing = 2; + ui_category = "Screen Geometry"; +> = geom_overscan_static; + +// ==== BORDER ==== +uniform float border_size < + #if !HIDE_HELP_SECTIONS + ui_text = "Apply a thin vignette to the edge of the screen.\n\n"; + #endif + ui_label = "Border Size"; + ui_category_closed = true; + ui_type = "slider"; + ui_min = 0.0; + ui_max = 0.5; + ui_step = 0.01; + + ui_category = "Screen Border"; +> = border_size_static; + +uniform float border_darkness < + ui_label = "Border Darkness"; + ui_type = "drag"; + ui_min = 0.0; + ui_step = 0.01; + + ui_category = "Screen Border"; +> = border_darkness_static; + +uniform float border_compress < + ui_label = "Border Compress"; + ui_type = "drag"; + ui_min = 0.0; + ui_step = 0.01; + + ui_category = "Screen Border"; +> = border_compress_static; + +// ==== PREBLUR ==== +#if ENABLE_PREBLUR + uniform float2 preblur_effect_radius < + #if !HIDE_HELP_SECTIONS + ui_text = "- Apply a linear blur to the input image. Kind of like an NTSC/Composite shader, but much faster.\n" + "- If you want to use an NTSC shader or don't like this effect, disable it by setting ENABLE_PREBLUR to 0\n" + "- If you leave all of these set to 0, then they don't do anything. Consider disabling the effect to improve performance.\n\n"; + #endif + ui_type = "drag"; + ui_min = 0; + ui_max = 100; + ui_step = 1; + ui_label = "Effect Radius XY"; + ui_tooltip = "The radius of the effect visible on the screen (measured in pixels)"; + + ui_category = "Pre-Blur"; + ui_category_closed = true; + > = 0; + uniform uint2 preblur_sampling_radius < + ui_type = "drag"; + ui_min = 0; + ui_max = 100; + ui_step = 1; + ui_label = "Sampling Radius XY"; + ui_tooltip = "The number of samples to take on either side of each pixel"; + + ui_category = "Pre-Blur"; + > = 0; +#else + static const float2 preblur_effect_radius = 0; + static const uint2 preblur_sampling_radius = 0; +#endif + +// Provide accessors for vector constants that pack scalar uniforms: +float2 get_aspect_vector(const float geom_aspect_ratio) +{ + // Get an aspect ratio vector. Enforce geom_max_aspect_ratio, and prevent + // the absolute scale from affecting the uv-mapping for curvature: + const float geom_clamped_aspect_ratio = + min(geom_aspect_ratio, geom_max_aspect_ratio); + const float2 geom_aspect = + normalize(float2(geom_clamped_aspect_ratio, 1.0)); + return geom_aspect; +} + +float2 get_geom_overscan_vector() +{ + return geom_overscan; +} + +float2 get_geom_tilt_angle_vector() +{ + return geom_tilt_angle; +} + +float3 get_convergence_offsets_x_vector() +{ + return convergence_offset_x; +} + +float3 get_convergence_offsets_y_vector() +{ + return convergence_offset_y; +} + +float2 get_convergence_offsets_r_vector() +{ + return float2(convergence_offset_x.r, convergence_offset_y.r); +} + +float2 get_convergence_offsets_g_vector() +{ + return float2(convergence_offset_x.g, convergence_offset_y.g); +} + +float2 get_convergence_offsets_b_vector() +{ + return float2(convergence_offset_x.b, convergence_offset_y.b); +} + +float2 get_aa_subpixel_r_offset() +{ + #if _RUNTIME_ANTIALIAS_WEIGHTS + #if _RUNTIME_ANTIALIAS_SUBPIXEL_OFFSETS + // WARNING: THIS IS EXTREMELY EXPENSIVE. + return aa_subpixel_r_offset_runtime; + #else + return aa_subpixel_r_offset_static; + #endif + #else + return aa_subpixel_r_offset_static; + #endif +} + +// Provide accessors settings which still need "cooking:" +float get_mask_amplify() +{ + static const float mask_grille_amplify = 1.0/mask_grille_avg_color; + static const float mask_slot_amplify = 1.0/mask_slot_avg_color; + static const float mask_shadow_amplify = 1.0/mask_shadow_avg_color; + + float mask_amplify; + [flatten] + switch (mask_type) { + case 0: + mask_amplify = mask_grille_amplify; + break; + case 1: + mask_amplify = mask_slot_amplify; + break; + case 2: + mask_amplify = mask_shadow_amplify; + break; + case 3: + mask_amplify = mask_grille_amplify; + break; + case 4: + mask_amplify = mask_slot_amplify; + break; + default: + mask_amplify = mask_shadow_amplify; + break; + + } + + return mask_amplify; +} + +#endif // _BIND_SHADER_PARAMS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/bloom-functions.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/bloom-functions.fxh new file mode 100644 index 000000000..2c6456ace --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/bloom-functions.fxh @@ -0,0 +1,320 @@ +#ifndef _BLOOM_FUNCTIONS_H +#define _BLOOM_FUNCTIONS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +///////////////////////////////// DESCRIPTION //////////////////////////////// + +// These utility functions and constants help several passes determine the +// size and center texel weight of the phosphor bloom in a uniform manner. + + +////////////////////////////////// INCLUDES ////////////////////////////////// + +// We need to calculate the correct blur sigma using some .cgp constants: +//#include "../user-settings.h" + + +#include "user-settings.fxh" +#include "derived-settings-and-constants.fxh" +#include "bind-shader-params.fxh" +#include "blur-functions.fxh" + +/////////////////////////////// BLOOM CONSTANTS ////////////////////////////// + +// Compute constants with manual inlines of the functions below: +static const float bloom_diff_thresh = 1.0/256.0; + + + +/////////////////////////////////// HELPERS ////////////////////////////////// + +float get_min_sigma_to_blur_triad(const float triad_size, + const float thresh) +{ + // Requires: 1.) triad_size is the final phosphor triad size in pixels + // 2.) thresh is the max desired pixel difference in the + // blurred triad (e.g. 1.0/256.0). + // Returns: Return the minimum sigma that will fully blur a phosphor + // triad on the screen to an even color, within thresh. + // This closed-form function was found by curve-fitting data. + // Estimate: max error = ~0.086036, mean sq. error = ~0.0013387: + return -0.05168 + 0.6113*triad_size - + 1.122*triad_size*sqrt(0.000416 + thresh); + // Estimate: max error = ~0.16486, mean sq. error = ~0.0041041: + //return 0.5985*triad_size - triad_size*sqrt(thresh) +} + +float get_absolute_scale_blur_sigma(const float thresh) +{ + // Requires: 1.) min_expected_triads must be a global float. The number + // of horizontal phosphor triads in the final image must be + // >= min_allowed_viewport_triads.x for realistic results. + // 2.) bloom_approx_scale_x must be a global float equal to the + // absolute horizontal scale of BLOOM_APPROX. + // 3.) bloom_approx_scale_x/min_allowed_viewport_triads.x + // should be <= 1.1658025090 to keep the final result < + // 0.62666015625 (the largest sigma ensuring the largest + // unused texel weight stays < 1.0/256.0 for a 3x3 blur). + // 4.) thresh is the max desired pixel difference in the + // blurred triad (e.g. 1.0/256.0). + // Returns: Return the minimum Gaussian sigma that will blur the pass + // output as much as it would have taken to blur away + // bloom_approx_scale_x horizontal phosphor triads. + // Description: + // BLOOM_APPROX should look like a downscaled phosphor blur. Ideally, we'd + // use the same blur sigma as the actual phosphor bloom and scale it down + // to the current resolution with (bloom_approx_scale_x/viewport_size_x), but + // we don't know the viewport size in this pass. Instead, we'll blur as + // much as it would take to blur away min_allowed_viewport_triads.x. This + // will blur "more than necessary" if the user actually uses more triads, + // but that's not terrible either, because blurring a constant fraction of + // the viewport may better resemble a true optical bloom anyway (since the + // viewport will generally be about the same fraction of each player's + // field of view, regardless of screen size and resolution). + // Assume an extremely large viewport size for asymptotic results. + return bloom_approx_scale_x/max_viewport_size_x * + get_min_sigma_to_blur_triad( + max_viewport_size_x/min_allowed_viewport_triads.x, thresh); +} + +float get_center_weight(const float sigma) +{ + // Given a Gaussian blur sigma, get the blur weight for the center texel. + #if _RUNTIME_PHOSPHOR_BLOOM_SIGMA + return get_fast_gaussian_weight_sum_inv(sigma); + #else + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float w5 = exp(-25.0 * denom_inv); + const float w6 = exp(-36.0 * denom_inv); + const float w7 = exp(-49.0 * denom_inv); + const float w8 = exp(-64.0 * denom_inv); + const float w9 = exp(-81.0 * denom_inv); + const float w10 = exp(-100.0 * denom_inv); + const float w11 = exp(-121.0 * denom_inv); + const float w12 = exp(-144.0 * denom_inv); + const float w13 = exp(-169.0 * denom_inv); + const float w14 = exp(-196.0 * denom_inv); + const float w15 = exp(-225.0 * denom_inv); + const float w16 = exp(-256.0 * denom_inv); + const float w17 = exp(-289.0 * denom_inv); + const float w18 = exp(-324.0 * denom_inv); + const float w19 = exp(-361.0 * denom_inv); + const float w20 = exp(-400.0 * denom_inv); + const float w21 = exp(-441.0 * denom_inv); + // Note: If the implementation uses a smaller blur than the max allowed, + // the worst case scenario is that the center weight will be overestimated, + // so we'll put a bit more energy into the brightpass...no huge deal. + // Then again, if the implementation uses a larger blur than the max + // "allowed" because of dynamic branching, the center weight could be + // underestimated, which is more of a problem...consider always using + #if PHOSPHOR_BLOOM_TRIAD_SIZE_MODE >= _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_12_PIXELS + // 43x blur: + const float weight_sum_inv = 1.0 / + (w0 + 2.0 * (w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8 + w9 + w10 + + w11 + w12 + w13 + w14 + w15 + w16 + w17 + w18 + w19 + w20 + w21)); + #else + #if PHOSPHOR_BLOOM_TRIAD_SIZE_MODE >= _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_9_PIXELS + // 31x blur: + const float weight_sum_inv = 1.0 / + (w0 + 2.0 * (w1 + w2 + w3 + w4 + w5 + w6 + w7 + + w8 + w9 + w10 + w11 + w12 + w13 + w14 + w15)); + #else + #if PHOSPHOR_BLOOM_TRIAD_SIZE_MODE >= _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_6_PIXELS + // 25x blur: + const float weight_sum_inv = 1.0 / (w0 + 2.0 * ( + w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8 + w9 + w10 + w11 + w12)); + #else + #if PHOSPHOR_BLOOM_TRIAD_SIZE_MODE >= _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_3_PIXELS + // 17x blur: + const float weight_sum_inv = 1.0 / (w0 + 2.0 * ( + w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8)); + #else + // 9x blur: + const float weight_sum_inv = 1.0 / (w0 + 2.0 * (w1 + w2 + w3 + w4)); + #endif // _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_3_PIXELS + #endif // _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_6_PIXELS + #endif // _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_9_PIXELS + #endif // _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_12_PIXELS + const float center_weight = weight_sum_inv * weight_sum_inv; + return center_weight; + #endif +} + +float3 tex2DblurNfast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // If sigma is static, we can safely branch and use the smallest blur + // that's big enough. Ignore #define hints, because we'll only use a + // large blur if we actually need it, and the branches cost nothing. + #if !_RUNTIME_PHOSPHOR_BLOOM_SIGMA + #define PHOSPHOR_BLOOM_BRANCH_FOR_BLUR_SIZE + #else + // It's still worth branching if the profile supports dynamic branches: + // It's much faster than using a hugely excessive blur, but each branch + // eats ~1% FPS. + #if _DRIVERS_ALLOW_DYNAMIC_BRANCHES + #define PHOSPHOR_BLOOM_BRANCH_FOR_BLUR_SIZE + #endif + #endif + // Failed optimization notes: + // I originally created a same-size mipmapped 5-tap separable blur10 that + // could handle any sigma by reaching into lower mip levels. It was + // as fast as blur25fast for runtime sigmas and a tad faster than + // blur31fast for static sigmas, but mipmapping two viewport-size passes + // ate 10% of FPS across all codepaths, so it wasn't worth it. + #ifdef PHOSPHOR_BLOOM_BRANCH_FOR_BLUR_SIZE + if(sigma <= blur9_std_dev) + { + return tex2Dblur9fast(tex, tex_uv, dxdy, sigma, input_gamma); + } + else if(sigma <= blur17_std_dev) + { + return tex2Dblur17fast(tex, tex_uv, dxdy, sigma, input_gamma); + } + else if(sigma <= blur25_std_dev) + { + return tex2Dblur25fast(tex, tex_uv, dxdy, sigma, input_gamma); + } + else if(sigma <= blur31_std_dev) + { + return tex2Dblur31fast(tex, tex_uv, dxdy, sigma, input_gamma); + } + else + { + return tex2Dblur43fast(tex, tex_uv, dxdy, sigma, input_gamma); + } + #else + // If we can't afford to branch, we can only guess at what blur + // size we need. Therefore, use the largest blur allowed. + #ifdef PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_12_PIXELS + return tex2Dblur43fast(tex, tex_uv, dxdy, sigma, input_gamma); + #else + #ifdef PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_9_PIXELS + return tex2Dblur31fast(tex, tex_uv, dxdy, sigma, input_gamma); + #else + #ifdef PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_6_PIXELS + return tex2Dblur25fast(tex, tex_uv, dxdy, sigma, input_gamma); + #else + #if PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_3_PIXELS + return tex2Dblur17fast(tex, tex_uv, dxdy, sigma, input_gamma); + #else + return tex2Dblur9fast(tex, tex_uv, dxdy, sigma, input_gamma); + #endif // PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_3_PIXELS + #endif // PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_6_PIXELS + #endif // PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_9_PIXELS + #endif // PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_12_PIXELS + #endif // PHOSPHOR_BLOOM_BRANCH_FOR_BLUR_SIZE +} + +float get_bloom_approx_sigma(const float output_size_x_runtime, + const float estimated_viewport_size_x) +{ + // Requires: 1.) output_size_x_runtime == BLOOM_APPROX.output_size.x. + // This is included for dynamic codepaths just in case the + // following two globals are incorrect: + // 2.) bloom_approx_size_x_for_skip should == the same + // if PHOSPHOR_BLOOM_FAKE is #defined + // 3.) bloom_approx_size_x should == the same otherwise + // Returns: For gaussian4x4, return a dynamic small bloom sigma that's + // as close to optimal as possible given available information. + // For blur3x3, return the a static small bloom sigma that + // works well for typical cases. Otherwise, we're using simple + // bilinear filtering, so use static calculations. + // Assume the default static value. This is a compromise that ensures + // typical triads are blurred, even if unusually large ones aren't. + static const float mask_num_triads_static = + max(min_allowed_viewport_triads.x, mask_num_triads_across_static); + const float mask_num_triads_from_size = + estimated_viewport_size_x/mask_triad_width; + const float mask_num_triads_runtime = max(min_allowed_viewport_triads.x, + lerp(mask_num_triads_from_size, mask_num_triads_across, + mask_size_param)); + // Assume an extremely large viewport size for asymptotic results: + static const float max_viewport_size_x = 1080.0*1024.0*(4.0/3.0); + if(bloom_approx_filter > 1.5) // 4x4 true Gaussian resize + { + // Use the runtime num triads and output size: + const float asymptotic_triad_size = + max_viewport_size_x/mask_num_triads_runtime; + const float asymptotic_sigma = get_min_sigma_to_blur_triad( + asymptotic_triad_size, bloom_diff_thresh); + const float bloom_approx_sigma = + asymptotic_sigma * output_size_x_runtime/max_viewport_size_x; + // The BLOOM_APPROX input has to be ORIG_LINEARIZED to avoid moire, but + // account for the Gaussian scanline sigma from the last pass too. + // The bloom will be too wide horizontally but tall enough vertically. + return length(float2(bloom_approx_sigma, gaussian_beam_max_sigma)); + } + else // 3x3 blur resize (the bilinear resize doesn't need a sigma) + { + // We're either using blur3x3 or bilinear filtering. The biggest + // reason to choose blur3x3 is to avoid dynamic weights, so use a + // static calculation. + #ifdef PHOSPHOR_BLOOM_FAKE + static const float output_size_x_static = + bloom_approx_size_x_for_fake; + #else + static const float output_size_x_static = bloom_approx_size_x; + #endif + static const float asymptotic_triad_size = + max_viewport_size_x/mask_num_triads_static; + const float asymptotic_sigma = get_min_sigma_to_blur_triad( + asymptotic_triad_size, bloom_diff_thresh); + const float bloom_approx_sigma = + asymptotic_sigma * output_size_x_static/max_viewport_size_x; + // The BLOOM_APPROX input has to be ORIG_LINEARIZED to avoid moire, but + // try accounting for the Gaussian scanline sigma from the last pass + // too; use the static default value: + return length(float2(bloom_approx_sigma, gaussian_beam_max_sigma_static)); + } +} + +float get_final_bloom_sigma(const float bloom_sigma_runtime) +{ + // Requires: 1.) bloom_sigma_runtime is a precalculated sigma that's + // optimal for the [known] triad size. + // 2.) Call this from a fragment shader (not a vertex shader), + // or blurring with static sigmas won't be constant-folded. + // Returns: Return the optimistic static sigma if the triad size is + // known at compile time. Otherwise return the optimal runtime + // sigma (10% slower) or an implementation-specific compromise + // between an optimistic or pessimistic static sigma. + // Notes: Call this from the fragment shader, NOT the vertex shader, + // so static sigmas can be constant-folded! + const float bloom_sigma_optimistic = get_min_sigma_to_blur_triad( + mask_triad_width_static, bloom_diff_thresh); + #if _RUNTIME_PHOSPHOR_BLOOM_SIGMA + return bloom_sigma_runtime; + #else + // Overblurring looks as bad as underblurring, so assume average-size + // triads, not worst-case huge triads: + return bloom_sigma_optimistic; + #endif +} + + +#endif // _BLOOM_FUNCTIONS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/blur-functions.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/blur-functions.fxh new file mode 100644 index 000000000..2d81bfc03 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/blur-functions.fxh @@ -0,0 +1,1966 @@ +#ifndef _BLUR_FUNCTIONS_H +#define _BLUR_FUNCTIONS_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2014 TroggleMonkey +// Copyright (C) 2020 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + +///////////////////////////////// DESCRIPTION //////////////////////////////// + +// This file provides reusable one-pass and separable (two-pass) blurs. +// Requires: All blurs share these requirements (dxdy requirement is split): +// 1.) All requirements of gamma-management.h must be satisfied! +// 2.) filter_linearN must == "true" in your .cgp preset unless +// you're using tex2DblurNresize at 1x scale. +// 3.) mipmap_inputN must == "true" in your .cgp preset if +// output_size < video_size. +// 4.) output_size == video_size / pow(2, M), where M is some +// positive integer. tex2Dblur*resize can resize arbitrarily +// (and the blur will be done after resizing), but arbitrary +// resizes "fail" with other blurs due to the way they mix +// static weights with bilinear sample exploitation. +// 5.) In general, dxdy should contain the uv pixel spacing: +// dxdy = (video_size/output_size)/texture_size +// 6.) For separable blurs (tex2DblurNresize and tex2DblurNfast), +// zero out the dxdy component in the unblurred dimension: +// dxdy = float2(dxdy.x, 0.0) or float2(0.0, dxdy.y) +// Many blurs share these requirements: +// 1.) One-pass blurs require scale_xN == scale_yN or scales > 1.0, +// or they will blur more in the lower-scaled dimension. +// 2.) One-pass shared sample blurs require ddx(), ddy(), and +// tex2Dlod() to be supported by the current Cg profile, and +// the drivers must support high-quality derivatives. +// 3.) One-pass shared sample blurs require: +// tex_uv.w == log2(video_size/output_size).y; +// Non-wrapper blurs share this requirement: +// 1.) sigma is the intended standard deviation of the blur +// Wrapper blurs share this requirement, which is automatically +// met (unless OVERRIDE_BLUR_STD_DEVS is #defined; see below): +// 1.) blurN_std_dev must be global static const float values +// specifying standard deviations for Nx blurs in units +// of destination pixels +// Optional: 1.) The including file (or an earlier included file) may +// optionally #define USE_BINOMIAL_BLUR_STD_DEVS to replace +// default standard deviations with those matching a binomial +// distribution. (See below for details/properties.) +// 2.) The including file (or an earlier included file) may +// optionally #define OVERRIDE_BLUR_STD_DEVS and override: +// static const float blur3_std_dev +// static const float blur4_std_dev +// static const float blur5_std_dev +// static const float blur6_std_dev +// static const float blur7_std_dev +// static const float blur8_std_dev +// static const float blur9_std_dev +// static const float blur10_std_dev +// static const float blur11_std_dev +// static const float blur12_std_dev +// static const float blur17_std_dev +// static const float blur25_std_dev +// static const float blur31_std_dev +// static const float blur43_std_dev +// 3.) The including file (or an earlier included file) may +// optionally #define OVERRIDE_ERROR_BLURRING and override: +// static const float error_blurring +// This tuning value helps mitigate weighting errors from one- +// pass shared-sample blurs sharing bilinear samples between +// fragments. Values closer to 0.0 have "correct" blurriness +// but allow more artifacts, and values closer to 1.0 blur away +// artifacts by sampling closer to halfway between texels. +// UPDATE 6/21/14: The above static constants may now be overridden +// by non-static uniform constants. This permits exposing blur +// standard deviations as runtime GUI shader parameters. However, +// using them keeps weights from being statically computed, and the +// speed hit depends on the blur: On my machine, uniforms kill over +// 53% of the framerate with tex2Dblur12x12shared, but they only +// drop the framerate by about 18% with tex2Dblur11fast. +// Quality and Performance Comparisons: +// For the purposes of the following discussion, "no sRGB" means +// GAMMA_ENCODE_EVERY_FBO is #defined, and "sRGB" means it isn't. +// 1.) tex2DblurNfast is always faster than tex2DblurNresize. +// 2.) tex2DblurNresize functions are the only ones that can arbitrarily resize +// well, because they're the only ones that don't exploit bilinear samples. +// This also means they're the only functions which can be truly gamma- +// correct without linear (or sRGB FBO) input, but only at 1x scale. +// 3.) One-pass shared sample blurs only have a speed advantage without sRGB. +// They also have some inaccuracies due to their shared-[bilinear-]sample +// design, which grow increasingly bothersome for smaller blurs and higher- +// frequency source images (relative to their resolution). I had high +// hopes for them, but their most realistic use case is limited to quickly +// reblurring an already blurred input at full resolution. Otherwise: +// a.) If you're blurring a low-resolution source, you want a better blur. +// b.) If you're blurring a lower mipmap, you want a better blur. +// c.) If you're blurring a high-resolution, high-frequency source, you +// want a better blur. +// 4.) The one-pass blurs without shared samples grow slower for larger blurs, +// but they're competitive with separable blurs at 5x5 and smaller, and +// even tex2Dblur7x7 isn't bad if you're wanting to conserve passes. +// Here are some framerates from a GeForce 8800GTS. The first pass resizes to +// viewport size (4x in this test) and linearizes for sRGB codepaths, and the +// remaining passes perform 6 full blurs. Mipmapped tests are performed at the +// same scale, so they just measure the cost of mipmapping each FBO (only every +// other FBO is mipmapped for separable blurs, to mimic realistic usage). +// Mipmap Neither sRGB+Mipmap sRGB Function +// 76.0 92.3 131.3 193.7 tex2Dblur3fast +// 63.2 74.4 122.4 175.5 tex2Dblur3resize +// 93.7 121.2 159.3 263.2 tex2Dblur3x3 +// 59.7 68.7 115.4 162.1 tex2Dblur3x3resize +// 63.2 74.4 122.4 175.5 tex2Dblur5fast +// 49.3 54.8 100.0 132.7 tex2Dblur5resize +// 59.7 68.7 115.4 162.1 tex2Dblur5x5 +// 64.9 77.2 99.1 137.2 tex2Dblur6x6shared +// 55.8 63.7 110.4 151.8 tex2Dblur7fast +// 39.8 43.9 83.9 105.8 tex2Dblur7resize +// 40.0 44.2 83.2 104.9 tex2Dblur7x7 +// 56.4 65.5 71.9 87.9 tex2Dblur8x8shared +// 49.3 55.1 99.9 132.5 tex2Dblur9fast +// 33.3 36.2 72.4 88.0 tex2Dblur9resize +// 27.8 29.7 61.3 72.2 tex2Dblur9x9 +// 37.2 41.1 52.6 60.2 tex2Dblur10x10shared +// 44.4 49.5 91.3 117.8 tex2Dblur11fast +// 28.8 30.8 63.6 75.4 tex2Dblur11resize +// 33.6 36.5 40.9 45.5 tex2Dblur12x12shared +// TODO: Fill in benchmarks for new untested blurs. +// tex2Dblur17fast +// tex2Dblur25fast +// tex2Dblur31fast +// tex2Dblur43fast +// tex2Dblur3x3resize + + +///////////////////////////// SETTINGS MANAGEMENT //////////////////////////// + +// Set static standard deviations, but allow users to override them with their +// own constants (even non-static uniforms if they're okay with the speed hit): +#ifndef OVERRIDE_BLUR_STD_DEVS + // blurN_std_dev values are specified in terms of dxdy strides. + #ifdef USE_BINOMIAL_BLUR_STD_DEVS + // By request, we can define standard deviations corresponding to a + // binomial distribution with p = 0.5 (related to Pascal's triangle). + // This distribution works such that blurring multiple times should + // have the same result as a single larger blur. These values are + // larger than default for blurs up to 6x and smaller thereafter. + static const float blur3_std_dev = 0.84931640625; + static const float blur4_std_dev = 0.84931640625; + static const float blur5_std_dev = 1.0595703125; + static const float blur6_std_dev = 1.06591796875; + static const float blur7_std_dev = 1.17041015625; + static const float blur8_std_dev = 1.1720703125; + static const float blur9_std_dev = 1.2259765625; + static const float blur10_std_dev = 1.21982421875; + static const float blur11_std_dev = 1.25361328125; + static const float blur12_std_dev = 1.2423828125; + static const float blur17_std_dev = 1.27783203125; + static const float blur25_std_dev = 1.2810546875; + static const float blur31_std_dev = 1.28125; + static const float blur43_std_dev = 1.28125; + #else + // The defaults are the largest values that keep the largest unused + // blur term on each side <= 1.0/256.0. (We could get away with more + // or be more conservative, but this compromise is pretty reasonable.) + static const float blur3_std_dev = 0.62666015625; + static const float blur4_std_dev = 0.66171875; + static const float blur5_std_dev = 0.9845703125; + static const float blur6_std_dev = 1.02626953125; + static const float blur7_std_dev = 1.36103515625; + static const float blur8_std_dev = 1.4080078125; + static const float blur9_std_dev = 1.7533203125; + static const float blur10_std_dev = 1.80478515625; + static const float blur11_std_dev = 2.15986328125; + static const float blur12_std_dev = 2.215234375; + static const float blur17_std_dev = 3.45535583496; + static const float blur25_std_dev = 5.3409576416; + static const float blur31_std_dev = 6.86488037109; + static const float blur43_std_dev = 10.1852050781; + #endif // USE_BINOMIAL_BLUR_STD_DEVS +#endif // OVERRIDE_BLUR_STD_DEVS + +#ifndef OVERRIDE_ERROR_BLURRING + // error_blurring should be in [0.0, 1.0]. Higher values reduce ringing + // in shared-sample blurs but increase blurring and feature shifting. + static const float error_blurring = 0.5; +#endif + + +////////////////////////////////// INCLUDES ////////////////////////////////// + +// gamma-management.h relies on pass-specific settings to guide its behavior: +// FIRST_PASS, LAST_PASS, GAMMA_ENCODE_EVERY_FBO, etc. See it for details. +//#include "gamma-management.h" + + +#include "gamma-management.fxh" +#include "quad-pixel-communication.fxh" +#include "special-functions.fxh" + +//////////////////////////////// END INCLUDES //////////////////////////////// + +/////////////////////////////////// HELPERS ////////////////////////////////// + +float4 uv2_to_uv4(float2 tex_uv) +{ + // Make a float2 uv offset safe for adding to float4 tex2Dlod coords: + return float4(tex_uv, 0.0, 0.0); +} + +// Make a length squared helper macro (for usage with static constants): +#define LENGTH_SQ(vec) (dot(vec, vec)) + +float get_fast_gaussian_weight_sum_inv(const float sigma) +{ + // We can use the Gaussian integral to calculate the asymptotic weight for + // the center pixel. Since the unnormalized center pixel weight is 1.0, + // the normalized weight is the same as the weight sum inverse. Given a + // large enough blur (9+), the asymptotic weight sum is close and faster: + // center_weight = 0.5 * + // (erf(0.5/(sigma*sqrt(2.0))) - erf(-0.5/(sigma*sqrt(2.0)))) + // erf(-x) == -erf(x), so we get 0.5 * (2.0 * erf(blah blah)): + // However, we can get even faster results with curve-fitting. These are + // also closer than the asymptotic results, because they were constructed + // from 64 blurs sizes from [3, 131) and 255 equally-spaced sigmas from + // (0, blurN_std_dev), so the results for smaller sigmas are biased toward + // smaller blurs. The max error is 0.0031793913. + // Relative FPS: 134.3 with erf, 135.8 with curve-fitting. + //static const float temp = 0.5/sqrt(2.0); + //return erf(temp/sigma); + return min(exp(exp(0.348348412457428/ + (sigma - 0.0860587260734721))), 0.399334576340352/sigma); +} + + +//////////////////// ARBITRARILY RESIZABLE SEPARABLE BLURS /////////////////// + +float3 tex2Dblur11resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Global requirements must be met (see file description). + // Returns: A 1D 11x Gaussian blurred texture lookup using a 11-tap blur. + // It may be mipmapped depending on settings and dxdy. + // Calculate Gaussian blur kernel weights and a normalization factor for + // distances of 0-4, ignoring constant factors (since we're normalizing). + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float w5 = exp(-25.0 * denom_inv); + const float weight_sum_inv = 1.0 / + (w0 + 2.0 * (w1 + w2 + w3 + w4 + w5)); + // Statically normalize weights, sum weighted samples, and return. Blurs are + // currently optimized for dynamic weights. + float3 sum = float3(0.0,0.0,0.0); + sum += w5 * tex2D_linearize(tex, tex_uv - 5.0 * dxdy, input_gamma).rgb; + sum += w4 * tex2D_linearize(tex, tex_uv - 4.0 * dxdy, input_gamma).rgb; + sum += w3 * tex2D_linearize(tex, tex_uv - 3.0 * dxdy, input_gamma).rgb; + sum += w2 * tex2D_linearize(tex, tex_uv - 2.0 * dxdy, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv - 1.0 * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv + 1.0 * dxdy, input_gamma).rgb; + sum += w2 * tex2D_linearize(tex, tex_uv + 2.0 * dxdy, input_gamma).rgb; + sum += w3 * tex2D_linearize(tex, tex_uv + 3.0 * dxdy, input_gamma).rgb; + sum += w4 * tex2D_linearize(tex, tex_uv + 4.0 * dxdy, input_gamma).rgb; + sum += w5 * tex2D_linearize(tex, tex_uv + 5.0 * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur9resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Global requirements must be met (see file description). + // Returns: A 1D 9x Gaussian blurred texture lookup using a 9-tap blur. + // It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float weight_sum_inv = 1.0 / (w0 + 2.0 * (w1 + w2 + w3 + w4)); + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w4 * tex2D_linearize(tex, tex_uv - 4.0 * dxdy, input_gamma).rgb; + sum += w3 * tex2D_linearize(tex, tex_uv - 3.0 * dxdy, input_gamma).rgb; + sum += w2 * tex2D_linearize(tex, tex_uv - 2.0 * dxdy, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv - 1.0 * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv + 1.0 * dxdy, input_gamma).rgb; + sum += w2 * tex2D_linearize(tex, tex_uv + 2.0 * dxdy, input_gamma).rgb; + sum += w3 * tex2D_linearize(tex, tex_uv + 3.0 * dxdy, input_gamma).rgb; + sum += w4 * tex2D_linearize(tex, tex_uv + 4.0 * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur7resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Global requirements must be met (see file description). + // Returns: A 1D 7x Gaussian blurred texture lookup using a 7-tap blur. + // It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float weight_sum_inv = 1.0 / (w0 + 2.0 * (w1 + w2 + w3)); + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w3 * tex2D_linearize(tex, tex_uv - 3.0 * dxdy, input_gamma).rgb; + sum += w2 * tex2D_linearize(tex, tex_uv - 2.0 * dxdy, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv - 1.0 * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv + 1.0 * dxdy, input_gamma).rgb; + sum += w2 * tex2D_linearize(tex, tex_uv + 2.0 * dxdy, input_gamma).rgb; + sum += w3 * tex2D_linearize(tex, tex_uv + 3.0 * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur5resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Global requirements must be met (see file description). + // Returns: A 1D 5x Gaussian blurred texture lookup using a 5-tap blur. + // It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float weight_sum_inv = 1.0 / (w0 + 2.0 * (w1 + w2)); + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w2 * tex2D_linearize(tex, tex_uv - 2.0 * dxdy, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv - 1.0 * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv + 1.0 * dxdy, input_gamma).rgb; + sum += w2 * tex2D_linearize(tex, tex_uv + 2.0 * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur3resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Global requirements must be met (see file description). + // Returns: A 1D 3x Gaussian blurred texture lookup using a 3-tap blur. + // It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float weight_sum_inv = 1.0 / (w0 + 2.0 * w1); + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w1 * tex2D_linearize(tex, tex_uv - 1.0 * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w1 * tex2D_linearize(tex, tex_uv + 1.0 * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + + +/////////////////////////// FAST SEPARABLE BLURS /////////////////////////// + +float3 tex2Dblur11fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: 1.) Global requirements must be met (see file description). + // 2.) filter_linearN must = "true" in your .cgp file. + // 3.) For gamma-correct bilinear filtering, global + // gamma_aware_bilinear == true (from gamma-management.h) + // Returns: A 1D 11x Gaussian blurred texture lookup using 6 linear + // taps. It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float w5 = exp(-25.0 * denom_inv); + const float weight_sum_inv = 1.0 / + (w0 + 2.0 * (w1 + w2 + w3 + w4 + w5)); + // Calculate combined weights and linear sample ratios between texel pairs. + // The center texel (with weight w0) is used twice, so halve its weight. + const float w01 = w0 * 0.5 + w1; + const float w23 = w2 + w3; + const float w45 = w4 + w5; + const float w01_ratio = w1/w01; + const float w23_ratio = w3/w23; + const float w45_ratio = w5/w45; + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w45 * tex2D_linearize(tex, tex_uv - (4.0 + w45_ratio) * dxdy, input_gamma).rgb; + sum += w23 * tex2D_linearize(tex, tex_uv - (2.0 + w23_ratio) * dxdy, input_gamma).rgb; + sum += w01 * tex2D_linearize(tex, tex_uv - w01_ratio * dxdy, input_gamma).rgb; + sum += w01 * tex2D_linearize(tex, tex_uv + w01_ratio * dxdy, input_gamma).rgb; + sum += w23 * tex2D_linearize(tex, tex_uv + (2.0 + w23_ratio) * dxdy, input_gamma).rgb; + sum += w45 * tex2D_linearize(tex, tex_uv + (4.0 + w45_ratio) * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur9fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Same as tex2Dblur11() + // Returns: A 1D 9x Gaussian blurred texture lookup using 1 nearest + // neighbor and 4 linear taps. It may be mipmapped depending + // on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float weight_sum_inv = 1.0 / (w0 + 2.0 * (w1 + w2 + w3 + w4)); + // Calculate combined weights and linear sample ratios between texel pairs. + const float w12 = w1 + w2; + const float w34 = w3 + w4; + const float w12_ratio = w2/w12; + const float w34_ratio = w4/w34; + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w34 * tex2D_linearize(tex, tex_uv - (3.0 + w34_ratio) * dxdy, input_gamma).rgb; + sum += w12 * tex2D_linearize(tex, tex_uv - (1.0 + w12_ratio) * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w12 * tex2D_linearize(tex, tex_uv + (1.0 + w12_ratio) * dxdy, input_gamma).rgb; + sum += w34 * tex2D_linearize(tex, tex_uv + (3.0 + w34_ratio) * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur7fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Same as tex2Dblur11() + // Returns: A 1D 7x Gaussian blurred texture lookup using 4 linear + // taps. It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float weight_sum_inv = 1.0 / (w0 + 2.0 * (w1 + w2 + w3)); + // Calculate combined weights and linear sample ratios between texel pairs. + // The center texel (with weight w0) is used twice, so halve its weight. + const float w01 = w0 * 0.5 + w1; + const float w23 = w2 + w3; + const float w01_ratio = w1/w01; + const float w23_ratio = w3/w23; + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w23 * tex2D_linearize(tex, tex_uv - (2.0 + w23_ratio) * dxdy, input_gamma).rgb; + sum += w01 * tex2D_linearize(tex, tex_uv - w01_ratio * dxdy, input_gamma).rgb; + sum += w01 * tex2D_linearize(tex, tex_uv + w01_ratio * dxdy, input_gamma).rgb; + sum += w23 * tex2D_linearize(tex, tex_uv + (2.0 + w23_ratio) * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur5fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Same as tex2Dblur11() + // Returns: A 1D 5x Gaussian blurred texture lookup using 1 nearest + // neighbor and 2 linear taps. It may be mipmapped depending + // on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float weight_sum_inv = 1.0 / (w0 + 2.0 * (w1 + w2)); + // Calculate combined weights and linear sample ratios between texel pairs. + const float w12 = w1 + w2; + const float w12_ratio = w2/w12; + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w12 * tex2D_linearize(tex, tex_uv - (1.0 + w12_ratio) * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w12 * tex2D_linearize(tex, tex_uv + (1.0 + w12_ratio) * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur3fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Same as tex2Dblur11() + // Returns: A 1D 3x Gaussian blurred texture lookup using 2 linear + // taps. It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float weight_sum_inv = 1.0 / (w0 + 2.0 * w1); + // Calculate combined weights and linear sample ratios between texel pairs. + // The center texel (with weight w0) is used twice, so halve its weight. + const float w01 = w0 * 0.5 + w1; + const float w01_ratio = w1/w01; + // Weights for all samples are the same, so just average them: + return 0.5 * ( + tex2D_linearize(tex, tex_uv - w01_ratio * dxdy, input_gamma).rgb + + tex2D_linearize(tex, tex_uv + w01_ratio * dxdy, input_gamma).rgb); +} + + +//////////////////////////// HUGE SEPARABLE BLURS //////////////////////////// + +// Huge separable blurs come only in "fast" versions. +float3 tex2Dblur43fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Same as tex2Dblur11() + // Returns: A 1D 43x Gaussian blurred texture lookup using 22 linear + // taps. It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float w5 = exp(-25.0 * denom_inv); + const float w6 = exp(-36.0 * denom_inv); + const float w7 = exp(-49.0 * denom_inv); + const float w8 = exp(-64.0 * denom_inv); + const float w9 = exp(-81.0 * denom_inv); + const float w10 = exp(-100.0 * denom_inv); + const float w11 = exp(-121.0 * denom_inv); + const float w12 = exp(-144.0 * denom_inv); + const float w13 = exp(-169.0 * denom_inv); + const float w14 = exp(-196.0 * denom_inv); + const float w15 = exp(-225.0 * denom_inv); + const float w16 = exp(-256.0 * denom_inv); + const float w17 = exp(-289.0 * denom_inv); + const float w18 = exp(-324.0 * denom_inv); + const float w19 = exp(-361.0 * denom_inv); + const float w20 = exp(-400.0 * denom_inv); + const float w21 = exp(-441.0 * denom_inv); + //const float weight_sum_inv = 1.0 / + // (w0 + 2.0 * (w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8 + w9 + w10 + w11 + + // w12 + w13 + w14 + w15 + w16 + w17 + w18 + w19 + w20 + w21)); + const float weight_sum_inv = get_fast_gaussian_weight_sum_inv(sigma); + // Calculate combined weights and linear sample ratios between texel pairs. + // The center texel (with weight w0) is used twice, so halve its weight. + const float w0_1 = w0 * 0.5 + w1; + const float w2_3 = w2 + w3; + const float w4_5 = w4 + w5; + const float w6_7 = w6 + w7; + const float w8_9 = w8 + w9; + const float w10_11 = w10 + w11; + const float w12_13 = w12 + w13; + const float w14_15 = w14 + w15; + const float w16_17 = w16 + w17; + const float w18_19 = w18 + w19; + const float w20_21 = w20 + w21; + const float w0_1_ratio = w1/w0_1; + const float w2_3_ratio = w3/w2_3; + const float w4_5_ratio = w5/w4_5; + const float w6_7_ratio = w7/w6_7; + const float w8_9_ratio = w9/w8_9; + const float w10_11_ratio = w11/w10_11; + const float w12_13_ratio = w13/w12_13; + const float w14_15_ratio = w15/w14_15; + const float w16_17_ratio = w17/w16_17; + const float w18_19_ratio = w19/w18_19; + const float w20_21_ratio = w21/w20_21; + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w20_21 * tex2D_linearize(tex, tex_uv - (20.0 + w20_21_ratio) * dxdy, input_gamma).rgb; + sum += w18_19 * tex2D_linearize(tex, tex_uv - (18.0 + w18_19_ratio) * dxdy, input_gamma).rgb; + sum += w16_17 * tex2D_linearize(tex, tex_uv - (16.0 + w16_17_ratio) * dxdy, input_gamma).rgb; + sum += w14_15 * tex2D_linearize(tex, tex_uv - (14.0 + w14_15_ratio) * dxdy, input_gamma).rgb; + sum += w12_13 * tex2D_linearize(tex, tex_uv - (12.0 + w12_13_ratio) * dxdy, input_gamma).rgb; + sum += w10_11 * tex2D_linearize(tex, tex_uv - (10.0 + w10_11_ratio) * dxdy, input_gamma).rgb; + sum += w8_9 * tex2D_linearize(tex, tex_uv - (8.0 + w8_9_ratio) * dxdy, input_gamma).rgb; + sum += w6_7 * tex2D_linearize(tex, tex_uv - (6.0 + w6_7_ratio) * dxdy, input_gamma).rgb; + sum += w4_5 * tex2D_linearize(tex, tex_uv - (4.0 + w4_5_ratio) * dxdy, input_gamma).rgb; + sum += w2_3 * tex2D_linearize(tex, tex_uv - (2.0 + w2_3_ratio) * dxdy, input_gamma).rgb; + sum += w0_1 * tex2D_linearize(tex, tex_uv - w0_1_ratio * dxdy, input_gamma).rgb; + sum += w0_1 * tex2D_linearize(tex, tex_uv + w0_1_ratio * dxdy, input_gamma).rgb; + sum += w2_3 * tex2D_linearize(tex, tex_uv + (2.0 + w2_3_ratio) * dxdy, input_gamma).rgb; + sum += w4_5 * tex2D_linearize(tex, tex_uv + (4.0 + w4_5_ratio) * dxdy, input_gamma).rgb; + sum += w6_7 * tex2D_linearize(tex, tex_uv + (6.0 + w6_7_ratio) * dxdy, input_gamma).rgb; + sum += w8_9 * tex2D_linearize(tex, tex_uv + (8.0 + w8_9_ratio) * dxdy, input_gamma).rgb; + sum += w10_11 * tex2D_linearize(tex, tex_uv + (10.0 + w10_11_ratio) * dxdy, input_gamma).rgb; + sum += w12_13 * tex2D_linearize(tex, tex_uv + (12.0 + w12_13_ratio) * dxdy, input_gamma).rgb; + sum += w14_15 * tex2D_linearize(tex, tex_uv + (14.0 + w14_15_ratio) * dxdy, input_gamma).rgb; + sum += w16_17 * tex2D_linearize(tex, tex_uv + (16.0 + w16_17_ratio) * dxdy, input_gamma).rgb; + sum += w18_19 * tex2D_linearize(tex, tex_uv + (18.0 + w18_19_ratio) * dxdy, input_gamma).rgb; + sum += w20_21 * tex2D_linearize(tex, tex_uv + (20.0 + w20_21_ratio) * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur31fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Same as tex2Dblur11() + // Returns: A 1D 31x Gaussian blurred texture lookup using 16 linear + // taps. It may be mipmapped depending on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float w5 = exp(-25.0 * denom_inv); + const float w6 = exp(-36.0 * denom_inv); + const float w7 = exp(-49.0 * denom_inv); + const float w8 = exp(-64.0 * denom_inv); + const float w9 = exp(-81.0 * denom_inv); + const float w10 = exp(-100.0 * denom_inv); + const float w11 = exp(-121.0 * denom_inv); + const float w12 = exp(-144.0 * denom_inv); + const float w13 = exp(-169.0 * denom_inv); + const float w14 = exp(-196.0 * denom_inv); + const float w15 = exp(-225.0 * denom_inv); + //const float weight_sum_inv = 1.0 / + // (w0 + 2.0 * (w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8 + + // w9 + w10 + w11 + w12 + w13 + w14 + w15)); + const float weight_sum_inv = get_fast_gaussian_weight_sum_inv(sigma); + // Calculate combined weights and linear sample ratios between texel pairs. + // The center texel (with weight w0) is used twice, so halve its weight. + const float w0_1 = w0 * 0.5 + w1; + const float w2_3 = w2 + w3; + const float w4_5 = w4 + w5; + const float w6_7 = w6 + w7; + const float w8_9 = w8 + w9; + const float w10_11 = w10 + w11; + const float w12_13 = w12 + w13; + const float w14_15 = w14 + w15; + const float w0_1_ratio = w1/w0_1; + const float w2_3_ratio = w3/w2_3; + const float w4_5_ratio = w5/w4_5; + const float w6_7_ratio = w7/w6_7; + const float w8_9_ratio = w9/w8_9; + const float w10_11_ratio = w11/w10_11; + const float w12_13_ratio = w13/w12_13; + const float w14_15_ratio = w15/w14_15; + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w14_15 * tex2D_linearize(tex, tex_uv - (14.0 + w14_15_ratio) * dxdy, input_gamma).rgb; + sum += w12_13 * tex2D_linearize(tex, tex_uv - (12.0 + w12_13_ratio) * dxdy, input_gamma).rgb; + sum += w10_11 * tex2D_linearize(tex, tex_uv - (10.0 + w10_11_ratio) * dxdy, input_gamma).rgb; + sum += w8_9 * tex2D_linearize(tex, tex_uv - (8.0 + w8_9_ratio) * dxdy, input_gamma).rgb; + sum += w6_7 * tex2D_linearize(tex, tex_uv - (6.0 + w6_7_ratio) * dxdy, input_gamma).rgb; + sum += w4_5 * tex2D_linearize(tex, tex_uv - (4.0 + w4_5_ratio) * dxdy, input_gamma).rgb; + sum += w2_3 * tex2D_linearize(tex, tex_uv - (2.0 + w2_3_ratio) * dxdy, input_gamma).rgb; + sum += w0_1 * tex2D_linearize(tex, tex_uv - w0_1_ratio * dxdy, input_gamma).rgb; + sum += w0_1 * tex2D_linearize(tex, tex_uv + w0_1_ratio * dxdy, input_gamma).rgb; + sum += w2_3 * tex2D_linearize(tex, tex_uv + (2.0 + w2_3_ratio) * dxdy, input_gamma).rgb; + sum += w4_5 * tex2D_linearize(tex, tex_uv + (4.0 + w4_5_ratio) * dxdy, input_gamma).rgb; + sum += w6_7 * tex2D_linearize(tex, tex_uv + (6.0 + w6_7_ratio) * dxdy, input_gamma).rgb; + sum += w8_9 * tex2D_linearize(tex, tex_uv + (8.0 + w8_9_ratio) * dxdy, input_gamma).rgb; + sum += w10_11 * tex2D_linearize(tex, tex_uv + (10.0 + w10_11_ratio) * dxdy, input_gamma).rgb; + sum += w12_13 * tex2D_linearize(tex, tex_uv + (12.0 + w12_13_ratio) * dxdy, input_gamma).rgb; + sum += w14_15 * tex2D_linearize(tex, tex_uv + (14.0 + w14_15_ratio) * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur25fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Same as tex2Dblur11() + // Returns: A 1D 25x Gaussian blurred texture lookup using 1 nearest + // neighbor and 12 linear taps. It may be mipmapped depending + // on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float w5 = exp(-25.0 * denom_inv); + const float w6 = exp(-36.0 * denom_inv); + const float w7 = exp(-49.0 * denom_inv); + const float w8 = exp(-64.0 * denom_inv); + const float w9 = exp(-81.0 * denom_inv); + const float w10 = exp(-100.0 * denom_inv); + const float w11 = exp(-121.0 * denom_inv); + const float w12 = exp(-144.0 * denom_inv); + //const float weight_sum_inv = 1.0 / (w0 + 2.0 * ( + // w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8 + w9 + w10 + w11 + w12)); + const float weight_sum_inv = get_fast_gaussian_weight_sum_inv(sigma); + // Calculate combined weights and linear sample ratios between texel pairs. + const float w1_2 = w1 + w2; + const float w3_4 = w3 + w4; + const float w5_6 = w5 + w6; + const float w7_8 = w7 + w8; + const float w9_10 = w9 + w10; + const float w11_12 = w11 + w12; + const float w1_2_ratio = w2/w1_2; + const float w3_4_ratio = w4/w3_4; + const float w5_6_ratio = w6/w5_6; + const float w7_8_ratio = w8/w7_8; + const float w9_10_ratio = w10/w9_10; + const float w11_12_ratio = w12/w11_12; + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w11_12 * tex2D_linearize(tex, tex_uv - (11.0 + w11_12_ratio) * dxdy, input_gamma).rgb; + sum += w9_10 * tex2D_linearize(tex, tex_uv - (9.0 + w9_10_ratio) * dxdy, input_gamma).rgb; + sum += w7_8 * tex2D_linearize(tex, tex_uv - (7.0 + w7_8_ratio) * dxdy, input_gamma).rgb; + sum += w5_6 * tex2D_linearize(tex, tex_uv - (5.0 + w5_6_ratio) * dxdy, input_gamma).rgb; + sum += w3_4 * tex2D_linearize(tex, tex_uv - (3.0 + w3_4_ratio) * dxdy, input_gamma).rgb; + sum += w1_2 * tex2D_linearize(tex, tex_uv - (1.0 + w1_2_ratio) * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w1_2 * tex2D_linearize(tex, tex_uv + (1.0 + w1_2_ratio) * dxdy, input_gamma).rgb; + sum += w3_4 * tex2D_linearize(tex, tex_uv + (3.0 + w3_4_ratio) * dxdy, input_gamma).rgb; + sum += w5_6 * tex2D_linearize(tex, tex_uv + (5.0 + w5_6_ratio) * dxdy, input_gamma).rgb; + sum += w7_8 * tex2D_linearize(tex, tex_uv + (7.0 + w7_8_ratio) * dxdy, input_gamma).rgb; + sum += w9_10 * tex2D_linearize(tex, tex_uv + (9.0 + w9_10_ratio) * dxdy, input_gamma).rgb; + sum += w11_12 * tex2D_linearize(tex, tex_uv + (11.0 + w11_12_ratio) * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + +float3 tex2Dblur17fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Same as tex2Dblur11() + // Returns: A 1D 17x Gaussian blurred texture lookup using 1 nearest + // neighbor and 8 linear taps. It may be mipmapped depending + // on settings and dxdy. + // First get the texel weights and normalization factor as above. + const float denom_inv = 0.5/(sigma*sigma); + const float w0 = 1.0; + const float w1 = exp(-1.0 * denom_inv); + const float w2 = exp(-4.0 * denom_inv); + const float w3 = exp(-9.0 * denom_inv); + const float w4 = exp(-16.0 * denom_inv); + const float w5 = exp(-25.0 * denom_inv); + const float w6 = exp(-36.0 * denom_inv); + const float w7 = exp(-49.0 * denom_inv); + const float w8 = exp(-64.0 * denom_inv); + //const float weight_sum_inv = 1.0 / (w0 + 2.0 * ( + // w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8)); + const float weight_sum_inv = get_fast_gaussian_weight_sum_inv(sigma); + // Calculate combined weights and linear sample ratios between texel pairs. + const float w1_2 = w1 + w2; + const float w3_4 = w3 + w4; + const float w5_6 = w5 + w6; + const float w7_8 = w7 + w8; + const float w1_2_ratio = w2/w1_2; + const float w3_4_ratio = w4/w3_4; + const float w5_6_ratio = w6/w5_6; + const float w7_8_ratio = w8/w7_8; + // Statically normalize weights, sum weighted samples, and return: + float3 sum = float3(0.0,0.0,0.0); + sum += w7_8 * tex2D_linearize(tex, tex_uv - (7.0 + w7_8_ratio) * dxdy, input_gamma).rgb; + sum += w5_6 * tex2D_linearize(tex, tex_uv - (5.0 + w5_6_ratio) * dxdy, input_gamma).rgb; + sum += w3_4 * tex2D_linearize(tex, tex_uv - (3.0 + w3_4_ratio) * dxdy, input_gamma).rgb; + sum += w1_2 * tex2D_linearize(tex, tex_uv - (1.0 + w1_2_ratio) * dxdy, input_gamma).rgb; + sum += w0 * tex2D_linearize(tex, tex_uv, input_gamma).rgb; + sum += w1_2 * tex2D_linearize(tex, tex_uv + (1.0 + w1_2_ratio) * dxdy, input_gamma).rgb; + sum += w3_4 * tex2D_linearize(tex, tex_uv + (3.0 + w3_4_ratio) * dxdy, input_gamma).rgb; + sum += w5_6 * tex2D_linearize(tex, tex_uv + (5.0 + w5_6_ratio) * dxdy, input_gamma).rgb; + sum += w7_8 * tex2D_linearize(tex, tex_uv + (7.0 + w7_8_ratio) * dxdy, input_gamma).rgb; + return sum * weight_sum_inv; +} + + +//////////////////// ARBITRARILY RESIZABLE ONE-PASS BLURS //////////////////// + +float3 tex2Dblur3x3resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Requires: Global requirements must be met (see file description). + // Returns: A 3x3 Gaussian blurred mipmapped texture lookup of the + // resized input. + // Description: + // This is the only arbitrarily resizable one-pass blur; tex2Dblur5x5resize + // would perform like tex2Dblur9x9, MUCH slower than tex2Dblur5resize. + const float denom_inv = 0.5/(sigma*sigma); + // Load each sample. We need all 3x3 samples. Quad-pixel communication + // won't help either: This should perform like tex2Dblur5x5, but sharing a + // 4x4 sample field would perform more like tex2Dblur8x8shared (worse). + const float2 sample4_uv = tex_uv; + const float2 dx = float2(dxdy.x, 0.0); + const float2 dy = float2(0.0, dxdy.y); + const float2 sample1_uv = sample4_uv - dy; + const float2 sample7_uv = sample4_uv + dy; + const float3 sample0 = tex2D_linearize(tex, sample1_uv - dx, input_gamma).rgb; + const float3 sample1 = tex2D_linearize(tex, sample1_uv, input_gamma).rgb; + const float3 sample2 = tex2D_linearize(tex, sample1_uv + dx, input_gamma).rgb; + const float3 sample3 = tex2D_linearize(tex, sample4_uv - dx, input_gamma).rgb; + const float3 sample4 = tex2D_linearize(tex, sample4_uv, input_gamma).rgb; + const float3 sample5 = tex2D_linearize(tex, sample4_uv + dx, input_gamma).rgb; + const float3 sample6 = tex2D_linearize(tex, sample7_uv - dx, input_gamma).rgb; + const float3 sample7 = tex2D_linearize(tex, sample7_uv, input_gamma).rgb; + const float3 sample8 = tex2D_linearize(tex, sample7_uv + dx, input_gamma).rgb; + // Statically compute Gaussian sample weights: + const float w4 = 1.0; + const float w1_3_5_7 = exp(-LENGTH_SQ(float2(1.0, 0.0)) * denom_inv); + const float w0_2_6_8 = exp(-LENGTH_SQ(float2(1.0, 1.0)) * denom_inv); + const float weight_sum_inv = 1.0/(w4 + 4.0 * (w1_3_5_7 + w0_2_6_8)); + // Weight and sum the samples: + const float3 sum = w4 * sample4 + + w1_3_5_7 * (sample1 + sample3 + sample5 + sample7) + + w0_2_6_8 * (sample0 + sample2 + sample6 + sample8); + return sum * weight_sum_inv; +} + + +//////////////////////////// FASTER ONE-PASS BLURS /////////////////////////// + +float3 tex2Dblur9x9(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Perform a 1-pass 9x9 blur with 5x5 bilinear samples. + // Requires: Same as tex2Dblur9() + // Returns: A 9x9 Gaussian blurred mipmapped texture lookup composed of + // 5x5 carefully selected bilinear samples. + // Description: + // Perform a 1-pass 9x9 blur with 5x5 bilinear samples. Adjust the + // bilinear sample location to reflect the true Gaussian weights for each + // underlying texel. The following diagram illustrates the relative + // locations of bilinear samples. Each sample with the same number has the + // same weight (notice the symmetry). The letters a, b, c, d distinguish + // quadrants, and the letters U, D, L, R, C (up, down, left, right, center) + // distinguish 1D directions along the line containing the pixel center: + // 6a 5a 2U 5b 6b + // 4a 3a 1U 3b 4b + // 2L 1L 0C 1R 2R + // 4c 3c 1D 3d 4d + // 6c 5c 2D 5d 6d + // The following diagram illustrates the underlying equally spaced texels, + // named after the sample that accesses them and subnamed by their location + // within their 2x2, 2x1, 1x2, or 1x1 texel block: + // 6a4 6a3 5a4 5a3 2U2 5b3 5b4 6b3 6b4 + // 6a2 6a1 5a2 5a1 2U1 5b1 5b2 6b1 6b2 + // 4a4 4a3 3a4 3a3 1U2 3b3 3b4 4b3 4b4 + // 4a2 4a1 3a2 3a1 1U1 3b1 3b2 4b1 4b2 + // 2L2 2L1 1L2 1L1 0C1 1R1 1R2 2R1 2R2 + // 4c2 4c1 3c2 3c1 1D1 3d1 3d2 4d1 4d2 + // 4c4 4c3 3c4 3c3 1D2 3d3 3d4 4d3 4d4 + // 6c2 6c1 5c2 5c1 2D1 5d1 5d2 6d1 6d2 + // 6c4 6c3 5c4 5c3 2D2 5d3 5d4 6d3 6d4 + // Note there is only one C texel and only two texels for each U, D, L, or + // R sample. The center sample is effectively a nearest neighbor sample, + // and the U/D/L/R samples use 1D linear filtering. All other texels are + // read with bilinear samples somewhere within their 2x2 texel blocks. + + // COMPUTE TEXTURE COORDS: + // Statically compute sampling offsets within each 2x2 texel block, based + // on 1D sampling ratios between texels [1, 2] and [3, 4] texels away from + // the center, and reuse them independently for both dimensions. Compute + // these offsets based on the relative 1D Gaussian weights of the texels + // in question. (w1off means "Gaussian weight for the texel 1.0 texels + // away from the pixel center," etc.). + const float denom_inv = 0.5/(sigma*sigma); + const float w1off = exp(-1.0 * denom_inv); + const float w2off = exp(-4.0 * denom_inv); + const float w3off = exp(-9.0 * denom_inv); + const float w4off = exp(-16.0 * denom_inv); + const float texel1to2ratio = w2off/(w1off + w2off); + const float texel3to4ratio = w4off/(w3off + w4off); + // Statically compute texel offsets from the fragment center to each + // bilinear sample in the bottom-right quadrant, including x-axis-aligned: + const float2 sample1R_texel_offset = float2(1.0, 0.0) + float2(texel1to2ratio, 0.0); + const float2 sample2R_texel_offset = float2(3.0, 0.0) + float2(texel3to4ratio, 0.0); + const float2 sample3d_texel_offset = float2(1.0, 1.0) + float2(texel1to2ratio, texel1to2ratio); + const float2 sample4d_texel_offset = float2(3.0, 1.0) + float2(texel3to4ratio, texel1to2ratio); + const float2 sample5d_texel_offset = float2(1.0, 3.0) + float2(texel1to2ratio, texel3to4ratio); + const float2 sample6d_texel_offset = float2(3.0, 3.0) + float2(texel3to4ratio, texel3to4ratio); + + // CALCULATE KERNEL WEIGHTS FOR ALL SAMPLES: + // Statically compute Gaussian texel weights for the bottom-right quadrant. + // Read underscores as "and." + const float w1R1 = w1off; + const float w1R2 = w2off; + const float w2R1 = w3off; + const float w2R2 = w4off; + const float w3d1 = exp(-LENGTH_SQ(float2(1.0, 1.0)) * denom_inv); + const float w3d2_3d3 = exp(-LENGTH_SQ(float2(2.0, 1.0)) * denom_inv); + const float w3d4 = exp(-LENGTH_SQ(float2(2.0, 2.0)) * denom_inv); + const float w4d1_5d1 = exp(-LENGTH_SQ(float2(3.0, 1.0)) * denom_inv); + const float w4d2_5d3 = exp(-LENGTH_SQ(float2(4.0, 1.0)) * denom_inv); + const float w4d3_5d2 = exp(-LENGTH_SQ(float2(3.0, 2.0)) * denom_inv); + const float w4d4_5d4 = exp(-LENGTH_SQ(float2(4.0, 2.0)) * denom_inv); + const float w6d1 = exp(-LENGTH_SQ(float2(3.0, 3.0)) * denom_inv); + const float w6d2_6d3 = exp(-LENGTH_SQ(float2(4.0, 3.0)) * denom_inv); + const float w6d4 = exp(-LENGTH_SQ(float2(4.0, 4.0)) * denom_inv); + // Statically add texel weights in each sample to get sample weights: + const float w0 = 1.0; + const float w1 = w1R1 + w1R2; + const float w2 = w2R1 + w2R2; + const float w3 = w3d1 + 2.0 * w3d2_3d3 + w3d4; + const float w4 = w4d1_5d1 + w4d2_5d3 + w4d3_5d2 + w4d4_5d4; + const float w5 = w4; + const float w6 = w6d1 + 2.0 * w6d2_6d3 + w6d4; + // Get the weight sum inverse (normalization factor): + const float weight_sum_inv = + 1.0/(w0 + 4.0 * (w1 + w2 + w3 + w4 + w5 + w6)); + + // LOAD TEXTURE SAMPLES: + // Load all 25 samples (1 nearest, 8 linear, 16 bilinear) using symmetry: + const float2 mirror_x = float2(-1.0, 1.0); + const float2 mirror_y = float2(1.0, -1.0); + const float2 mirror_xy = float2(-1.0, -1.0); + const float2 dxdy_mirror_x = dxdy * mirror_x; + const float2 dxdy_mirror_y = dxdy * mirror_y; + const float2 dxdy_mirror_xy = dxdy * mirror_xy; + // Sampling order doesn't seem to affect performance, so just be clear: + const float3 sample0C = tex2D_linearize(tex, tex_uv, input_gamma).rgb; + const float3 sample1R = tex2D_linearize(tex, tex_uv + dxdy * sample1R_texel_offset, input_gamma).rgb; + const float3 sample1D = tex2D_linearize(tex, tex_uv + dxdy * sample1R_texel_offset.yx, input_gamma).rgb; + const float3 sample1L = tex2D_linearize(tex, tex_uv - dxdy * sample1R_texel_offset, input_gamma).rgb; + const float3 sample1U = tex2D_linearize(tex, tex_uv - dxdy * sample1R_texel_offset.yx, input_gamma).rgb; + const float3 sample2R = tex2D_linearize(tex, tex_uv + dxdy * sample2R_texel_offset, input_gamma).rgb; + const float3 sample2D = tex2D_linearize(tex, tex_uv + dxdy * sample2R_texel_offset.yx, input_gamma).rgb; + const float3 sample2L = tex2D_linearize(tex, tex_uv - dxdy * sample2R_texel_offset, input_gamma).rgb; + const float3 sample2U = tex2D_linearize(tex, tex_uv - dxdy * sample2R_texel_offset.yx, input_gamma).rgb; + const float3 sample3d = tex2D_linearize(tex, tex_uv + dxdy * sample3d_texel_offset, input_gamma).rgb; + const float3 sample3c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample3d_texel_offset, input_gamma).rgb; + const float3 sample3b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample3d_texel_offset, input_gamma).rgb; + const float3 sample3a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample3d_texel_offset, input_gamma).rgb; + const float3 sample4d = tex2D_linearize(tex, tex_uv + dxdy * sample4d_texel_offset, input_gamma).rgb; + const float3 sample4c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample4d_texel_offset, input_gamma).rgb; + const float3 sample4b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample4d_texel_offset, input_gamma).rgb; + const float3 sample4a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample4d_texel_offset, input_gamma).rgb; + const float3 sample5d = tex2D_linearize(tex, tex_uv + dxdy * sample5d_texel_offset, input_gamma).rgb; + const float3 sample5c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample5d_texel_offset, input_gamma).rgb; + const float3 sample5b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample5d_texel_offset, input_gamma).rgb; + const float3 sample5a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample5d_texel_offset, input_gamma).rgb; + const float3 sample6d = tex2D_linearize(tex, tex_uv + dxdy * sample6d_texel_offset, input_gamma).rgb; + const float3 sample6c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample6d_texel_offset, input_gamma).rgb; + const float3 sample6b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample6d_texel_offset, input_gamma).rgb; + const float3 sample6a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample6d_texel_offset, input_gamma).rgb; + + // SUM WEIGHTED SAMPLES: + // Statically normalize weights (so total = 1.0), and sum weighted samples. + float3 sum = w0 * sample0C; + sum += w1 * (sample1R + sample1D + sample1L + sample1U); + sum += w2 * (sample2R + sample2D + sample2L + sample2U); + sum += w3 * (sample3d + sample3c + sample3b + sample3a); + sum += w4 * (sample4d + sample4c + sample4b + sample4a); + sum += w5 * (sample5d + sample5c + sample5b + sample5a); + sum += w6 * (sample6d + sample6c + sample6b + sample6a); + return sum * weight_sum_inv; +} + +float3 tex2Dblur7x7(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Perform a 1-pass 7x7 blur with 5x5 bilinear samples. + // Requires: Same as tex2Dblur9() + // Returns: A 7x7 Gaussian blurred mipmapped texture lookup composed of + // 4x4 carefully selected bilinear samples. + // Description: + // First see the descriptions for tex2Dblur9x9() and tex2Dblur7(). This + // blur mixes concepts from both. The sample layout is as follows: + // 4a 3a 3b 4b + // 2a 1a 1b 2b + // 2c 1c 1d 2d + // 4c 3c 3d 4d + // The texel layout is as follows. Note that samples 3a/3b, 1a/1b, 1c/1d, + // and 3c/3d share a vertical column of texels, and samples 2a/2c, 1a/1c, + // 1b/1d, and 2b/2d share a horizontal row of texels (all sample1's share + // the center texel): + // 4a4 4a3 3a4 3ab3 3b4 4b3 4b4 + // 4a2 4a1 3a2 3ab1 3b2 4b1 4b2 + // 2a4 2a3 1a4 1ab3 1b4 2b3 2b4 + // 2ac2 2ac1 1ac2 1* 1bd2 2bd1 2bd2 + // 2c4 2c3 1c4 1cd3 1d4 2d3 2d4 + // 4c2 4c1 3c2 3cd1 3d2 4d1 4d2 + // 4c4 4c3 3c4 3cd3 3d4 4d3 4d4 + + // COMPUTE TEXTURE COORDS: + // Statically compute bilinear sampling offsets (details in tex2Dblur9x9). + const float denom_inv = 0.5/(sigma*sigma); + const float w0off = 1.0; + const float w1off = exp(-1.0 * denom_inv); + const float w2off = exp(-4.0 * denom_inv); + const float w3off = exp(-9.0 * denom_inv); + const float texel0to1ratio = w1off/(w0off * 0.5 + w1off); + const float texel2to3ratio = w3off/(w2off + w3off); + // Statically compute texel offsets from the fragment center to each + // bilinear sample in the bottom-right quadrant, including axis-aligned: + const float2 sample1d_texel_offset = float2(texel0to1ratio, texel0to1ratio); + const float2 sample2d_texel_offset = float2(2.0, 0.0) + float2(texel2to3ratio, texel0to1ratio); + const float2 sample3d_texel_offset = float2(0.0, 2.0) + float2(texel0to1ratio, texel2to3ratio); + const float2 sample4d_texel_offset = float2(2.0, 2.0) + float2(texel2to3ratio, texel2to3ratio); + + // CALCULATE KERNEL WEIGHTS FOR ALL SAMPLES: + // Statically compute Gaussian texel weights for the bottom-right quadrant. + // Read underscores as "and." + const float w1abcd = 1.0; + const float w1bd2_1cd3 = exp(-LENGTH_SQ(float2(1.0, 0.0)) * denom_inv); + const float w2bd1_3cd1 = exp(-LENGTH_SQ(float2(2.0, 0.0)) * denom_inv); + const float w2bd2_3cd2 = exp(-LENGTH_SQ(float2(3.0, 0.0)) * denom_inv); + const float w1d4 = exp(-LENGTH_SQ(float2(1.0, 1.0)) * denom_inv); + const float w2d3_3d2 = exp(-LENGTH_SQ(float2(2.0, 1.0)) * denom_inv); + const float w2d4_3d4 = exp(-LENGTH_SQ(float2(3.0, 1.0)) * denom_inv); + const float w4d1 = exp(-LENGTH_SQ(float2(2.0, 2.0)) * denom_inv); + const float w4d2_4d3 = exp(-LENGTH_SQ(float2(3.0, 2.0)) * denom_inv); + const float w4d4 = exp(-LENGTH_SQ(float2(3.0, 3.0)) * denom_inv); + // Statically add texel weights in each sample to get sample weights. + // Split weights for shared texels between samples sharing them: + const float w1 = w1abcd * 0.25 + w1bd2_1cd3 + w1d4; + const float w2_3 = (w2bd1_3cd1 + w2bd2_3cd2) * 0.5 + w2d3_3d2 + w2d4_3d4; + const float w4 = w4d1 + 2.0 * w4d2_4d3 + w4d4; + // Get the weight sum inverse (normalization factor): + const float weight_sum_inv = + 1.0/(4.0 * (w1 + 2.0 * w2_3 + w4)); + + // LOAD TEXTURE SAMPLES: + // Load all 16 samples using symmetry: + const float2 mirror_x = float2(-1.0, 1.0); + const float2 mirror_y = float2(1.0, -1.0); + const float2 mirror_xy = float2(-1.0, -1.0); + const float2 dxdy_mirror_x = dxdy * mirror_x; + const float2 dxdy_mirror_y = dxdy * mirror_y; + const float2 dxdy_mirror_xy = dxdy * mirror_xy; + const float3 sample1a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample1d_texel_offset, input_gamma).rgb; + const float3 sample2a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample2d_texel_offset, input_gamma).rgb; + const float3 sample3a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample3d_texel_offset, input_gamma).rgb; + const float3 sample4a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample4d_texel_offset, input_gamma).rgb; + const float3 sample1b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample1d_texel_offset, input_gamma).rgb; + const float3 sample2b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample2d_texel_offset, input_gamma).rgb; + const float3 sample3b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample3d_texel_offset, input_gamma).rgb; + const float3 sample4b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample4d_texel_offset, input_gamma).rgb; + const float3 sample1c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample1d_texel_offset, input_gamma).rgb; + const float3 sample2c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample2d_texel_offset, input_gamma).rgb; + const float3 sample3c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample3d_texel_offset, input_gamma).rgb; + const float3 sample4c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample4d_texel_offset, input_gamma).rgb; + const float3 sample1d = tex2D_linearize(tex, tex_uv + dxdy * sample1d_texel_offset, input_gamma).rgb; + const float3 sample2d = tex2D_linearize(tex, tex_uv + dxdy * sample2d_texel_offset, input_gamma).rgb; + const float3 sample3d = tex2D_linearize(tex, tex_uv + dxdy * sample3d_texel_offset, input_gamma).rgb; + const float3 sample4d = tex2D_linearize(tex, tex_uv + dxdy * sample4d_texel_offset, input_gamma).rgb; + + // SUM WEIGHTED SAMPLES: + // Statically normalize weights (so total = 1.0), and sum weighted samples. + float3 sum = float3(0.0,0.0,0.0); + sum += w1 * (sample1a + sample1b + sample1c + sample1d); + sum += w2_3 * (sample2a + sample2b + sample2c + sample2d); + sum += w2_3 * (sample3a + sample3b + sample3c + sample3d); + sum += w4 * (sample4a + sample4b + sample4c + sample4d); + return sum * weight_sum_inv; +} + +float3 tex2Dblur5x5(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Perform a 1-pass 5x5 blur with 3x3 bilinear samples. + // Requires: Same as tex2Dblur9() + // Returns: A 5x5 Gaussian blurred mipmapped texture lookup composed of + // 3x3 carefully selected bilinear samples. + // Description: + // First see the description for tex2Dblur9x9(). This blur uses the same + // concept and sample/texel locations except on a smaller scale. Samples: + // 2a 1U 2b + // 1L 0C 1R + // 2c 1D 2d + // Texels: + // 2a4 2a3 1U2 2b3 2b4 + // 2a2 2a1 1U1 2b1 2b2 + // 1L2 1L1 0C1 1R1 1R2 + // 2c2 2c1 1D1 2d1 2d2 + // 2c4 2c3 1D2 2d3 2d4 + + // COMPUTE TEXTURE COORDS: + // Statically compute bilinear sampling offsets (details in tex2Dblur9x9). + const float denom_inv = 0.5/(sigma*sigma); + const float w1off = exp(-1.0 * denom_inv); + const float w2off = exp(-4.0 * denom_inv); + const float texel1to2ratio = w2off/(w1off + w2off); + // Statically compute texel offsets from the fragment center to each + // bilinear sample in the bottom-right quadrant, including x-axis-aligned: + const float2 sample1R_texel_offset = float2(1.0, 0.0) + float2(texel1to2ratio, 0.0); + const float2 sample2d_texel_offset = float2(1.0, 1.0) + float2(texel1to2ratio, texel1to2ratio); + + // CALCULATE KERNEL WEIGHTS FOR ALL SAMPLES: + // Statically compute Gaussian texel weights for the bottom-right quadrant. + // Read underscores as "and." + const float w1R1 = w1off; + const float w1R2 = w2off; + const float w2d1 = exp(-LENGTH_SQ(float2(1.0, 1.0)) * denom_inv); + const float w2d2_3 = exp(-LENGTH_SQ(float2(2.0, 1.0)) * denom_inv); + const float w2d4 = exp(-LENGTH_SQ(float2(2.0, 2.0)) * denom_inv); + // Statically add texel weights in each sample to get sample weights: + const float w0 = 1.0; + const float w1 = w1R1 + w1R2; + const float w2 = w2d1 + 2.0 * w2d2_3 + w2d4; + // Get the weight sum inverse (normalization factor): + const float weight_sum_inv = 1.0/(w0 + 4.0 * (w1 + w2)); + + // LOAD TEXTURE SAMPLES: + // Load all 9 samples (1 nearest, 4 linear, 4 bilinear) using symmetry: + const float2 mirror_x = float2(-1.0, 1.0); + const float2 mirror_y = float2(1.0, -1.0); + const float2 mirror_xy = float2(-1.0, -1.0); + const float2 dxdy_mirror_x = dxdy * mirror_x; + const float2 dxdy_mirror_y = dxdy * mirror_y; + const float2 dxdy_mirror_xy = dxdy * mirror_xy; + const float3 sample0C = tex2D_linearize(tex, tex_uv, input_gamma).rgb; + const float3 sample1R = tex2D_linearize(tex, tex_uv + dxdy * sample1R_texel_offset, input_gamma).rgb; + const float3 sample1D = tex2D_linearize(tex, tex_uv + dxdy * sample1R_texel_offset.yx, input_gamma).rgb; + const float3 sample1L = tex2D_linearize(tex, tex_uv - dxdy * sample1R_texel_offset, input_gamma).rgb; + const float3 sample1U = tex2D_linearize(tex, tex_uv - dxdy * sample1R_texel_offset.yx, input_gamma).rgb; + const float3 sample2d = tex2D_linearize(tex, tex_uv + dxdy * sample2d_texel_offset, input_gamma).rgb; + const float3 sample2c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample2d_texel_offset, input_gamma).rgb; + const float3 sample2b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample2d_texel_offset, input_gamma).rgb; + const float3 sample2a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample2d_texel_offset, input_gamma).rgb; + + // SUM WEIGHTED SAMPLES: + // Statically normalize weights (so total = 1.0), and sum weighted samples. + float3 sum = w0 * sample0C; + sum += w1 * (sample1R + sample1D + sample1L + sample1U); + sum += w2 * (sample2a + sample2b + sample2c + sample2d); + return sum * weight_sum_inv; +} + +float3 tex2Dblur3x3(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, const float sigma, + const float input_gamma) +{ + // Perform a 1-pass 3x3 blur with 5x5 bilinear samples. + // Requires: Same as tex2Dblur9() + // Returns: A 3x3 Gaussian blurred mipmapped texture lookup composed of + // 2x2 carefully selected bilinear samples. + // Description: + // First see the descriptions for tex2Dblur9x9() and tex2Dblur7(). This + // blur mixes concepts from both. The sample layout is as follows: + // 0a 0b + // 0c 0d + // The texel layout is as follows. Note that samples 0a/0b and 0c/0d share + // a vertical column of texels, and samples 0a/0c and 0b/0d share a + // horizontal row of texels (all samples share the center texel): + // 0a3 0ab2 0b3 + // 0ac1 0*0 0bd1 + // 0c3 0cd2 0d3 + + // COMPUTE TEXTURE COORDS: + // Statically compute bilinear sampling offsets (details in tex2Dblur9x9). + const float denom_inv = 0.5/(sigma*sigma); + const float w0off = 1.0; + const float w1off = exp(-1.0 * denom_inv); + const float texel0to1ratio = w1off/(w0off * 0.5 + w1off); + // Statically compute texel offsets from the fragment center to each + // bilinear sample in the bottom-right quadrant, including axis-aligned: + const float2 sample0d_texel_offset = float2(texel0to1ratio, texel0to1ratio); + + // LOAD TEXTURE SAMPLES: + // Load all 4 samples using symmetry: + const float2 mirror_x = float2(-1.0, 1.0); + const float2 mirror_y = float2(1.0, -1.0); + const float2 mirror_xy = float2(-1.0, -1.0); + const float2 dxdy_mirror_x = dxdy * mirror_x; + const float2 dxdy_mirror_y = dxdy * mirror_y; + const float2 dxdy_mirror_xy = dxdy * mirror_xy; + const float3 sample0a = tex2D_linearize(tex, tex_uv + dxdy_mirror_xy * sample0d_texel_offset, input_gamma).rgb; + const float3 sample0b = tex2D_linearize(tex, tex_uv + dxdy_mirror_y * sample0d_texel_offset, input_gamma).rgb; + const float3 sample0c = tex2D_linearize(tex, tex_uv + dxdy_mirror_x * sample0d_texel_offset, input_gamma).rgb; + const float3 sample0d = tex2D_linearize(tex, tex_uv + dxdy * sample0d_texel_offset, input_gamma).rgb; + + // SUM WEIGHTED SAMPLES: + // Weights for all samples are the same, so just average them: + return 0.25 * (sample0a + sample0b + sample0c + sample0d); +} + + +////////////////// LINEAR ONE-PASS BLURS WITH SHARED SAMPLES ///////////////// + +float3 tex2Dblur12x12shared(const sampler2D tex, + const float4 tex_uv, const float2 dxdy, const float4 quad_vector, + const float sigma, + const float input_gamma) +{ + // Perform a 1-pass mipmapped blur with shared samples across a pixel quad. + // Requires: 1.) Same as tex2Dblur9() + // 2.) ddx() and ddy() are present in the current Cg profile. + // 3.) The GPU driver is using fine/high-quality derivatives. + // 4.) quad_vector *correctly* describes the current fragment's + // location in its pixel quad, by the conventions noted in + // get_quad_vector[_naive]. + // 5.) tex_uv.w = log2(video_size/output_size).y + // 6.) tex2Dlod() is present in the current Cg profile. + // Optional: Tune artifacts vs. excessive blurriness with the global + // float error_blurring. + // Returns: A blurred texture lookup using a "virtual" 12x12 Gaussian + // blur (a 6x6 blur of carefully selected bilinear samples) + // of the given mip level. There will be subtle inaccuracies, + // especially for small or high-frequency detailed sources. + // Description: + // Perform a 1-pass blur with shared texture lookups across a pixel quad. + // We'll get neighboring samples with high-quality ddx/ddy derivatives, as + // in GPU Pro 2, Chapter VI.2, "Shader Amortization using Pixel Quad + // Message Passing" by Eric Penner. + // + // Our "virtual" 12x12 blur will be comprised of ((6 - 1)^2)/4 + 3 = 12 + // bilinear samples, where bilinear sampling positions are computed from + // the relative Gaussian weights of the 4 surrounding texels. The catch is + // that the appropriate texel weights and sample coords differ for each + // fragment, but we're reusing most of the same samples across a quad of + // destination fragments. (We do use unique coords for the four nearest + // samples at each fragment.) Mixing bilinear filtering and sample-sharing + // therefore introduces some error into the weights, and this can get nasty + // when the source image is small or high-frequency. Computing bilinear + // ratios based on weights at the sample field center results in sharpening + // and ringing artifacts, but we can move samples closer to halfway between + // texels to try blurring away the error (which can move features around by + // a texel or so). Tune this with the global float "error_blurring". + // + // The pixel quad's sample field covers 12x12 texels, accessed through 6x6 + // bilinear (2x2 texel) taps. Each fragment depends on a window of 10x10 + // texels (5x5 bilinear taps), and each fragment is responsible for loading + // a 6x6 texel quadrant as a 3x3 block of bilinear taps, plus 3 more taps + // to use unique bilinear coords for sample0* for each fragment. This + // diagram illustrates the relative locations of bilinear samples 1-9 for + // each quadrant a, b, c, d (note samples will not be equally spaced): + // 8a 7a 6a 6b 7b 8b + // 5a 4a 3a 3b 4b 5b + // 2a 1a 0a 0b 1b 2b + // 2c 1c 0c 0d 1d 2d + // 5c 4c 3c 3d 4d 5d + // 8c 7c 6c 6d 7d 8d + // The following diagram illustrates the underlying equally spaced texels, + // named after the sample that accesses them and subnamed by their location + // within their 2x2 texel block: + // 8a3 8a2 7a3 7a2 6a3 6a2 6b2 6b3 7b2 7b3 8b2 8b3 + // 8a1 8a0 7a1 7a0 6a1 6a0 6b0 6b1 7b0 7b1 8b0 8b1 + // 5a3 5a2 4a3 4a2 3a3 3a2 3b2 3b3 4b2 4b3 5b2 5b3 + // 5a1 5a0 4a1 4a0 3a1 3a0 3b0 3b1 4b0 4b1 5b0 5b1 + // 2a3 2a2 1a3 1a2 0a3 0a2 0b2 0b3 1b2 1b3 2b2 2b3 + // 2a1 2a0 1a1 1a0 0a1 0a0 0b0 0b1 1b0 1b1 2b0 2b1 + // 2c1 2c0 1c1 1c0 0c1 0c0 0d0 0d1 1d0 1d1 2d0 2d1 + // 2c3 2c2 1c3 1c2 0c3 0c2 0d2 0d3 1d2 1d3 2d2 2d3 + // 5c1 5c0 4c1 4c0 3c1 3c0 3d0 3d1 4d0 4d1 5d0 5d1 + // 5c3 5c2 4c3 4c2 3c3 3c2 3d2 3d3 4d2 4d3 5d2 5d3 + // 8c1 8c0 7c1 7c0 6c1 6c0 6d0 6d1 7d0 7d1 8d0 8d1 + // 8c3 8c2 7c3 7c2 6c3 6c2 6d2 6d3 7d2 7d3 8d2 8d3 + // With this symmetric arrangement, we don't have to know which absolute + // quadrant a sample lies in to assign kernel weights; it's enough to know + // the sample number and the relative quadrant of the sample (relative to + // the current quadrant): + // {current, adjacent x, adjacent y, diagonal} + + // COMPUTE COORDS FOR TEXTURE SAMPLES THIS FRAGMENT IS RESPONSIBLE FOR: + // Statically compute sampling offsets within each 2x2 texel block, based + // on appropriate 1D Gaussian sampling ratio between texels [0, 1], [2, 3], + // and [4, 5] away from the fragment, and reuse them independently for both + // dimensions. Use the sample field center as the estimated destination, + // but nudge the result closer to halfway between texels to blur error. + const float denom_inv = 0.5/(sigma*sigma); + const float w0off = 1.0; + const float w0_5off = exp(-(0.5*0.5) * denom_inv); + const float w1off = exp(-(1.0*1.0) * denom_inv); + const float w1_5off = exp(-(1.5*1.5) * denom_inv); + const float w2off = exp(-(2.0*2.0) * denom_inv); + const float w2_5off = exp(-(2.5*2.5) * denom_inv); + const float w3_5off = exp(-(3.5*3.5) * denom_inv); + const float w4_5off = exp(-(4.5*4.5) * denom_inv); + const float w5_5off = exp(-(5.5*5.5) * denom_inv); + const float texel0to1ratio = lerp(w1_5off/(w0_5off + w1_5off), 0.5, error_blurring); + const float texel2to3ratio = lerp(w3_5off/(w2_5off + w3_5off), 0.5, error_blurring); + const float texel4to5ratio = lerp(w5_5off/(w4_5off + w5_5off), 0.5, error_blurring); + // We don't share sample0*, so use the nearest destination fragment: + const float texel0to1ratio_nearest = w1off/(w0off + w1off); + const float texel1to2ratio_nearest = w2off/(w1off + w2off); + // Statically compute texel offsets from the bottom-right fragment to each + // bilinear sample in the bottom-right quadrant: + const float2 sample0curr_texel_offset = float2(0.0, 0.0) + float2(texel0to1ratio_nearest, texel0to1ratio_nearest); + const float2 sample0adjx_texel_offset = float2(-1.0, 0.0) + float2(-texel1to2ratio_nearest, texel0to1ratio_nearest); + const float2 sample0adjy_texel_offset = float2(0.0, -1.0) + float2(texel0to1ratio_nearest, -texel1to2ratio_nearest); + const float2 sample0diag_texel_offset = float2(-1.0, -1.0) + float2(-texel1to2ratio_nearest, -texel1to2ratio_nearest); + const float2 sample1_texel_offset = float2(2.0, 0.0) + float2(texel2to3ratio, texel0to1ratio); + const float2 sample2_texel_offset = float2(4.0, 0.0) + float2(texel4to5ratio, texel0to1ratio); + const float2 sample3_texel_offset = float2(0.0, 2.0) + float2(texel0to1ratio, texel2to3ratio); + const float2 sample4_texel_offset = float2(2.0, 2.0) + float2(texel2to3ratio, texel2to3ratio); + const float2 sample5_texel_offset = float2(4.0, 2.0) + float2(texel4to5ratio, texel2to3ratio); + const float2 sample6_texel_offset = float2(0.0, 4.0) + float2(texel0to1ratio, texel4to5ratio); + const float2 sample7_texel_offset = float2(2.0, 4.0) + float2(texel2to3ratio, texel4to5ratio); + const float2 sample8_texel_offset = float2(4.0, 4.0) + float2(texel4to5ratio, texel4to5ratio); + + // CALCULATE KERNEL WEIGHTS: + // Statically compute bilinear sample weights at each destination fragment + // based on the sum of their 4 underlying texel weights. Assume a same- + // resolution blur, so each symmetrically named sample weight will compute + // the same at every fragment in the pixel quad: We can therefore compute + // texel weights based only on the bottom-right quadrant (fragment at 0d0). + // Too avoid too much boilerplate code, use a macro to get all 4 texel + // weights for a bilinear sample based on the offset of its top-left texel: + #define GET_TEXEL_QUAD_WEIGHTS(xoff, yoff) \ + (exp(-LENGTH_SQ(float2(xoff, yoff)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff + 1.0, yoff)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff, yoff + 1.0)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff + 1.0, yoff + 1.0)) * denom_inv)) + const float w8diag = GET_TEXEL_QUAD_WEIGHTS(-6.0, -6.0); + const float w7diag = GET_TEXEL_QUAD_WEIGHTS(-4.0, -6.0); + const float w6diag = GET_TEXEL_QUAD_WEIGHTS(-2.0, -6.0); + const float w6adjy = GET_TEXEL_QUAD_WEIGHTS(0.0, -6.0); + const float w7adjy = GET_TEXEL_QUAD_WEIGHTS(2.0, -6.0); + const float w8adjy = GET_TEXEL_QUAD_WEIGHTS(4.0, -6.0); + const float w5diag = GET_TEXEL_QUAD_WEIGHTS(-6.0, -4.0); + const float w4diag = GET_TEXEL_QUAD_WEIGHTS(-4.0, -4.0); + const float w3diag = GET_TEXEL_QUAD_WEIGHTS(-2.0, -4.0); + const float w3adjy = GET_TEXEL_QUAD_WEIGHTS(0.0, -4.0); + const float w4adjy = GET_TEXEL_QUAD_WEIGHTS(2.0, -4.0); + const float w5adjy = GET_TEXEL_QUAD_WEIGHTS(4.0, -4.0); + const float w2diag = GET_TEXEL_QUAD_WEIGHTS(-6.0, -2.0); + const float w1diag = GET_TEXEL_QUAD_WEIGHTS(-4.0, -2.0); + const float w0diag = GET_TEXEL_QUAD_WEIGHTS(-2.0, -2.0); + const float w0adjy = GET_TEXEL_QUAD_WEIGHTS(0.0, -2.0); + const float w1adjy = GET_TEXEL_QUAD_WEIGHTS(2.0, -2.0); + const float w2adjy = GET_TEXEL_QUAD_WEIGHTS(4.0, -2.0); + const float w2adjx = GET_TEXEL_QUAD_WEIGHTS(-6.0, 0.0); + const float w1adjx = GET_TEXEL_QUAD_WEIGHTS(-4.0, 0.0); + const float w0adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 0.0); + const float w0curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 0.0); + const float w1curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 0.0); + const float w2curr = GET_TEXEL_QUAD_WEIGHTS(4.0, 0.0); + const float w5adjx = GET_TEXEL_QUAD_WEIGHTS(-6.0, 2.0); + const float w4adjx = GET_TEXEL_QUAD_WEIGHTS(-4.0, 2.0); + const float w3adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 2.0); + const float w3curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 2.0); + const float w4curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 2.0); + const float w5curr = GET_TEXEL_QUAD_WEIGHTS(4.0, 2.0); + const float w8adjx = GET_TEXEL_QUAD_WEIGHTS(-6.0, 4.0); + const float w7adjx = GET_TEXEL_QUAD_WEIGHTS(-4.0, 4.0); + const float w6adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 4.0); + const float w6curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 4.0); + const float w7curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 4.0); + const float w8curr = GET_TEXEL_QUAD_WEIGHTS(4.0, 4.0); + #undef GET_TEXEL_QUAD_WEIGHTS + // Statically pack weights for runtime: + const float4 w0 = float4(w0curr, w0adjx, w0adjy, w0diag); + const float4 w1 = float4(w1curr, w1adjx, w1adjy, w1diag); + const float4 w2 = float4(w2curr, w2adjx, w2adjy, w2diag); + const float4 w3 = float4(w3curr, w3adjx, w3adjy, w3diag); + const float4 w4 = float4(w4curr, w4adjx, w4adjy, w4diag); + const float4 w5 = float4(w5curr, w5adjx, w5adjy, w5diag); + const float4 w6 = float4(w6curr, w6adjx, w6adjy, w6diag); + const float4 w7 = float4(w7curr, w7adjx, w7adjy, w7diag); + const float4 w8 = float4(w8curr, w8adjx, w8adjy, w8diag); + // Get the weight sum inverse (normalization factor): + const float4 weight_sum4 = w0 + w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8; + const float2 weight_sum2 = weight_sum4.xy + weight_sum4.zw; + const float weight_sum = weight_sum2.x + weight_sum2.y; + const float weight_sum_inv = 1.0/(weight_sum); + + // LOAD TEXTURE SAMPLES THIS FRAGMENT IS RESPONSIBLE FOR: + // Get a uv vector from texel 0q0 of this quadrant to texel 0q3: + const float2 dxdy_curr = dxdy * quad_vector.xy; + // Load bilinear samples for the current quadrant (for this fragment): + const float3 sample0curr = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0curr_texel_offset, input_gamma).rgb; + const float3 sample0adjx = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0adjx_texel_offset, input_gamma).rgb; + const float3 sample0adjy = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0adjy_texel_offset, input_gamma).rgb; + const float3 sample0diag = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0diag_texel_offset, input_gamma).rgb; + const float3 sample1curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample1_texel_offset), input_gamma).rgb; + const float3 sample2curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample2_texel_offset), input_gamma).rgb; + const float3 sample3curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample3_texel_offset), input_gamma).rgb; + const float3 sample4curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample4_texel_offset), input_gamma).rgb; + const float3 sample5curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample5_texel_offset), input_gamma).rgb; + const float3 sample6curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample6_texel_offset), input_gamma).rgb; + const float3 sample7curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample7_texel_offset), input_gamma).rgb; + const float3 sample8curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample8_texel_offset), input_gamma).rgb; + + // GATHER NEIGHBORING SAMPLES AND SUM WEIGHTED SAMPLES: + // Fetch the samples from other fragments in the 2x2 quad: + float3 sample1adjx, sample1adjy, sample1diag; + float3 sample2adjx, sample2adjy, sample2diag; + float3 sample3adjx, sample3adjy, sample3diag; + float3 sample4adjx, sample4adjy, sample4diag; + float3 sample5adjx, sample5adjy, sample5diag; + float3 sample6adjx, sample6adjy, sample6diag; + float3 sample7adjx, sample7adjy, sample7diag; + float3 sample8adjx, sample8adjy, sample8diag; + quad_gather(quad_vector, sample1curr, sample1adjx, sample1adjy, sample1diag); + quad_gather(quad_vector, sample2curr, sample2adjx, sample2adjy, sample2diag); + quad_gather(quad_vector, sample3curr, sample3adjx, sample3adjy, sample3diag); + quad_gather(quad_vector, sample4curr, sample4adjx, sample4adjy, sample4diag); + quad_gather(quad_vector, sample5curr, sample5adjx, sample5adjy, sample5diag); + quad_gather(quad_vector, sample6curr, sample6adjx, sample6adjy, sample6diag); + quad_gather(quad_vector, sample7curr, sample7adjx, sample7adjy, sample7diag); + quad_gather(quad_vector, sample8curr, sample8adjx, sample8adjy, sample8diag); + // Statically normalize weights (so total = 1.0), and sum weighted samples. + // Fill each row of a matrix with an rgb sample and pre-multiply by the + // weights to obtain a weighted result: + float3 sum = float3(0.0,0.0,0.0); + sum += mul(w0, float4x3(sample0curr, sample0adjx, sample0adjy, sample0diag)); + sum += mul(w1, float4x3(sample1curr, sample1adjx, sample1adjy, sample1diag)); + sum += mul(w2, float4x3(sample2curr, sample2adjx, sample2adjy, sample2diag)); + sum += mul(w3, float4x3(sample3curr, sample3adjx, sample3adjy, sample3diag)); + sum += mul(w4, float4x3(sample4curr, sample4adjx, sample4adjy, sample4diag)); + sum += mul(w5, float4x3(sample5curr, sample5adjx, sample5adjy, sample5diag)); + sum += mul(w6, float4x3(sample6curr, sample6adjx, sample6adjy, sample6diag)); + sum += mul(w7, float4x3(sample7curr, sample7adjx, sample7adjy, sample7diag)); + sum += mul(w8, float4x3(sample8curr, sample8adjx, sample8adjy, sample8diag)); + return sum * weight_sum_inv; +} + +float3 tex2Dblur10x10shared(const sampler2D tex, + const float4 tex_uv, const float2 dxdy, const float4 quad_vector, + const float sigma, + const float input_gamma) +{ + // Perform a 1-pass mipmapped blur with shared samples across a pixel quad. + // Requires: Same as tex2Dblur12x12shared() + // Returns: A blurred texture lookup using a "virtual" 10x10 Gaussian + // blur (a 5x5 blur of carefully selected bilinear samples) + // of the given mip level. There will be subtle inaccuracies, + // especially for small or high-frequency detailed sources. + // Description: + // First see the description for tex2Dblur12x12shared(). This + // function shares the same concept and sample placement, but each fragment + // only uses 25 of the 36 samples taken across the pixel quad (to cover a + // 5x5 sample area, or 10x10 texel area), and it uses a lower standard + // deviation to compensate. Thanks to symmetry, the 11 omitted samples + // are always the "same:" + // 8adjx, 2adjx, 5adjx, + // 6adjy, 7adjy, 8adjy, + // 2diag, 5diag, 6diag, 7diag, 8diag + + // COMPUTE COORDS FOR TEXTURE SAMPLES THIS FRAGMENT IS RESPONSIBLE FOR: + // Statically compute bilinear sampling offsets (details in tex2Dblur12x12shared). + const float denom_inv = 0.5/(sigma*sigma); + const float w0off = 1.0; + const float w0_5off = exp(-(0.5*0.5) * denom_inv); + const float w1off = exp(-(1.0*1.0) * denom_inv); + const float w1_5off = exp(-(1.5*1.5) * denom_inv); + const float w2off = exp(-(2.0*2.0) * denom_inv); + const float w2_5off = exp(-(2.5*2.5) * denom_inv); + const float w3_5off = exp(-(3.5*3.5) * denom_inv); + const float w4_5off = exp(-(4.5*4.5) * denom_inv); + const float w5_5off = exp(-(5.5*5.5) * denom_inv); + const float texel0to1ratio = lerp(w1_5off/(w0_5off + w1_5off), 0.5, error_blurring); + const float texel2to3ratio = lerp(w3_5off/(w2_5off + w3_5off), 0.5, error_blurring); + const float texel4to5ratio = lerp(w5_5off/(w4_5off + w5_5off), 0.5, error_blurring); + // We don't share sample0*, so use the nearest destination fragment: + const float texel0to1ratio_nearest = w1off/(w0off + w1off); + const float texel1to2ratio_nearest = w2off/(w1off + w2off); + // Statically compute texel offsets from the bottom-right fragment to each + // bilinear sample in the bottom-right quadrant: + const float2 sample0curr_texel_offset = float2(0.0, 0.0) + float2(texel0to1ratio_nearest, texel0to1ratio_nearest); + const float2 sample0adjx_texel_offset = float2(-1.0, 0.0) + float2(-texel1to2ratio_nearest, texel0to1ratio_nearest); + const float2 sample0adjy_texel_offset = float2(0.0, -1.0) + float2(texel0to1ratio_nearest, -texel1to2ratio_nearest); + const float2 sample0diag_texel_offset = float2(-1.0, -1.0) + float2(-texel1to2ratio_nearest, -texel1to2ratio_nearest); + const float2 sample1_texel_offset = float2(2.0, 0.0) + float2(texel2to3ratio, texel0to1ratio); + const float2 sample2_texel_offset = float2(4.0, 0.0) + float2(texel4to5ratio, texel0to1ratio); + const float2 sample3_texel_offset = float2(0.0, 2.0) + float2(texel0to1ratio, texel2to3ratio); + const float2 sample4_texel_offset = float2(2.0, 2.0) + float2(texel2to3ratio, texel2to3ratio); + const float2 sample5_texel_offset = float2(4.0, 2.0) + float2(texel4to5ratio, texel2to3ratio); + const float2 sample6_texel_offset = float2(0.0, 4.0) + float2(texel0to1ratio, texel4to5ratio); + const float2 sample7_texel_offset = float2(2.0, 4.0) + float2(texel2to3ratio, texel4to5ratio); + const float2 sample8_texel_offset = float2(4.0, 4.0) + float2(texel4to5ratio, texel4to5ratio); + + // CALCULATE KERNEL WEIGHTS: + // Statically compute bilinear sample weights at each destination fragment + // from the sum of their 4 texel weights (details in tex2Dblur12x12shared). + #define GET_TEXEL_QUAD_WEIGHTS(xoff, yoff) \ + (exp(-LENGTH_SQ(float2(xoff, yoff)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff + 1.0, yoff)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff, yoff + 1.0)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff + 1.0, yoff + 1.0)) * denom_inv)) + // We only need 25 of the 36 sample weights. Skip the following weights: + // 8adjx, 2adjx, 5adjx, + // 6adjy, 7adjy, 8adjy, + // 2diag, 5diag, 6diag, 7diag, 8diag + const float w4diag = GET_TEXEL_QUAD_WEIGHTS(-4.0, -4.0); + const float w3diag = GET_TEXEL_QUAD_WEIGHTS(-2.0, -4.0); + const float w3adjy = GET_TEXEL_QUAD_WEIGHTS(0.0, -4.0); + const float w4adjy = GET_TEXEL_QUAD_WEIGHTS(2.0, -4.0); + const float w5adjy = GET_TEXEL_QUAD_WEIGHTS(4.0, -4.0); + const float w1diag = GET_TEXEL_QUAD_WEIGHTS(-4.0, -2.0); + const float w0diag = GET_TEXEL_QUAD_WEIGHTS(-2.0, -2.0); + const float w0adjy = GET_TEXEL_QUAD_WEIGHTS(0.0, -2.0); + const float w1adjy = GET_TEXEL_QUAD_WEIGHTS(2.0, -2.0); + const float w2adjy = GET_TEXEL_QUAD_WEIGHTS(4.0, -2.0); + const float w1adjx = GET_TEXEL_QUAD_WEIGHTS(-4.0, 0.0); + const float w0adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 0.0); + const float w0curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 0.0); + const float w1curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 0.0); + const float w2curr = GET_TEXEL_QUAD_WEIGHTS(4.0, 0.0); + const float w4adjx = GET_TEXEL_QUAD_WEIGHTS(-4.0, 2.0); + const float w3adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 2.0); + const float w3curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 2.0); + const float w4curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 2.0); + const float w5curr = GET_TEXEL_QUAD_WEIGHTS(4.0, 2.0); + const float w7adjx = GET_TEXEL_QUAD_WEIGHTS(-4.0, 4.0); + const float w6adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 4.0); + const float w6curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 4.0); + const float w7curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 4.0); + const float w8curr = GET_TEXEL_QUAD_WEIGHTS(4.0, 4.0); + #undef GET_TEXEL_QUAD_WEIGHTS + // Get the weight sum inverse (normalization factor): + const float weight_sum_inv = 1.0/(w0curr + w1curr + w2curr + w3curr + + w4curr + w5curr + w6curr + w7curr + w8curr + + w0adjx + w1adjx + w3adjx + w4adjx + w6adjx + w7adjx + + w0adjy + w1adjy + w2adjy + w3adjy + w4adjy + w5adjy + + w0diag + w1diag + w3diag + w4diag); + // Statically pack most weights for runtime. Note the mixed packing: + const float4 w0 = float4(w0curr, w0adjx, w0adjy, w0diag); + const float4 w1 = float4(w1curr, w1adjx, w1adjy, w1diag); + const float4 w3 = float4(w3curr, w3adjx, w3adjy, w3diag); + const float4 w4 = float4(w4curr, w4adjx, w4adjy, w4diag); + const float4 w2and5 = float4(w2curr, w2adjy, w5curr, w5adjy); + const float4 w6and7 = float4(w6curr, w6adjx, w7curr, w7adjx); + + // LOAD TEXTURE SAMPLES THIS FRAGMENT IS RESPONSIBLE FOR: + // Get a uv vector from texel 0q0 of this quadrant to texel 0q3: + const float2 dxdy_curr = dxdy * quad_vector.xy; + // Load bilinear samples for the current quadrant (for this fragment): + const float3 sample0curr = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0curr_texel_offset, input_gamma).rgb; + const float3 sample0adjx = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0adjx_texel_offset, input_gamma).rgb; + const float3 sample0adjy = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0adjy_texel_offset, input_gamma).rgb; + const float3 sample0diag = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0diag_texel_offset, input_gamma).rgb; + const float3 sample1curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample1_texel_offset), input_gamma).rgb; + const float3 sample2curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample2_texel_offset), input_gamma).rgb; + const float3 sample3curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample3_texel_offset), input_gamma).rgb; + const float3 sample4curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample4_texel_offset), input_gamma).rgb; + const float3 sample5curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample5_texel_offset), input_gamma).rgb; + const float3 sample6curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample6_texel_offset), input_gamma).rgb; + const float3 sample7curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample7_texel_offset), input_gamma).rgb; + const float3 sample8curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample8_texel_offset), input_gamma).rgb; + + // GATHER NEIGHBORING SAMPLES AND SUM WEIGHTED SAMPLES: + // Fetch the samples from other fragments in the 2x2 quad in order of need: + float3 sample1adjx, sample1adjy, sample1diag; + float3 sample2adjx, sample2adjy, sample2diag; + float3 sample3adjx, sample3adjy, sample3diag; + float3 sample4adjx, sample4adjy, sample4diag; + float3 sample5adjx, sample5adjy, sample5diag; + float3 sample6adjx, sample6adjy, sample6diag; + float3 sample7adjx, sample7adjy, sample7diag; + quad_gather(quad_vector, sample1curr, sample1adjx, sample1adjy, sample1diag); + quad_gather(quad_vector, sample2curr, sample2adjx, sample2adjy, sample2diag); + quad_gather(quad_vector, sample3curr, sample3adjx, sample3adjy, sample3diag); + quad_gather(quad_vector, sample4curr, sample4adjx, sample4adjy, sample4diag); + quad_gather(quad_vector, sample5curr, sample5adjx, sample5adjy, sample5diag); + quad_gather(quad_vector, sample6curr, sample6adjx, sample6adjy, sample6diag); + quad_gather(quad_vector, sample7curr, sample7adjx, sample7adjy, sample7diag); + // Statically normalize weights (so total = 1.0), and sum weighted samples. + // Fill each row of a matrix with an rgb sample and pre-multiply by the + // weights to obtain a weighted result. First do the simple ones: + float3 sum = float3(0.0,0.0,0.0); + sum += mul(w0, float4x3(sample0curr, sample0adjx, sample0adjy, sample0diag)); + sum += mul(w1, float4x3(sample1curr, sample1adjx, sample1adjy, sample1diag)); + sum += mul(w3, float4x3(sample3curr, sample3adjx, sample3adjy, sample3diag)); + sum += mul(w4, float4x3(sample4curr, sample4adjx, sample4adjy, sample4diag)); + // Now do the mixed-sample ones: + sum += mul(w2and5, float4x3(sample2curr, sample2adjy, sample5curr, sample5adjy)); + sum += mul(w6and7, float4x3(sample6curr, sample6adjx, sample7curr, sample7adjx)); + sum += w8curr * sample8curr; + // Normalize the sum (so the weights add to 1.0) and return: + return sum * weight_sum_inv; +} + +float3 tex2Dblur8x8shared(const sampler2D tex, + const float4 tex_uv, const float2 dxdy, const float4 quad_vector, + const float sigma, + const float input_gamma) +{ + // Perform a 1-pass mipmapped blur with shared samples across a pixel quad. + // Requires: Same as tex2Dblur12x12shared() + // Returns: A blurred texture lookup using a "virtual" 8x8 Gaussian + // blur (a 4x4 blur of carefully selected bilinear samples) + // of the given mip level. There will be subtle inaccuracies, + // especially for small or high-frequency detailed sources. + // Description: + // First see the description for tex2Dblur12x12shared(). This function + // shares the same concept and a similar sample placement, except each + // quadrant contains 4x4 texels and 2x2 samples instead of 6x6 and 3x3 + // respectively. There could be a total of 16 samples, 4 of which each + // fragment is responsible for, but each fragment loads 0a/0b/0c/0d with + // its own offset to reduce shared sample artifacts, bringing the sample + // count for each fragment to 7. Sample placement: + // 3a 2a 2b 3b + // 1a 0a 0b 1b + // 1c 0c 0d 1d + // 3c 2c 2d 3d + // Texel placement: + // 3a3 3a2 2a3 2a2 2b2 2b3 3b2 3b3 + // 3a1 3a0 2a1 2a0 2b0 2b1 3b0 3b1 + // 1a3 1a2 0a3 0a2 0b2 0b3 1b2 1b3 + // 1a1 1a0 0a1 0a0 0b0 0b1 1b0 1b1 + // 1c1 1c0 0c1 0c0 0d0 0d1 1d0 1d1 + // 1c3 1c2 0c3 0c2 0d2 0d3 1d2 1d3 + // 3c1 3c0 2c1 2c0 2d0 2d1 3d0 4d1 + // 3c3 3c2 2c3 2c2 2d2 2d3 3d2 4d3 + + // COMPUTE COORDS FOR TEXTURE SAMPLES THIS FRAGMENT IS RESPONSIBLE FOR: + // Statically compute bilinear sampling offsets (details in tex2Dblur12x12shared). + const float denom_inv = 0.5/(sigma*sigma); + const float w0off = 1.0; + const float w0_5off = exp(-(0.5*0.5) * denom_inv); + const float w1off = exp(-(1.0*1.0) * denom_inv); + const float w1_5off = exp(-(1.5*1.5) * denom_inv); + const float w2off = exp(-(2.0*2.0) * denom_inv); + const float w2_5off = exp(-(2.5*2.5) * denom_inv); + const float w3_5off = exp(-(3.5*3.5) * denom_inv); + const float texel0to1ratio = lerp(w1_5off/(w0_5off + w1_5off), 0.5, error_blurring); + const float texel2to3ratio = lerp(w3_5off/(w2_5off + w3_5off), 0.5, error_blurring); + // We don't share sample0*, so use the nearest destination fragment: + const float texel0to1ratio_nearest = w1off/(w0off + w1off); + const float texel1to2ratio_nearest = w2off/(w1off + w2off); + // Statically compute texel offsets from the bottom-right fragment to each + // bilinear sample in the bottom-right quadrant: + const float2 sample0curr_texel_offset = float2(0.0, 0.0) + float2(texel0to1ratio_nearest, texel0to1ratio_nearest); + const float2 sample0adjx_texel_offset = float2(-1.0, 0.0) + float2(-texel1to2ratio_nearest, texel0to1ratio_nearest); + const float2 sample0adjy_texel_offset = float2(0.0, -1.0) + float2(texel0to1ratio_nearest, -texel1to2ratio_nearest); + const float2 sample0diag_texel_offset = float2(-1.0, -1.0) + float2(-texel1to2ratio_nearest, -texel1to2ratio_nearest); + const float2 sample1_texel_offset = float2(2.0, 0.0) + float2(texel2to3ratio, texel0to1ratio); + const float2 sample2_texel_offset = float2(0.0, 2.0) + float2(texel0to1ratio, texel2to3ratio); + const float2 sample3_texel_offset = float2(2.0, 2.0) + float2(texel2to3ratio, texel2to3ratio); + + // CALCULATE KERNEL WEIGHTS: + // Statically compute bilinear sample weights at each destination fragment + // from the sum of their 4 texel weights (details in tex2Dblur12x12shared). + #define GET_TEXEL_QUAD_WEIGHTS(xoff, yoff) \ + (exp(-LENGTH_SQ(float2(xoff, yoff)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff + 1.0, yoff)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff, yoff + 1.0)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff + 1.0, yoff + 1.0)) * denom_inv)) + const float w3diag = GET_TEXEL_QUAD_WEIGHTS(-4.0, -4.0); + const float w2diag = GET_TEXEL_QUAD_WEIGHTS(-2.0, -4.0); + const float w2adjy = GET_TEXEL_QUAD_WEIGHTS(0.0, -4.0); + const float w3adjy = GET_TEXEL_QUAD_WEIGHTS(2.0, -4.0); + const float w1diag = GET_TEXEL_QUAD_WEIGHTS(-4.0, -2.0); + const float w0diag = GET_TEXEL_QUAD_WEIGHTS(-2.0, -2.0); + const float w0adjy = GET_TEXEL_QUAD_WEIGHTS(0.0, -2.0); + const float w1adjy = GET_TEXEL_QUAD_WEIGHTS(2.0, -2.0); + const float w1adjx = GET_TEXEL_QUAD_WEIGHTS(-4.0, 0.0); + const float w0adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 0.0); + const float w0curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 0.0); + const float w1curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 0.0); + const float w3adjx = GET_TEXEL_QUAD_WEIGHTS(-4.0, 2.0); + const float w2adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 2.0); + const float w2curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 2.0); + const float w3curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 2.0); + #undef GET_TEXEL_QUAD_WEIGHTS + // Statically pack weights for runtime: + const float4 w0 = float4(w0curr, w0adjx, w0adjy, w0diag); + const float4 w1 = float4(w1curr, w1adjx, w1adjy, w1diag); + const float4 w2 = float4(w2curr, w2adjx, w2adjy, w2diag); + const float4 w3 = float4(w3curr, w3adjx, w3adjy, w3diag); + // Get the weight sum inverse (normalization factor): + const float4 weight_sum4 = w0 + w1 + w2 + w3; + const float2 weight_sum2 = weight_sum4.xy + weight_sum4.zw; + const float weight_sum = weight_sum2.x + weight_sum2.y; + const float weight_sum_inv = 1.0/(weight_sum); + + // LOAD TEXTURE SAMPLES THIS FRAGMENT IS RESPONSIBLE FOR: + // Get a uv vector from texel 0q0 of this quadrant to texel 0q3: + const float2 dxdy_curr = dxdy * quad_vector.xy; + // Load bilinear samples for the current quadrant (for this fragment): + const float3 sample0curr = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0curr_texel_offset, input_gamma).rgb; + const float3 sample0adjx = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0adjx_texel_offset, input_gamma).rgb; + const float3 sample0adjy = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0adjy_texel_offset, input_gamma).rgb; + const float3 sample0diag = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0diag_texel_offset, input_gamma).rgb; + const float3 sample1curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample1_texel_offset), input_gamma).rgb; + const float3 sample2curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample2_texel_offset), input_gamma).rgb; + const float3 sample3curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample3_texel_offset), input_gamma).rgb; + + // GATHER NEIGHBORING SAMPLES AND SUM WEIGHTED SAMPLES: + // Fetch the samples from other fragments in the 2x2 quad: + float3 sample1adjx, sample1adjy, sample1diag; + float3 sample2adjx, sample2adjy, sample2diag; + float3 sample3adjx, sample3adjy, sample3diag; + quad_gather(quad_vector, sample1curr, sample1adjx, sample1adjy, sample1diag); + quad_gather(quad_vector, sample2curr, sample2adjx, sample2adjy, sample2diag); + quad_gather(quad_vector, sample3curr, sample3adjx, sample3adjy, sample3diag); + // Statically normalize weights (so total = 1.0), and sum weighted samples. + // Fill each row of a matrix with an rgb sample and pre-multiply by the + // weights to obtain a weighted result: + float3 sum = float3(0.0,0.0,0.0); + sum += mul(w0, float4x3(sample0curr, sample0adjx, sample0adjy, sample0diag)); + sum += mul(w1, float4x3(sample1curr, sample1adjx, sample1adjy, sample1diag)); + sum += mul(w2, float4x3(sample2curr, sample2adjx, sample2adjy, sample2diag)); + sum += mul(w3, float4x3(sample3curr, sample3adjx, sample3adjy, sample3diag)); + return sum * weight_sum_inv; +} + +float3 tex2Dblur6x6shared(const sampler2D tex, + const float4 tex_uv, const float2 dxdy, const float4 quad_vector, + const float sigma, + const float input_gamma) +{ + // Perform a 1-pass mipmapped blur with shared samples across a pixel quad. + // Requires: Same as tex2Dblur12x12shared() + // Returns: A blurred texture lookup using a "virtual" 6x6 Gaussian + // blur (a 3x3 blur of carefully selected bilinear samples) + // of the given mip level. There will be some inaccuracies,subtle inaccuracies, + // especially for small or high-frequency detailed sources. + // Description: + // First see the description for tex2Dblur8x8shared(). This + // function shares the same concept and sample placement, but each fragment + // only uses 9 of the 16 samples taken across the pixel quad (to cover a + // 3x3 sample area, or 6x6 texel area), and it uses a lower standard + // deviation to compensate. Thanks to symmetry, the 7 omitted samples + // are always the "same:" + // 1adjx, 3adjx + // 2adjy, 3adjy + // 1diag, 2diag, 3diag + + // COMPUTE COORDS FOR TEXTURE SAMPLES THIS FRAGMENT IS RESPONSIBLE FOR: + // Statically compute bilinear sampling offsets (details in tex2Dblur12x12shared). + const float denom_inv = 0.5/(sigma*sigma); + const float w0off = 1.0; + const float w0_5off = exp(-(0.5*0.5) * denom_inv); + const float w1off = exp(-(1.0*1.0) * denom_inv); + const float w1_5off = exp(-(1.5*1.5) * denom_inv); + const float w2off = exp(-(2.0*2.0) * denom_inv); + const float w2_5off = exp(-(2.5*2.5) * denom_inv); + const float w3_5off = exp(-(3.5*3.5) * denom_inv); + const float texel0to1ratio = lerp(w1_5off/(w0_5off + w1_5off), 0.5, error_blurring); + const float texel2to3ratio = lerp(w3_5off/(w2_5off + w3_5off), 0.5, error_blurring); + // We don't share sample0*, so use the nearest destination fragment: + const float texel0to1ratio_nearest = w1off/(w0off + w1off); + const float texel1to2ratio_nearest = w2off/(w1off + w2off); + // Statically compute texel offsets from the bottom-right fragment to each + // bilinear sample in the bottom-right quadrant: + const float2 sample0curr_texel_offset = float2(0.0, 0.0) + float2(texel0to1ratio_nearest, texel0to1ratio_nearest); + const float2 sample0adjx_texel_offset = float2(-1.0, 0.0) + float2(-texel1to2ratio_nearest, texel0to1ratio_nearest); + const float2 sample0adjy_texel_offset = float2(0.0, -1.0) + float2(texel0to1ratio_nearest, -texel1to2ratio_nearest); + const float2 sample0diag_texel_offset = float2(-1.0, -1.0) + float2(-texel1to2ratio_nearest, -texel1to2ratio_nearest); + const float2 sample1_texel_offset = float2(2.0, 0.0) + float2(texel2to3ratio, texel0to1ratio); + const float2 sample2_texel_offset = float2(0.0, 2.0) + float2(texel0to1ratio, texel2to3ratio); + const float2 sample3_texel_offset = float2(2.0, 2.0) + float2(texel2to3ratio, texel2to3ratio); + + // CALCULATE KERNEL WEIGHTS: + // Statically compute bilinear sample weights at each destination fragment + // from the sum of their 4 texel weights (details in tex2Dblur12x12shared). + #define GET_TEXEL_QUAD_WEIGHTS(xoff, yoff) \ + (exp(-LENGTH_SQ(float2(xoff, yoff)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff + 1.0, yoff)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff, yoff + 1.0)) * denom_inv) + \ + exp(-LENGTH_SQ(float2(xoff + 1.0, yoff + 1.0)) * denom_inv)) + // We only need 9 of the 16 sample weights. Skip the following weights: + // 1adjx, 3adjx + // 2adjy, 3adjy + // 1diag, 2diag, 3diag + const float w0diag = GET_TEXEL_QUAD_WEIGHTS(-2.0, -2.0); + const float w0adjy = GET_TEXEL_QUAD_WEIGHTS(0.0, -2.0); + const float w1adjy = GET_TEXEL_QUAD_WEIGHTS(2.0, -2.0); + const float w0adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 0.0); + const float w0curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 0.0); + const float w1curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 0.0); + const float w2adjx = GET_TEXEL_QUAD_WEIGHTS(-2.0, 2.0); + const float w2curr = GET_TEXEL_QUAD_WEIGHTS(0.0, 2.0); + const float w3curr = GET_TEXEL_QUAD_WEIGHTS(2.0, 2.0); + #undef GET_TEXEL_QUAD_WEIGHTS + // Get the weight sum inverse (normalization factor): + const float weight_sum_inv = 1.0/(w0curr + w1curr + w2curr + w3curr + + w0adjx + w2adjx + w0adjy + w1adjy + w0diag); + // Statically pack some weights for runtime: + const float4 w0 = float4(w0curr, w0adjx, w0adjy, w0diag); + + // LOAD TEXTURE SAMPLES THIS FRAGMENT IS RESPONSIBLE FOR: + // Get a uv vector from texel 0q0 of this quadrant to texel 0q3: + const float2 dxdy_curr = dxdy * quad_vector.xy; + // Load bilinear samples for the current quadrant (for this fragment): + const float3 sample0curr = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0curr_texel_offset, input_gamma).rgb; + const float3 sample0adjx = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0adjx_texel_offset, input_gamma).rgb; + const float3 sample0adjy = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0adjy_texel_offset, input_gamma).rgb; + const float3 sample0diag = tex2D_linearize(tex, tex_uv.xy + dxdy_curr * sample0diag_texel_offset, input_gamma).rgb; + const float3 sample1curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample1_texel_offset), input_gamma).rgb; + const float3 sample2curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample2_texel_offset), input_gamma).rgb; + const float3 sample3curr = tex2Dlod_linearize(tex, tex_uv + uv2_to_uv4(dxdy_curr * sample3_texel_offset), input_gamma).rgb; + + // GATHER NEIGHBORING SAMPLES AND SUM WEIGHTED SAMPLES: + // Fetch the samples from other fragments in the 2x2 quad: + float3 sample1adjx, sample1adjy, sample1diag; + float3 sample2adjx, sample2adjy, sample2diag; + quad_gather(quad_vector, sample1curr, sample1adjx, sample1adjy, sample1diag); + quad_gather(quad_vector, sample2curr, sample2adjx, sample2adjy, sample2diag); + // Statically normalize weights (so total = 1.0), and sum weighted samples. + // Fill each row of a matrix with an rgb sample and pre-multiply by the + // weights to obtain a weighted result for sample1*, and handle the rest + // of the weights more directly/verbosely: + float3 sum = float3(0.0,0.0,0.0); + sum += mul(w0, float4x3(sample0curr, sample0adjx, sample0adjy, sample0diag)); + sum += w1curr * sample1curr + w1adjy * sample1adjy + w2curr * sample2curr + + w2adjx * sample2adjx + w3curr * sample3curr; + return sum * weight_sum_inv; +} + + +/////////////////////// MAX OPTIMAL SIGMA BLUR WRAPPERS ////////////////////// + +// The following blurs are static wrappers around the dynamic blurs above. +// HOPEFULLY, the compiler will be smart enough to do constant-folding. + +// Resizable separable blurs: +float3 tex2Dblur11resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur11resize(tex, tex_uv, dxdy, blur11_std_dev, input_gamma); +} +float3 tex2Dblur9resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur9resize(tex, tex_uv, dxdy, blur9_std_dev, input_gamma); +} +float3 tex2Dblur7resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur7resize(tex, tex_uv, dxdy, blur7_std_dev, input_gamma); +} +float3 tex2Dblur5resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur5resize(tex, tex_uv, dxdy, blur5_std_dev, input_gamma); +} +float3 tex2Dblur3resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur3resize(tex, tex_uv, dxdy, blur3_std_dev, input_gamma); +} +// Fast separable blurs: +float3 tex2Dblur11fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur11fast(tex, tex_uv, dxdy, blur11_std_dev, input_gamma); +} +float3 tex2Dblur9fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur9fast(tex, tex_uv, dxdy, blur9_std_dev, input_gamma); +} +float3 tex2Dblur7fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur7fast(tex, tex_uv, dxdy, blur7_std_dev, input_gamma); +} +float3 tex2Dblur5fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur5fast(tex, tex_uv, dxdy, blur5_std_dev, input_gamma); +} +float3 tex2Dblur3fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur3fast(tex, tex_uv, dxdy, blur3_std_dev, input_gamma); +} +// Huge, "fast" separable blurs: +float3 tex2Dblur43fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur43fast(tex, tex_uv, dxdy, blur43_std_dev, input_gamma); +} +float3 tex2Dblur31fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur31fast(tex, tex_uv, dxdy, blur31_std_dev, input_gamma); +} +float3 tex2Dblur25fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur25fast(tex, tex_uv, dxdy, blur25_std_dev, input_gamma); +} +float3 tex2Dblur17fast(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur17fast(tex, tex_uv, dxdy, blur17_std_dev, input_gamma); +} +// Resizable one-pass blurs: +float3 tex2Dblur3x3resize(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur3x3resize(tex, tex_uv, dxdy, blur3_std_dev, input_gamma); +} +// "Fast" one-pass blurs: +float3 tex2Dblur9x9(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur9x9(tex, tex_uv, dxdy, blur9_std_dev, input_gamma); +} +float3 tex2Dblur7x7(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur7x7(tex, tex_uv, dxdy, blur7_std_dev, input_gamma); +} +float3 tex2Dblur5x5(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur5x5(tex, tex_uv, dxdy, blur5_std_dev, input_gamma); +} +float3 tex2Dblur3x3(const sampler2D tex, const float2 tex_uv, + const float2 dxdy, + const float input_gamma) +{ + return tex2Dblur3x3(tex, tex_uv, dxdy, blur3_std_dev, input_gamma); +} +// "Fast" shared-sample one-pass blurs: +float3 tex2Dblur12x12shared(const sampler2D tex, + const float4 tex_uv, const float2 dxdy, const float4 quad_vector, + const float input_gamma) +{ + return tex2Dblur12x12shared(tex, tex_uv, dxdy, quad_vector, blur12_std_dev, input_gamma); +} +float3 tex2Dblur10x10shared(const sampler2D tex, + const float4 tex_uv, const float2 dxdy, const float4 quad_vector, + const float input_gamma) +{ + return tex2Dblur10x10shared(tex, tex_uv, dxdy, quad_vector, blur10_std_dev, input_gamma); +} +float3 tex2Dblur8x8shared(const sampler2D tex, + const float4 tex_uv, const float2 dxdy, const float4 quad_vector, + const float input_gamma) +{ + return tex2Dblur8x8shared(tex, tex_uv, dxdy, quad_vector, blur8_std_dev, input_gamma); +} +float3 tex2Dblur6x6shared(const sampler2D tex, + const float4 tex_uv, const float2 dxdy, const float4 quad_vector, + const float input_gamma) +{ + return tex2Dblur6x6shared(tex, tex_uv, dxdy, quad_vector, blur6_std_dev, input_gamma); +} + + +#endif // _BLUR_FUNCTIONS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/derived-settings-and-constants.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/derived-settings-and-constants.fxh new file mode 100644 index 000000000..8443c321c --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/derived-settings-and-constants.fxh @@ -0,0 +1,405 @@ +#ifndef _DERIVED_SETTINGS_AND_CONSTANTS_H +#define _DERIVED_SETTINGS_AND_CONSTANTS_H + +#include "helper-functions-and-macros.fxh" +#include "user-settings.fxh" + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +///////////////////////////////// DESCRIPTION //////////////////////////////// + +// These macros and constants can be used across the whole codebase. +// Unlike the values in user-settings.cgh, end users shouldn't modify these. + + +/////////////////////////////// BEGIN INCLUDES /////////////////////////////// + +//#include "../user-settings.h" + +//#include "user-cgp-constants.h" + +///////////////////////// BEGIN USER-CGP-CONSTANTS ///////////////////////// + +#ifndef _USER_CGP_CONSTANTS_H +#define _USER_CGP_CONSTANTS_H + +// IMPORTANT: +// These constants MUST be set appropriately for the settings in crt-royale.cgp +// (or whatever related .cgp file you're using). If they aren't, you're likely +// to get artifacts, the wrong phosphor mask size, etc. I wish these could be +// set directly in the .cgp file to make things easier, but...they can't. + +// PASS SCALES AND RELATED CONSTANTS: +// Copy the absolute scale_x for BLOOM_APPROX. There are two major versions of +// this shader: One does a viewport-scale bloom, and the other skips it. The +// latter benefits from a higher bloom_approx_scale_x, so save both separately: +static const float bloom_approx_scale_x = 4.0 / 3.0; +static const float max_viewport_size_x = 1080.0*1024.0*(4.0/3.0); +static const float bloom_diff_thresh_ = 1.0/256.0; + +static const float bloom_approx_size_x = 320.0; +static const float bloom_approx_size_x_for_fake = 400.0; +// Copy the viewport-relative scales of the phosphor mask resize passes +// (MASK_RESIZE and the pass immediately preceding it): +static const float2 mask_resize_viewport_scale = float2(0.0625, 0.0625); +// Copy the geom_max_aspect_ratio used to calculate the MASK_RESIZE scales, etc.: +static const float geom_max_aspect_ratio = 4.0/3.0; + +// PHOSPHOR MASK TEXTURE CONSTANTS: +// Set the following constants to reflect the properties of the phosphor mask +// texture named in crt-royale.cgp. The shader optionally resizes a mask tile +// based on user settings, then repeats a single tile until filling the screen. +// The shader must know the input texture size (default 64x64), and to manually +// resize, it must also know the horizontal triads per tile (default 8). +static const float2 mask_texture_small_size = float2(64.0, 64.0); +static const float2 mask_texture_large_size = float2(512.0, 512.0); +static const float mask_triads_per_tile = 8.0; +// We need the average brightness of the phosphor mask to compensate for the +// dimming it causes. The following four values are roughly correct for the +// masks included with the shader. Update the value for any LUT texture you +// change. [Un]comment "#define PHOSPHOR_MASK_GRILLE14" depending on whether +// the loaded aperture grille uses 14-pixel or 15-pixel stripes (default 15). +// #ifndef PHOSPHOR_MASK_GRILLE14 +// #define PHOSPHOR_MASK_GRILLE14 0 +// #endif +static const float mask_grille14_avg_color = 50.6666666/255.0; + // TileableLinearApertureGrille14Wide7d33Spacing*.png + // TileableLinearApertureGrille14Wide10And6Spacing*.png +static const float mask_grille15_avg_color = 53.0/255.0; + // TileableLinearApertureGrille15Wide6d33Spacing*.png + // TileableLinearApertureGrille15Wide8And5d5Spacing*.png +static const float mask_slot_avg_color = 46.0/255.0; + // TileableLinearSlotMask15Wide9And4d5Horizontal8VerticalSpacing*.png + // TileableLinearSlotMaskTall15Wide9And4d5Horizontal9d14VerticalSpacing*.png +static const float mask_shadow_avg_color = 41.0/255.0; + // TileableLinearShadowMask*.png + // TileableLinearShadowMaskEDP*.png + +// #if PHOSPHOR_MASK_GRILLE14 +// static const float mask_grille_avg_color = mask_grille14_avg_color; +// #else + static const float mask_grille_avg_color = mask_grille15_avg_color; +// #endif + + +#endif // _USER_CGP_CONSTANTS_H + +////////////////////////// END USER-CGP-CONSTANTS ////////////////////////// + +//////////////////////////////// END INCLUDES //////////////////////////////// + +/////////////////////////////// FIXED SETTINGS /////////////////////////////// + + + +#define _SIMULATE_CRT_ON_LCD 1 +#define _SIMULATE_GBA_ON_LCD 2 +#define _SIMULATE_LCD_ON_CRT 3 +#define _SIMULATE_GBA_ON_CRT 4 + +// Ensure the first pass decodes CRT gamma and the last encodes LCD gamma. +#define GAMMA_SIMULATION_MODE _SIMULATE_CRT_ON_LCD + +// Manually tiling a manually resized texture creates texture coord derivative +// discontinuities and confuses anisotropic filtering, causing discolored tile +// seams in the phosphor mask. Workarounds: +// a.) Using tex2Dlod disables anisotropic filtering for tiled masks. It's +// downgraded to tex2Dbias without _DRIVERS_ALLOW_TEX2DLOD #defined and +// disabled without _DRIVERS_ALLOW_TEX2DBIAS #defined either. +// b.) "Tile flat twice" requires drawing two full tiles without border padding +// to the resized mask FBO, and it's incompatible with same-pass curvature. +// (Same-pass curvature isn't used but could be in the future...maybe.) +// c.) "Fix discontinuities" requires derivatives and drawing one tile with +// border padding to the resized mask FBO, but it works with same-pass +// curvature. It's disabled without _DRIVERS_ALLOW_DERIVATIVES #defined. +// Precedence: a, then, b, then c (if multiple strategies are #defined). +// #ifndef ANISOTROPIC_TILING_COMPAT_TEX2DLOD +// #define ANISOTROPIC_TILING_COMPAT_TEX2DLOD 1 // 129.7 FPS, 4x, flat; 101.8 at fullscreen +// #endif +// #ifndef ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE +// #define ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE 1 // 128.1 FPS, 4x, flat; 101.5 at fullscreen +// #endif +// #ifndef ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES +// #define ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES 1 // 124.4 FPS, 4x, flat; 97.4 at fullscreen +// #endif +// Also, manually resampling the phosphor mask is slightly blurrier with +// anisotropic filtering. (Resampling with mipmapping is even worse: It +// creates artifacts, but only with the fully bloomed shader.) The difference +// is subtle with small triads, but you can fix it for a small cost. +// #ifndef ANISOTROPIC_RESAMPLING_COMPAT_TEX2DLOD +// #define ANISOTROPIC_RESAMPLING_COMPAT_TEX2DLOD 0 +// #endif + + +////////////////////////////// DERIVED SETTINGS ////////////////////////////// + +// Intel HD 4000 GPU's can't handle manual mask resizing (for now), setting the +// geometry mode at runtime, or a 4x4 true Gaussian resize. Disable +// incompatible settings ASAP. (_INTEGRATED_GRAPHICS_COMPATIBILITY_MODE may be +// #defined by either user-settings.h or a wrapper .cg that #includes the +// current .cg pass.) +#if _INTEGRATED_GRAPHICS_COMPATIBILITY_MODE + #if _PHOSPHOR_MASK_MANUALLY_RESIZE + #undef _PHOSPHOR_MASK_MANUALLY_RESIZE + #define _PHOSPHOR_MASK_MANUALLY_RESIZE 0 + #endif + #if _RUNTIME_GEOMETRY_MODE + #undef _RUNTIME_GEOMETRY_MODE + #define _RUNTIME_GEOMETRY_MODE 0 + #endif + // Mode 2 (4x4 Gaussian resize) won't work, and mode 1 (3x3 blur) is + // inferior in most cases, so replace 2.0 with 0.0: + static const float bloom_approx_filter = macro_cond( + bloom_approx_filter_static > 1.5, + 0.0, + bloom_approx_filter_static + ); +#else + static const float bloom_approx_filter = bloom_approx_filter_static; +#endif + +// Disable slow runtime paths if static parameters are used. Most of these +// won't be a problem anyway once the params are disabled, but some will. +#if !_RUNTIME_SHADER_PARAMS_ENABLE + #if _RUNTIME_PHOSPHOR_BLOOM_SIGMA + #undef _RUNTIME_PHOSPHOR_BLOOM_SIGMA + #define _RUNTIME_PHOSPHOR_BLOOM_SIGMA 0 + #endif + #if _RUNTIME_ANTIALIAS_WEIGHTS + #undef _RUNTIME_ANTIALIAS_WEIGHTS + #define _RUNTIME_ANTIALIAS_WEIGHTS 0 + #endif + #if _RUNTIME_ANTIALIAS_SUBPIXEL_OFFSETS + #undef _RUNTIME_ANTIALIAS_SUBPIXEL_OFFSETS + #define _RUNTIME_ANTIALIAS_SUBPIXEL_OFFSETS 0 + #endif + #if _RUNTIME_SCANLINES_HORIZ_FILTER_COLORSPACE + #undef _RUNTIME_SCANLINES_HORIZ_FILTER_COLORSPACE + #define _RUNTIME_SCANLINES_HORIZ_FILTER_COLORSPACE 0 + #endif + #if _RUNTIME_GEOMETRY_TILT + #undef _RUNTIME_GEOMETRY_TILT + #define _RUNTIME_GEOMETRY_TILT 0 + #endif + #if _RUNTIME_GEOMETRY_MODE + #undef _RUNTIME_GEOMETRY_MODE + #define _RUNTIME_GEOMETRY_MODE 0 + #endif + // #if FORCE_RUNTIME_PHOSPHOR_MASK_MODE_TYPE_SELECT + // #undef FORCE_RUNTIME_PHOSPHOR_MASK_MODE_TYPE_SELECT + // #define FORCE_RUNTIME_PHOSPHOR_MASK_MODE_TYPE_SELECT 0 + // #endif +#endif + +// Make tex2Dbias a backup for tex2Dlod for wider compatibility. +// #if ANISOTROPIC_TILING_COMPAT_TEX2DLOD +// #define ANISOTROPIC_TILING_COMPAT_TEX2DBIAS +// #endif +// #if ANISOTROPIC_RESAMPLING_COMPAT_TEX2DLOD +// #define ANISOTROPIC_RESAMPLING_COMPAT_TEX2DBIAS +// #endif +// Rule out unavailable anisotropic compatibility strategies: +#if !_DRIVERS_ALLOW_DERIVATIVES + // #if ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES + // #undef ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES + // #define ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES 0 + // #endif +#endif +// #if !_DRIVERS_ALLOW_TEX2DLOD + // #if ANISOTROPIC_TILING_COMPAT_TEX2DLOD + // #undef ANISOTROPIC_TILING_COMPAT_TEX2DLOD + // #define ANISOTROPIC_TILING_COMPAT_TEX2DLOD 0 + // #endif + // #if ANISOTROPIC_RESAMPLING_COMPAT_TEX2DLOD + // #undef ANISOTROPIC_RESAMPLING_COMPAT_TEX2DLOD + // #define ANISOTROPIC_RESAMPLING_COMPAT_TEX2DLOD 0 + // #endif + // #ifdef ANTIALIAS_DISABLE_ANISOTROPIC + // #undef ANTIALIAS_DISABLE_ANISOTROPIC + // #endif +// #endif +// #if !_DRIVERS_ALLOW_TEX2DBIAS + // #ifdef ANISOTROPIC_TILING_COMPAT_TEX2DBIAS + // #undef ANISOTROPIC_TILING_COMPAT_TEX2DBIAS + // #endif + // #ifdef ANISOTROPIC_RESAMPLING_COMPAT_TEX2DBIAS + // #undef ANISOTROPIC_RESAMPLING_COMPAT_TEX2DBIAS + // #endif +// #endif +// Prioritize anisotropic tiling compatibility strategies by performance and +// disable unused strategies. This concentrates all the nesting in one place. +// #if ANISOTROPIC_TILING_COMPAT_TEX2DLOD +// #ifdef ANISOTROPIC_TILING_COMPAT_TEX2DBIAS +// #undef ANISOTROPIC_TILING_COMPAT_TEX2DBIAS +// #endif +// #if ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE +// #undef ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE +// #define ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE 0 +// #endif +// #if ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES +// #undef ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES +// #define ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES 0 +// #endif +// #else +// #ifdef ANISOTROPIC_TILING_COMPAT_TEX2DBIAS +// #if ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE +// #undef ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE +// #define ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE 0 +// #endif +// #if ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES +// #undef ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES +// #define ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES 0 +// #endif +// #else +// // ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE is only compatible with +// // flat texture coords in the same pass, but that's all we use. +// #if ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE +// #if ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES +// #undef ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES +// #define ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES 0 +// #endif +// #endif +// #endif +// #endif +// The tex2Dlod and tex2Dbias strategies share a lot in common, and we can +// reduce some #ifdef nesting in the next section by essentially OR'ing them: +// #if ANISOTROPIC_TILING_COMPAT_TEX2DLOD +// #define ANISOTROPIC_TILING_COMPAT_TEX2DLOD_FAMILY +// #endif +// #ifdef ANISOTROPIC_TILING_COMPAT_TEX2DBIAS +// #define ANISOTROPIC_TILING_COMPAT_TEX2DLOD_FAMILY +// #endif +// Prioritize anisotropic resampling compatibility strategies the same way: +// #if ANISOTROPIC_RESAMPLING_COMPAT_TEX2DLOD +// #ifdef ANISOTROPIC_RESAMPLING_COMPAT_TEX2DBIAS +// #undef ANISOTROPIC_RESAMPLING_COMPAT_TEX2DBIAS +// #endif +// #endif + + +/////////////////////// DERIVED PHOSPHOR MASK CONSTANTS ////////////////////// + +// If we can use the large mipmapped LUT without mipmapping artifacts, we +// should: It gives us more options for using fewer samples. +// #if USE_LARGE_PHOSPHOR_MASK + // #if ANISOTROPIC_RESAMPLING_COMPAT_TEX2DLOD + // // TODO: Take advantage of this! + // #define PHOSPHOR_MASK_RESIZE_MIPMAPPED_LUT + // static const float2 mask_resize_src_lut_size = mask_texture_large_size; + // #else +static const float2 mask_resize_src_lut_size = mask_texture_large_size; + // #endif +// #else +// static const float2 mask_resize_src_lut_size = mask_texture_small_size; +// #endif + +static const float tile_aspect_inv = mask_resize_src_lut_size.y/mask_resize_src_lut_size.x; + + +// tex2D's sampler2D parameter MUST be a uniform global, a uniform input to +// main_fragment, or a static alias of one of the above. This makes it hard +// to select the phosphor mask at runtime: We can't even assign to a uniform +// global in the vertex shader or select a sampler2D in the vertex shader and +// pass it to the fragment shader (even with explicit TEXUNIT# bindings), +// because it just gives us the input texture or a black screen. However, we +// can get around these limitations by calling tex2D three times with different +// uniform samplers (or resizing the phosphor mask three times altogether). +// With dynamic branches, we can process only one of these branches on top of +// quickly discarding fragments we don't need (cgc seems able to overcome +// limigations around dependent texture fetches inside of branches). Without +// dynamic branches, we have to process every branch for every fragment...which +// is slower. Runtime sampling mode selection is slower without dynamic +// branches as well. Let the user's static #defines decide if it's worth it. +#if _DRIVERS_ALLOW_DYNAMIC_BRANCHES + #define _RUNTIME_PHOSPHOR_MASK_MODE_TYPE_SELECT +// #else + // #if FORCE_RUNTIME_PHOSPHOR_MASK_MODE_TYPE_SELECT + // #define _RUNTIME_PHOSPHOR_MASK_MODE_TYPE_SELECT + // #endif +#endif + +// We need to render some minimum number of tiles in the resize passes. +// We need at least 1.0 just to repeat a single tile, and we need extra +// padding beyond that for anisotropic filtering, discontinuitity fixing, +// antialiasing, same-pass curvature (not currently used), etc. First +// determine how many border texels and tiles we need, based on how the result +// will be sampled: +#ifdef GEOMETRY_EARLY + static const float max_subpixel_offset = aa_subpixel_r_offset_static.x; + // Most antialiasing filters have a base radius of 4.0 pixels: + static const float max_aa_base_pixel_border = 4.0 + + max_subpixel_offset; +#else + static const float max_aa_base_pixel_border = 0.0; +#endif +// Anisotropic filtering adds about 0.5 to the pixel border: +// #ifndef ANISOTROPIC_TILING_COMPAT_TEX2DLOD_FAMILY +static const float max_aniso_pixel_border = max_aa_base_pixel_border + 0.5; +// #else +// static const float max_aniso_pixel_border = max_aa_base_pixel_border; +// #endif +// Fixing discontinuities adds 1.0 more to the pixel border: +// #if ANISOTROPIC_TILING_COMPAT_FIX_DISCONTINUITIES +// static const float max_tiled_pixel_border = max_aniso_pixel_border + 1.0; +// #else + static const float max_tiled_pixel_border = max_aniso_pixel_border; +// #endif +// Convert the pixel border to an integer texel border. Assume same-pass +// curvature about triples the texel frequency: +#ifdef GEOMETRY_EARLY + #define max_mask_texel_border macro_ceil(max_tiled_pixel_border * 3.0f) +#else + #define max_mask_texel_border macro_ceil(max_tiled_pixel_border) +#endif +// Convert the texel border to a tile border using worst-case assumptions: +static const float max_mask_tile_border = max_mask_texel_border/ +(mask_min_allowed_triad_size * mask_triads_per_tile); + +// Finally, set the number of resized tiles to render to MASK_RESIZE, and set +// the starting texel (inside borders) for sampling it. +#ifndef GEOMETRY_EARLY + // #if ANISOTROPIC_TILING_COMPAT_TILE_FLAT_TWICE + // Special case: Render two tiles without borders. Anisotropic + // filtering doesn't seem to be a problem here. + // static const float mask_resize_num_tiles = 1.0 + 1.0; + // static const float mask_start_texels = 0.0; + // #else + static const float mask_resize_num_tiles = 1.0 + 2.0 * max_mask_tile_border; + static const float mask_start_texels = max_mask_texel_border; + // #endif +#else + static const float mask_resize_num_tiles = 1.0 + 2.0*max_mask_tile_border; + static const float mask_start_texels = max_mask_texel_border; +#endif + +// We have to fit mask_resize_num_tiles into an FBO with a viewport scale of +// mask_resize_viewport_scale. This limits the maximum final triad size. +// Estimate the minimum number of triads we can split the screen into in each +// dimension (we'll be as correct as mask_resize_viewport_scale is): +static const float mask_resize_num_triads = mask_resize_num_tiles * mask_triads_per_tile; +static const float2 min_allowed_viewport_triads = +float2(mask_resize_num_triads, mask_resize_num_triads) / mask_resize_viewport_scale; + + + +#endif // _DERIVED_SETTINGS_AND_CONSTANTS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/downsampling-functions.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/downsampling-functions.fxh new file mode 100644 index 000000000..7e8404994 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/downsampling-functions.fxh @@ -0,0 +1,84 @@ +#ifndef _DOWNSAMPLING_FUNCTIONS_H +#define _DOWNSAMPLING_FUNCTIONS_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2020 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + +float3 opaque_linear_downsample( + const sampler2D tex, + const float2 texcoord, + const uint num_pairs, + const float2 delta_uv +) { + const uint total_num_samples = num_pairs * 2 + 1; + const float2 coord_left = texcoord - delta_uv * num_pairs; + + float3 acc = 0; + for(int i = 0; i < total_num_samples; i++) { + const float2 coord = coord_left + i * delta_uv; + acc += tex2D_nograd(tex, coord).rgb; + } + + return acc / total_num_samples; +} + + +float3 opaque_lanczos_downsample( + const sampler2D tex, + const float2 texcoord, + const uint num_pairs, + const float2 delta_uv, + const float num_sinc_lobes, + const float weight_at_center +) { + const uint total_num_samples = num_pairs * 2 + 1; + const float2 coord_left = texcoord - delta_uv * num_pairs; + const float sinc_dx = num_sinc_lobes / num_pairs; // 2 * num_sinc_lobes / (total_num_samples - 1) + + float3 acc = 0; + float w_sum = 0; + for(int i = 0; i < total_num_samples; i++) { + const float2 coord = coord_left + i * delta_uv; + const float sinc_x = i * sinc_dx; + + const float weight = (i != num_pairs) ? + num_sinc_lobes * sin(pi*sinc_x) * sin(pi*sinc_x/num_sinc_lobes) / (pi*pi * sinc_x*sinc_x) : + weight_at_center; + + acc += weight * tex2D_nograd(tex, coord).rgb; + w_sum += weight; + } + + return acc / w_sum; +} + +float3 opaque_lanczos_downsample( + const sampler2D tex, + const float2 texcoord, + const uint num_pairs, + const float2 delta_uv, + const float num_sinc_lobes +) { + return opaque_lanczos_downsample(tex, texcoord, num_pairs, delta_uv, num_sinc_lobes, 1); +} + +#endif // _DOWNSAMPLING_FUNCTIONS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/gamma-management.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/gamma-management.fxh new file mode 100644 index 000000000..a0ce35ff0 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/gamma-management.fxh @@ -0,0 +1,225 @@ +#ifndef _GAMMA_MANAGEMENT_H +#define _GAMMA_MANAGEMENT_H + + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2014 TroggleMonkey +// Copyright (C) 2020 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + +#include "helper-functions-and-macros.fxh" + + +/////////////////////////////// BASE CONSTANTS /////////////////////////////// + +// Set standard gamma constants, but allow users to override them: +#ifndef OVERRIDE_STANDARD_GAMMA + // Standard encoding gammas: + static const float ntsc_gamma = 2.2; // Best to use NTSC for PAL too? + static const float pal_gamma = 2.8; // Never actually 2.8 in practice + // Typical device decoding gammas (only use for emulating devices): + // CRT/LCD reference gammas are higher than NTSC and Rec.709 video standard + // gammas: The standards purposely undercorrected for an analog CRT's + // assumed 2.5 reference display gamma to maintain contrast in assumed + // [dark] viewing conditions: http://www.poynton.com/PDFs/GammaFAQ.pdf + // These unstated assumptions about display gamma and perceptual rendering + // intent caused a lot of confusion, and more modern CRT's seemed to target + // NTSC 2.2 gamma with circuitry. LCD displays seem to have followed suit + // (they struggle near black with 2.5 gamma anyway), especially PC/laptop + // displays designed to view sRGB in bright environments. (Standards are + // also in flux again with BT.1886, but it's underspecified for displays.) + static const float crt_reference_gamma_high = 2.5; // In (2.35, 2.55) + static const float crt_reference_gamma_low = 2.35; // In (2.35, 2.55) + static const float lcd_reference_gamma = 2.5; // To match CRT + static const float crt_office_gamma = 2.2; // Circuitry-adjusted for NTSC + static const float lcd_office_gamma = 2.2; // Approximates sRGB +#endif // OVERRIDE_STANDARD_GAMMA + +// Assuming alpha == 1.0 might make it easier for users to avoid some bugs, +// but only if they're aware of it. +#ifndef OVERRIDE_ALPHA_ASSUMPTIONS + static const bool assume_opaque_alpha = false; +#endif + + +/////////////////////// DERIVED CONSTANTS AS FUNCTIONS /////////////////////// + +// gamma-management.h should be compatible with overriding gamma values with +// runtime user parameters, but we can only define other global constants in +// terms of static constants, not uniform user parameters. To get around this +// limitation, we need to define derived constants using functions. + +// Set device gamma constants, but allow users to override them: +#if _OVERRIDE_DEVICE_GAMMA + // The user promises to globally define the appropriate constants: + float get_crt_gamma() { return crt_gamma; } + float get_gba_gamma() { return gba_gamma; } + float get_lcd_gamma() { return lcd_gamma; } +#else + float get_crt_gamma() { return crt_reference_gamma_high; } + float get_gba_gamma() { return 3.5; } // Game Boy Advance; in (3.0, 4.0) + float get_lcd_gamma() { return lcd_office_gamma; } +#endif // _OVERRIDE_DEVICE_GAMMA + +// Set decoding/encoding gammas for the first/lass passes, but allow overrides: +#ifdef OVERRIDE_FINAL_GAMMA + // The user promises to globally define the appropriate constants: + float get_intermediate_gamma() { return intermediate_gamma; } + float get_input_gamma() { return input_gamma; } + float get_output_gamma() { return output_gamma; } +#else + // If we gamma-correct every pass, always use ntsc_gamma between passes to + // ensure middle passes don't need to care if anything is being simulated: + + // TODO: Figure out the correct way to configure this now that intermediate + // FBOs all use get_intermediate_gamma() directly. Also refer to the + // original code to confirm when a shader uses ntsc_gamma despite + // GAMMA_ENCODE_EVERY_FBO being undefined. + // float get_intermediate_gamma() { return ntsc_gamma; } + float get_intermediate_gamma() { return 1.0; } + + #if GAMMA_SIMULATION_MODE == _SIMULATE_CRT_ON_LCD + float get_input_gamma() { return get_crt_gamma(); } + float get_output_gamma() { return get_lcd_gamma(); } + #else + #if GAMMA_SIMULATION_MODE == _SIMULATE_GBA_ON_LCD + float get_input_gamma() { return get_gba_gamma(); } + float get_output_gamma() { return get_lcd_gamma(); } + #else + #if GAMMA_SIMULATION_MODE == _SIMULATE_LCD_ON_CRT + float get_input_gamma() { return get_lcd_gamma(); } + float get_output_gamma() { return get_crt_gamma(); } + #else + #if GAMMA_SIMULATION_MODE == _SIMULATE_GBA_ON_CRT + float get_input_gamma() { return get_gba_gamma(); } + float get_output_gamma() { return get_crt_gamma(); } + #else // Don't simulate anything: + float get_input_gamma() { return ntsc_gamma; } + float get_output_gamma() { return ntsc_gamma; } + #endif // _SIMULATE_GBA_ON_CRT + #endif // _SIMULATE_LCD_ON_CRT + #endif // _SIMULATE_GBA_ON_LCD + #endif // _SIMULATE_CRT_ON_LCD +#endif // OVERRIDE_FINAL_GAMMA + + +// Set decoding/encoding gammas for the current pass. Use static constants for +// linearize_input and gamma_encode_output, because they aren't derived, and +// they let the compiler do dead-code elimination. +// #ifndef GAMMA_ENCODE_EVERY_FBO +// #ifdef FIRST_PASS +// static const bool linearize_input = true; +// float get_pass_input_gamma() { return get_input_gamma(); } +// #else +// static const bool linearize_input = false; +// float get_pass_input_gamma() { return 1.0; } +// #endif +// #ifdef LAST_PASS +// static const bool gamma_encode_output = true; +// float get_pass_output_gamma() { return get_output_gamma(); } +// #else +// static const bool gamma_encode_output = false; +// float get_pass_output_gamma() { return 1.0; } +// #endif +// #else +// static const bool linearize_input = true; +// static const bool gamma_encode_output = true; +// #ifdef FIRST_PASS +// float get_pass_input_gamma() { return get_input_gamma(); } +// #else +// float get_pass_input_gamma() { return get_intermediate_gamma(); } +// #endif +// #ifdef LAST_PASS +// float get_pass_output_gamma() { return get_output_gamma(); } +// #else +// float get_pass_output_gamma() { return get_intermediate_gamma(); } +// #endif +// #endif + +// Users might want to know if bilinear filtering will be gamma-correct: +// static const bool gamma_aware_bilinear = !linearize_input; + + +////////////////////// COLOR ENCODING/DECODING FUNCTIONS ///////////////////// + +float4 encode_output_opaque(const float4 color, const float gamma) +{ + static const float3 g = 1.0 / float3(gamma, gamma, gamma); + return float4(pow(color.rgb, g), 1); +} + +float4 decode_input_opaque(const float4 color, const float gamma) +{ + static const float3 g = float3(gamma, gamma, gamma); + return float4(pow(color.rgb, g), 1); +} + +float4 encode_output(const float4 color, const float gamma) +{ + static const float3 g = 1.0 / float3(gamma, gamma, gamma); + return float4(pow(color.rgb, g), color.a); +} + +float4 decode_input(const float4 color, const float gamma) +{ + static const float3 g = float3(gamma, gamma, gamma); + return float4(pow(color.rgb, g), color.a); +} + +/////////////////////////// TEXTURE LOOKUP WRAPPERS ////////////////////////// + +// "SMART" LINEARIZING TEXTURE LOOKUP FUNCTIONS: +// Provide a wide array of linearizing texture lookup wrapper functions. The +// Cg shader spec Retroarch uses only allows for 2D textures, but 1D and 3D +// lookups are provided for completeness in case that changes someday. Nobody +// is likely to use the *fetch and *proj functions, but they're included just +// in case. The only tex*D texture sampling functions omitted are: +// - tex*Dcmpbias +// - tex*Dcmplod +// - tex*DARRAY* +// - tex*DMS* +// - Variants returning integers +// Standard line length restrictions are ignored below for vertical brevity. + +// tex2D: +float4 tex2D_linearize(const sampler2D tex, const float2 tex_coords, const float gamma) +{ return decode_input(tex2D(tex, tex_coords), gamma); } + +float4 tex2D_linearize(const sampler2D tex, const float3 tex_coords, const float gamma) +{ return decode_input(tex2D(tex, tex_coords.xy), gamma); } + +// float4 tex2D_linearize(const sampler2D tex, const float2 tex_coords, const int texel_off, const float gamma) +// { return decode_input(tex2Dlod(tex, float4(tex_coords.x, tex_coords.y, 0, 0), texel_off), gamma); } + +// float4 tex2D_linearize(const sampler2D tex, const float3 tex_coords, const int texel_off, const float gamma) +// { return decode_input(tex2Dlod(tex, float4(tex_coords.x, tex_coords.y, 0, 0), texel_off), gamma); } + +// tex2Dlod: +float4 tex2Dlod_linearize(const sampler2D tex, const float2 tex_coords, const float gamma) +{ return decode_input(tex2Dlod(tex, float4(tex_coords, 0, 0), 0.0), gamma); } + +float4 tex2Dlod_linearize(const sampler2D tex, const float4 tex_coords, const float gamma) +{ return decode_input(tex2Dlod(tex, float4(tex_coords.xy, 0, 0), 0.0), gamma); } + +// float4 tex2Dlod_linearize(const sampler2D tex, const float4 tex_coords, const int texel_off, const float gamma) +// { return decode_input(tex2Dlod(tex, float4(tex_coords.x, tex_coords.y, 0, 0), texel_off), gamma); } + +#endif // _GAMMA_MANAGEMENT_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/geometry-functions.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/geometry-functions.fxh new file mode 100644 index 000000000..c16d03f9a --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/geometry-functions.fxh @@ -0,0 +1,715 @@ +#ifndef _GEOMETRY_FUNCTIONS_H +#define _GEOMETRY_FUNCTIONS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +////////////////////////////////// INCLUDES ////////////////////////////////// + +#include "user-settings.fxh" +#include "derived-settings-and-constants.fxh" +#include "bind-shader-params.fxh" + + +//////////////////////////// MACROS AND CONSTANTS //////////////////////////// + +// Curvature-related constants: +#define MAX_POINT_CLOUD_SIZE 9 + + +///////////////////////////// CURVATURE FUNCTIONS ///////////////////////////// + +float2 quadratic_solve(const float a, const float b_over_2, const float c) +{ + // Requires: 1.) a, b, and c are quadratic formula coefficients + // 2.) b_over_2 = b/2.0 (simplifies terms to factor 2 out) + // 3.) b_over_2 must be guaranteed < 0.0 (avoids a branch) + // Returns: Returns float2(first_solution, discriminant), so the caller + // can choose how to handle the "no intersection" case. The + // Kahan or Citardauq formula is used for numerical robustness. + const float discriminant = b_over_2*b_over_2 - a*c; + const float solution0 = c/(-b_over_2 + sqrt(discriminant)); + return float2(solution0, discriminant); +} + +float2 intersect_sphere(const float3 view_vec, const float3 eye_pos_vec) +{ + // Requires: 1.) view_vec and eye_pos_vec are 3D vectors in the sphere's + // local coordinate frame (eye_pos_vec is a position, i.e. + // a vector from the origin to the eye/camera) + // 2.) geom_radius is a global containing the sphere's radius + // Returns: Cast a ray of direction view_vec from eye_pos_vec at a + // sphere of radius geom_radius, and return the distance to + // the first intersection in units of length(view_vec). + // http://wiki.cgsociety.org/index.php/Ray_Sphere_Intersection + // Quadratic formula coefficients (b_over_2 is guaranteed negative): + const float a = dot(view_vec, view_vec); + const float b_over_2 = dot(view_vec, eye_pos_vec); // * 2.0 factored out + const float c = dot(eye_pos_vec, eye_pos_vec) - geom_radius*geom_radius; + return quadratic_solve(a, b_over_2, c); +} + +float2 intersect_cylinder(const float3 view_vec, const float3 eye_pos_vec) +{ + // Requires: 1.) view_vec and eye_pos_vec are 3D vectors in the sphere's + // local coordinate frame (eye_pos_vec is a position, i.e. + // a vector from the origin to the eye/camera) + // 2.) geom_radius is a global containing the cylinder's radius + // Returns: Cast a ray of direction view_vec from eye_pos_vec at a + // cylinder of radius geom_radius, and return the distance to + // the first intersection in units of length(view_vec). The + // derivation of the coefficients is in Christer Ericson's + // Real-Time Collision Detection, p. 195-196, and this version + // uses LaGrange's identity to reduce operations. + // Arbitrary "cylinder top" reference point for an infinite cylinder: + const float3 cylinder_top_vec = float3(0.0, geom_radius, 0.0); + const float3 cylinder_axis_vec = float3(0.0, 1.0, 0.0);//float3(0.0, 2.0*geom_radius, 0.0); + const float3 top_to_eye_vec = eye_pos_vec - cylinder_top_vec; + const float3 axis_x_view = cross(cylinder_axis_vec, view_vec); + const float3 axis_x_top_to_eye = cross(cylinder_axis_vec, top_to_eye_vec); + // Quadratic formula coefficients (b_over_2 is guaranteed negative): + const float a = dot(axis_x_view, axis_x_view); + const float b_over_2 = dot(axis_x_top_to_eye, axis_x_view); + const float c = dot(axis_x_top_to_eye, axis_x_top_to_eye) - + geom_radius*geom_radius;//*dot(cylinder_axis_vec, cylinder_axis_vec); + return quadratic_solve(a, b_over_2, c); +} + +float2 cylinder_xyz_to_uv(const float3 intersection_pos_local, + const float2 geom_aspect) +{ + // Requires: An xyz intersection position on a cylinder. + // Returns: video_uv coords mapped to range [-0.5, 0.5] + // Mapping: Define square_uv.x to be the signed arc length in xz-space, + // and define square_uv.y = -intersection_pos_local.y (+v = -y). + // Start with a numerically robust arc length calculation. + const float angle_from_image_center = atan2(intersection_pos_local.x, + intersection_pos_local.z); + const float signed_arc_len = angle_from_image_center * geom_radius; + // Get a uv-mapping where [-0.5, 0.5] maps to a "square" area, then divide + // by the aspect ratio to stretch the mapping appropriately: + const float2 square_uv = float2(signed_arc_len, -intersection_pos_local.y); + const float2 video_uv = square_uv / geom_aspect; + return video_uv; +} + +float3 cylinder_uv_to_xyz(const float2 video_uv, const float2 geom_aspect) +{ + // Requires: video_uv coords mapped to range [-0.5, 0.5] + // Returns: An xyz intersection position on a cylinder. This is the + // inverse of cylinder_xyz_to_uv(). + // Expand video_uv by the aspect ratio to get proportionate x/y lengths, + // then calculate an xyz position for the cylindrical mapping above. + const float2 square_uv = video_uv * geom_aspect; + const float arc_len = square_uv.x; + const float angle_from_image_center = arc_len / geom_radius; + const float x_pos = sin(angle_from_image_center) * geom_radius; + const float z_pos = cos(angle_from_image_center) * geom_radius; + // Or: z = sqrt(geom_radius**2 - x**2) + // Or: z = geom_radius/sqrt(1.0 + tan(angle)**2), x = z * tan(angle) + const float3 intersection_pos_local = float3(x_pos, -square_uv.y, z_pos); + return intersection_pos_local; +} + +float2 sphere_xyz_to_uv(const float3 intersection_pos_local, + const float2 geom_aspect) +{ + // Requires: An xyz intersection position on a sphere. + // Returns: video_uv coords mapped to range [-0.5, 0.5] + // Mapping: First define square_uv.x/square_uv.y == + // intersection_pos_local.x/intersection_pos_local.y. Then, + // length(square_uv) is the arc length from the image center + // at (0.0, 0.0, geom_radius) along the tangent great circle. + // Credit for this mapping goes to cgwg: I never managed to + // understand his code, but he told me his mapping was based on + // great circle distances when I asked him about it, which + // informed this very similar (almost identical) mapping. + // Start with a numerically robust arc length calculation between the ray- + // sphere intersection point and the image center using a method posted by + // Roger Stafford on comp.soft-sys.matlab: + // https://groups.google.com/d/msg/comp.soft-sys.matlab/zNbUui3bjcA/c0HV_bHSx9cJ + const float3 image_center_pos_local = float3(0.0, 0.0, geom_radius); + const float cp_len = + length(cross(intersection_pos_local, image_center_pos_local)); + const float dp = dot(intersection_pos_local, image_center_pos_local); + const float angle_from_image_center = atan2(cp_len, dp); + const float arc_len = angle_from_image_center * geom_radius; + // Get a uv-mapping where [-0.5, 0.5] maps to a "square" area, then divide + // by the aspect ratio to stretch the mapping appropriately: + const float2 square_uv_unit = normalize(float2(intersection_pos_local.x, + -intersection_pos_local.y)); + const float2 square_uv = arc_len * square_uv_unit; + const float2 video_uv = square_uv / geom_aspect; + return video_uv; +} + +float3 sphere_uv_to_xyz(const float2 video_uv, const float2 geom_aspect) +{ + // Requires: video_uv coords mapped to range [-0.5, 0.5] + // Returns: An xyz intersection position on a sphere. This is the + // inverse of sphere_xyz_to_uv(). + // Expand video_uv by the aspect ratio to get proportionate x/y lengths, + // then calculate an xyz position for the spherical mapping above. + if (video_uv.x != 0 && video_uv.y != 0) { + const float2 square_uv = video_uv * geom_aspect; + // Using length or sqrt here butchers the framerate on my 8800GTS if + // this function is called too many times, and so does taking the max + // component of square_uv/square_uv_unit (program length threshold?). + //float arc_len = length(square_uv); + const float2 square_uv_unit = normalize(square_uv); + const float arc_len = square_uv.y/square_uv_unit.y; + const float angle_from_image_center = arc_len / geom_radius; + const float xy_dist_from_sphere_center = + sin(angle_from_image_center) * geom_radius; + //float2 xy_pos = xy_dist_from_sphere_center * (square_uv/FIX_ZERO(arc_len)); + const float2 xy_pos = xy_dist_from_sphere_center * square_uv_unit; + const float z_pos = cos(angle_from_image_center) * geom_radius; + const float3 intersection_pos_local = float3(xy_pos.x, -xy_pos.y, z_pos); + return intersection_pos_local; + } + else if (video_uv.x != 0) { + const float2 square_uv = video_uv * geom_aspect; + // Using length or sqrt here butchers the framerate on my 8800GTS if + // this function is called too many times, and so does taking the max + // component of square_uv/square_uv_unit (program length threshold?). + //float arc_len = length(square_uv); + const float2 square_uv_unit = normalize(square_uv); + const float angle_from_image_center = 0; + const float xy_dist_from_sphere_center = sin(angle_from_image_center) * geom_radius; + const float2 xy_pos = xy_dist_from_sphere_center * square_uv_unit; + const float z_pos = cos(angle_from_image_center) * geom_radius; + const float3 intersection_pos_local = float3(xy_pos.x, -xy_pos.y, z_pos); + return intersection_pos_local; + } + else { + const float2 xy_pos = float2(0, 0); + const float z_pos = geom_radius; + const float3 intersection_pos_local = float3(xy_pos.x, -xy_pos.y, z_pos); + return intersection_pos_local; + } +} + +float2 sphere_alt_xyz_to_uv(const float3 intersection_pos_local, + const float2 geom_aspect) +{ + // Requires: An xyz intersection position on a cylinder. + // Returns: video_uv coords mapped to range [-0.5, 0.5] + // Mapping: Define square_uv.x to be the signed arc length in xz-space, + // and define square_uv.y == signed arc length in yz-space. + // See cylinder_xyz_to_uv() for implementation details (very similar). + const float2 angle_from_image_center = atan2( + float2(intersection_pos_local.x, -intersection_pos_local.y), + intersection_pos_local.zz); + const float2 signed_arc_len = angle_from_image_center * geom_radius; + const float2 video_uv = signed_arc_len / geom_aspect; + return video_uv; +} + +float3 sphere_alt_uv_to_xyz(const float2 video_uv, const float2 geom_aspect) +{ + // Requires: video_uv coords mapped to range [-0.5, 0.5] + // Returns: An xyz intersection position on a sphere. This is the + // inverse of sphere_alt_xyz_to_uv(). + // See cylinder_uv_to_xyz() for implementation details (very similar). + const float2 square_uv = video_uv * geom_aspect; + const float2 arc_len = square_uv; + const float2 angle_from_image_center = arc_len / geom_radius; + const float2 xy_pos = sin(angle_from_image_center) * geom_radius; + const float z_pos = sqrt(geom_radius*geom_radius - dot(xy_pos, xy_pos)); + return float3(xy_pos.x, -xy_pos.y, z_pos); +} + +float2 intersect(const float3 view_vec_local, const float3 eye_pos_local, + const float geom_mode) +{ + return geom_mode < 2.5 ? intersect_sphere(view_vec_local, eye_pos_local) : + intersect_cylinder(view_vec_local, eye_pos_local); +} + +float2 xyz_to_uv(const float3 intersection_pos_local, + const float2 geom_aspect, const float geom_mode) +{ + return geom_mode < 1.5 ? + sphere_xyz_to_uv(intersection_pos_local, geom_aspect) : + geom_mode < 2.5 ? + sphere_alt_xyz_to_uv(intersection_pos_local, geom_aspect) : + cylinder_xyz_to_uv(intersection_pos_local, geom_aspect); +} + +float3 uv_to_xyz(const float2 uv, const float2 geom_aspect, + const float geom_mode) +{ + return geom_mode < 1.5 ? sphere_uv_to_xyz(uv, geom_aspect) : + geom_mode < 2.5 ? sphere_alt_uv_to_xyz(uv, geom_aspect) : + cylinder_uv_to_xyz(uv, geom_aspect); +} + +float2 view_vec_to_uv(const float3 view_vec_local, const float3 eye_pos_local, + const float2 geom_aspect, const float geom_mode, out float3 intersection_pos) +{ + // Get the intersection point on the primitive, given an eye position + // and view vector already in its local coordinate frame: + const float2 intersect_dist_and_discriminant = intersect(view_vec_local, + eye_pos_local, geom_mode); + const float3 intersection_pos_local = eye_pos_local + + view_vec_local * intersect_dist_and_discriminant.x; + // Save the intersection position to an output parameter: + intersection_pos = intersection_pos_local; + // Transform into uv coords, but give out-of-range coords if the + // view ray doesn't intersect the primitive in the first place: + return intersect_dist_and_discriminant.y > 0.005 ? + xyz_to_uv(intersection_pos_local, geom_aspect, geom_mode) : float2(1.0, 1.0); +} + +float3 get_ideal_global_eye_pos_for_points(float3 eye_pos, + const float2 geom_aspect, const float3 global_coords[MAX_POINT_CLOUD_SIZE], + const int num_points) +{ + // Requires: Parameters: + // 1.) Starting eye_pos is a global 3D position at which the + // camera contains all points in global_coords[] in its FOV + // 2.) geom_aspect = get_aspect_vector( + // IN.output_size.x / IN.output_size.y); + // 3.) global_coords is a point cloud containing global xyz + // coords of extreme points on the simulated CRT screen. + // Globals: + // 1.) geom_view_dist must be > 0.0. It controls the "near + // plane" used to interpret flat_video_uv as a view + // vector, which controls the field of view (FOV). + // Eyespace coordinate frame: +x = right, +y = up, +z = back + // Returns: Return an eye position at which the point cloud spans as + // much of the screen as possible (given the FOV controlled by + // geom_view_dist) without being cropped or sheared. + // Algorithm: + // 1.) Move the eye laterally to a point which attempts to maximize the + // the amount we can move forward without clipping the CRT screen. + // 2.) Move forward by as much as possible without clipping the CRT. + // Get the allowed movement range by solving for the eye_pos offsets + // that result in each point being projected to a screen edge/corner in + // pseudo-normalized device coords (where xy ranges from [-0.5, 0.5] + // and z = eyespace z): + // pndc_coord = float3(float2(eyespace_xyz.x, -eyespace_xyz.y)* + // geom_view_dist / (geom_aspect * -eyespace_xyz.z), eyespace_xyz.z); + // Notes: + // The field of view is controlled by geom_view_dist's magnitude relative to + // the view vector's x and y components: + // view_vec.xy ranges from [-0.5, 0.5] * geom_aspect + // view_vec.z = -geom_view_dist + // But for the purposes of perspective divide, it should be considered: + // view_vec.xy ranges from [-0.5, 0.5] * geom_aspect / geom_view_dist + // view_vec.z = -1.0 + static const int max_centering_iters = 1; // Keep for easy testing. + for(int iter = 0; iter < max_centering_iters; iter++) + { + // 0.) Get the eyespace coordinates of our point cloud: + float3 eyespace_coords[MAX_POINT_CLOUD_SIZE]; + for(int i = 0; i < num_points; i++) + { + eyespace_coords[i] = global_coords[i] - eye_pos; + } + // 1a.)For each point, find out how far we can move eye_pos in each + // lateral direction without the point clipping the frustum. + // Eyespace +y = up, screenspace +y = down, so flip y after + // applying the eyespace offset (on the way to "clip space"). + // Solve for two offsets per point based on: + // (eyespace_xyz.xy - offset_dr) * float2(1.0, -1.0) * + // geom_view_dist / (geom_aspect * -eyespace_xyz.z) = float2(-0.5) + // (eyespace_xyz.xy - offset_dr) * float2(1.0, -1.0) * + // geom_view_dist / (geom_aspect * -eyespace_xyz.z) = float2(0.5) + // offset_ul and offset_dr represent the farthest we can move the + // eye_pos up-left and down-right. Save the min of all offset_dr's + // and the max of all offset_ul's (since it's negative). + float abs_radius = abs(geom_radius); // In case anyone gets ideas. ;) + float2 offset_dr_min = float2(10.0 * abs_radius, 10.0 * abs_radius); + float2 offset_ul_max = float2(-10.0 * abs_radius, -10.0 * abs_radius); + for(int i = 0; i < num_points; i++) + { + static const float2 flipy = float2(1.0, -1.0); + float3 eyespace_xyz = eyespace_coords[i]; + float2 offset_dr = eyespace_xyz.xy - float2(-0.5, -0.5) * + (geom_aspect * -eyespace_xyz.z) / (geom_view_dist * flipy); + float2 offset_ul = eyespace_xyz.xy - float2(0.5, 0.5) * + (geom_aspect * -eyespace_xyz.z) / (geom_view_dist * flipy); + offset_dr_min = min(offset_dr_min, offset_dr); + offset_ul_max = max(offset_ul_max, offset_ul); + } + // 1b.)Update eye_pos: Adding the average of offset_ul_max and + // offset_dr_min gives it equal leeway on the top vs. bottom + // and left vs. right. Recalculate eyespace_coords accordingly. + float2 center_offset = 0.5 * (offset_ul_max + offset_dr_min); + eye_pos.xy += center_offset; + for(int i = 0; i < num_points; i++) + { + eyespace_coords[i] = global_coords[i] - eye_pos; + } + // 2a.)For each point, find out how far we can move eye_pos forward + // without the point clipping the frustum. Flip the y + // direction in advance (matters for a later step, not here). + // Solve for four offsets per point based on: + // eyespace_xyz_flipy.x * geom_view_dist / + // (geom_aspect.x * (offset_z - eyespace_xyz_flipy.z)) =-0.5 + // eyespace_xyz_flipy.y * geom_view_dist / + // (geom_aspect.y * (offset_z - eyespace_xyz_flipy.z)) =-0.5 + // eyespace_xyz_flipy.x * geom_view_dist / + // (geom_aspect.x * (offset_z - eyespace_xyz_flipy.z)) = 0.5 + // eyespace_xyz_flipy.y * geom_view_dist / + // (geom_aspect.y * (offset_z - eyespace_xyz_flipy.z)) = 0.5 + // We'll vectorize the actual computation. Take the maximum of + // these four for a single offset, and continue taking the max + // for every point (use max because offset.z is negative). + float offset_z_max = -10.0 * geom_radius * geom_view_dist; + for(int i = 0; i < num_points; i++) + { + float3 eyespace_xyz_flipy = eyespace_coords[i] * + float3(1.0, -1.0, 1.0); + float4 offset_zzzz = eyespace_xyz_flipy.zzzz + + (eyespace_xyz_flipy.xyxy * geom_view_dist) / + (float4(-0.5, -0.5, 0.5, 0.5) * float4(geom_aspect, geom_aspect)); + // Ignore offsets that push positive x/y values to opposite + // boundaries, and vice versa, and don't let the camera move + // past a point in the dead center of the screen: + offset_z_max = (eyespace_xyz_flipy.x < 0.0) ? + max(offset_z_max, offset_zzzz.x) : offset_z_max; + offset_z_max = (eyespace_xyz_flipy.y < 0.0) ? + max(offset_z_max, offset_zzzz.y) : offset_z_max; + offset_z_max = (eyespace_xyz_flipy.x > 0.0) ? + max(offset_z_max, offset_zzzz.z) : offset_z_max; + offset_z_max = (eyespace_xyz_flipy.y > 0.0) ? + max(offset_z_max, offset_zzzz.w) : offset_z_max; + offset_z_max = max(offset_z_max, eyespace_xyz_flipy.z); + } + // 2b.)Update eye_pos: Add the maximum (smallest negative) z offset. + eye_pos.z += offset_z_max; + } + return eye_pos; +} + +float3 get_ideal_global_eye_pos(const float3x3 local_to_global, + const float2 geom_aspect, const float geom_mode) +{ + // Start with an initial eye_pos that includes the entire primitive + // (sphere or cylinder) in its field-of-view: + const float3 high_view = float3(0.0, geom_aspect.y, -geom_view_dist); + const float3 low_view = high_view * float3(1.0, -1.0, 1.0); + const float len_sq = dot(high_view, high_view); + const float fov = abs(acos(dot(high_view, low_view)/len_sq)); + // Trigonometry/similar triangles say distance = geom_radius/sin(fov/2): + const float eye_z_spherical = geom_radius/sin(fov*0.5); + const float3 eye_pos = geom_mode < 2.5 ? + float3(0.0, 0.0, eye_z_spherical) : + float3(0.0, 0.0, max(geom_view_dist, eye_z_spherical)); + + // Get global xyz coords of extreme sample points on the simulated CRT + // screen. Start with the center, edge centers, and corners of the + // video image. We can't ignore backfacing points: They're occluded + // by closer points on the primitive, but they may NOT be occluded by + // the convex hull of the remaining samples (i.e. the remaining convex + // hull might not envelope points that do occlude a back-facing point.) + static const int num_points = MAX_POINT_CLOUD_SIZE; + float3 global_coords[MAX_POINT_CLOUD_SIZE]; + global_coords[0] = mul(local_to_global, uv_to_xyz(float2(0.0, 0.0), geom_aspect, geom_mode)); + global_coords[1] = mul(local_to_global, uv_to_xyz(float2(0.0, -0.5), geom_aspect, geom_mode)); + global_coords[2] = mul(local_to_global, uv_to_xyz(float2(0.0, 0.5), geom_aspect, geom_mode)); + global_coords[3] = mul(local_to_global, uv_to_xyz(float2(-0.5, 0.0), geom_aspect, geom_mode)); + global_coords[4] = mul(local_to_global, uv_to_xyz(float2(0.5, 0.0), geom_aspect, geom_mode)); + global_coords[5] = mul(local_to_global, uv_to_xyz(float2(-0.5, -0.5), geom_aspect, geom_mode)); + global_coords[6] = mul(local_to_global, uv_to_xyz(float2(0.5, -0.5), geom_aspect, geom_mode)); + global_coords[7] = mul(local_to_global, uv_to_xyz(float2(-0.5, 0.5), geom_aspect, geom_mode)); + global_coords[8] = mul(local_to_global, uv_to_xyz(float2(0.5, 0.5), geom_aspect, geom_mode)); + // Adding more inner image points could help in extreme cases, but too many + // points will kille the framerate. For safety, default to the initial + // eye_pos if any z coords are negative: + float num_negative_z_coords = 0.0; + for(int i = 0; i < num_points; i++) + { + num_negative_z_coords += float(global_coords[0].z < 0.0); + } + // Outsource the optimized eye_pos calculation: + return num_negative_z_coords > 0.5 ? eye_pos : + get_ideal_global_eye_pos_for_points(eye_pos, geom_aspect, + global_coords, num_points); +} + +float3x3 get_pixel_to_object_matrix(const float3x3 global_to_local, + const float3 eye_pos_local, const float3 view_vec_global, + const float3 intersection_pos_local, const float3 normal, + const float2 output_size_inv) +{ + // Requires: See get_curved_video_uv_coords_and_tangent_matrix for + // descriptions of each parameter. + // Returns: Return a transformation matrix from 2D pixel-space vectors + // (where (+1.0, +1.0) is a vector to one pixel down-right, + // i.e. same directionality as uv texels) to 3D object-space + // vectors in the CRT's local coordinate frame (right-handed) + // ***which are tangent to the CRT surface at the intersection + // position.*** (Basically, we want to convert pixel-space + // vectors to 3D vectors along the CRT's surface, for later + // conversion to uv vectors.) + // Shorthand inputs: + const float3 pos = intersection_pos_local; + const float3 eye_pos = eye_pos_local; + // Get a piecewise-linear matrix transforming from "pixelspace" offset + // vectors (1.0 = one pixel) to object space vectors in the tangent + // plane (faster than finding 3 view-object intersections). + // 1.) Get the local view vecs for the pixels to the right and down: + const float3 view_vec_right_global = view_vec_global + + float3(output_size_inv.x, 0.0, 0.0); + const float3 view_vec_down_global = view_vec_global + + float3(0.0, -output_size_inv.y, 0.0); + const float3 view_vec_right_local = + mul(global_to_local, view_vec_right_global); + const float3 view_vec_down_local = + mul(global_to_local, view_vec_down_global); + // 2.) Using the true intersection point, intersect the neighboring + // view vectors with the tangent plane: + const float3 intersection_vec_dot_normal = float3(dot(pos - eye_pos, normal), dot(pos - eye_pos, normal), dot(pos - eye_pos, normal)); + const float3 right_pos = eye_pos + (intersection_vec_dot_normal / + dot(view_vec_right_local, normal))*view_vec_right_local; + const float3 down_pos = eye_pos + (intersection_vec_dot_normal / + dot(view_vec_down_local, normal))*view_vec_down_local; + // 3.) Subtract the original intersection pos from its neighbors; the + // resulting vectors are object-space vectors tangent to the plane. + // These vectors are the object-space transformations of (1.0, 0.0) + // and (0.0, 1.0) pixel offsets, so they form the first two basis + // vectors of a pixelspace to object space transformation. This + // transformation is 2D to 3D, so use (0, 0, 0) for the third vector. + const float3 object_right_vec = right_pos - pos; + const float3 object_down_vec = down_pos - pos; + const float3x3 pixel_to_object = float3x3( + object_right_vec.x, object_down_vec.x, 0.0, + object_right_vec.y, object_down_vec.y, 0.0, + object_right_vec.z, object_down_vec.z, 0.0); + return pixel_to_object; +} + +float3x3 get_object_to_tangent_matrix(const float3 intersection_pos_local, + const float3 normal, const float2 geom_aspect, const float geom_mode) +{ + // Requires: See get_curved_video_uv_coords_and_tangent_matrix for + // descriptions of each parameter. + // Returns: Return a transformation matrix from 3D object-space vectors + // in the CRT's local coordinate frame (right-handed, +y = up) + // to 2D video_uv vectors (+v = down). + // Description: + // The TBN matrix formed by the [tangent, bitangent, normal] basis + // vectors transforms ordinary vectors from tangent->object space. + // The cotangent matrix formed by the [cotangent, cobitangent, normal] + // basis vectors transforms normal vectors (covectors) from + // tangent->object space. It's the inverse-transpose of the TBN matrix. + // We want the inverse of the TBN matrix (transpose of the cotangent + // matrix), which transforms ordinary vectors from object->tangent space. + // Start by calculating the relevant basis vectors in accordance with + // Christian Schüler's blog post "Followup: Normal Mapping Without + // Precomputed Tangents": http://www.thetenthplanet.de/archives/1180 + // With our particular uv mapping, the scale of the u and v directions + // is determined entirely by the aspect ratio for cylindrical and ordinary + // spherical mappings, and so tangent and bitangent lengths are also + // determined by it (the alternate mapping is more complex). Therefore, we + // must ensure appropriate cotangent and cobitangent lengths as well. + // Base these off the uv<=>xyz mappings for each primitive. + const float3 pos = intersection_pos_local; + static const float3 x_vec = float3(1.0, 0.0, 0.0); + static const float3 y_vec = float3(0.0, 1.0, 0.0); + // The tangent and bitangent vectors correspond with increasing u and v, + // respectively. Mathematically we'd base the cotangent/cobitangent on + // those, but we'll compute the cotangent/cobitangent directly when we can. + float3 cotangent_unscaled, cobitangent_unscaled; + // geom_mode should be constant-folded without _RUNTIME_GEOMETRY_MODE. + if(geom_mode < 1.5) + { + // Sphere: + // tangent = normalize(cross(normal, cross(x_vec, pos))) * geom_aspect.x + // bitangent = normalize(cross(cross(y_vec, pos), normal)) * geom_aspect.y + // inv_determinant = 1.0/length(cross(bitangent, tangent)) + // cotangent = cross(normal, bitangent) * inv_determinant + // == normalize(cross(y_vec, pos)) * geom_aspect.y * inv_determinant + // cobitangent = cross(tangent, normal) * inv_determinant + // == normalize(cross(x_vec, pos)) * geom_aspect.x * inv_determinant + // Simplified (scale by inv_determinant below): + cotangent_unscaled = normalize(cross(y_vec, pos)) * geom_aspect.y; + cobitangent_unscaled = normalize(cross(x_vec, pos)) * geom_aspect.x; + } + else if(geom_mode < 2.5) + { + // Sphere, alternate mapping: + // This mapping works a bit like the cylindrical mapping in two + // directions, which makes the lengths and directions more complex. + // Unfortunately, I can't find much of a shortcut: + const float3 tangent = normalize( + cross(y_vec, float3(pos.x, 0.0, pos.z))) * geom_aspect.x; + const float3 bitangent = normalize( + cross(x_vec, float3(0.0, pos.yz))) * geom_aspect.y; + cotangent_unscaled = cross(normal, bitangent); + cobitangent_unscaled = cross(tangent, normal); + } + else + { + // Cylinder: + // tangent = normalize(cross(y_vec, normal)) * geom_aspect.x; + // bitangent = float3(0.0, -geom_aspect.y, 0.0); + // inv_determinant = 1.0/length(cross(bitangent, tangent)) + // cotangent = cross(normal, bitangent) * inv_determinant + // == normalize(cross(y_vec, pos)) * geom_aspect.y * inv_determinant + // cobitangent = cross(tangent, normal) * inv_determinant + // == float3(0.0, -geom_aspect.x, 0.0) * inv_determinant + cotangent_unscaled = cross(y_vec, normal) * geom_aspect.y; + cobitangent_unscaled = float3(0.0, -geom_aspect.x, 0.0); + } + const float3 computed_normal = + cross(cobitangent_unscaled, cotangent_unscaled); + const float inv_determinant = rsqrt(dot(computed_normal, computed_normal)); + const float3 cotangent = cotangent_unscaled * inv_determinant; + const float3 cobitangent = cobitangent_unscaled * inv_determinant; + // The [cotangent, cobitangent, normal] column vecs form the cotangent + // frame, i.e. the inverse-transpose TBN matrix. Get its transpose: + const float3x3 object_to_tangent = float3x3(cotangent, cobitangent, normal); + return object_to_tangent; +} + +float2 get_curved_video_uv_coords_and_tangent_matrix( + const float2 flat_video_uv, const float3 eye_pos_local, + const float2 output_size_inv, const float2 geom_aspect, + const float geom_mode, const float3x3 global_to_local, + out float2x2 pixel_to_tangent_video_uv) +{ + // Requires: Parameters: + // 1.) flat_video_uv coords are in range [0.0, 1.0], where + // (0.0, 0.0) is the top-left corner of the screen and + // (1.0, 1.0) is the bottom-right corner. + // 2.) eye_pos_local is the 3D camera position in the simulated + // CRT's local coordinate frame. For best results, it must + // be computed based on the same geom_view_dist used here. + // 3.) output_size_inv = float2(1.0)/IN.output_size + // 4.) geom_aspect = get_aspect_vector( + // IN.output_size.x / IN.output_size.y); + // 5.) geom_mode is a static or runtime mode setting: + // 0 = off, 1 = sphere, 2 = sphere alt., 3 = cylinder + // 6.) global_to_local is a 3x3 matrix transforming (ordinary) + // worldspace vectors to the CRT's local coordinate frame + // Globals: + // 1.) geom_view_dist must be > 0.0. It controls the "near + // plane" used to interpret flat_video_uv as a view + // vector, which controls the field of view (FOV). + // Returns: Return final uv coords in [0.0, 1.0], and return a pixel- + // space to video_uv tangent-space matrix in the out parameter. + // (This matrix assumes pixel-space +y = down, like +v = down.) + // We'll transform flat_video_uv into a view vector, project + // the view vector from the camera/eye, intersect with a sphere + // or cylinder representing the simulated CRT, and convert the + // intersection position into final uv coords and a local + // transformation matrix. + // First get the 3D view vector (geom_aspect and geom_view_dist are globals): + // 1.) Center uv around (0.0, 0.0) and make (-0.5, -0.5) and (0.5, 0.5) + // correspond to the top-left/bottom-right output screen corners. + // 2.) Multiply by geom_aspect to preemptively "undo" Retroarch's screen- + // space 2D aspect correction. We'll reapply it in uv-space. + // 3.) (x, y) = (u, -v), because +v is down in 2D screenspace, but +y + // is up in 3D worldspace (enforce a right-handed system). + // 4.) The view vector z controls the "near plane" distance and FOV. + // For the effect of "looking through a window" at a CRT, it should be + // set equal to the user's distance from their physical screen, in + // units of the viewport's physical diagonal size. + const float2 view_uv = (flat_video_uv - float2(0.5, 0.5)) * geom_aspect; + const float3 view_vec_global = + float3(view_uv.x, -view_uv.y, -geom_view_dist); + // Transform the view vector into the CRT's local coordinate frame, convert + // to video_uv coords, and get the local 3D intersection position: + const float3 view_vec_local = mul(global_to_local, view_vec_global); + float3 pos; + const float2 centered_uv = view_vec_to_uv( + view_vec_local, eye_pos_local, geom_aspect, geom_mode, pos); + const float2 video_uv = centered_uv + float2(0.5, 0.5); + // Get a pixel-to-tangent-video-uv matrix. The caller could deal with + // all but one of these cases, but that would be more complicated. + #if _DRIVERS_ALLOW_DERIVATIVES + // Derivatives obtain a matrix very fast, but the direction of pixel- + // space +y seems to depend on the pass. Enforce the correct direction + // on a best-effort basis (but it shouldn't matter for antialiasing). + const float2 duv_dx = ddx(video_uv); + const float2 duv_dy = ddy(video_uv); + #ifdef LAST_PASS + pixel_to_tangent_video_uv = float2x2( + duv_dx.x, duv_dy.x, + -duv_dx.y, -duv_dy.y); + #else + pixel_to_tangent_video_uv = float2x2( + duv_dx.x, duv_dy.x, + duv_dx.y, duv_dy.y); + #endif + #else + // Manually define a transformation matrix. We'll assume pixel-space + // +y = down, just like +v = down. + if(geom_force_correct_tangent_matrix) + { + // Get the surface normal based on the local intersection position: + const float3 normal_base = geom_mode < 2.5 ? pos : + float3(pos.x, 0.0, pos.z); + const float3 normal = normalize(normal_base); + // Get pixel-to-object and object-to-tangent matrices and combine + // them into a 2x2 pixel-to-tangent matrix for video_uv offsets: + const float3x3 pixel_to_object = get_pixel_to_object_matrix( + global_to_local, eye_pos_local, view_vec_global, pos, normal, + output_size_inv); + const float3x3 object_to_tangent = get_object_to_tangent_matrix( + pos, normal, geom_aspect, geom_mode); + const float3x3 pixel_to_tangent3x3 = + mul(object_to_tangent, pixel_to_object); + pixel_to_tangent_video_uv = float2x2( + pixel_to_tangent3x3[0][0], pixel_to_tangent3x3[0][1], pixel_to_tangent3x3[1][0], pixel_to_tangent3x3[1][1]);//._m00_m01_m10_m11); + } + else + { + // Ignore curvature, and just consider flat scaling. The + // difference is only apparent with strong curvature: + pixel_to_tangent_video_uv = float2x2( + output_size_inv.x, 0.0, 0.0, output_size_inv.y); + } + #endif + return video_uv; +} + +float get_border_dim_factor(const float2 video_uv, const float2 geom_aspect) +{ + // COPYRIGHT NOTE FOR THIS FUNCTION: + // Copyright (C) 2010-2012 cgwg, 2014 TroggleMonkey + // This function uses an algorithm first coded in several of cgwg's GPL- + // licensed lines in crt-geom-curved.cg and its ancestors. The line + // between algorithm and code is nearly indistinguishable here, so it's + // unclear whether I could even release this project under a non-GPL + // license with this function included. + + // Calculate border_dim_factor from the proximity to uv-space image + // borders; geom_aspect/border_size/border/darkness/border_compress are globals: + const float2 edge_dists = min(video_uv, float2(1.0, 1.0) - video_uv) * + geom_aspect; + const float2 border_penetration = + max(float2(border_size, border_size) - edge_dists, float2(0.0, 0.0)); + const float penetration_ratio = border_size > 0 ? length(border_penetration)/border_size : 0; + const float border_escape_ratio = max(1.0 - penetration_ratio, 0.0); + const float border_dim_factor = + pow(border_escape_ratio, border_darkness) * max(1.0, border_compress); + return min(border_dim_factor, 1.0); +} + + + +#endif // _GEOMETRY_FUNCTIONS_H + + + diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/helper-functions-and-macros.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/helper-functions-and-macros.fxh new file mode 100644 index 000000000..d9e1820df --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/helper-functions-and-macros.fxh @@ -0,0 +1,76 @@ +#ifndef _HELPER_FUNCTIONS_AND_MACROS_H +#define _HELPER_FUNCTIONS_AND_MACROS_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2020 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +float4 tex2D_nograd(sampler2D tex, float2 tex_coords) +{ + return tex2Dlod(tex, float4(tex_coords, 0, 0), 0.0); +} + +// ReShade 4 does not permit the use of functions or the ternary operator +// outside of a function definition. This is a problem for this port +// because the original crt-royale shader makes heavy use of these +// constructs at the root level. + +// These preprocessor definitions are a workaround for this limitation. +// Note that they are strictly intended for defining complex global +// constants. I doubt they're more performant than the built-in +// equivalents, so I recommend using the built-ins whenever you can. + + +#define macro_sign(c) -((int) ((c) != 0)) * -((int) ((c) > 0)) +#define macro_abs(c) (c) * macro_sign(c) + +#define macro_min(c, d) (c) * ((int) ((c) <= (d))) + (d) * ((int) ((c) > (d))) +#define macro_max(c, d) (c) * ((int) ((c) >= (d))) + (d) * ((int) ((c) < (d))) +#define macro_clamp(c, l, u) macro_min(macro_max(c, l), u) + +#define macro_ceil(c) (float) ((int) (c) + (int) (((int) (c)) < (c))) + +#define macro_cond(c, a, b) float(c) * (a) + float(!(c)) * (b) + + + +//////////////////////// COMMON MATHEMATICAL CONSTANTS /////////////////////// + +static const float pi = 3.141592653589; +// We often want to find the location of the previous texel, e.g.: +// const float2 curr_texel = uv * texture_size; +// const float2 prev_texel = floor(curr_texel - float2(0.5)) + float2(0.5); +// const float2 prev_texel_uv = prev_texel / texture_size; +// However, many GPU drivers round incorrectly around exact texel locations. +// We need to subtract a little less than 0.5 before flooring, and some GPU's +// require this value to be farther from 0.5 than others; define it here. +// const float2 prev_texel = +// floor(curr_texel - float2(under_half)) + float2(0.5); +static const float under_half = 0.4995; + +// Avoid dividing by zero; using a macro overloads for float, float2, etc.: +#define FIX_ZERO(c) (macro_max(macro_abs(c), 0.0000152587890625)) // 2^-16 + +// #define fmod(x, y) ((x) - (y) * floor((x)/(y) + FIX_ZERO(0.0))) +#define fmod(x, y) (frac((x) / (y)) * (y)) + +#endif // _HELPER_FUNCTIONS_AND_MACROS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/phosphor-mask-calculations.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/phosphor-mask-calculations.fxh new file mode 100644 index 000000000..99b1d021c --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/phosphor-mask-calculations.fxh @@ -0,0 +1,624 @@ +#ifndef _PHOSHOR_MASK_CALCULATIONS_H +#define _PHOSHOR_MASK_CALCULATIONS_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2020 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +/* + * Our goal is to use arithmetic to generate the phosphor mask. + * Phosphor masks are regular patterns, so we want something periodic. + * We need to avoid integer arithmetic because it tends to cause rounding errors. + * + * For all masks, we want to approximate a pulse wave in at least one dimension. This pulse wave + * will have narrow peaks, wide troughs, and constant periodicity. + * GRILLE will have a pulse wave along the x-axis and will be constant along the y-axis. + * SLOT and SHADOW will likely have a superposition of two out-of-phase pulse waves along each axis. + * For SHADOW, the width of the peaks will vary such that they generate ellipsoids on the screen. + * + * We can get a periodic function by starting with a triangle wave: T(t, f) = abs(1 - 2*frac(t * f)). + * This function gives us a triangle wave with f cycles in the domain [0, 1]. + * Note that T(0, f) = 1. + * + * Then we can compose this with a sigmoid curve to squish the triangle wave into a pulse wave. + * P(s, p, q) = exp(q s - q/2) / (exp(q s - q/2) + exp(-p)) + * s(t, f, o) = T(t*f - o, 1) + * + * f is the number of pulses to render along the given axis. + * o is the channel's horizontal ofset along the given axis, normalized via the quotient raw_offset / raw_triad width. + * p and q control how closely P resembles an ideal pulse wave and also how wide the peaks and troughs are. + * + * The interaction between p and q is rather complicated and difficult to describe, so they're not a good pair + * of parameters for users. But we have the info necessary to solve for p in terms of q. + * We know the width of a phosphor and the width of a triad, and we know the domain and range of P. + * We can choose a coordinate (t0, y0) that will denote the edge of the phosphor. + * Note that y0 = P(t0, p, q) for some p and q. + * We let t0 = raw_phosphor_width / raw_triad_width, since we need to respect the shape of the phosphor. + * We let the user define P(t0). + * Technically, this means the user is defining the brightness of the phosphor's furthest edge. + * Visually, this looks like the user is defining the width of the phosphor. + * We'll call this the Phosphor Thickness. + * We let the user define q. + * Technically, this means the user is defining the squareness of the pulse wave. + * Visually, this looks like the user is defining the sharpness of the phosphor. + * We'll call this the Phosphor Sharpness. + * + * We can solve for p in terms of q very efficiently. + * p = (ln(y0 / (1 - y0)) - q) / (0.5 - 2 t0) + * + * Note that, if you work through the algebra, you get a denominator of (t0 - 0.5). + * Using (0.5 - 2 t0) actually works better. It also matches up when you try plotting P and (t0, y0). + * + * For the GRILLE and SLOT masks, we can compute p once and recycle it. + * For the SHADOW mask, we can either compute p on each iteration or find a way to interpolate between min_p and max_p. + * + * One might expect it'd be way better to use a clamped triangle wave rather than a sigmoid or exponentiated cosine wave. + * As far as I can tell, this ends up being incorrect surprisingly enough. Although it's a good bit faster, + * it has terrible aliasing artifacts at small scales. The other implementations are slower, but they produce + * evenly-sized RGB phosphors for a variety of configurations even when the triad width is 3 pixels. At that + * scale, the triangle wave approach produces triads where one of the phosphors is thicker than the others. + * Taking into account the compute_mask_factor trick, the triangle wave approach would be a negligible + * performance improvement at the cost of a large drop in visual quality and user friendliness. + */ + + +#include "bind-shader-params.fxh" +#include "scanline-functions.fxh" + +/* + * The GRILLE mask consists of an array of vertical stripes, so each channel will vary along the x-axis and will be constant + * along the y-axis. + * + * It has the following dimensions: + * Phosphors are 18 units wide with unbounded height. + * Phosphors in a triad are 2 units apart. + * Triads are 6 units apart. + * Triad centers are 64 units apart. + * The phosphors follow an RGB pattern. + * The left-most phosphor is red and offset by 3 units to the right. + */ +static const float grille_raw_phosphor_width = 18; +static const float grille_raw_phosphor_gap = 2; +static const float grille_raw_triad_horiz_gap = 6; +static const float grille_raw_triad_width = 3*grille_raw_phosphor_width + 2*grille_raw_phosphor_gap + grille_raw_triad_horiz_gap; + +static const float grille_raw_r_offset = (grille_raw_triad_horiz_gap + grille_raw_phosphor_width) / 2; +static const float grille_raw_g_offset = grille_raw_r_offset + grille_raw_phosphor_width + grille_raw_phosphor_gap; +static const float grille_raw_b_offset = grille_raw_g_offset + grille_raw_phosphor_width + grille_raw_phosphor_gap; +static const float3 grille_norm_center_offsets = float3( + grille_raw_r_offset, + grille_raw_g_offset, + grille_raw_b_offset +) / grille_raw_triad_width; + +static const float grille_edge_t = grille_raw_phosphor_width / 2; +static const float grille_edge_norm_t = grille_edge_t / grille_raw_triad_width; + + +/* + * The SLOT mask consists of an array of rectangles, so each channel will vary along both the x- and y-axes. + * + * It has the following dimensions: + * Phosphors are 18 units wide and 66 units tall. + * Phosphors in a triad are 2 units apart. + * Triads are 6 units apart horizontally and 6 units apart vertically. + * Triad centers are 64 units apart horizontally and 73 units apart vertically. + * The phosphors follow an RGB pattern. + * The upper-left-most phosphor is red and offset by 3 units to the right and 3 units down. + */ +static const float slot_raw_phosphor_width = 18; +static const float slot_raw_phosphor_gap = 2; +static const float slot_raw_triad_horiz_gap = 6; +static const float slot_raw_triad_width = 3*slot_raw_phosphor_width + 2*slot_raw_phosphor_gap + slot_raw_triad_horiz_gap; + +static const float slot_raw_phosphor_height = 66; +static const float slot_raw_triad_vert_gap = 6; +static const float slot_raw_triad_height = slot_raw_phosphor_height + slot_raw_triad_vert_gap; + +static const float slot_aspect_ratio = slot_raw_triad_height / slot_raw_triad_width; + +static const float slot_raw_r_offset_x = (slot_raw_triad_horiz_gap + slot_raw_phosphor_width) / 2; +static const float slot_raw_g_offset_x = slot_raw_r_offset_x + slot_raw_phosphor_width + slot_raw_phosphor_gap; +static const float slot_raw_b_offset_x = slot_raw_g_offset_x + slot_raw_phosphor_width + slot_raw_phosphor_gap; +static const float3 slot_norm_center_offsets_x = float3( + slot_raw_r_offset_x, + slot_raw_g_offset_x, + slot_raw_b_offset_x +) / slot_raw_triad_width; +static const float3 slot_norm_center_offsets_y = float3(0.5, 0.5, 0.5); + +static const float slot_edge_tx = slot_raw_phosphor_width / 2; +// We draw the slot mask as two sets of columns. To do that, we have to pretend the horizontal gap is the size of a whole triad. +// Then we need to halve the position of the phosphor edge. +static const float slot_edge_norm_tx = 0.5 * slot_edge_tx / slot_raw_triad_width; +static const float slot_edge_ty = slot_raw_phosphor_height / 2; +static const float slot_edge_norm_ty = slot_edge_ty / slot_raw_triad_height; + +/* + * The SHADOW mask consists of an array of circles, so each channel will vary along both the x- and y-axes. + * + * It has the following dimensions: + * Phosphors are 21 units in diameter. + * All phosphors are 0 units apart. + * Triad centers are 63 units apart horizontally and 21 units apart vertically. + * The phosphors follow a GBR pattern on odd rows and RBG on even rows. + * The upper-left-most phosphor is green and centered on the corner of the screen. + */ +static const float shadow_raw_phosphor_diam = 21; +static const float shadow_raw_phosphor_gap = 0; +static const float shadow_raw_triad_horiz_gap = 0; +static const float shadow_raw_triad_vert_gap = 0; + +static const float shadow_raw_triad_width = 3*shadow_raw_phosphor_diam + 2*shadow_raw_phosphor_gap + shadow_raw_triad_horiz_gap; +static const float shadow_raw_triad_height = shadow_raw_phosphor_diam + shadow_raw_triad_vert_gap; + +static const float shadow_aspect_ratio = shadow_raw_triad_height / shadow_raw_triad_width; + +static const float shadow_raw_g_offset_x = 0; +static const float shadow_raw_b_offset_x = shadow_raw_g_offset_x + shadow_raw_phosphor_diam + shadow_raw_phosphor_gap; +static const float shadow_raw_r_offset_x = shadow_raw_b_offset_x + shadow_raw_phosphor_diam + shadow_raw_phosphor_gap; +static const float3 shadow_norm_center_offsets_x = float3( + shadow_raw_r_offset_x, + shadow_raw_g_offset_x, + shadow_raw_b_offset_x +) / shadow_raw_triad_width; + +static const float3 shadow_norm_center_offsets_y = float3(0.0, 0.0, 0.0); + +static const float shadow_edge_tx = shadow_raw_phosphor_diam / 2; +static const float shadow_edge_norm_tx = shadow_edge_tx / shadow_raw_triad_width; +static const float shadow_edge_ty = shadow_raw_phosphor_diam / 2; +// We draw the shadow mask as two sets of rows. To do that, we have to pretend the vertical gap is the size of a whole triad. +// Then we need to halve the position of the phosphor edge. +static const float shadow_edge_norm_ty = 0.5 * shadow_edge_ty / shadow_raw_triad_height; +static const float shadow_norm_phosphor_rad = (shadow_raw_phosphor_diam/2) / shadow_raw_triad_width; + + +/* + * The SMALL GRILLE mask is composed of magenta and green stripes. + * Sourced from http://filthypants.blogspot.com/2020/02/crt-shader-masks.html + * + * It has the following dimensions: + * Stripes are 32 units wide. + * Stripes in a triad are 0 units apart. + * Triads are 0 units apart horizontally. + * + * Each triad has two quads, side-by-side and aligned. + * Neighboring triads are offset vertically. + * Below is an array of 2 triads. + * x's denote magenta stripes, and o's denote green ones. + * + * xxooxxoo + * xxooxxoo + * xxooxxoo + * xxooxxoo + * xxooxxoo + * xxooxxoo + * + * The phosphors follow a MG pattern. + * The left-most phosphor is magenta and offset by 16 units to the right. + */ + +static const float smallgrille_raw_stripe_width = 32; +static const float smallgrille_raw_triad_width = 2*smallgrille_raw_stripe_width; + +static const float smallgrille_raw_r_offset_x = 0.5 * smallgrille_raw_stripe_width; +static const float smallgrille_raw_g_offset_x = smallgrille_raw_r_offset_x + smallgrille_raw_stripe_width; +static const float smallgrille_raw_b_offset_x = smallgrille_raw_r_offset_x; +static const float3 smallgrille_norm_center_offsets_x = float3( + smallgrille_raw_r_offset_x, + smallgrille_raw_g_offset_x, + smallgrille_raw_b_offset_x +) / smallgrille_raw_triad_width; + +static const float smallgrille_edge_t = 0.5 * smallgrille_raw_stripe_width; +static const float smallgrille_edge_norm_t = smallgrille_edge_t / smallgrille_raw_triad_width; + + +/* + * The SMALL SLOT mask is composed of magenta and green quads. + * Sourced from http://filthypants.blogspot.com/2020/02/crt-shader-masks.html + * + * It has the following dimensions: + * Quads are 32 units wide and 48 units tall. + * Quads in a triad are 0 units apart. + * Triads are 0 units apart horizontally and 16 units apart vertically. + * + * Each triad has two quads, side-by-side and aligned. + * Neighboring triads are offset vertically. + * Below is a 2x2 matrix of 4 triads. + * x's denote magenta quads, and o's denote green ones. + * + * xxoo + * xxooxxoo + * xxooxxoo + * xxoo + * xxoo + * xxooxxoo + * xxooxxoo + * xxoo + * + * The phosphors follow a MG pattern. + * The upper-left-most phosphor is magenta and offset by 16 units to the right and 16 units down. + */ + +static const float smallslot_raw_quad_width = 32; +static const float smallslot_raw_triad_width = 2*smallslot_raw_quad_width; + +static const float smallslot_raw_quad_height = 1.5 * smallslot_raw_quad_width; +static const float smallslot_raw_triad_vert_gap = 0.5 * smallslot_raw_quad_width; +static const float smallslot_raw_triad_height = smallslot_raw_quad_height + smallslot_raw_triad_vert_gap; + +static const float smallslot_aspect_ratio = smallslot_raw_triad_height / smallslot_raw_triad_width; + +static const float smallslot_raw_r_offset_x = 0.5 * smallslot_raw_quad_width; +static const float smallslot_raw_g_offset_x = smallslot_raw_r_offset_x + smallslot_raw_quad_width; +static const float smallslot_raw_b_offset_x = smallslot_raw_r_offset_x; +static const float3 smallslot_norm_center_offsets_x = float3( + smallslot_raw_r_offset_x, + smallslot_raw_g_offset_x, + smallslot_raw_b_offset_x +) / smallslot_raw_triad_width; + +static const float3 smallslot_norm_center_offsets_y1 = 0.5 * smallslot_raw_quad_height / smallslot_raw_triad_height; +static const float3 smallslot_norm_center_offsets_y2 = smallslot_norm_center_offsets_y1 + smallslot_raw_triad_vert_gap / smallslot_raw_triad_height; + +static const float smallslot_edge_tx = 0.5 * smallslot_raw_quad_width; +// We draw the slot mask as two sets of columns. To do that, we have to pretend the horizontal gap is the size of a whole triad. +// Then we need to halve the position of the phosphor edge. +static const float smallslot_edge_norm_tx = 0.5 * smallslot_edge_tx / smallslot_raw_triad_width; +static const float smallslot_edge_ty = smallslot_raw_quad_height / 2; +static const float smallslot_edge_norm_ty = smallslot_edge_ty / smallslot_raw_triad_height; + +/* + * The SMALL SHADOW mask is composed of magenta and green quads. + * Sourced from http://filthypants.blogspot.com/2020/02/crt-shader-masks.html + * + * It has the following dimensions: + * Quads are 17 units wide and 17 units tall. + * Quads in a triad are 0 units apart. + * Triads are 0 units apart horizontally and 0 units apart vertically. + * + * Each triad has two quads, side-by-side and aligned. + * Neighboring triads are offset vertically. + * Below is a 2x2 matrix of 4 triads. + * x's denote magenta quads, and o's denote green ones. + * + * xxooxxoo + * xxooxxoo + * ooxxooxx + * ooxxooxx + * + * The phosphors follow a MG pattern. + * The upper-left-most phosphor is magenta and offset by 16 units to the right and 16 units down. + */ + +static const float smallshadow_raw_quad_width = 17; +static const float smallshadow_raw_triad_width = 2 * smallshadow_raw_quad_width; + +static const float smallshadow_raw_quad_height = 17; +static const float smallshadow_raw_triad_height = smallshadow_raw_quad_height; + +static const float smallshadow_aspect_ratio = smallshadow_raw_triad_height / smallshadow_raw_triad_width; + +static const float smallshadow_raw_r_offset_x = 0.5 * smallshadow_raw_quad_width; +static const float smallshadow_raw_g_offset_x = smallshadow_raw_r_offset_x + smallshadow_raw_quad_width; +static const float smallshadow_raw_b_offset_x = smallshadow_raw_r_offset_x; +static const float3 smallshadow_norm_center_offsets_x = float3( + smallshadow_raw_r_offset_x, + smallshadow_raw_g_offset_x, + smallshadow_raw_b_offset_x +) / smallshadow_raw_triad_width; + +static const float3 smallshadow_norm_center_offsets_y = 0.5 * smallshadow_raw_triad_height; + +static const float smallshadow_edge_tx = 0.5 * smallshadow_raw_quad_width; +static const float smallshadow_edge_norm_tx = smallshadow_edge_tx / smallshadow_raw_triad_width; +static const float smallshadow_edge_ty = 0.5 * smallshadow_raw_quad_height; +// We draw the shadow mask as two sets of rows. To do that, we have to pretend the vertical gap is the size of a whole triad. +// Then we need to halve the position of the phosphor edge. +static const float smallshadow_edge_norm_ty = 0.5 * smallshadow_edge_ty / smallshadow_raw_triad_height; + + + + +float get_selected_aspect_ratio() { + float aspect_ratio; + [flatten] + if (mask_type == 0 || mask_type == 3) { + aspect_ratio = scale_triad_height; + } + else if (mask_type == 1 || mask_type == 4) { + aspect_ratio = scale_triad_height * slot_aspect_ratio; + } + else { + aspect_ratio = scale_triad_height * shadow_aspect_ratio; + } + [flatten] + switch (mask_type) { + case 0: + aspect_ratio = scale_triad_height; + break; + case 1: + aspect_ratio = scale_triad_height * slot_aspect_ratio; + break; + case 2: + aspect_ratio = scale_triad_height * shadow_aspect_ratio; + break; + case 3: + aspect_ratio = scale_triad_height; + break; + case 4: + aspect_ratio = scale_triad_height * smallslot_aspect_ratio; + break; + default: + aspect_ratio = scale_triad_height * smallshadow_aspect_ratio; + break; + } + + return aspect_ratio; +} + +float2 calc_triad_size() { + const float aspect_ratio = get_selected_aspect_ratio(); + + [branch] + if (mask_size_param == 0) { + return float2(1, aspect_ratio) * mask_triad_width; + } + else { + float triad_width = content_size.x * rcp(mask_num_triads_across); + return float2(1, aspect_ratio) * triad_width; + } + +} + +float2 calc_phosphor_viewport_frequency_factor() { + const float aspect_ratio = get_selected_aspect_ratio(); + + float2 triad_size_factor; + float2 num_triads_factor; + [branch] + if (geom_rotation_mode == 0 || geom_rotation_mode == 2) { + triad_size_factor = content_size * rcp(mask_triad_width * float2(1, aspect_ratio)); + num_triads_factor = mask_num_triads_across * float2(1, content_size.y * rcp(content_size.x) * rcp(aspect_ratio)); + } + else { + triad_size_factor = content_size * rcp(mask_triad_width * float2(1, aspect_ratio)).yx; + num_triads_factor = mask_num_triads_across * float2(1, content_size.y * rcp(content_size.x) * rcp(aspect_ratio)).yx; + } + + return ((mask_size_param == 0) ? triad_size_factor : num_triads_factor); +} + + +/* + * We have a pulse wave f(t0_norm, p, q) = y0 with unknown p. + * This function solves for p. + */ +#define calculate_phosphor_p_value(t0_norm, y0, q) (log((y0) * rcp(1 - (y0))) - (q) * (0.5 - 2*(t0_norm))) + +/* + * If we don't rescale the phosphor_thickness parameter, it has a logarithmic effect on the phosphor shape. + * Rescaling it makes it look closer to a linear effect. + */ +#define linearize_phosphor_thickness_param(p) (1 - exp(-(p))) + + +/* + * Generates a grille mask with the desired resolution and sharpness. + */ +float3 get_phosphor_intensity_grille( + const float2 texcoord, + const float2 viewport_frequency_factor, + const float2 grille_pq +) { + float3 center_offsets = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + grille_norm_center_offsets.bgr : grille_norm_center_offsets; + + center_offsets += phosphor_offset_x * 0.5; + + float3 theta = triangle_wave(texcoord.x * viewport_frequency_factor.x - center_offsets, 1); + float3 alpha = exp((theta - 0.5) * grille_pq.y); + return alpha * rcp(alpha + grille_pq.x); +} + + +/* + * Generates a slot mask with the desired resolution and sharpness. + */ +float3 get_phosphor_intensity_slot( + const float2 texcoord, + const float2 viewport_frequency_factor, + const float2 slot_pq_x, + const float2 slot_pq_y +) { + float3 center_offsets_x = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + slot_norm_center_offsets_x.bgr : slot_norm_center_offsets_x; + float3 center_offsets_y = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + slot_norm_center_offsets_y.bgr : slot_norm_center_offsets_y; + + center_offsets_x += phosphor_offset_x * 0.5; + center_offsets_y += phosphor_offset_y * 0.5; + + float3 theta_x1 = triangle_wave(texcoord.x * viewport_frequency_factor.x - center_offsets_x, 0.5); + float3 alpha_x1 = exp((theta_x1 - 0.5) * slot_pq_x.y); + alpha_x1 *= rcp(alpha_x1 + slot_pq_x.x); + + float3 theta_x2 = triangle_wave(texcoord.x * viewport_frequency_factor.x - center_offsets_x + 1, 0.5); + float3 alpha_x2 = exp((theta_x2 - 0.5) * slot_pq_x.y); + alpha_x2 *= rcp(alpha_x2 + slot_pq_x.x); + + float3 theta_y1 = triangle_wave(texcoord.y * viewport_frequency_factor.y - center_offsets_y, 1); + float3 alpha_y1 = exp((theta_y1 - 0.5) * slot_pq_y.y); + alpha_y1 *= rcp(alpha_y1 + slot_pq_y.x); + + float3 theta_y2 = triangle_wave(texcoord.y * viewport_frequency_factor.y - center_offsets_y + 0.5, 1); + float3 alpha_y2 = exp((theta_y2 - 0.5) * slot_pq_y.y); + alpha_y2 *= rcp(alpha_y2 + slot_pq_y.x); + + return alpha_x1 * alpha_y1 + alpha_x2 * alpha_y2; +} + +/* + * Generates a shadow mask with the desired resolution and sharpness. + */ +float3 get_phosphor_intensity_shadow( + const float2 texcoord, + const float2 viewport_frequency_factor, + const float2 shadow_q +) { + float3 center_offsets_x = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + shadow_norm_center_offsets_x.bgr : shadow_norm_center_offsets_x; + float3 center_offsets_y = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + shadow_norm_center_offsets_y.bgr : shadow_norm_center_offsets_y; + + center_offsets_x += phosphor_offset_x * 0.5; + center_offsets_y += phosphor_offset_y * 0.5; + + const float2 thickness_scaled = linearize_phosphor_thickness_param(phosphor_thickness); + + const float3 x_adj = texcoord.x * viewport_frequency_factor.x - center_offsets_x; + const float3 y_adj = texcoord.y * viewport_frequency_factor.y - center_offsets_y; + + const float3 texcoord_x_periodic1 = shadow_norm_phosphor_rad * triangle_wave(x_adj * 3 - 0.5, 1.0); + const float3 texcoord_x_periodic2 = shadow_norm_phosphor_rad * triangle_wave(x_adj * 3, 1.0); + const float3 ty1 = sqrt( + shadow_norm_phosphor_rad*shadow_norm_phosphor_rad - texcoord_x_periodic1*texcoord_x_periodic1 + ); + const float3 ty2 = sqrt( + shadow_norm_phosphor_rad*shadow_norm_phosphor_rad - texcoord_x_periodic2*texcoord_x_periodic2 + ); + + const float shadow_px = exp(-calculate_phosphor_p_value(shadow_edge_norm_tx, thickness_scaled.x, shadow_q.x)); + const float3 shadow_py1 = exp(-calculate_phosphor_p_value(ty1 * 0.5 * rcp(shadow_aspect_ratio), thickness_scaled.y, shadow_q.y)); + const float3 shadow_py2 = exp(-calculate_phosphor_p_value(ty2 * 0.5 * rcp(shadow_aspect_ratio), thickness_scaled.y, shadow_q.y)); + + float3 theta_x1 = triangle_wave(x_adj, 1); + float3 alpha_x1 = exp((theta_x1 - 0.5) * shadow_q.x); + alpha_x1 *= rcp(alpha_x1 + shadow_px); + + float3 theta_x2 = triangle_wave(x_adj + 0.5, 1); + float3 alpha_x2 = exp((theta_x2 - 0.5) * shadow_q.x); + alpha_x2 *= rcp(alpha_x2 + shadow_px); + + float3 theta_y1 = triangle_wave(y_adj, 0.5); + float3 alpha_y1 = exp((theta_y1 - 0.5) * shadow_q.y); + alpha_y1 *= rcp(alpha_y1 + shadow_py1); + + float3 theta_y2 = triangle_wave(y_adj + 1, 0.5); + float3 alpha_y2 = exp((theta_y2 - 0.5) * shadow_q.y); + alpha_y2 *= rcp(alpha_y2 + shadow_py2); + + return alpha_x1 * alpha_y1 + alpha_x2 * alpha_y2; +} + +float3 get_phosphor_intensity_grille_small( + const float2 texcoord, + const float2 viewport_frequency_factor, + const float2 grille_pq_x +) { + float3 center_offsets_x = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + smallgrille_norm_center_offsets_x.grg : smallgrille_norm_center_offsets_x; + + center_offsets_x += phosphor_offset_x * 0.5; + + float3 theta = triangle_wave(texcoord.x * viewport_frequency_factor.x - center_offsets_x, 1); + float3 alpha = exp((theta - 0.5) * grille_pq_x.y); + alpha *= rcp(alpha + grille_pq_x.x); + + // Taking a sqrt here helps hide the gaps between the pixels when the triad size is small + return sqrt(alpha); +} + +float3 get_phosphor_intensity_slot_small( + const float2 texcoord, + const float2 viewport_frequency_factor, + const float2 slot_pq_x, + const float2 slot_pq_y +) { + float3 center_offsets_x = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + smallslot_norm_center_offsets_x.grg : smallslot_norm_center_offsets_x; + float3 center_offsets_y1 = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + smallslot_norm_center_offsets_y1.grg : smallslot_norm_center_offsets_y1; + float3 center_offsets_y2 = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + smallslot_norm_center_offsets_y2.grg : smallslot_norm_center_offsets_y2; + + center_offsets_x += phosphor_offset_x * 0.5; + center_offsets_y1 += phosphor_offset_y * 0.5; + center_offsets_y2 += phosphor_offset_y * 0.5; + + float3 theta_x1 = triangle_wave(texcoord.x * viewport_frequency_factor.x - center_offsets_x, 0.5); + float3 alpha_x1 = exp((theta_x1 - 0.5) * slot_pq_x.y); + alpha_x1 *= rcp(alpha_x1 + slot_pq_x.x); + + float3 theta_x2 = triangle_wave(texcoord.x * viewport_frequency_factor.x - center_offsets_x + 1, 0.5); + float3 alpha_x2 = exp((theta_x2 - 0.5) * slot_pq_x.y); + alpha_x2 *= rcp(alpha_x2 + slot_pq_x.x); + + float3 theta_y1 = triangle_wave(texcoord.y * viewport_frequency_factor.y - center_offsets_y1, 1); + float3 alpha_y1 = exp((theta_y1 - 0.5) * slot_pq_y.y); + alpha_y1 *= rcp(alpha_y1 + slot_pq_y.x); + + float3 theta_y2 = triangle_wave(texcoord.y * viewport_frequency_factor.y - center_offsets_y2 + 0.5, 1); + float3 alpha_y2 = exp((theta_y2 - 0.5) * slot_pq_y.y); + alpha_y2 *= rcp(alpha_y2 + slot_pq_y.x); + + // Taking a sqrt here helps hide the gaps between the pixels when the triad size is small + return (alpha_x1 * alpha_y1 + alpha_x2 * alpha_y2); +} + +float3 get_phosphor_intensity_shadow_small( + const float2 texcoord, + const float2 viewport_frequency_factor, + const float2 shadow_pq_x, + const float2 shadow_pq_y +) { + float3 center_offsets_x = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + smallshadow_norm_center_offsets_x.grg : smallshadow_norm_center_offsets_x; + float3 center_offsets_y = (geom_rotation_mode == 2 || geom_rotation_mode == 3) ? + smallshadow_norm_center_offsets_y.grg : smallshadow_norm_center_offsets_y; + + center_offsets_x += phosphor_offset_x * 0.5; + center_offsets_y += phosphor_offset_y * 0.5; + + float3 theta_x1 = triangle_wave(texcoord.x * viewport_frequency_factor.x - center_offsets_x, 1); + float3 alpha_x1 = exp((theta_x1 - 0.5) * shadow_pq_x.y); + alpha_x1 *= rcp(alpha_x1 + shadow_pq_x.x); + + float3 theta_x2 = triangle_wave(texcoord.x * viewport_frequency_factor.x - center_offsets_x + 0.5, 1); + float3 alpha_x2 = exp((theta_x2 - 0.5) * shadow_pq_x.y); + alpha_x2 *= rcp(alpha_x2 + shadow_pq_x.x); + + float3 theta_y1 = triangle_wave(texcoord.y * viewport_frequency_factor.y - center_offsets_y, 0.5); + float3 alpha_y1 = exp((theta_y1 - 0.5) * shadow_pq_y.y); + alpha_y1 *= rcp(alpha_y1 + shadow_pq_y.x); + + float3 theta_y2 = triangle_wave(texcoord.y * viewport_frequency_factor.y - center_offsets_y + 1, 0.5); + float3 alpha_y2 = exp((theta_y2 - 0.5) * shadow_pq_y.y); + alpha_y2 *= rcp(alpha_y2 + shadow_pq_y.x); + + // Taking a sqrt here helps hide the gaps between the pixels when the triad size is small + return sqrt(alpha_x1 * alpha_y1 + alpha_x2 * alpha_y2); +} + +#endif // _PHOSHOR_MASK_CALCULATIONS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/quad-pixel-communication.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/quad-pixel-communication.fxh new file mode 100644 index 000000000..8e44b4e3f --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/quad-pixel-communication.fxh @@ -0,0 +1,243 @@ + +#ifndef _QUAD_PIXEL_COMMUNICATION_H +#define _QUAD_PIXEL_COMMUNICATION_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2014 TroggleMonkey* +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + +///////////////////////////////// DISCLAIMER ///////////////////////////////// + +// *This code was inspired by "Shader Amortization using Pixel Quad Message +// Passing" by Eric Penner, published in GPU Pro 2, Chapter VI.2. My intent +// is not to plagiarize his fundamentally similar code and assert my own +// copyright, but the algorithmic helper functions require so little code that +// implementations can't vary by much except bugfixes and conventions. I just +// wanted to license my own particular code here to avoid ambiguity and make it +// clear that as far as I'm concerned, people can do as they please with it. + +///////////////////////////////// DESCRIPTION //////////////////////////////// + +// Given screen pixel numbers, derive a "quad vector" describing a fragment's +// position in its 2x2 pixel quad. Given that vector, obtain the values of any +// variable at neighboring fragments. +// Requires: Using this file in general requires: +// 1.) ddx() and ddy() are present in the current Cg profile. +// 2.) The GPU driver is using fine/high-quality derivatives. +// Functions will give incorrect results if this is not true, +// so a test function is included. + + +///////////////////// QUAD-PIXEL COMMUNICATION PRIMITIVES //////////////////// + +float4 get_quad_vector_naive(float4 output_pixel_num_wrt_uvxy) +{ + // Requires: Two measures of the current fragment's output pixel number + // in the range ([0, output_size.x), [0, output_size.y)): + // 1.) output_pixel_num_wrt_uvxy.xy increase with uv coords. + // 2.) output_pixel_num_wrt_uvxy.zw increase with screen xy. + // Returns: Two measures of the fragment's position in its 2x2 quad: + // 1.) The .xy components are its 2x2 placement with respect to + // uv direction (the origin (0, 0) is at the top-left): + // top-left = (-1.0, -1.0) top-right = ( 1.0, -1.0) + // bottom-left = (-1.0, 1.0) bottom-right = ( 1.0, 1.0) + // You need this to arrange/weight shared texture samples. + // 2.) The .zw components are its 2x2 placement with respect to + // screen xy direction (position); the origin varies. + // quad_gather needs this measure to work correctly. + // Note: quad_vector.zw = quad_vector.xy * float2( + // ddx(output_pixel_num_wrt_uvxy.x), + // ddy(output_pixel_num_wrt_uvxy.y)); + // Caveats: This function assumes the GPU driver always starts 2x2 pixel + // quads at even pixel numbers. This assumption can be wrong + // for odd output resolutions (nondeterministically so). + float4 pixel_odd = frac(output_pixel_num_wrt_uvxy * 0.5) * 2.0; + float4 quad_vector = pixel_odd * 2.0 - float4(1.0, 1.0, 1.0, 1.0); + return quad_vector; +} + +float4 get_quad_vector(float4 output_pixel_num_wrt_uvxy) +{ + // Requires: Same as get_quad_vector_naive() (see that first). + // Returns: Same as get_quad_vector_naive() (see that first), but it's + // correct even if the 2x2 pixel quad starts at an odd pixel, + // which can occur at odd resolutions. + float4 quad_vector_guess = + get_quad_vector_naive(output_pixel_num_wrt_uvxy); + // If quad_vector_guess.zw doesn't increase with screen xy, we know + // the 2x2 pixel quad starts at an odd pixel: + float2 odd_start_mirror = 0.5 * float2(ddx(quad_vector_guess.z), + ddy(quad_vector_guess.w)); + return quad_vector_guess * odd_start_mirror.xyxy; +} + +float4 get_quad_vector(float2 output_pixel_num_wrt_uv) +{ + // Requires: 1.) ddx() and ddy() are present in the current Cg profile. + // 2.) output_pixel_num_wrt_uv must increase with uv coords and + // measure the current fragment's output pixel number in: + // ([0, output_size.x), [0, output_size.y)) + // Returns: Same as get_quad_vector_naive() (see that first), but it's + // correct even if the 2x2 pixel quad starts at an odd pixel, + // which can occur at odd resolutions. + // Caveats: This function requires less information than the version + // taking a float4, but it's potentially slower. + // Do screen coords increase with or against uv? Get the direction + // with respect to (uv.x, uv.y) for (screen.x, screen.y) in {-1, 1}. + float2 screen_uv_mirror = float2(ddx(output_pixel_num_wrt_uv.x), + ddy(output_pixel_num_wrt_uv.y)); + float2 pixel_odd_wrt_uv = frac(output_pixel_num_wrt_uv * 0.5) * 2.0; + float2 quad_vector_uv_guess = (pixel_odd_wrt_uv - float2(0.5, 0.5)) * 2.0; + float2 quad_vector_screen_guess = quad_vector_uv_guess * screen_uv_mirror; + // If quad_vector_screen_guess doesn't increase with screen xy, we know + // the 2x2 pixel quad starts at an odd pixel: + float2 odd_start_mirror = 0.5 * float2(ddx(quad_vector_screen_guess.x), + ddy(quad_vector_screen_guess.y)); + float4 quad_vector_guess = float4( + quad_vector_uv_guess, quad_vector_screen_guess); + return quad_vector_guess * odd_start_mirror.xyxy; +} + +void quad_gather(float4 quad_vector, float4 curr, + out float4 adjx, out float4 adjy, out float4 diag) +{ + // Requires: 1.) ddx() and ddy() are present in the current Cg profile. + // 2.) The GPU driver is using fine/high-quality derivatives. + // 3.) quad_vector describes the current fragment's location in + // its 2x2 pixel quad using get_quad_vector()'s conventions. + // 4.) curr is any vector you wish to get neighboring values of. + // Returns: Values of an input vector (curr) at neighboring fragments + // adjacent x, adjacent y, and diagonal (via out parameters). + adjx = curr - ddx(curr) * quad_vector.z; + adjy = curr - ddy(curr) * quad_vector.w; + diag = adjx - ddy(adjx) * quad_vector.w; +} + +void quad_gather(float4 quad_vector, float3 curr, + out float3 adjx, out float3 adjy, out float3 diag) +{ + // Float3 version + adjx = curr - ddx(curr) * quad_vector.z; + adjy = curr - ddy(curr) * quad_vector.w; + diag = adjx - ddy(adjx) * quad_vector.w; +} + +void quad_gather(float4 quad_vector, float2 curr, + out float2 adjx, out float2 adjy, out float2 diag) +{ + // Float2 version + adjx = curr - ddx(curr) * quad_vector.z; + adjy = curr - ddy(curr) * quad_vector.w; + diag = adjx - ddy(adjx) * quad_vector.w; +} + +float4 quad_gather(float4 quad_vector, float curr) +{ + // Float version: + // Returns: return.x == current + // return.y == adjacent x + // return.z == adjacent y + // return.w == diagonal + float4 all = float4(curr, curr, curr, curr); + all.y = all.x - ddx(all.x) * quad_vector.z; + all.zw = all.xy - ddy(all.xy) * quad_vector.w; + return all; +} + +float4 quad_gather_sum(float4 quad_vector, float4 curr) +{ + // Requires: Same as quad_gather() + // Returns: Sum of an input vector (curr) at all fragments in a quad. + float4 adjx, adjy, diag; + quad_gather(quad_vector, curr, adjx, adjy, diag); + return (curr + adjx + adjy + diag); +} + +float3 quad_gather_sum(float4 quad_vector, float3 curr) +{ + // Float3 version: + float3 adjx, adjy, diag; + quad_gather(quad_vector, curr, adjx, adjy, diag); + return (curr + adjx + adjy + diag); +} + +float2 quad_gather_sum(float4 quad_vector, float2 curr) +{ + // Float2 version: + float2 adjx, adjy, diag; + quad_gather(quad_vector, curr, adjx, adjy, diag); + return (curr + adjx + adjy + diag); +} + +float quad_gather_sum(float4 quad_vector, float curr) +{ + // Float version: + float4 all_values = quad_gather(quad_vector, curr); + return (all_values.x + all_values.y + all_values.z + all_values.w); +} + +bool fine_derivatives_working(float4 quad_vector, float4 curr) +{ + // Requires: 1.) ddx() and ddy() are present in the current Cg profile. + // 2.) quad_vector describes the current fragment's location in + // its 2x2 pixel quad using get_quad_vector()'s conventions. + // 3.) curr must be a test vector with non-constant derivatives + // (its value should change nonlinearly across fragments). + // Returns: true if fine/hybrid/high-quality derivatives are used, or + // false if coarse derivatives are used or inconclusive + // Usage: Test whether quad-pixel communication is working! + // Method: We can confirm fine derivatives are used if the following + // holds (ever, for any value at any fragment): + // (ddy(curr) != ddy(adjx)) or (ddx(curr) != ddx(adjy)) + // The more values we test (e.g. test a float4 two ways), the + // easier it is to demonstrate fine derivatives are working. + // TODO: Check for floating point exact comparison issues! + float4 ddx_curr = ddx(curr); + float4 ddy_curr = ddy(curr); + float4 adjx = curr - ddx_curr * quad_vector.z; + float4 adjy = curr - ddy_curr * quad_vector.w; + bool ddy_different = any(bool4(ddy_curr.x != ddy(adjx).x, ddy_curr.y != ddy(adjx).y, ddy_curr.z != ddy(adjx).z, ddy_curr.w != ddy(adjx).w)); + bool ddx_different = any(bool4(ddx_curr.x != ddx(adjy).x, ddx_curr.y != ddx(adjy).y, ddx_curr.z != ddx(adjy).z, ddx_curr.w != ddx(adjy).w)); + return any(bool2(ddy_different, ddx_different)); +} + +bool fine_derivatives_working_fast(float4 quad_vector, float curr) +{ + // Requires: Same as fine_derivatives_working() + // Returns: Same as fine_derivatives_working() + // Usage: This is faster than fine_derivatives_working() but more + // likely to return false negatives, so it's less useful for + // offline testing/debugging. It's also useless as the basis + // for dynamic runtime branching as of May 2014: Derivatives + // (and quad-pixel communication) are currently disallowed in + // branches. However, future GPU's may allow you to use them + // in dynamic branches if you promise the branch condition + // evaluates the same for every fragment in the quad (and/or if + // the driver enforces that promise by making a single fragment + // control branch decisions). If that ever happens, this + // version may become a more economical choice. + float ddx_curr = ddx(curr); + float ddy_curr = ddy(curr); + float adjx = curr - ddx_curr * quad_vector.z; + return (ddy_curr != ddy(adjx)); +} + +#endif // _QUAD_PIXEL_COMMUNICATION_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/scanline-functions.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/scanline-functions.fxh new file mode 100644 index 000000000..9f796c59c --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/scanline-functions.fxh @@ -0,0 +1,501 @@ +#ifndef _SCANLINE_FUNCTIONS_H +#define _SCANLINE_FUNCTIONS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +/////////////////////////////// BEGIN INCLUDES /////////////////////////////// + +#include "bind-shader-params.fxh" +#include "gamma-management.fxh" +#include "special-functions.fxh" + +//////////////////////////////// END INCLUDES //////////////////////////////// + +///////////////////////////// SCANLINE FUNCTIONS ///////////////////////////// + +float2 round_coord( + const float2 c, + const float2 starting_position, + const float2 bin_size +) { + const float2 adj_c = c - starting_position; + return c - fmod(adj_c, bin_size) + bin_size * 0.5; +} + + +// Use preproc defs for these, so they work for arbitrary choices of float1/2/3/4 +#define triangle_wave(t, f) abs(1 - 2*frac((t) * (f))) + +#define sawtooth_incr_wave(t, f) frac((t) * (f)) + +// using fmod(-t*f, 1.0) outputs 0 at t == 0, but I want it to output 1 +#define sawtooth_decr_wave(t, f) 1 - frac((t) * (f)) + + +struct InterpolationFieldData { + float triangle_wave_freq; + bool field_parity; + bool scanline_parity; + bool wrong_field; +}; + +InterpolationFieldData precalc_interpolation_field_data(float2 texcoord) { + InterpolationFieldData data; + + data.triangle_wave_freq = 2; + + const float field_wave = triangle_wave(texcoord.y + rcp(2*data.triangle_wave_freq), data.triangle_wave_freq * 0.5) * 2 - 1; + data.scanline_parity = field_wave >= 0; + + return data; +} + +InterpolationFieldData calc_interpolation_field_data(float2 texcoord, float scale) { + InterpolationFieldData data; + + data.triangle_wave_freq = scale * rcp(scanline_thickness); + // data.triangle_wave_freq = content_size.y * rcp(scanline_thickness); + + const bool frame_count_parity = (frame_count % 2 == 1) && (scanline_deinterlacing_mode != 1); + data.field_parity = (frame_count_parity && !interlace_back_field_first) || (!frame_count_parity && interlace_back_field_first); + + const float field_wave = triangle_wave(texcoord.y + rcp(2*data.triangle_wave_freq), data.triangle_wave_freq * 0.5) * 2 - 1; + data.scanline_parity = field_wave >= 0; + + const bool wrong_field_raw = (data.scanline_parity && !data.field_parity) || (!data.scanline_parity && data.field_parity); + data.wrong_field = enable_interlacing && wrong_field_raw; + + return data; +} + +float get_gaussian_sigma(const float color, const float sigma_range) +{ + // Requires: Globals: + // 1.) gaussian_beam_min_sigma and gaussian_beam_max_sigma are global floats + // containing the desired minimum and maximum beam standard + // deviations, for dim and bright colors respectively. + // 2.) gaussian_beam_max_sigma must be > 0.0 + // 3.) gaussian_beam_min_sigma must be in (0.0, gaussian_beam_max_sigma] + // 4.) gaussian_beam_spot_power must be defined as a global float. + // Parameters: + // 1.) color is the underlying source color along a scanline + // 2.) sigma_range = gaussian_beam_max_sigma - gaussian_beam_min_sigma; we take + // sigma_range as a parameter to avoid repeated computation + // when beam_{min, max}_sigma are runtime shader parameters + // Optional: Users may set beam_spot_shape_function to 1 to define the + // inner f(color) subfunction (see below) as: + // f(color) = sqrt(1.0 - (color - 1.0)*(color - 1.0)) + // Otherwise (technically, if beam_spot_shape_function < 0.5): + // f(color) = pow(color, gaussian_beam_spot_power) + // Returns: The standard deviation of the Gaussian beam for "color:" + // sigma = gaussian_beam_min_sigma + sigma_range * f(color) + // Details/Discussion: + // The beam's spot shape vaguely resembles an aspect-corrected f() in the + // range [0, 1] (not quite, but it's related). f(color) = color makes + // spots look like diamonds, and a spherical function or cube balances + // between variable width and a soft/realistic shape. A gaussian_beam_spot_power + // > 1.0 can produce an ugly spot shape and more initial clipping, but the + // final shape also differs based on the horizontal resampling filter and + // the phosphor bloom. For instance, resampling horizontally in nonlinear + // light and/or with a sharp (e.g. Lanczos) filter will sharpen the spot + // shape, but a sixth root is still quite soft. A power function (default + // 1.0/3.0 gaussian_beam_spot_power) is most flexible, but a fixed spherical curve + // has the highest variability without an awful spot shape. + // + // gaussian_beam_min_sigma affects scanline sharpness/aliasing in dim areas, and its + // difference from gaussian_beam_max_sigma affects beam width variability. It only + // affects clipping [for pure Gaussians] if gaussian_beam_spot_power > 1.0 (which is + // a conservative estimate for a more complex constraint). + // + // gaussian_beam_max_sigma affects clipping and increasing scanline width/softness + // as color increases. The wider this is, the more scanlines need to be + // evaluated to avoid distortion. For a pure Gaussian, the max_beam_sigma + // at which the first unused scanline always has a weight < 1.0/255.0 is: + // num scanlines = 2, max_beam_sigma = 0.2089; distortions begin ~0.34 + // num scanlines = 3, max_beam_sigma = 0.3879; distortions begin ~0.52 + // num scanlines = 4, max_beam_sigma = 0.5723; distortions begin ~0.70 + // num scanlines = 5, max_beam_sigma = 0.7591; distortions begin ~0.89 + // num scanlines = 6, max_beam_sigma = 0.9483; distortions begin ~1.08 + // Generalized Gaussians permit more leeway here as steepness increases. + if(beam_spot_shape_function < 0.5) + { + // Use a power function: + return gaussian_beam_min_sigma + sigma_range * pow(color, gaussian_beam_spot_power); + } + else + { + // Use a spherical function: + const float color_minus_1 = color - 1; + return gaussian_beam_min_sigma + sigma_range * sqrt(1.0 - color_minus_1*color_minus_1); + } +} + +float get_generalized_gaussian_beta(const float color, const float shape_range) +{ + // Requires: Globals: + // 1.) gaussian_beam_min_shape and gaussian_beam_max_shape are global floats + // containing the desired min/max generalized Gaussian + // beta parameters, for dim and bright colors respectively. + // 2.) gaussian_beam_max_shape must be >= 2.0 + // 3.) gaussian_beam_min_shape must be in [2.0, gaussian_beam_max_shape] + // 4.) gaussian_beam_shape_power must be defined as a global float. + // Parameters: + // 1.) color is the underlying source color along a scanline + // 2.) shape_range = gaussian_beam_max_shape - gaussian_beam_min_shape; we take + // shape_range as a parameter to avoid repeated computation + // when beam_{min, max}_shape are runtime shader parameters + // Returns: The type-I generalized Gaussian "shape" parameter beta for + // the given color. + // Details/Discussion: + // Beta affects the scanline distribution as follows: + // a.) beta < 2.0 narrows the peak to a spike with a discontinuous slope + // b.) beta == 2.0 just degenerates to a Gaussian + // c.) beta > 2.0 flattens and widens the peak, then drops off more steeply + // than a Gaussian. Whereas high sigmas widen and soften peaks, high + // beta widen and sharpen peaks at the risk of aliasing. + // Unlike high gaussian_beam_spot_powers, high gaussian_beam_shape_powers actually soften shape + // transitions, whereas lower ones sharpen them (at the risk of aliasing). + return gaussian_beam_min_shape + shape_range * pow(color, gaussian_beam_shape_power); +} + +float3 get_raw_interpolated_color(const float3 color0, + const float3 color1, const float3 color2, const float3 color3, + const float4 weights) +{ + // Use max to avoid bizarre artifacts from negative colors: + const float4x3 mtrx = float4x3(color0, color1, color2, color3); + const float3 m = mul(weights, mtrx); + return max(m, 0.0); +} + +float3 get_interpolated_linear_color(const float3 color0, const float3 color1, + const float3 color2, const float3 color3, const float4 weights) +{ + // Requires: 1.) Requirements of include/gamma-management.h must be met: + // intermediate_gamma must be globally defined, and input + // colors are interpreted as linear RGB unless you #define + // GAMMA_ENCODE_EVERY_FBO (in which case they are + // interpreted as gamma-encoded with intermediate_gamma). + // 2.) color0-3 are colors sampled from a texture with tex2D(). + // They are interpreted as defined in requirement 1. + // 3.) weights contains weights for each color, summing to 1.0. + // 4.) beam_horiz_linear_rgb_weight must be defined as a global + // float in [0.0, 1.0] describing how much blending should + // be done in linear RGB (rest is gamma-corrected RGB). + // 5.) _RUNTIME_SCANLINES_HORIZ_FILTER_COLORSPACE must be #defined + // if beam_horiz_linear_rgb_weight is anything other than a + // static constant, or we may try branching at runtime + // without dynamic branches allowed (slow). + // Returns: Return an interpolated color lookup between the four input + // colors based on the weights in weights. The final color will + // be a linear RGB value, but the blending will be done as + // indicated above. + const float intermediate_gamma = get_intermediate_gamma(); + const float inv_intermediate_gamma = 1.0 / intermediate_gamma; + // Branch if beam_horiz_linear_rgb_weight is static (for free) or if the + // profile allows dynamic branches (faster than computing extra pows): + #if !_RUNTIME_SCANLINES_HORIZ_FILTER_COLORSPACE + #define SCANLINES_BRANCH_FOR_LINEAR_RGB_WEIGHT + #else + #if _DRIVERS_ALLOW_DYNAMIC_BRANCHES + #define SCANLINES_BRANCH_FOR_LINEAR_RGB_WEIGHT + #endif + #endif + #ifdef SCANLINES_BRANCH_FOR_LINEAR_RGB_WEIGHT + // beam_horiz_linear_rgb_weight is static, so we can branch: + #ifdef GAMMA_ENCODE_EVERY_FBO + const float3 gamma_mixed_color = pow( + get_raw_interpolated_color(color0, color1, color2, color3, weights), + intermediate_gamma); + if(beam_horiz_linear_rgb_weight > 0.0) + { + const float3 linear_mixed_color = get_raw_interpolated_color( + pow(color0, intermediate_gamma), + pow(color1, intermediate_gamma), + pow(color2, intermediate_gamma), + pow(color3, intermediate_gamma), + weights); + return lerp(gamma_mixed_color, linear_mixed_color, beam_horiz_linear_rgb_weight); + } + else + { + return gamma_mixed_color; + } + #else + const float3 linear_mixed_color = get_raw_interpolated_color( + color0, color1, color2, color3, weights); + if(beam_horiz_linear_rgb_weight < 1.0) + { + const float3 gamma_mixed_color = get_raw_interpolated_color( + pow(color0, inv_intermediate_gamma), + pow(color1, inv_intermediate_gamma), + pow(color2, inv_intermediate_gamma), + pow(color3, inv_intermediate_gamma), + weights); + return lerp(gamma_mixed_color, linear_mixed_color, beam_horiz_linear_rgb_weight); + } + else + { + return linear_mixed_color; + } + #endif // GAMMA_ENCODE_EVERY_FBO + #else + #ifdef GAMMA_ENCODE_EVERY_FBO + // Inputs: color0-3 are colors in gamma-encoded RGB. + const float3 gamma_mixed_color = pow(get_raw_interpolated_color( + color0, color1, color2, color3, weights), intermediate_gamma); + const float3 linear_mixed_color = get_raw_interpolated_color( + pow(color0, intermediate_gamma), + pow(color1, intermediate_gamma), + pow(color2, intermediate_gamma), + pow(color3, intermediate_gamma), + weights); + return lerp(gamma_mixed_color, linear_mixed_color, beam_horiz_linear_rgb_weight); + #else + // Inputs: color0-3 are colors in linear RGB. + const float3 linear_mixed_color = get_raw_interpolated_color( + color0, color1, color2, color3, weights); + const float3 gamma_mixed_color = get_raw_interpolated_color( + pow(color0, inv_intermediate_gamma), + pow(color1, inv_intermediate_gamma), + pow(color2, inv_intermediate_gamma), + pow(color3, inv_intermediate_gamma), + weights); + // wtf fixme +// const float beam_horiz_linear_rgb_weight1 = 1.0; + return lerp(gamma_mixed_color, linear_mixed_color, + beam_horiz_linear_rgb_weight); + #endif // GAMMA_ENCODE_EVERY_FBO + #endif // SCANLINES_BRANCH_FOR_LINEAR_RGB_WEIGHT +} + +float3 get_scanline_color(const sampler2D tex, const float2 scanline_uv, + const float2 uv_step_x, const float4 weights) +{ + // Requires: 1.) scanline_uv must be vertically snapped to the caller's + // desired line or scanline and horizontally snapped to the + // texel just left of the output pixel (color1) + // 2.) uv_step_x must contain the horizontal uv distance + // between texels. + // 3.) weights must contain interpolation filter weights for + // color0, color1, color2, and color3, where color1 is just + // left of the output pixel. + // Returns: Return a horizontally interpolated texture lookup using 2-4 + // nearby texels, according to weights and the conventions of + // get_interpolated_linear_color(). + // We can ignore the outside texture lookups for Quilez resampling. + const float3 color1 = tex2D_linearize(tex, scanline_uv, get_input_gamma()).rgb; + const float3 color2 = tex2D_linearize(tex, scanline_uv + uv_step_x, get_input_gamma()).rgb; + float3 color0 = float3(0.0, 0.0, 0.0); + float3 color3 = float3(0.0, 0.0, 0.0); + if(beam_horiz_filter > 0.5) + { + color0 = tex2D_linearize(tex, scanline_uv - uv_step_x, get_input_gamma()).rgb; + color3 = tex2D_linearize(tex, scanline_uv + 2.0 * uv_step_x, get_input_gamma()).rgb; + } + // Sample the texture as-is, whether it's linear or gamma-encoded: + // get_interpolated_linear_color() will handle the difference. + return get_interpolated_linear_color(color0, color1, color2, color3, weights); +} + +float3 sample_single_scanline_horizontal(const sampler2D tex, + const float2 tex_uv, const float2 tex_size, + const float2 texture_size_inv) +{ + // TODO: Add function requirements. + // Snap to the previous texel and get sample dists from 2/4 nearby texels: + const float2 curr_texel = tex_uv * tex_size; + // Use under_half to fix a rounding bug right around exact texel locations. + const float2 prev_texel = floor(curr_texel - under_half) + 0.5; + const float2 prev_texel_hor = float2(prev_texel.x, curr_texel.y); + const float2 prev_texel_hor_uv = prev_texel_hor * texture_size_inv; + const float prev_dist = curr_texel.x - prev_texel_hor.x; + const float4 sample_dists = float4(1.0 + prev_dist, prev_dist, + 1.0 - prev_dist, 2.0 - prev_dist); + // Get Quilez, Lanczos2, or Gaussian resize weights for 2/4 nearby texels: + float4 weights; + if (beam_horiz_filter < 0.5) { + // None: + weights = float4(0, 1, 0, 0); + } + else if(beam_horiz_filter < 1.5) + { + // Quilez: + const float x = sample_dists.y; + const float w2 = x*x*x*(x*(x*6.0 - 15.0) + 10.0); + weights = float4(0.0, 1.0 - w2, w2, 0.0); + } + else if(beam_horiz_filter < 2.5) + { + // Gaussian: + float inner_denom_inv = 1.0/(2.0*beam_horiz_sigma*beam_horiz_sigma); + weights = exp(-(sample_dists*sample_dists)*inner_denom_inv); + } + else + { + // Lanczos2: + const float4 pi_dists = FIX_ZERO(sample_dists * pi); + weights = 2.0 * sin(pi_dists) * sin(pi_dists * 0.5) / + (pi_dists * pi_dists); + } + // Ensure the weight sum == 1.0: + const float4 final_weights = weights/dot(weights, float4(1.0, 1.0, 1.0, 1.0)); + // Get the interpolated horizontal scanline color: + const float2 uv_step_x = float2(texture_size_inv.x, 0.0); + return get_scanline_color( + tex, prev_texel_hor_uv, uv_step_x, final_weights); +} + +float3 sample_rgb_scanline( + const sampler2D tex, + const float2 tex_uv, const float2 tex_size, + const float2 texture_size_inv +) { + if (beam_misconvergence) { + const float3 convergence_offsets_rgb_x = get_convergence_offsets_x_vector(); + const float3 convergence_offsets_rgb_y = get_convergence_offsets_y_vector(); + + const float3 offset_u_rgb = convergence_offsets_rgb_x * texture_size_inv.x; + const float3 offset_v_rgb = convergence_offsets_rgb_y * texture_size_inv.y; + + const float2 scanline_uv_r = tex_uv - float2(offset_u_rgb.r, offset_v_rgb.r); + const float2 scanline_uv_g = tex_uv - float2(offset_u_rgb.g, offset_v_rgb.g); + const float2 scanline_uv_b = tex_uv - float2(offset_u_rgb.b, offset_v_rgb.b); + + /**/ + const float4 sample_r = tex2D(tex, scanline_uv_r); + const float4 sample_g = tex2D(tex, scanline_uv_g); + const float4 sample_b = tex2D(tex, scanline_uv_b); + /**/ + + /* + const float3 sample_r = sample_single_scanline_horizontal( + tex, scanline_uv_r, tex_size, texture_size_inv); + const float3 sample_g = sample_single_scanline_horizontal( + tex, scanline_uv_g, tex_size, texture_size_inv); + const float3 sample_b = sample_single_scanline_horizontal( + tex, scanline_uv_b, tex_size, texture_size_inv); + */ + + return float3(sample_r.r, sample_g.g, sample_b.b); + } + else { + // return tex2D(tex, tex_uv).rgb; + return sample_single_scanline_horizontal(tex, tex_uv, tex_size, texture_size_inv); + } +} + +float3 sample_rgb_scanline_horizontal(const sampler2D tex, + const float2 tex_uv, const float2 tex_size, + const float2 texture_size_inv) +{ + // TODO: Add function requirements. + // Rely on a helper to make convergence easier. + if(beam_misconvergence) + { + const float3 convergence_offsets_rgb = get_convergence_offsets_x_vector(); + const float3 offset_u_rgb = convergence_offsets_rgb * texture_size_inv.xxx; + const float2 scanline_uv_r = tex_uv - float2(offset_u_rgb.r, 0.0); + const float2 scanline_uv_g = tex_uv - float2(offset_u_rgb.g, 0.0); + const float2 scanline_uv_b = tex_uv - float2(offset_u_rgb.b, 0.0); + const float3 sample_r = sample_single_scanline_horizontal( + tex, scanline_uv_r, tex_size, texture_size_inv); + const float3 sample_g = sample_single_scanline_horizontal( + tex, scanline_uv_g, tex_size, texture_size_inv); + const float3 sample_b = sample_single_scanline_horizontal( + tex, scanline_uv_b, tex_size, texture_size_inv); + return float3(sample_r.r, sample_g.g, sample_b.b); + } + else + { + return sample_single_scanline_horizontal(tex, tex_uv, tex_size, texture_size_inv); + } +} + +float3 get_averaged_scanline_sample( + sampler2D tex, const float2 texcoord, + const float scanline_start_y, const float v_step_y, + const float input_gamma +) { + // Sample `scanline_thickness` vertically-contiguous pixels and average them. + float3 interpolated_line = 0.0; + for (int i = 0; i < scanline_thickness; i++) { + float4 coord = float4(texcoord.x, scanline_start_y + i * v_step_y, 0, 0); + interpolated_line += tex2Dlod_linearize(tex, coord, input_gamma).rgb; + } + interpolated_line /= float(scanline_thickness); + + return interpolated_line; +} + +float get_beam_strength(float dist, float color, + const float sigma_range, const float shape_range) +{ + // entry point in original is scanline_contrib() + // this is based on scanline_gaussian_sampled_contrib() from original + + // See scanline_gaussian_integral_contrib() for detailed comments! + // gaussian sample = 1/(sigma*sqrt(2*pi)) * e**(-(x**2)/(2*sigma**2)) + const float sigma = get_gaussian_sigma(color, sigma_range); + // Avoid repeated divides: + const float sigma_inv = 1.0 / sigma; + const float inner_denom_inv = 0.5 * sigma_inv * sigma_inv; + const float outer_denom_inv = sigma_inv/sqrt(2.0*pi); + + return color*exp(-(dist*dist)*inner_denom_inv)*outer_denom_inv; +} + +float get_gaussian_beam_strength( + float dist, + float color, + const float sigma_range, + const float shape_range +) { + // entry point in original is scanline_contrib() + // this is based on scanline_generalized_gaussian_sampled_contrib() from original + + // See scanline_generalized_gaussian_integral_contrib() for details! + // generalized sample = + // beta/(2*alpha*gamma(1/beta)) * e**(-(|x|/alpha)**beta) + const float alpha = sqrt(2.0) * get_gaussian_sigma(color, sigma_range); + const float beta = get_generalized_gaussian_beta(color, shape_range); + // Avoid repeated divides: + const float alpha_inv = 1.0 / alpha; + const float beta_inv = 1.0 / beta; + const float scale = color * beta * 0.5 * alpha_inv / gamma_impl(beta_inv, beta); + + return scale * exp(-pow(abs(dist*alpha_inv), beta)); +} + +float get_linear_beam_strength( + const float dist, + const float color, + const float num_pixels, + const bool interlaced +) { + const float p = color * (1 - abs(dist)); + return clamp(p, 0, color); +} + + +#endif // _SCANLINE_FUNCTIONS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/special-functions.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/special-functions.fxh new file mode 100644 index 000000000..1808223a7 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/special-functions.fxh @@ -0,0 +1,504 @@ +#ifndef _SPECIAL_FUNCTIONS_H +#define _SPECIAL_FUNCTIONS_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2014 TroggleMonkey +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +///////////////////////////////// DESCRIPTION //////////////////////////////// + +// This file implements the following mathematical special functions: +// 1.) erf() = 2/sqrt(pi) * indefinite_integral(e**(-x**2)) +// 2.) gamma(s), a real-numbered extension of the integer factorial function +// It also implements normalized_ligamma(s, z), a normalized lower incomplete +// gamma function for s < 0.5 only. Both gamma() and normalized_ligamma() can +// be called with an _impl suffix to use an implementation version with a few +// extra precomputed parameters (which may be useful for the caller to reuse). +// See below for details. +// +// Design Rationale: +// Pretty much every line of code in this file is duplicated four times for +// different input types (float4/float3/float2/float). This is unfortunate, +// but Cg doesn't allow function templates. Macros would be far less verbose, +// but they would make the code harder to document and read. I don't expect +// these functions will require a whole lot of maintenance changes unless +// someone ever has need for more robust incomplete gamma functions, so code +// duplication seems to be the lesser evil in this case. + + +/////////////////////////// GAUSSIAN ERROR FUNCTION ////////////////////////// + +float4 erf6(float4 x) +{ + // Requires: x is the standard parameter to erf(). + // Returns: Return an Abramowitz/Stegun approximation of erf(), where: + // erf(x) = 2/sqrt(pi) * integral(e**(-x**2)) + // This approximation has a max absolute error of 2.5*10**-5 + // with solid numerical robustness and efficiency. See: + // https://en.wikipedia.org/wiki/Error_function#Approximation_with_elementary_functions + const float4 sign_x = sign(x); + const float4 t = 1.0/(1.0 + 0.47047*abs(x)); + const float4 result = 1.0 - t*(0.3480242 + t*(-0.0958798 + t*0.7478556))* + exp(-(x*x)); + return result * sign_x; +} + +float3 erf6(const float3 x) +{ + // Float3 version: + const float3 sign_x = sign(x); + const float3 t = 1.0/(1.0 + 0.47047*abs(x)); + const float3 result = 1.0 - t*(0.3480242 + t*(-0.0958798 + t*0.7478556))* + exp(-(x*x)); + return result * sign_x; +} + +float2 erf6(const float2 x) +{ + // Float2 version: + const float2 sign_x = sign(x); + const float2 t = 1.0/(1.0 + 0.47047*abs(x)); + const float2 result = 1.0 - t*(0.3480242 + t*(-0.0958798 + t*0.7478556))* + exp(-(x*x)); + return result * sign_x; +} + +float erf6(const float x) +{ + // Float version: + const float sign_x = sign(x); + const float t = 1.0/(1.0 + 0.47047*abs(x)); + const float result = 1.0 - t*(0.3480242 + t*(-0.0958798 + t*0.7478556))* + exp(-(x*x)); + return result * sign_x; +} + +float4 erft(const float4 x) +{ + // Requires: x is the standard parameter to erf(). + // Returns: Approximate erf() with the hyperbolic tangent. The error is + // visually noticeable, but it's blazing fast and perceptually + // close...at least on ATI hardware. See: + // http://www.maplesoft.com/applications/view.aspx?SID=5525&view=html + // Warning: Only use this if your hardware drivers correctly implement + // tanh(): My nVidia 8800GTS returns garbage output. + return tanh(1.202760580 * x); +} + +float3 erft(const float3 x) +{ + // Float3 version: + return tanh(1.202760580 * x); +} + +float2 erft(const float2 x) +{ + // Float2 version: + return tanh(1.202760580 * x); +} + +float erft(const float x) +{ + // Float version: + return tanh(1.202760580 * x); +} + +float4 erf(const float4 x) +{ + // Requires: x is the standard parameter to erf(). + // Returns: Some approximation of erf(x), depending on user settings. + #ifdef ERF_FAST_APPROXIMATION + return erft(x); + #else + return erf6(x); + #endif +} + +float3 erf(const float3 x) +{ + // Float3 version: + #ifdef ERF_FAST_APPROXIMATION + return erft(x); + #else + return erf6(x); + #endif +} + +float2 erf(const float2 x) +{ + // Float2 version: + #ifdef ERF_FAST_APPROXIMATION + return erft(x); + #else + return erf6(x); + #endif +} + +float erf(const float x) +{ + // Float version: + #ifdef ERF_FAST_APPROXIMATION + return erft(x); + #else + return erf6(x); + #endif +} + + +/////////////////////////// COMPLETE GAMMA FUNCTION ////////////////////////// + +float4 gamma_impl(const float4 s, const float4 s_inv) +{ + // Requires: 1.) s is the standard parameter to the gamma function, and + // it should lie in the [0, 36] range. + // 2.) s_inv = 1.0/s. This implementation function requires + // the caller to precompute this value, giving users the + // opportunity to reuse it. + // Returns: Return approximate gamma function (real-numbered factorial) + // output using the Lanczos approximation with two coefficients + // calculated using Paul Godfrey's method here: + // http://my.fit.edu/~gabdo/gamma.txt + // An optimal g value for s in [0, 36] is ~1.12906830989, with + // a maximum relative error of 0.000463 for 2**16 equally + // evals. We could use three coeffs (0.0000346 error) without + // hurting latency, but this allows more parallelism with + // outside instructions. + static const float g = 1.12906830989; + static const float c0 = 0.8109119309638332633713423362694399653724431; + static const float c1 = 0.4808354605142681877121661197951496120000040; + static const float e = 2.71828182845904523536028747135266249775724709; + const float4 sph = s + 0.5; + const float4 lanczos_sum = c0 + c1/(s + 1.0); + const float4 base = (sph + g)/e; // or (s + g + float4(0.5))/e + // gamma(s + 1) = base**sph * lanczos_sum; divide by s for gamma(s). + // This has less error for small s's than (s -= 1.0) at the beginning. + return (pow(base, sph) * lanczos_sum) * s_inv; +} + +float3 gamma_impl(const float3 s, const float3 s_inv) +{ + // Float3 version: + static const float g = 1.12906830989; + static const float c0 = 0.8109119309638332633713423362694399653724431; + static const float c1 = 0.4808354605142681877121661197951496120000040; + static const float e = 2.71828182845904523536028747135266249775724709; + const float3 sph = s + 0.5; + const float3 lanczos_sum = c0 + c1/(s + 1.0); + const float3 base = (sph + g)/e; + return (pow(base, sph) * lanczos_sum) * s_inv; +} + +float2 gamma_impl(const float2 s, const float2 s_inv) +{ + // Float2 version: + static const float g = 1.12906830989; + static const float c0 = 0.8109119309638332633713423362694399653724431; + static const float c1 = 0.4808354605142681877121661197951496120000040; + static const float e = 2.71828182845904523536028747135266249775724709; + const float2 sph = s + 0.5; + const float2 lanczos_sum = c0 + c1/(s + 1.0); + const float2 base = (sph + g)/e; + return (pow(base, sph) * lanczos_sum) * s_inv; +} + +float gamma_impl(const float s, const float s_inv) +{ + // Float version: + static const float g = 1.12906830989; + static const float c0 = 0.8109119309638332633713423362694399653724431; + static const float c1 = 0.4808354605142681877121661197951496120000040; + static const float e = 2.71828182845904523536028747135266249775724709; + const float sph = s + 0.5; + const float lanczos_sum = c0 + c1/(s + 1.0); + const float base = (sph + g)/e; + return (pow(base, sph) * lanczos_sum) * s_inv; +} + +float4 gamma(const float4 s) +{ + // Requires: s is the standard parameter to the gamma function, and it + // should lie in the [0, 36] range. + // Returns: Return approximate gamma function output with a maximum + // relative error of 0.000463. See gamma_impl for details. + return gamma_impl(s, 1.0/s); +} + +float3 gamma(const float3 s) +{ + // Float3 version: + return gamma_impl(s, 1.0/s); +} + +float2 gamma(const float2 s) +{ + // Float2 version: + return gamma_impl(s, 1.0/s); +} + +float gamma(const float s) +{ + // Float version: + return gamma_impl(s, 1.0/s); +} + + +//////////////// INCOMPLETE GAMMA FUNCTIONS (RESTRICTED INPUT) /////////////// + +// Lower incomplete gamma function for small s and z (implementation): +float4 ligamma_small_z_impl(const float4 s, const float4 z, const float4 s_inv) +{ + // Requires: 1.) s < ~0.5 + // 2.) z <= ~0.775075 + // 3.) s_inv = 1.0/s (precomputed for outside reuse) + // Returns: A series representation for the lower incomplete gamma + // function for small s and small z (4 terms). + // The actual "rolled up" summation looks like: + // last_sign = 1.0; last_pow = 1.0; last_factorial = 1.0; + // sum = last_sign * last_pow / ((s + k) * last_factorial) + // for(int i = 0; i < 4; ++i) + // { + // last_sign *= -1.0; last_pow *= z; last_factorial *= i; + // sum += last_sign * last_pow / ((s + k) * last_factorial); + // } + // Unrolled, constant-unfolded and arranged for madds and parallelism: + const float4 scale = pow(z, s); + float4 sum = s_inv; // Summation iteration 0 result + // Summation iterations 1, 2, and 3: + const float4 z_sq = z*z; + const float4 denom1 = s + 1.0; + const float4 denom2 = 2.0*s + 4.0; + const float4 denom3 = 6.0*s + 18.0; + //float4 denom4 = 24.0*s + float4(96.0); + sum -= z/denom1; + sum += z_sq/denom2; + sum -= z * z_sq/denom3; + //sum += z_sq * z_sq / denom4; + // Scale and return: + return scale * sum; +} + +float3 ligamma_small_z_impl(const float3 s, const float3 z, const float3 s_inv) +{ + // Float3 version: + const float3 scale = pow(z, s); + float3 sum = s_inv; + const float3 z_sq = z*z; + const float3 denom1 = s + 1.0; + const float3 denom2 = 2.0*s + 4.0; + const float3 denom3 = 6.0*s + 18.0; + sum -= z/denom1; + sum += z_sq/denom2; + sum -= z * z_sq/denom3; + return scale * sum; +} + +float2 ligamma_small_z_impl(const float2 s, const float2 z, const float2 s_inv) +{ + // Float2 version: + const float2 scale = pow(z, s); + float2 sum = s_inv; + const float2 z_sq = z*z; + const float2 denom1 = s + 1.0; + const float2 denom2 = 2.0*s + 4.0; + const float2 denom3 = 6.0*s + 18.0; + sum -= z/denom1; + sum += z_sq/denom2; + sum -= z * z_sq/denom3; + return scale * sum; +} + +float ligamma_small_z_impl(const float s, const float z, const float s_inv) +{ + // Float version: + const float scale = pow(z, s); + float sum = s_inv; + const float z_sq = z*z; + const float denom1 = s + 1.0; + const float denom2 = 2.0*s + 4.0; + const float denom3 = 6.0*s + 18.0; + sum -= z/denom1; + sum += z_sq/denom2; + sum -= z * z_sq/denom3; + return scale * sum; +} + +// Upper incomplete gamma function for small s and large z (implementation): +float4 uigamma_large_z_impl(const float4 s, const float4 z) +{ + // Requires: 1.) s < ~0.5 + // 2.) z > ~0.775075 + // Returns: Gauss's continued fraction representation for the upper + // incomplete gamma function (4 terms). + // The "rolled up" continued fraction looks like this. The denominator + // is truncated, and it's calculated "from the bottom up:" + // denom = float4('inf'); + // float4 one = float4(1.0); + // for(int i = 4; i > 0; --i) + // { + // denom = ((i * 2.0) - one) + z - s + (i * (s - i))/denom; + // } + // Unrolled and constant-unfolded for madds and parallelism: + const float4 numerator = pow(z, s) * exp(-z); + float4 denom = 7.0 + z - s; + denom = 5.0 + z - s + (3.0*s - 9.0)/denom; + denom = 3.0 + z - s + (2.0*s - 4.0)/denom; + denom = 1.0 + z - s + (s - 1.0)/denom; + return numerator / denom; +} + +float3 uigamma_large_z_impl(const float3 s, const float3 z) +{ + // Float3 version: + const float3 numerator = pow(z, s) * exp(-z); + float3 denom = 7.0 + z - s; + denom = 5.0 + z - s + (3.0*s - 9.0)/denom; + denom = 3.0 + z - s + (2.0*s - 4.0)/denom; + denom = 1.0 + z - s + (s - 1.0)/denom; + return numerator / denom; +} + +float2 uigamma_large_z_impl(const float2 s, const float2 z) +{ + // Float2 version: + const float2 numerator = pow(z, s) * exp(-z); + float2 denom = 7.0 + z - s; + denom = 5.0 + z - s + (3.0*s - 9.0)/denom; + denom = 3.0 + z - s + (2.0*s - 4.0)/denom; + denom = 1.0 + z - s + (s - 1.0)/denom; + return numerator / denom; +} + +float uigamma_large_z_impl(const float s, const float z) +{ + // Float version: + const float numerator = pow(z, s) * exp(-z); + float denom = 7.0 + z - s; + denom = 5.0 + z - s + (3.0*s - 9.0)/denom; + denom = 3.0 + z - s + (2.0*s - 4.0)/denom; + denom = 1.0 + z - s + (s - 1.0)/denom; + return numerator / denom; +} + +// Normalized lower incomplete gamma function for small s (implementation): +float4 normalized_ligamma_impl(const float4 s, const float4 z, + const float4 s_inv, const float4 gamma_s_inv) +{ + // Requires: 1.) s < ~0.5 + // 2.) s_inv = 1/s (precomputed for outside reuse) + // 3.) gamma_s_inv = 1/gamma(s) (precomputed for outside reuse) + // Returns: Approximate the normalized lower incomplete gamma function + // for s < 0.5. Since we only care about s < 0.5, we only need + // to evaluate two branches (not four) based on z. Each branch + // uses four terms, with a max relative error of ~0.00182. The + // branch threshold and specifics were adapted for fewer terms + // from Gil/Segura/Temme's paper here: + // http://oai.cwi.nl/oai/asset/20433/20433B.pdf + // Evaluate both branches: Real branches test slower even when available. + static const float thresh = 0.775075; + int4 z_is_large; + z_is_large.x = int(z.x > thresh); + z_is_large.y = int(z.y > thresh); + z_is_large.z = int(z.z > thresh); + z_is_large.w = int(z.w > thresh); + const float4 large_z = 1.0 - uigamma_large_z_impl(s, z) * gamma_s_inv; + const float4 small_z = ligamma_small_z_impl(s, z, s_inv) * gamma_s_inv; + // Combine the results from both branches: + int4 inverse_z_is_large = saturate(~(z_is_large)); + return large_z * float4(z_is_large) + small_z * float4(inverse_z_is_large); +} + +float3 normalized_ligamma_impl(const float3 s, const float3 z, + const float3 s_inv, const float3 gamma_s_inv) +{ + // Float3 version: + static const float thresh = 0.775075; + int3 z_is_large; + z_is_large.x = int(z.x > thresh); + z_is_large.y = int(z.y > thresh); + z_is_large.z = int(z.z > thresh); + const float3 large_z = 1.0 - uigamma_large_z_impl(s, z) * gamma_s_inv; + const float3 small_z = ligamma_small_z_impl(s, z, s_inv) * gamma_s_inv; + int3 inverse_z_is_large = saturate(~(z_is_large)); + return large_z * float3(z_is_large) + small_z * float3(inverse_z_is_large); +} + +float2 normalized_ligamma_impl(const float2 s, const float2 z, + const float2 s_inv, const float2 gamma_s_inv) +{ + // Float2 version: + static const float thresh = 0.775075; + int2 z_is_large; + z_is_large.x = int(z.x > thresh); + z_is_large.y = int(z.y > thresh); + const float2 large_z = 1.0 - uigamma_large_z_impl(s, z) * gamma_s_inv; + const float2 small_z = ligamma_small_z_impl(s, z, s_inv) * gamma_s_inv; + int2 inverse_z_is_large = saturate(~(z_is_large)); + return large_z * float2(z_is_large) + small_z * float2(inverse_z_is_large); +} + +float normalized_ligamma_impl(const float s, const float z, + const float s_inv, const float gamma_s_inv) +{ + // Float version: + static const float thresh = 0.775075; + const bool z_is_large = z > thresh; + const float large_z = 1.0 - uigamma_large_z_impl(s, z) * gamma_s_inv; + const float small_z = ligamma_small_z_impl(s, z, s_inv) * gamma_s_inv; + return large_z * float(z_is_large) + small_z * float(!z_is_large); +} + +// Normalized lower incomplete gamma function for small s: +float4 normalized_ligamma(const float4 s, const float4 z) +{ + // Requires: s < ~0.5 + // Returns: Approximate the normalized lower incomplete gamma function + // for s < 0.5. See normalized_ligamma_impl() for details. + const float4 s_inv = 1.0/s; + const float4 gamma_s_inv = 1.0/gamma_impl(s, s_inv); + return normalized_ligamma_impl(s, z, s_inv, gamma_s_inv); +} + +float3 normalized_ligamma(const float3 s, const float3 z) +{ + // Float3 version: + const float3 s_inv = 1.0/s; + const float3 gamma_s_inv = 1.0/gamma_impl(s, s_inv); + return normalized_ligamma_impl(s, z, s_inv, gamma_s_inv); +} + +float2 normalized_ligamma(const float2 s, const float2 z) +{ + // Float2 version: + const float2 s_inv = 1.0/s; + const float2 gamma_s_inv = 1.0/gamma_impl(s, s_inv); + return normalized_ligamma_impl(s, z, s_inv, gamma_s_inv); +} + +float normalized_ligamma(const float s, const float z) +{ + // Float version: + const float s_inv = 1.0/s; + const float gamma_s_inv = 1.0/gamma_impl(s, s_inv); + return normalized_ligamma_impl(s, z, s_inv, gamma_s_inv); +} + +#endif // _SPECIAL_FUNCTIONS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/tex2Dantialias.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/tex2Dantialias.fxh new file mode 100644 index 000000000..65ea4f04b --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/tex2Dantialias.fxh @@ -0,0 +1,1393 @@ +#ifndef _TEX2DANTIALIAS_H +#define _TEX2DANTIALIAS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +///////////////////////////////// DESCRIPTION //////////////////////////////// + +// This file provides antialiased and subpixel-aware tex2D lookups. +// Requires: All functions share these requirements: +// 1.) All requirements of gamma-management.h must be satisfied! +// 2.) pixel_to_tex_uv must be a 2x2 matrix that transforms pixe- +// space offsets to texture uv offsets. You can get this with: +// const float2 duv_dx = ddx(tex_uv); +// const float2 duv_dy = ddy(tex_uv); +// const float2x2 pixel_to_tex_uv = float2x2( +// duv_dx.x, duv_dy.x, +// duv_dx.y, duv_dy.y); +// This is left to the user in case the current Cg profile +// doesn't support ddx()/ddy(). Ideally, the user could find +// calculate a distorted tangent-space mapping analytically. +// If not, a simple flat mapping can be obtained with: +// const float2 xy_to_uv_scale = IN.output_size * +// IN.video_size/IN.texture_size; +// const float2x2 pixel_to_tex_uv = float2x2( +// xy_to_uv_scale.x, 0.0, +// 0.0, xy_to_uv_scale.y); +// Optional: To set basic AA settings, #define ANTIALIAS_OVERRIDE_BASICS and: +// 1.) Set an antialiasing level: +// static const float antialias_level = {0 (none), +// 1 (sample subpixels), 4, 5, 6, 7, 8, 12, 16, 20, 24} +// 2.) Set a filter type: +// static const float aa_filter = { +// 0 (Box, Separable), 1 (Box, Cylindrical), +// 2 (Tent, Separable), 3 (Tent, Cylindrical) +// 4 (Gaussian, Separable), 5 (Gaussian, Cylindrical) +// 6 (Cubic, Separable), 7 (Cubic, Cylindrical) +// 8 (Lanczos Sinc, Separable), +// 9 (Lanczos Jinc, Cylindrical)} +// If the input is unknown, a separable box filter is used. +// Note: Lanczos Jinc is terrible for sparse sampling, and +// using aa_axis_importance (see below) defeats the purpose. +// 3.) Mirror the sample pattern on odd frames? +// static const bool aa_temporal = {true, false] +// This helps rotational invariance but can look "fluttery." +// The user may #define ANTIALIAS_OVERRIDE_PARAMETERS to override +// (all of) the following default parameters with static or uniform +// constants (or an accessor function for subpixel offsets): +// 1.) Cubic parameters: +// static const float aa_cubic_c = 0.5; +// See http://www.imagemagick.org/Usage/filter/#mitchell +// 2.) Gaussian parameters: +// static const float aa_gauss_sigma = +// 0.5/aa_pixel_diameter; +// 3.) Set subpixel offsets. This requires an accessor function +// for compatibility with scalar runtime shader params. Return +// a float2 pixel offset in [-0.5, 0.5] for the red subpixel: +// float2 get_aa_subpixel_r_offset() +// The user may also #define ANTIALIAS_OVERRIDE_STATIC_CONSTANTS to +// override (all of) the following default static values. However, +// the file's structure requires them to be declared static const: +// 1.) static const float aa_lanczos_lobes = 3.0; +// 2.) static const float aa_gauss_support = 1.0/aa_pixel_diameter; +// Note the default tent/Gaussian support radii may appear +// arbitrary, but extensive testing found them nearly optimal +// for tough cases like strong distortion at low AA levels. +// (The Gaussian default is only best for practical gauss_sigma +// values; much larger gauss_sigmas ironically prefer slightly +// smaller support given sparse sampling, and vice versa.) +// 3.) static const float aa_tent_support = 1.0 / aa_pixel_diameter; +// 4.) static const float2 aa_xy_axis_importance: +// The sparse N-queens sampling grid interacts poorly with +// negative-lobed 2D filters. However, if aliasing is much +// stronger in one direction (e.g. horizontally with a phosphor +// mask), it can be useful to downplay sample offsets along the +// other axis. The support radius in each direction scales with +// aa_xy_axis_importance down to a minimum of 0.5 (box support), +// after which point only the offsets used for calculating +// weights continue to scale downward. This works as follows: +// If aa_xy_axis_importance = float2(1.0, 1.0/support_radius), +// the vertical support radius will drop to 1.0, and we'll just +// filter vertical offsets with the first filter lobe, while +// horizontal offsets go through the full multi-lobe filter. +// If aa_xy_axis_importance = float2(1.0, 0.0), the vertical +// support radius will drop to box support, and the vertical +// offsets will be ignored entirely (essentially giving us a +// box filter vertically). The former is potentially smoother +// (but less predictable) and the default behavior of Lanczos +// jinc, whereas the latter is sharper and the default behavior +// of cubics and Lanczos sinc. +// 5.) static const float aa_pixel_diameter: You can expand the +// pixel diameter to e.g. sqrt(2.0), which may be a better +// support range for cylindrical filters (they don't +// currently discard out-of-circle samples though). +// Finally, there are two miscellaneous options: +// 1.) If you want to antialias a manually tiled texture, you can +// #define ANTIALIAS_DISABLE_ANISOTROPIC to use tex2Dlod() to +// fix incompatibilities with anisotropic filtering. This is +// slower, and the Cg profile must support tex2Dlod(). +// 2.) If aa_cubic_c is a runtime uniform, you can #define +// _RUNTIME_ANTIALIAS_WEIGHTS to evaluate cubic weights once per +// fragment instead of at the usage site (which is used by +// default, because it enables static evaluation). +// Description: +// Each antialiased lookup follows these steps: +// 1.) Define a sample pattern of pixel offsets in the range of [-0.5, 0.5] +// pixels, spanning the diameter of a rectangular box filter. +// 2.) Scale these offsets by the support diameter of the user's chosen filter. +// 3.) Using these pixel offsets from the pixel center, compute the offsets to +// predefined subpixel locations. +// 4.) Compute filter weights based on subpixel offsets. +// Much of that can often be done at compile-time. At runtime: +// 1.) Project pixel-space offsets into uv-space with a matrix multiplication +// to get the uv offsets for each sample. Rectangular pixels have a +// diameter of 1.0. Circular pixels are not currently supported, but they +// might be better with a diameter of sqrt(2.0) to ensure there are no gaps +// between them. +// 2.) Load, weight, and sum samples. +// We use a sparse bilinear sampling grid, so there are two major implications: +// 1.) We can directly project the pixel-space support box into uv-space even +// if we're upsizing. This wouldn't be the case for nearest neighbor, +// where we'd have to expand the uv-space diameter to at least the support +// size to ensure sufficient filter support. In our case, this allows us +// to treat upsizing the same as downsizing and use static weighting. :) +// 2.) For decent results, negative-lobed filters must be computed based on +// separable weights, not radial distances, because the sparse sampling +// makes no guarantees about radial distributions. Even then, it's much +// better to set aa_xy_axis_importance to e.g. float2(1.0, 0.0) to use e.g. +// Lanczos2 horizontally and a box filter vertically. This is mainly due +// to the sparse N-queens sampling and a statistically enormous positive or +// negative covariance between horizontal and vertical weights. +// +// Design Decision Comments: +// "aa_temporal" mirrors the sample pattern on odd frames along the axis that +// keeps subpixel weights constant. This helps with rotational invariance, but +// it can cause distracting fluctuations, and horizontal and vertical edges +// will look the same. Using a different pattern on a shifted grid would +// exploit temporal AA better, but it would require a dynamic branch or a lot +// of conditional moves, so it's prohibitively slow for the minor benefit. + + +#include "helper-functions-and-macros.fxh" + +///////////////////////////// SETTINGS MANAGEMENT //////////////////////////// + +// #if !ANTIALIAS_OVERRIDE_BASICS +// // The following settings must be static constants: +// static const float antialias_level = 12.0; +// static const float aa_filter = 0.0; +// static const bool aa_temporal = false; +// #endif + +#ifndef ANTIALIAS_OVERRIDE_STATIC_CONSTANTS + // Users may override these parameters, but the file structure requires + // them to be static constants; see the descriptions above. + static const float aa_pixel_diameter = 1.0; + static const float aa_lanczos_lobes = 3.0; + static const float aa_gauss_support = 1.0 / aa_pixel_diameter; + static const float aa_tent_support = 1.0 / aa_pixel_diameter; + + // If we're using a negative-lobed filter, default to using it horizontally + // only, and use only the first lobe vertically or a box filter, over a + // correspondingly smaller range. This compensates for the sparse sampling + // grid's typically large positive/negative x/y covariance. + static const float2 aa_xy_axis_importance = macro_cond( + aa_filter < 5.5, + float2(1.0, 1.0), // Box, tent, Gaussian + macro_cond( + aa_filter < 8.5, + float2(1.0, 0.0), // Cubic and Lanczos sinc + macro_cond( + aa_filter < 9.5, + float2(1.0, 1.0/aa_lanczos_lobes), // Lanczos jinc + float2(1.0, 1.0) // Default to box + ) + ) + ); +#endif + +// #if !ANTIALIAS_OVERRIDE_PARAMETERS +// // Users may override these values with their own uniform or static consts. +// // Cubics: See http://www.imagemagick.org/Usage/filter/#mitchell +// // 1.) "Keys cubics" with B = 1 - 2C are considered the highest quality. +// // 2.) C = 0.5 (default) is Catmull-Rom; higher C's apply sharpening. +// // 3.) C = 1.0/3.0 is the Mitchell-Netravali filter. +// // 4.) C = 0.0 is a soft spline filter. +// static const float aa_cubic_c = 0.5; +// static const float aa_gauss_sigma = 0.5 / aa_pixel_diameter; +// // Users may override the subpixel offset accessor function with their own. +// // A function is used for compatibility with scalar runtime shader params. +// float2 get_aa_subpixel_r_offset() +// { +// return float2(0.0, 0.0); +// } +// #endif + + +////////////////////////////////// INCLUDES ////////////////////////////////// + +#include "gamma-management.fxh" + + +////////////////////////////////// CONSTANTS ///////////////////////////////// + +static const float aa_box_support = 0.5; +static const float aa_cubic_support = 2.0; + + +//////////////////////////// GLOBAL NON-CONSTANTS //////////////////////////// + + +// We'll want to define these only once per fragment at most. +// Compute cubic coefficients on demand at runtime, and save them to global +// uniforms. The B parameter is computed from C, because "Keys cubics" +// with B = 1 - 2C are considered the highest quality. +static const float aa_cubic_b = 1.0 - 2.0*aa_cubic_c; +static const float cubic_branch1_x3_coeff = 12.0 - 9.0*aa_cubic_b - 6.0*aa_cubic_c; +static const float cubic_branch1_x2_coeff = -18.0 + 12.0*aa_cubic_b + 6.0*aa_cubic_c; +static const float cubic_branch1_x0_coeff = 6.0 - 2.0 * aa_cubic_b; +static const float cubic_branch2_x3_coeff = -aa_cubic_b - 6.0 * aa_cubic_c; +static const float cubic_branch2_x2_coeff = 6.0*aa_cubic_b + 30.0*aa_cubic_c; +static const float cubic_branch2_x1_coeff = -12.0*aa_cubic_b - 48.0*aa_cubic_c; +static const float cubic_branch2_x0_coeff = 8.0*aa_cubic_b + 24.0*aa_cubic_c; + + +/////////////////////////////////// HELPERS ////////////////////////////////// + +// In the RetroArch version, we can optionally implement aa_cubic_c as a uniform. +// I've disabled that for now because the asssociated mutable singleton mess was a +// pain to port. So for now, this function does absolutely nothing. Maybe I'll reintroduce +// the uniform implementation later, but for now the statics will do. +void assign_aa_cubic_constants() +{ + return; +} + +float4 get_subpixel_support_diam_and_final_axis_importance() +{ + // Statically select the base support radius: + static const float base_support_radius = + aa_filter < 1.5 ? aa_box_support : + aa_filter < 3.5 ? aa_tent_support : + aa_filter < 5.5 ? aa_gauss_support : + aa_filter < 7.5 ? aa_cubic_support : + aa_filter < 9.5 ? aa_lanczos_lobes : + aa_box_support; // Default to box + // Expand the filter support for subpixel filtering. + const float2 subpixel_support_radius_raw = + float2(base_support_radius, base_support_radius) + abs(get_aa_subpixel_r_offset()); + if(aa_filter < 1.5) + { + // Ignore aa_xy_axis_importance for box filtering. + const float2 subpixel_support_diam = + 2.0 * subpixel_support_radius_raw; + const float2 final_axis_importance = float2(1.0, 1.0); + return float4(subpixel_support_diam, final_axis_importance); + } + else + { + // Scale the support window by aa_xy_axis_importance, but don't narrow + // it further than box support. This allows decent vertical AA without + // messing up horizontal weights or using something silly like Lanczos4 + // horizontally with a huge vertical average over an 8-pixel radius. + const float2 subpixel_support_radius = max(float2(aa_box_support, aa_box_support), + subpixel_support_radius_raw * aa_xy_axis_importance); + // Adjust aa_xy_axis_importance to compensate for what's already done: + const float2 final_axis_importance = aa_xy_axis_importance * + subpixel_support_radius_raw/subpixel_support_radius; + const float2 subpixel_support_diam = 2.0 * subpixel_support_radius; + return float4(subpixel_support_diam, final_axis_importance); + } +} + + +/////////////////////////// FILTER WEIGHT FUNCTIONS ////////////////////////// + +float eval_box_filter(const float dist) +{ + return float(abs(dist) <= aa_box_support); +} + +float eval_separable_box_filter(const float2 offset) +{ + return float(all(bool2((abs(offset.x) <= aa_box_support), (abs(offset.y) <= aa_box_support)))); +} + +float eval_tent_filter(const float dist) +{ + return saturate((aa_tent_support - dist) / aa_tent_support); +} + +float eval_gaussian_filter(const float dist) +{ + return exp(-(dist*dist) / (2.0*aa_gauss_sigma*aa_gauss_sigma)); +} + +float eval_cubic_filter(const float dist) +{ + // Compute coefficients like assign_aa_cubic_constants(), but statically. + #if _RUNTIME_ANTIALIAS_WEIGHTS + // When runtime weights are used, these values are instead written to + // global uniforms at the beginning of each tex2Daa* call. + const float aa_cubic_b = 1.0 - 2.0*aa_cubic_c; + const float cubic_branch1_x3_coeff = 12.0 - 9.0*aa_cubic_b - 6.0*aa_cubic_c; + const float cubic_branch1_x2_coeff = -18.0 + 12.0*aa_cubic_b + 6.0*aa_cubic_c; + const float cubic_branch1_x0_coeff = 6.0 - 2.0 * aa_cubic_b; + const float cubic_branch2_x3_coeff = -aa_cubic_b - 6.0 * aa_cubic_c; + const float cubic_branch2_x2_coeff = 6.0*aa_cubic_b + 30.0*aa_cubic_c; + const float cubic_branch2_x1_coeff = -12.0*aa_cubic_b - 48.0*aa_cubic_c; + const float cubic_branch2_x0_coeff = 8.0*aa_cubic_b + 24.0*aa_cubic_c; + #endif + const float abs_dist = abs(dist); + // Compute the cubic based on the Horner's method formula in: + // http://www.cs.utexas.edu/users/fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf + return (abs_dist < 1.0 ? + (cubic_branch1_x3_coeff*abs_dist + + cubic_branch1_x2_coeff)*abs_dist*abs_dist + + cubic_branch1_x0_coeff : + abs_dist < 2.0 ? + ((cubic_branch2_x3_coeff*abs_dist + + cubic_branch2_x2_coeff)*abs_dist + + cubic_branch2_x1_coeff)*abs_dist + cubic_branch2_x0_coeff : + 0.0)/6.0; +} + +float eval_separable_cubic_filter(const float2 offset) +{ + // This is faster than using a specific float2 version: + return eval_cubic_filter(offset.x) * + eval_cubic_filter(offset.y); +} + +float2 eval_sinc_filter(const float2 offset) +{ + // It's faster to let the caller handle the zero case, or at least it + // was when I used macros and the shader preset took a full minute to load. + const float2 pi_offset = pi * offset; + return sin(pi_offset)/pi_offset; +} + +float eval_separable_lanczos_sinc_filter(const float2 offset_unsafe) +{ + // Note: For sparse sampling, you really need to pick an axis to use + // Lanczos along (e.g. set aa_xy_axis_importance = float2(1.0, 0.0)). + const float2 offset = FIX_ZERO(offset_unsafe); + const float2 xy_weights = eval_sinc_filter(offset) * + eval_sinc_filter(offset/aa_lanczos_lobes); + return xy_weights.x * xy_weights.y; +} + +float eval_jinc_filter_unorm(const float x) +{ + // This is a Jinc approximation for x in [0, 45). We'll use x in range + // [0, 4*pi) or so. There are faster/closer approximations based on + // piecewise cubics from [0, 45) and asymptotic approximations beyond that, + // but this has a maximum absolute error < 1/512, and it's simpler/faster + // for shaders...not that it's all that useful for sparse sampling anyway. + const float point3845_x = 0.38448566093564*x; + const float exp_term = exp(-(point3845_x*point3845_x)); + const float point8154_plus_x = 0.815362332840791 + x; + const float cos_term = cos(point8154_plus_x); + return ( + 0.0264727330997042*min(x, 6.83134964622778) + + 0.680823557250528*exp_term + + -0.0597255978950933*min(7.41043194481873, x)*cos_term / + (point8154_plus_x + 0.0646074538634482*(x*x) + + cos(x)*max(exp_term, cos(x) + cos_term)) - + 0.180837503591406); +} + +float eval_jinc_filter(const float dist) +{ + return eval_jinc_filter_unorm(pi * dist); +} + +float eval_lanczos_jinc_filter(const float dist) +{ + return eval_jinc_filter(dist) * eval_jinc_filter(dist/aa_lanczos_lobes); +} + + +float3 eval_unorm_rgb_weights(const float2 offset, + const float2 final_axis_importance) +{ + // Requires: 1.) final_axis_impportance must be computed according to + // get_subpixel_support_diam_and_final_axis_importance(). + // 2.) aa_filter must be a global constant. + // 3.) offset must be an xy pixel offset in the range: + // ([-subpixel_support_diameter.x/2, + // subpixel_support_diameter.x/2], + // [-subpixel_support_diameter.y/2, + // subpixel_support_diameter.y/2]) + // Returns: Sample weights at R/G/B destination subpixels for the + // given xy pixel offset. + const float2 offset_g = offset * final_axis_importance; + const float2 aa_r_offset = get_aa_subpixel_r_offset(); + const float2 offset_r = offset_g - aa_r_offset * final_axis_importance; + const float2 offset_b = offset_g + aa_r_offset * final_axis_importance; + // Statically select a filter: + if(aa_filter < 0.5) + { + return float3(eval_separable_box_filter(offset_r), + eval_separable_box_filter(offset_g), + eval_separable_box_filter(offset_b)); + } + else if(aa_filter < 1.5) + { + return float3(eval_box_filter(length(offset_r)), + eval_box_filter(length(offset_g)), + eval_box_filter(length(offset_b))); + } + else if(aa_filter < 2.5) + { + return float3( + eval_tent_filter(offset_r.x) * eval_tent_filter(offset_r.y), + eval_tent_filter(offset_g.x) * eval_tent_filter(offset_g.y), + eval_tent_filter(offset_b.x) * eval_tent_filter(offset_b.y)); + } + else if(aa_filter < 3.5) + { + return float3(eval_tent_filter(length(offset_r)), + eval_tent_filter(length(offset_g)), + eval_tent_filter(length(offset_b))); + } + else if(aa_filter < 4.5) + { + return float3( + eval_gaussian_filter(offset_r.x) * eval_gaussian_filter(offset_r.y), + eval_gaussian_filter(offset_g.x) * eval_gaussian_filter(offset_g.y), + eval_gaussian_filter(offset_b.x) * eval_gaussian_filter(offset_b.y)); + } + else if(aa_filter < 5.5) + { + return float3(eval_gaussian_filter(length(offset_r)), + eval_gaussian_filter(length(offset_g)), + eval_gaussian_filter(length(offset_b))); + } + else if(aa_filter < 6.5) + { + return float3( + eval_cubic_filter(offset_r.x) * eval_cubic_filter(offset_r.y), + eval_cubic_filter(offset_g.x) * eval_cubic_filter(offset_g.y), + eval_cubic_filter(offset_b.x) * eval_cubic_filter(offset_b.y)); + } + else if(aa_filter < 7.5) + { + return float3(eval_cubic_filter(length(offset_r)), + eval_cubic_filter(length(offset_g)), + eval_cubic_filter(length(offset_b))); + } + else if(aa_filter < 8.5) + { + return float3(eval_separable_lanczos_sinc_filter(offset_r), + eval_separable_lanczos_sinc_filter(offset_g), + eval_separable_lanczos_sinc_filter(offset_b)); + } + else if(aa_filter < 9.5) + { + return float3(eval_lanczos_jinc_filter(length(offset_r)), + eval_lanczos_jinc_filter(length(offset_g)), + eval_lanczos_jinc_filter(length(offset_b))); + } + else + { + // Default to a box, because Lanczos Jinc is so bad. ;) + return float3(eval_separable_box_filter(offset_r), + eval_separable_box_filter(offset_g), + eval_separable_box_filter(offset_b)); + } +} + + +////////////////////////////// HELPER FUNCTIONS ////////////////////////////// + +float4 tex2Daa_tiled_linearize(const sampler2D samp, const float2 s, const float input_gamma) +{ + // If we're manually tiling a texture, anisotropic filtering can get + // confused. This is one workaround: + #ifdef ANTIALIAS_DISABLE_ANISOTROPIC + // TODO: Use tex2Dlod_linearize with a calculated mip level. + return tex2Dlod_linearize(samp, float4(s, 0.0, 0.0), input_gamma); + #else + return tex2D_linearize(samp, s, input_gamma); + #endif +} + +float2 get_frame_sign(const float frame) +{ + if(aa_temporal) + { + // Mirror the sampling pattern for odd frames in a direction that + // lets us keep the same subpixel sample weights: + const float frame_odd = float(fmod(frame, 2.0) > 0.5); + const float2 aa_r_offset = get_aa_subpixel_r_offset(); + const float2 mirror = -float2(abs(aa_r_offset.x) < (FIX_ZERO(0.0)), abs(aa_r_offset.y) < (FIX_ZERO(0.0))); + return mirror; + } + else + { + return float2(1.0, 1.0); + } +} + + +///////////////////////// ANTIALIASED TEXTURE LOOKUPS //////////////////////// + +float3 tex2Daa_subpixel_weights_only(const sampler2D tex, + const float2 tex_uv, const float2x2 pixel_to_tex_uv, const float input_gamma) +{ + // This function is unlike the others: Just perform a single independent + // lookup for each subpixel. It may be very aliased. + const float2 aa_r_offset = get_aa_subpixel_r_offset(); + const float2 aa_r_offset_uv_offset = mul(pixel_to_tex_uv, aa_r_offset); + const float color_g = tex2D_linearize(tex, tex_uv, input_gamma).g; + const float color_r = tex2D_linearize(tex, tex_uv + aa_r_offset_uv_offset, input_gamma).r; + const float color_b = tex2D_linearize(tex, tex_uv - aa_r_offset_uv_offset, input_gamma).b; + return float3(color_r, color_g, color_b); +} + +// The tex2Daa* functions compile very slowly due to all the macros and +// compile-time math, so only include the ones we'll actually use! +float3 tex2Daa4x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use an RGMS4 pattern (4-queens): + // . . Q . : off =(-1.5, -1.5)/4 + (2.0, 0.0)/4 + // Q . . . : off =(-1.5, -1.5)/4 + (0.0, 1.0)/4 + // . . . Q : off =(-1.5, -1.5)/4 + (3.0, 2.0)/4 + // . Q . . : off =(-1.5, -1.5)/4 + (1.0, 3.0)/4 + // Static screenspace sample offsets (compute some implicitly): + static const float grid_size = 4.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(2.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(0.0, 1.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = w1.bgr; + const float3 w3 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1; + const float3 w_sum = half_sum + half_sum.bgr; + const float3 w_sum_inv = 1.0/w_sum; + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = pixel_to_tex_uv * aa_pixel_diameter; + // Get uv sample offsets, mirror on odd frames if directed, and exploit + // diagonal symmetry: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * (w0 * sample0 + w1 * sample1 + + w2 * sample2 + w3 * sample3); +} + +float3 tex2Daa5x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use a diagonally symmetric 5-queens pattern: + // . Q . . . : off =(-2.0, -2.0)/5 + (1.0, 0.0)/5 + // . . . . Q : off =(-2.0, -2.0)/5 + (4.0, 1.0)/5 + // . . Q . . : off =(-2.0, -2.0)/5 + (2.0, 2.0)/5 + // Q . . . . : off =(-2.0, -2.0)/5 + (0.0, 3.0)/5 + // . . . Q . : off =(-2.0, -2.0)/5 + (3.0, 4.0)/5 + // Static screenspace sample offsets (compute some implicitly): + static const float grid_size = 5.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(1.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(4.0, 1.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(2.0, 2.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = w1.bgr; + const float3 w4 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 w_sum_inv = 1.0/(w0 + w1 + w2 + w3 + w4); + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, mirror on odd frames if directed, and exploit + // diagonal symmetry: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * (w0 * sample0 + w1 * sample1 + + w2 * sample2 + w3 * sample3 + w4 * sample4); + + // return (w0 + w1 + w2 + w3 + w4); +} + +float3 tex2Daa6x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use a diagonally symmetric 6-queens pattern with a stronger horizontal + // than vertical slant: + // . . . . Q . : off =(-2.5, -2.5)/6 + (4.0, 0.0)/6 + // . . Q . . . : off =(-2.5, -2.5)/6 + (2.0, 1.0)/6 + // Q . . . . . : off =(-2.5, -2.5)/6 + (0.0, 2.0)/6 + // . . . . . Q : off =(-2.5, -2.5)/6 + (5.0, 3.0)/6 + // . . . Q . . : off =(-2.5, -2.5)/6 + (3.0, 4.0)/6 + // . Q . . . . : off =(-2.5, -2.5)/6 + (1.0, 5.0)/6 + // Static screenspace sample offsets (compute some implicitly): + static const float grid_size = 6.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(4.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(2.0, 1.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(0.0, 2.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = w2.bgr; + const float3 w4 = w1.bgr; + const float3 w5 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1 + w2; + const float3 w_sum = half_sum + half_sum.bgr; + const float3 w_sum_inv = 1.0/(w_sum); + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, mirror on odd frames if directed, and exploit + // diagonal symmetry: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + const float2 uv_offset2 = mul(true_pixel_to_tex_uv, xy_offset2 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset2, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset2, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample5 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * (w0 * sample0 + w1 * sample1 + w2 * sample2 + + w3 * sample3 + w4 * sample4 + w5 * sample5); +} + +float3 tex2Daa7x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use a diagonally symmetric 7-queens pattern with a queen in the center: + // . Q . . . . . : off =(-3.0, -3.0)/7 + (1.0, 0.0)/7 + // . . . . Q . . : off =(-3.0, -3.0)/7 + (4.0, 1.0)/7 + // Q . . . . . . : off =(-3.0, -3.0)/7 + (0.0, 2.0)/7 + // . . . Q . . . : off =(-3.0, -3.0)/7 + (3.0, 3.0)/7 + // . . . . . . Q : off =(-3.0, -3.0)/7 + (6.0, 4.0)/7 + // . . Q . . . . : off =(-3.0, -3.0)/7 + (2.0, 5.0)/7 + // . . . . . Q . : off =(-3.0, -3.0)/7 + (5.0, 6.0)/7 + static const float grid_size = 7.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(1.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(4.0, 1.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(0.0, 2.0) * xy_step; + const float2 xy_offset3 = xy_start_offset + float2(3.0, 3.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = eval_unorm_rgb_weights(xy_offset3, final_axis_importance); + const float3 w4 = w2.bgr; + const float3 w5 = w1.bgr; + const float3 w6 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1 + w2; + const float3 w_sum = half_sum + half_sum.bgr + w3; + const float3 w_sum_inv = 1.0/(w_sum); + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, mirror on odd frames if directed, and exploit + // diagonal symmetry: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + const float2 uv_offset2 = mul(true_pixel_to_tex_uv, xy_offset2 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset2, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset2, input_gamma).rgb; + const float3 sample5 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample6 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * ( + w0 * sample0 + w1 * sample1 + w2 * sample2 + w3 * sample3 + + w4 * sample4 + w5 * sample5 + w6 * sample6); +} + +float3 tex2Daa8x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use a diagonally symmetric 8-queens pattern. + // . . Q . . . . . : off =(-3.5, -3.5)/8 + (2.0, 0.0)/8 + // . . . . Q . . . : off =(-3.5, -3.5)/8 + (4.0, 1.0)/8 + // . Q . . . . . . : off =(-3.5, -3.5)/8 + (1.0, 2.0)/8 + // . . . . . . . Q : off =(-3.5, -3.5)/8 + (7.0, 3.0)/8 + // Q . . . . . . . : off =(-3.5, -3.5)/8 + (0.0, 4.0)/8 + // . . . . . . Q . : off =(-3.5, -3.5)/8 + (6.0, 5.0)/8 + // . . . Q . . . . : off =(-3.5, -3.5)/8 + (3.0, 6.0)/8 + // . . . . . Q . . : off =(-3.5, -3.5)/8 + (5.0, 7.0)/8 + static const float grid_size = 8.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(2.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(4.0, 1.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(1.0, 2.0) * xy_step; + const float2 xy_offset3 = xy_start_offset + float2(7.0, 3.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = eval_unorm_rgb_weights(xy_offset3, final_axis_importance); + const float3 w4 = w3.bgr; + const float3 w5 = w2.bgr; + const float3 w6 = w1.bgr; + const float3 w7 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1 + w2 + w3; + const float3 w_sum = half_sum + half_sum.bgr; + const float3 w_sum_inv = 1.0/(w_sum); + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, and mirror on odd frames if directed: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + const float2 uv_offset2 = mul(true_pixel_to_tex_uv, xy_offset2 * frame_sign); + const float2 uv_offset3 = mul(true_pixel_to_tex_uv, xy_offset3 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset2, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset3, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset3, input_gamma).rgb; + const float3 sample5 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset2, input_gamma).rgb; + const float3 sample6 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample7 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * ( + w0 * sample0 + w1 * sample1 + w2 * sample2 + w3 * sample3 + + w4 * sample4 + w5 * sample5 + w6 * sample6 + w7 * sample7); +} + +float3 tex2Daa12x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use a diagonally symmetric 12-superqueens pattern where no 3 points are + // exactly collinear. + // . . . Q . . . . . . . . : off =(-5.5, -5.5)/12 + (3.0, 0.0)/12 + // . . . . . . . . . Q . . : off =(-5.5, -5.5)/12 + (9.0, 1.0)/12 + // . . . . . . Q . . . . . : off =(-5.5, -5.5)/12 + (6.0, 2.0)/12 + // . Q . . . . . . . . . . : off =(-5.5, -5.5)/12 + (1.0, 3.0)/12 + // . . . . . . . . . . . Q : off =(-5.5, -5.5)/12 + (11.0, 4.0)/12 + // . . . . Q . . . . . . . : off =(-5.5, -5.5)/12 + (4.0, 5.0)/12 + // . . . . . . . Q . . . . : off =(-5.5, -5.5)/12 + (7.0, 6.0)/12 + // Q . . . . . . . . . . . : off =(-5.5, -5.5)/12 + (0.0, 7.0)/12 + // . . . . . . . . . . Q . : off =(-5.5, -5.5)/12 + (10.0, 8.0)/12 + // . . . . . Q . . . . . . : off =(-5.5, -5.5)/12 + (5.0, 9.0)/12 + // . . Q . . . . . . . . . : off =(-5.5, -5.5)/12 + (2.0, 10.0)/12 + // . . . . . . . . Q . . . : off =(-5.5, -5.5)/12 + (8.0, 11.0)/12 + static const float grid_size = 12.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(3.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(9.0, 1.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(6.0, 2.0) * xy_step; + const float2 xy_offset3 = xy_start_offset + float2(1.0, 3.0) * xy_step; + const float2 xy_offset4 = xy_start_offset + float2(11.0, 4.0) * xy_step; + const float2 xy_offset5 = xy_start_offset + float2(4.0, 5.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = eval_unorm_rgb_weights(xy_offset3, final_axis_importance); + const float3 w4 = eval_unorm_rgb_weights(xy_offset4, final_axis_importance); + const float3 w5 = eval_unorm_rgb_weights(xy_offset5, final_axis_importance); + const float3 w6 = w5.bgr; + const float3 w7 = w4.bgr; + const float3 w8 = w3.bgr; + const float3 w9 = w2.bgr; + const float3 w10 = w1.bgr; + const float3 w11 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1 + w2 + w3 + w4 + w5; + const float3 w_sum = half_sum + half_sum.bgr; + const float3 w_sum_inv = 1.0/w_sum; + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, mirror on odd frames if directed, and exploit + // diagonal symmetry: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + const float2 uv_offset2 = mul(true_pixel_to_tex_uv, xy_offset2 * frame_sign); + const float2 uv_offset3 = mul(true_pixel_to_tex_uv, xy_offset3 * frame_sign); + const float2 uv_offset4 = mul(true_pixel_to_tex_uv, xy_offset4 * frame_sign); + const float2 uv_offset5 = mul(true_pixel_to_tex_uv, xy_offset5 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset2, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset3, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset4, input_gamma).rgb; + const float3 sample5 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset5, input_gamma).rgb; + const float3 sample6 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset5, input_gamma).rgb; + const float3 sample7 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset4, input_gamma).rgb; + const float3 sample8 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset3, input_gamma).rgb; + const float3 sample9 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset2, input_gamma).rgb; + const float3 sample10 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample11 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * ( + w0 * sample0 + w1 * sample1 + w2 * sample2 + w3 * sample3 + + w4 * sample4 + w5 * sample5 + w6 * sample6 + w7 * sample7 + + w8 * sample8 + w9 * sample9 + w10 * sample10 + w11 * sample11); +} + +float3 tex2Daa16x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use a diagonally symmetric 16-superqueens pattern where no 3 points are + // exactly collinear. + // . . Q . . . . . . . . . . . . . : off =(-7.5, -7.5)/16 + (2.0, 0.0)/16 + // . . . . . . . . . Q . . . . . . : off =(-7.5, -7.5)/16 + (9.0, 1.0)/16 + // . . . . . . . . . . . . Q . . . : off =(-7.5, -7.5)/16 + (12.0, 2.0)/16 + // . . . . Q . . . . . . . . . . . : off =(-7.5, -7.5)/16 + (4.0, 3.0)/16 + // . . . . . . . . Q . . . . . . . : off =(-7.5, -7.5)/16 + (8.0, 4.0)/16 + // . . . . . . . . . . . . . . Q . : off =(-7.5, -7.5)/16 + (14.0, 5.0)/16 + // Q . . . . . . . . . . . . . . . : off =(-7.5, -7.5)/16 + (0.0, 6.0)/16 + // . . . . . . . . . . Q . . . . . : off =(-7.5, -7.5)/16 + (10.0, 7.0)/16 + // . . . . . Q . . . . . . . . . . : off =(-7.5, -7.5)/16 + (5.0, 8.0)/16 + // . . . . . . . . . . . . . . . Q : off =(-7.5, -7.5)/16 + (15.0, 9.0)/16 + // . Q . . . . . . . . . . . . . . : off =(-7.5, -7.5)/16 + (1.0, 10.0)/16 + // . . . . . . . Q . . . . . . . . : off =(-7.5, -7.5)/16 + (7.0, 11.0)/16 + // . . . . . . . . . . . Q . . . . : off =(-7.5, -7.5)/16 + (11.0, 12.0)/16 + // . . . Q . . . . . . . . . . . . : off =(-7.5, -7.5)/16 + (3.0, 13.0)/16 + // . . . . . . Q . . . . . . . . . : off =(-7.5, -7.5)/16 + (6.0, 14.0)/16 + // . . . . . . . . . . . . . Q . . : off =(-7.5, -7.5)/16 + (13.0, 15.0)/16 + static const float grid_size = 16.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(2.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(9.0, 1.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(12.0, 2.0) * xy_step; + const float2 xy_offset3 = xy_start_offset + float2(4.0, 3.0) * xy_step; + const float2 xy_offset4 = xy_start_offset + float2(8.0, 4.0) * xy_step; + const float2 xy_offset5 = xy_start_offset + float2(14.0, 5.0) * xy_step; + const float2 xy_offset6 = xy_start_offset + float2(0.0, 6.0) * xy_step; + const float2 xy_offset7 = xy_start_offset + float2(10.0, 7.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = eval_unorm_rgb_weights(xy_offset3, final_axis_importance); + const float3 w4 = eval_unorm_rgb_weights(xy_offset4, final_axis_importance); + const float3 w5 = eval_unorm_rgb_weights(xy_offset5, final_axis_importance); + const float3 w6 = eval_unorm_rgb_weights(xy_offset6, final_axis_importance); + const float3 w7 = eval_unorm_rgb_weights(xy_offset7, final_axis_importance); + const float3 w8 = w7.bgr; + const float3 w9 = w6.bgr; + const float3 w10 = w5.bgr; + const float3 w11 = w4.bgr; + const float3 w12 = w3.bgr; + const float3 w13 = w2.bgr; + const float3 w14 = w1.bgr; + const float3 w15 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1 + w2 + w3 + w4 + w5 + w6 + w7; + const float3 w_sum = half_sum + half_sum.bgr; + const float3 w_sum_inv = 1.0/(w_sum); + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, mirror on odd frames if directed, and exploit + // diagonal symmetry: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + const float2 uv_offset2 = mul(true_pixel_to_tex_uv, xy_offset2 * frame_sign); + const float2 uv_offset3 = mul(true_pixel_to_tex_uv, xy_offset3 * frame_sign); + const float2 uv_offset4 = mul(true_pixel_to_tex_uv, xy_offset4 * frame_sign); + const float2 uv_offset5 = mul(true_pixel_to_tex_uv, xy_offset5 * frame_sign); + const float2 uv_offset6 = mul(true_pixel_to_tex_uv, xy_offset6 * frame_sign); + const float2 uv_offset7 = mul(true_pixel_to_tex_uv, xy_offset7 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset2, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset3, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset4, input_gamma).rgb; + const float3 sample5 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset5, input_gamma).rgb; + const float3 sample6 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset6, input_gamma).rgb; + const float3 sample7 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset7, input_gamma).rgb; + const float3 sample8 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset7, input_gamma).rgb; + const float3 sample9 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset6, input_gamma).rgb; + const float3 sample10 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset5, input_gamma).rgb; + const float3 sample11 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset4, input_gamma).rgb; + const float3 sample12 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset3, input_gamma).rgb; + const float3 sample13 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset2, input_gamma).rgb; + const float3 sample14 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample15 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * ( + w0 * sample0 + w1 * sample1 + w2 * sample2 + w3 * sample3 + + w4 * sample4 + w5 * sample5 + w6 * sample6 + w7 * sample7 + + w8 * sample8 + w9 * sample9 + w10 * sample10 + w11 * sample11 + + w12 * sample12 + w13 * sample13 + w14 * sample14 + w15 * sample15); +} + +float3 tex2Daa20x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use a diagonally symmetric 20-superqueens pattern where no 3 points are + // exactly collinear and superqueens have a squared attack radius of 13. + // . . . . . . . Q . . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (7.0, 0.0)/20 + // . . . . . . . . . . . . . . . . Q . . . : off =(-9.5, -9.5)/20 + (16.0, 1.0)/20 + // . . . . . . . . . . . Q . . . . . . . . : off =(-9.5, -9.5)/20 + (11.0, 2.0)/20 + // . Q . . . . . . . . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (1.0, 3.0)/20 + // . . . . . Q . . . . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (5.0, 4.0)/20 + // . . . . . . . . . . . . . . . Q . . . . : off =(-9.5, -9.5)/20 + (15.0, 5.0)/20 + // . . . . . . . . . . Q . . . . . . . . . : off =(-9.5, -9.5)/20 + (10.0, 6.0)/20 + // . . . . . . . . . . . . . . . . . . . Q : off =(-9.5, -9.5)/20 + (19.0, 7.0)/20 + // . . Q . . . . . . . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (2.0, 8.0)/20 + // . . . . . . Q . . . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (6.0, 9.0)/20 + // . . . . . . . . . . . . . Q . . . . . . : off =(-9.5, -9.5)/20 + (13.0, 10.0)/20 + // . . . . . . . . . . . . . . . . . Q . . : off =(-9.5, -9.5)/20 + (17.0, 11.0)/20 + // Q . . . . . . . . . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (0.0, 12.0)/20 + // . . . . . . . . . Q . . . . . . . . . . : off =(-9.5, -9.5)/20 + (9.0, 13.0)/20 + // . . . . Q . . . . . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (4.0, 14.0)/20 + // . . . . . . . . . . . . . . Q . . . . . : off =(-9.5, -9.5)/20 + (14.0, 15.0)/20 + // . . . . . . . . . . . . . . . . . . Q . : off =(-9.5, -9.5)/20 + (18.0, 16.0)/20 + // . . . . . . . . Q . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (8.0, 17.0)/20 + // . . . Q . . . . . . . . . . . . . . . . : off =(-9.5, -9.5)/20 + (3.0, 18.0)/20 + // . . . . . . . . . . . . Q . . . . . . . : off =(-9.5, -9.5)/20 + (12.0, 19.0)/20 + static const float grid_size = 20.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(7.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(16.0, 1.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(11.0, 2.0) * xy_step; + const float2 xy_offset3 = xy_start_offset + float2(1.0, 3.0) * xy_step; + const float2 xy_offset4 = xy_start_offset + float2(5.0, 4.0) * xy_step; + const float2 xy_offset5 = xy_start_offset + float2(15.0, 5.0) * xy_step; + const float2 xy_offset6 = xy_start_offset + float2(10.0, 6.0) * xy_step; + const float2 xy_offset7 = xy_start_offset + float2(19.0, 7.0) * xy_step; + const float2 xy_offset8 = xy_start_offset + float2(2.0, 8.0) * xy_step; + const float2 xy_offset9 = xy_start_offset + float2(6.0, 9.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = eval_unorm_rgb_weights(xy_offset3, final_axis_importance); + const float3 w4 = eval_unorm_rgb_weights(xy_offset4, final_axis_importance); + const float3 w5 = eval_unorm_rgb_weights(xy_offset5, final_axis_importance); + const float3 w6 = eval_unorm_rgb_weights(xy_offset6, final_axis_importance); + const float3 w7 = eval_unorm_rgb_weights(xy_offset7, final_axis_importance); + const float3 w8 = eval_unorm_rgb_weights(xy_offset8, final_axis_importance); + const float3 w9 = eval_unorm_rgb_weights(xy_offset9, final_axis_importance); + const float3 w10 = w9.bgr; + const float3 w11 = w8.bgr; + const float3 w12 = w7.bgr; + const float3 w13 = w6.bgr; + const float3 w14 = w5.bgr; + const float3 w15 = w4.bgr; + const float3 w16 = w3.bgr; + const float3 w17 = w2.bgr; + const float3 w18 = w1.bgr; + const float3 w19 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1 + w2 + w3 + w4 + w5 + w6 + w7 + w8 + w9; + const float3 w_sum = half_sum + half_sum.bgr; + const float3 w_sum_inv = 1.0/(w_sum); + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, mirror on odd frames if directed, and exploit + // diagonal symmetry: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + const float2 uv_offset2 = mul(true_pixel_to_tex_uv, xy_offset2 * frame_sign); + const float2 uv_offset3 = mul(true_pixel_to_tex_uv, xy_offset3 * frame_sign); + const float2 uv_offset4 = mul(true_pixel_to_tex_uv, xy_offset4 * frame_sign); + const float2 uv_offset5 = mul(true_pixel_to_tex_uv, xy_offset5 * frame_sign); + const float2 uv_offset6 = mul(true_pixel_to_tex_uv, xy_offset6 * frame_sign); + const float2 uv_offset7 = mul(true_pixel_to_tex_uv, xy_offset7 * frame_sign); + const float2 uv_offset8 = mul(true_pixel_to_tex_uv, xy_offset8 * frame_sign); + const float2 uv_offset9 = mul(true_pixel_to_tex_uv, xy_offset9 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset2, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset3, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset4, input_gamma).rgb; + const float3 sample5 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset5, input_gamma).rgb; + const float3 sample6 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset6, input_gamma).rgb; + const float3 sample7 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset7, input_gamma).rgb; + const float3 sample8 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset8, input_gamma).rgb; + const float3 sample9 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset9, input_gamma).rgb; + const float3 sample10 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset9, input_gamma).rgb; + const float3 sample11 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset8, input_gamma).rgb; + const float3 sample12 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset7, input_gamma).rgb; + const float3 sample13 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset6, input_gamma).rgb; + const float3 sample14 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset5, input_gamma).rgb; + const float3 sample15 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset4, input_gamma).rgb; + const float3 sample16 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset3, input_gamma).rgb; + const float3 sample17 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset2, input_gamma).rgb; + const float3 sample18 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample19 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * ( + w0 * sample0 + w1 * sample1 + w2 * sample2 + w3 * sample3 + + w4 * sample4 + w5 * sample5 + w6 * sample6 + w7 * sample7 + + w8 * sample8 + w9 * sample9 + w10 * sample10 + w11 * sample11 + + w12 * sample12 + w13 * sample13 + w14 * sample14 + w15 * sample15 + + w16 * sample16 + w17 * sample17 + w18 * sample18 + w19 * sample19); +} + +float3 tex2Daa24x(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Use a diagonally symmetric 24-superqueens pattern where no 3 points are + // exactly collinear and superqueens have a squared attack radius of 13. + // . . . . . . Q . . . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (6.0, 0.0)/24 + // . . . . . . . . . . . . . . . . Q . . . . . . . : off =(-11.5, -11.5)/24 + (16.0, 1.0)/24 + // . . . . . . . . . . Q . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (10.0, 2.0)/24 + // . . . . . . . . . . . . . . . . . . . . . Q . . : off =(-11.5, -11.5)/24 + (21.0, 3.0)/24 + // . . . . . Q . . . . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (5.0, 4.0)/24 + // . . . . . . . . . . . . . . . Q . . . . . . . . : off =(-11.5, -11.5)/24 + (15.0, 5.0)/24 + // . Q . . . . . . . . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (1.0, 6.0)/24 + // . . . . . . . . . . . Q . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (11.0, 7.0)/24 + // . . . . . . . . . . . . . . . . . . . Q . . . . : off =(-11.5, -11.5)/24 + (19.0, 8.0)/24 + // . . . . . . . . . . . . . . . . . . . . . . . Q : off =(-11.5, -11.5)/24 + (23.0, 9.0)/24 + // . . . Q . . . . . . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (3.0, 10.0)/24 + // . . . . . . . . . . . . . . Q . . . . . . . . . : off =(-11.5, -11.5)/24 + (14.0, 11.0)/24 + // . . . . . . . . . Q . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (9.0, 12.0)/24 + // . . . . . . . . . . . . . . . . . . . . Q . . . : off =(-11.5, -11.5)/24 + (20.0, 13.0)/24 + // Q . . . . . . . . . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (0.0, 14.0)/24 + // . . . . Q . . . . . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (4.0, 15.0)/24 + // . . . . . . . . . . . . Q . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (12.0, 16.0)/24 + // . . . . . . . . . . . . . . . . . . . . . . Q . : off =(-11.5, -11.5)/24 + (22.0, 17.0)/24 + // . . . . . . . . Q . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (8.0, 18.0)/24 + // . . . . . . . . . . . . . . . . . . Q . . . . . : off =(-11.5, -11.5)/24 + (18.0, 19.0)/24 + // . . Q . . . . . . . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (2.0, 20.0)/24 + // . . . . . . . . . . . . . Q . . . . . . . . . . : off =(-11.5, -11.5)/24 + (13.0, 21.0)/24 + // . . . . . . . Q . . . . . . . . . . . . . . . . : off =(-11.5, -11.5)/24 + (7.0, 22.0)/24 + // . . . . . . . . . . . . . . . . . Q . . . . . . : off =(-11.5, -11.5)/24 + (17.0, 23.0)/24 + static const float grid_size = 24.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample. Exploit diagonal symmetry: + const float2 xy_offset0 = xy_start_offset + float2(6.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(16.0, 1.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(10.0, 2.0) * xy_step; + const float2 xy_offset3 = xy_start_offset + float2(21.0, 3.0) * xy_step; + const float2 xy_offset4 = xy_start_offset + float2(5.0, 4.0) * xy_step; + const float2 xy_offset5 = xy_start_offset + float2(15.0, 5.0) * xy_step; + const float2 xy_offset6 = xy_start_offset + float2(1.0, 6.0) * xy_step; + const float2 xy_offset7 = xy_start_offset + float2(11.0, 7.0) * xy_step; + const float2 xy_offset8 = xy_start_offset + float2(19.0, 8.0) * xy_step; + const float2 xy_offset9 = xy_start_offset + float2(23.0, 9.0) * xy_step; + const float2 xy_offset10 = xy_start_offset + float2(3.0, 10.0) * xy_step; + const float2 xy_offset11 = xy_start_offset + float2(14.0, 11.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = eval_unorm_rgb_weights(xy_offset3, final_axis_importance); + const float3 w4 = eval_unorm_rgb_weights(xy_offset4, final_axis_importance); + const float3 w5 = eval_unorm_rgb_weights(xy_offset5, final_axis_importance); + const float3 w6 = eval_unorm_rgb_weights(xy_offset6, final_axis_importance); + const float3 w7 = eval_unorm_rgb_weights(xy_offset7, final_axis_importance); + const float3 w8 = eval_unorm_rgb_weights(xy_offset8, final_axis_importance); + const float3 w9 = eval_unorm_rgb_weights(xy_offset9, final_axis_importance); + const float3 w10 = eval_unorm_rgb_weights(xy_offset10, final_axis_importance); + const float3 w11 = eval_unorm_rgb_weights(xy_offset11, final_axis_importance); + const float3 w12 = w11.bgr; + const float3 w13 = w10.bgr; + const float3 w14 = w9.bgr; + const float3 w15 = w8.bgr; + const float3 w16 = w7.bgr; + const float3 w17 = w6.bgr; + const float3 w18 = w5.bgr; + const float3 w19 = w4.bgr; + const float3 w20 = w3.bgr; + const float3 w21 = w2.bgr; + const float3 w22 = w1.bgr; + const float3 w23 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1 + w2 + w3 + w4 + + w5 + w6 + w7 + w8 + w9 + w10 + w11; + const float3 w_sum = half_sum + half_sum.bgr; + const float3 w_sum_inv = 1.0/(w_sum); + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, mirror on odd frames if directed, and exploit + // diagonal symmetry: + const float2 frame_sign = get_frame_sign(frame); + const float2 uv_offset0 = mul(true_pixel_to_tex_uv, xy_offset0 * frame_sign); + const float2 uv_offset1 = mul(true_pixel_to_tex_uv, xy_offset1 * frame_sign); + const float2 uv_offset2 = mul(true_pixel_to_tex_uv, xy_offset2 * frame_sign); + const float2 uv_offset3 = mul(true_pixel_to_tex_uv, xy_offset3 * frame_sign); + const float2 uv_offset4 = mul(true_pixel_to_tex_uv, xy_offset4 * frame_sign); + const float2 uv_offset5 = mul(true_pixel_to_tex_uv, xy_offset5 * frame_sign); + const float2 uv_offset6 = mul(true_pixel_to_tex_uv, xy_offset6 * frame_sign); + const float2 uv_offset7 = mul(true_pixel_to_tex_uv, xy_offset7 * frame_sign); + const float2 uv_offset8 = mul(true_pixel_to_tex_uv, xy_offset8 * frame_sign); + const float2 uv_offset9 = mul(true_pixel_to_tex_uv, xy_offset9 * frame_sign); + const float2 uv_offset10 = mul(true_pixel_to_tex_uv, xy_offset10 * frame_sign); + const float2 uv_offset11 = mul(true_pixel_to_tex_uv, xy_offset11 * frame_sign); + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset0, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset1, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset2, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset3, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset4, input_gamma).rgb; + const float3 sample5 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset5, input_gamma).rgb; + const float3 sample6 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset6, input_gamma).rgb; + const float3 sample7 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset7, input_gamma).rgb; + const float3 sample8 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset8, input_gamma).rgb; + const float3 sample9 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset9, input_gamma).rgb; + const float3 sample10 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset10, input_gamma).rgb; + const float3 sample11 = tex2Daa_tiled_linearize(tex, tex_uv + uv_offset11, input_gamma).rgb; + const float3 sample12 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset11, input_gamma).rgb; + const float3 sample13 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset10, input_gamma).rgb; + const float3 sample14 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset9, input_gamma).rgb; + const float3 sample15 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset8, input_gamma).rgb; + const float3 sample16 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset7, input_gamma).rgb; + const float3 sample17 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset6, input_gamma).rgb; + const float3 sample18 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset5, input_gamma).rgb; + const float3 sample19 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset4, input_gamma).rgb; + const float3 sample20 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset3, input_gamma).rgb; + const float3 sample21 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset2, input_gamma).rgb; + const float3 sample22 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset1, input_gamma).rgb; + const float3 sample23 = tex2Daa_tiled_linearize(tex, tex_uv - uv_offset0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * ( + w0 * sample0 + w1 * sample1 + w2 * sample2 + w3 * sample3 + + w4 * sample4 + w5 * sample5 + w6 * sample6 + w7 * sample7 + + w8 * sample8 + w9 * sample9 + w10 * sample10 + w11 * sample11 + + w12 * sample12 + w13 * sample13 + w14 * sample14 + w15 * sample15 + + w16 * sample16 + w17 * sample17 + w18 * sample18 + w19 * sample19 + + w20 * sample20 + w21 * sample21 + w22 * sample22 + w23 * sample23); +} + +float3 tex2Daa_debug_16x_regular(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Sample on a regular 4x4 grid. This is mainly for testing. + static const float grid_size = 4.0; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float2 xy_step = 1.0/grid_size * subpixel_support_diameter; + const float2 xy_start_offset = (0.5 - grid_size*0.5) * xy_step; + // Get the xy offset of each sample: + const float2 xy_offset0 = xy_start_offset + float2(0.0, 0.0) * xy_step; + const float2 xy_offset1 = xy_start_offset + float2(1.0, 0.0) * xy_step; + const float2 xy_offset2 = xy_start_offset + float2(2.0, 0.0) * xy_step; + const float2 xy_offset3 = xy_start_offset + float2(3.0, 0.0) * xy_step; + const float2 xy_offset4 = xy_start_offset + float2(0.0, 1.0) * xy_step; + const float2 xy_offset5 = xy_start_offset + float2(1.0, 1.0) * xy_step; + const float2 xy_offset6 = xy_start_offset + float2(2.0, 1.0) * xy_step; + const float2 xy_offset7 = xy_start_offset + float2(3.0, 1.0) * xy_step; + // Compute subpixel weights, and exploit diagonal symmetry for speed. + // (We can't exploit vertical or horizontal symmetry due to uncertain + // subpixel offsets. We could fix that by rotating xy offsets with the + // subpixel structure, but...no.) + const float3 w0 = eval_unorm_rgb_weights(xy_offset0, final_axis_importance); + const float3 w1 = eval_unorm_rgb_weights(xy_offset1, final_axis_importance); + const float3 w2 = eval_unorm_rgb_weights(xy_offset2, final_axis_importance); + const float3 w3 = eval_unorm_rgb_weights(xy_offset3, final_axis_importance); + const float3 w4 = eval_unorm_rgb_weights(xy_offset4, final_axis_importance); + const float3 w5 = eval_unorm_rgb_weights(xy_offset5, final_axis_importance); + const float3 w6 = eval_unorm_rgb_weights(xy_offset6, final_axis_importance); + const float3 w7 = eval_unorm_rgb_weights(xy_offset7, final_axis_importance); + const float3 w8 = w7.bgr; + const float3 w9 = w6.bgr; + const float3 w10 = w5.bgr; + const float3 w11 = w4.bgr; + const float3 w12 = w3.bgr; + const float3 w13 = w2.bgr; + const float3 w14 = w1.bgr; + const float3 w15 = w0.bgr; + // Get the weight sum to normalize the total to 1.0 later: + const float3 half_sum = w0 + w1 + w2 + w3 + w4 + w5 + w6 + w7; + const float3 w_sum = half_sum + half_sum.bgr; + const float3 w_sum_inv = 1.0/(w_sum); + // Scale the pixel-space to texture offset matrix by the pixel diameter. + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + // Get uv sample offsets, taking advantage of row alignment: + const float2 uv_step_x = mul(true_pixel_to_tex_uv, float2(xy_step.x, 0.0)); + const float2 uv_step_y = mul(true_pixel_to_tex_uv, float2(0.0, xy_step.y)); + const float2 uv_offset0 = -1.5 * (uv_step_x + uv_step_y); + const float2 sample0_uv = tex_uv + uv_offset0; + const float2 sample4_uv = sample0_uv + uv_step_y; + const float2 sample8_uv = sample0_uv + uv_step_y * 2.0; + const float2 sample12_uv = sample0_uv + uv_step_y * 3.0; + // Load samples, linearizing if necessary, etc.: + const float3 sample0 = tex2Daa_tiled_linearize(tex, sample0_uv, input_gamma).rgb; + const float3 sample1 = tex2Daa_tiled_linearize(tex, sample0_uv + uv_step_x, input_gamma).rgb; + const float3 sample2 = tex2Daa_tiled_linearize(tex, sample0_uv + uv_step_x * 2.0, input_gamma).rgb; + const float3 sample3 = tex2Daa_tiled_linearize(tex, sample0_uv + uv_step_x * 3.0, input_gamma).rgb; + const float3 sample4 = tex2Daa_tiled_linearize(tex, sample4_uv, input_gamma).rgb; + const float3 sample5 = tex2Daa_tiled_linearize(tex, sample4_uv + uv_step_x, input_gamma).rgb; + const float3 sample6 = tex2Daa_tiled_linearize(tex, sample4_uv + uv_step_x * 2.0, input_gamma).rgb; + const float3 sample7 = tex2Daa_tiled_linearize(tex, sample4_uv + uv_step_x * 3.0, input_gamma).rgb; + const float3 sample8 = tex2Daa_tiled_linearize(tex, sample8_uv, input_gamma).rgb; + const float3 sample9 = tex2Daa_tiled_linearize(tex, sample8_uv + uv_step_x, input_gamma).rgb; + const float3 sample10 = tex2Daa_tiled_linearize(tex, sample8_uv + uv_step_x * 2.0, input_gamma).rgb; + const float3 sample11 = tex2Daa_tiled_linearize(tex, sample8_uv + uv_step_x * 3.0, input_gamma).rgb; + const float3 sample12 = tex2Daa_tiled_linearize(tex, sample12_uv, input_gamma).rgb; + const float3 sample13 = tex2Daa_tiled_linearize(tex, sample12_uv + uv_step_x, input_gamma).rgb; + const float3 sample14 = tex2Daa_tiled_linearize(tex, sample12_uv + uv_step_x * 2.0, input_gamma).rgb; + const float3 sample15 = tex2Daa_tiled_linearize(tex, sample12_uv + uv_step_x * 3.0, input_gamma).rgb; + // Sum weighted samples (weight sum must equal 1.0 for each channel): + return w_sum_inv * ( + w0 * sample0 + w1 * sample1 + w2 * sample2 + w3 * sample3 + + w4 * sample4 + w5 * sample5 + w6 * sample6 + w7 * sample7 + + w8 * sample8 + w9 * sample9 + w10 * sample10 + w11 * sample11 + + w12 * sample12 + w13 * sample13 + w14 * sample14 + w15 * sample15); +} + +float3 tex2Daa_debug_dynamic(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // This function is for testing only: Use an NxN grid with dynamic weights. + static const int grid_size = 8; + assign_aa_cubic_constants(); + const float4 ssd_fai = get_subpixel_support_diam_and_final_axis_importance(); + const float2 subpixel_support_diameter = ssd_fai.xy; + const float2 final_axis_importance = ssd_fai.zw; + const float grid_radius_in_samples = (float(grid_size) - 1.0)/2.0; + const float2 filter_space_offset_step = + subpixel_support_diameter / grid_size; + const float2 sample0_filter_space_offset = + -grid_radius_in_samples * filter_space_offset_step; + // Compute xy sample offsets and subpixel weights: + float3 weights[64]; // grid_size * grid_size + float3 weight_sum = float3(0.0, 0.0, 0.0); + for(int i = 0; i < grid_size; ++i) + { + for(int j = 0; j < grid_size; ++j) + { + // Weights based on xy distances: + const float2 offset = sample0_filter_space_offset + + float2(j, i) * filter_space_offset_step; + const float3 weight = eval_unorm_rgb_weights(offset, final_axis_importance); + weights[i*grid_size + j] = weight; + weight_sum += weight; + } + } + // Get uv offset vectors along x and y directions: + const float2x2 true_pixel_to_tex_uv = + float2x2((pixel_to_tex_uv * aa_pixel_diameter)); + const float2 uv_offset_step_x = mul(true_pixel_to_tex_uv, + float2(filter_space_offset_step.x, 0.0)); + const float2 uv_offset_step_y = mul(true_pixel_to_tex_uv, + float2(0.0, filter_space_offset_step.y)); + // Get a starting sample location: + const float2 sample0_uv_offset = -grid_radius_in_samples * + (uv_offset_step_x + uv_offset_step_y); + const float2 sample0_uv = tex_uv + sample0_uv_offset; + // Load, weight, and sum [linearized] samples: + float3 sum = float3(0.0, 0.0, 0.0); + const float3 weight_sum_inv = 1.0/weight_sum; + for(int i = 0; i < grid_size; ++i) + { + const float2 row_i_first_sample_uv = + sample0_uv + i * uv_offset_step_y; + for(int j = 0; j < grid_size; ++j) + { + const float2 sample_uv = + row_i_first_sample_uv + j * uv_offset_step_x; + sum += weights[i*grid_size + j] * + tex2Daa_tiled_linearize(tex, sample_uv, input_gamma).rgb; + } + } + return sum * weight_sum_inv; +} + + +/////////////////////// ANTIALIASING CODEPATH SELECTION ////////////////////// + +float3 tex2Daa(const sampler2D tex, const float2 tex_uv, + const float2x2 pixel_to_tex_uv, const float frame, const float input_gamma) +{ + // Statically switch between antialiasing modes/levels: + return (antialias_level < 0.5) ? tex2D_linearize(tex, tex_uv, input_gamma).rgb : + (antialias_level < 3.5) ? tex2Daa_subpixel_weights_only( + tex, tex_uv, pixel_to_tex_uv, input_gamma) : + (antialias_level < 4.5) ? tex2Daa4x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 5.5) ? tex2Daa5x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 6.5) ? tex2Daa6x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 7.5) ? tex2Daa7x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 11.5) ? tex2Daa8x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 15.5) ? tex2Daa12x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 19.5) ? tex2Daa16x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 23.5) ? tex2Daa20x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 253.5) ? tex2Daa24x(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + (antialias_level < 254.5) ? tex2Daa_debug_16x_regular( + tex, tex_uv, pixel_to_tex_uv, frame, input_gamma) : + tex2Daa_debug_dynamic(tex, tex_uv, pixel_to_tex_uv, frame, input_gamma); +} + + +#endif // _TEX2DANTIALIAS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/lib/user-settings.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/lib/user-settings.fxh new file mode 100644 index 000000000..90ffda8f9 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/lib/user-settings.fxh @@ -0,0 +1,428 @@ +#ifndef _USER_SETTINGS_H +#define _USER_SETTINGS_H + +///////////////////////////// DRIVER CAPABILITIES //////////////////////////// + +// The Cg compiler uses different "profiles" with different capabilities. +// This shader requires a Cg compilation profile >= arbfp1, but a few options +// require higher profiles like fp30 or fp40. The shader can't detect profile +// or driver capabilities, so instead you must comment or uncomment the lines +// below with "//" before "#define." Disable an option if you get compilation +// errors resembling those listed. Generally speaking, all of these options +// will run on nVidia cards, but only _DRIVERS_ALLOW_TEX2DBIAS (if that) is +// likely to run on ATI/AMD, due to the Cg compiler's profile limitations. + +// Derivatives: Unsupported on fp20, ps_1_1, ps_1_2, ps_1_3, and arbfp1. +// Among other things, derivatives help us fix anisotropic filtering artifacts +// with curved manually tiled phosphor mask coords. Related errors: +// error C3004: function "float2 ddx(float2);" not supported in this profile +// error C3004: function "float2 ddy(float2);" not supported in this profile +#ifndef _DRIVERS_ALLOW_DERIVATIVES + #define _DRIVERS_ALLOW_DERIVATIVES 0 +#endif + +// Fine derivatives: Unsupported on older ATI cards. +// Fine derivatives enable 2x2 fragment block communication, letting us perform +// fast single-pass blur operations. If your card uses coarse derivatives and +// these are enabled, blurs could look broken. Derivatives are a prerequisite. +#if _DRIVERS_ALLOW_DERIVATIVES + #define _DRIVERS_ALLOW_FINE_DERIVATIVES +#endif + +// Dynamic looping: Requires an fp30 or newer profile. +// This makes phosphor mask resampling faster in some cases. Related errors: +// error C5013: profile does not support "for" statements and "for" could not +// be unrolled +#ifndef _DRIVERS_ALLOW_DYNAMIC_BRANCHES + #define _DRIVERS_ALLOW_DYNAMIC_BRANCHES 0 +#endif + +// Without _DRIVERS_ALLOW_DYNAMIC_BRANCHES, we need to use unrollable loops. +// Using one static loop avoids overhead if the user is right, but if the user +// is wrong (loops are allowed), breaking a loop into if-blocked pieces with a +// binary search can potentially save some iterations. However, it may fail: +// error C6001: Temporary register limit of 32 exceeded; 35 registers +// needed to compile program +#ifndef _ACCOMODATE_POSSIBLE_DYNAMIC_LOOPS + #define _ACCOMODATE_POSSIBLE_DYNAMIC_LOOPS 0 +#endif + +// tex2Dlod: Requires an fp40 or newer profile. This can be used to disable +// anisotropic filtering, thereby fixing related artifacts. Related errors: +// error C3004: function "float4 tex2Dlod(sampler2D, float4);" not supported in +// this profile +// #ifndef _DRIVERS_ALLOW_TEX2DLOD +// #define _DRIVERS_ALLOW_TEX2DLOD 1 +// #endif + +// tex2Dbias: Requires an fp30 or newer profile. This can be used to alleviate +// artifacts from anisotropic filtering and mipmapping. Related errors: +// error C3004: function "float4 tex2Dbias(sampler2D, float4);" not supported +// in this profile +// #ifndef _DRIVERS_ALLOW_TEX2DBIAS +// #define _DRIVERS_ALLOW_TEX2DBIAS 0 +// #endif + +// Integrated graphics compatibility: Integrated graphics like Intel HD 4000 +// impose stricter limitations on register counts and instructions. Enable +// _INTEGRATED_GRAPHICS_COMPATIBILITY_MODE if you still see error C6001 or: +// error C6002: Instruction limit of 1024 exceeded: 1523 instructions needed +// to compile program. +// Enabling integrated graphics compatibility mode will automatically disable: +// 1.) _PHOSPHOR_MASK_MANUALLY_RESIZE: The phosphor mask will be softer. +// (This may be reenabled in a later release.) +// 2.) _RUNTIME_GEOMETRY_MODE +// 3.) The high-quality 4x4 Gaussian resize for the bloom approximation +#ifndef _INTEGRATED_GRAPHICS_COMPATIBILITY_MODE + #define _INTEGRATED_GRAPHICS_COMPATIBILITY_MODE 0 +#endif + + +//////////////////////////// USER CODEPATH OPTIONS /////////////////////////// + +// To disable a #define option, turn its line into a comment with "//." + +// RUNTIME VS. COMPILE-TIME OPTIONS (Major Performance Implications): +// Enable runtime shader parameters in the Retroarch (etc.) GUI? They override +// many of the options in this file and allow real-time tuning, but many of +// them are slower. Disabling them and using this text file will boost FPS. +#ifndef _RUNTIME_SHADER_PARAMS_ENABLE + #define _RUNTIME_SHADER_PARAMS_ENABLE 1 +#endif +// Specify the phosphor bloom sigma at runtime? This option is 10% slower, but +// it's the only way to do a wide-enough full bloom with a runtime dot pitch. +#ifndef _RUNTIME_PHOSPHOR_BLOOM_SIGMA + #define _RUNTIME_PHOSPHOR_BLOOM_SIGMA 1 +#endif +// Specify antialiasing weight parameters at runtime? (Costs ~20% with cubics) +#ifndef _RUNTIME_ANTIALIAS_WEIGHTS + #define _RUNTIME_ANTIALIAS_WEIGHTS 1 +#endif +// Specify subpixel offsets at runtime? (WARNING: EXTREMELY EXPENSIVE!) +#ifndef _RUNTIME_ANTIALIAS_SUBPIXEL_OFFSETS + #define _RUNTIME_ANTIALIAS_SUBPIXEL_OFFSETS 0 +#endif +// Make beam_horiz_filter and beam_horiz_linear_rgb_weight into runtime shader +// parameters? This will require more math or dynamic branching. +#ifndef _RUNTIME_SCANLINES_HORIZ_FILTER_COLORSPACE + #define _RUNTIME_SCANLINES_HORIZ_FILTER_COLORSPACE 1 +#endif +// Specify the tilt at runtime? This makes things about 3% slower. +// akgunter: +// This is used in crt-royale-geometry-aa-last-pass.fxh. +// I've hard-coded it to 1 and hidden it from the UI in the ReShade version because +// I don't know a good way to port that logic. If anyone ever does figure that +// out, we can uncomment and port that logic and then unhide this definition. +#define _RUNTIME_GEOMETRY_TILT 1 + +// Specify the geometry mode at runtime? +#ifndef _RUNTIME_GEOMETRY_MODE + #define _RUNTIME_GEOMETRY_MODE 1 +#endif +// Specify the phosphor mask type (aperture grille, slot mask, shadow mask) and +// mode (Lanczos-resize, hardware resize, or tile 1:1) at runtime, even without +// dynamic branches? This is cheap if mask_resize_viewport_scale is small. +// #ifndef FORCE_RUNTIME_PHOSPHOR_MASK_MODE_TYPE_SELECT +// #define FORCE_RUNTIME_PHOSPHOR_MASK_MODE_TYPE_SELECT 1 +// #endif + +// PHOSPHOR MASK: +// Choose between a 64x64 or 512x512 source for the phosphor mask +// Mainly affects Sample Mode 1 +// #ifndef USE_LARGE_PHOSPHOR_MASK +// #define USE_LARGE_PHOSPHOR_MASK 1 +// #endif + +// Manually resize the phosphor mask for best results (slower)? Disabling this +// removes the option to do so, but it may be faster without dynamic branches. +#ifndef _PHOSPHOR_MASK_MANUALLY_RESIZE + #define _PHOSPHOR_MASK_MANUALLY_RESIZE 1 +#endif +// If we sinc-resize the mask, should we Lanczos-window it (slower but better)? +// #ifndef PHOSPHOR_MASK_RESIZE_LANCZOS_WINDOW +// #define PHOSPHOR_MASK_RESIZE_LANCZOS_WINDOW 1 +// #endif +// Larger blurs are expensive, but we need them to blur larger triads. We can +// detect the right blur if the triad size is static or our profile allows +// dynamic branches, but otherwise we use the largest blur the user indicates +// they might need: + +#define _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_3_PIXELS 1 +#define _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_6_PIXELS 2 +#define _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_9_PIXELS 3 +#define _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_12_PIXELS 4 + +#if !_RUNTIME_PHOSPHOR_BLOOM_SIGMA + #ifndef PHOSPHOR_BLOOM_TRIAD_SIZE_MODE + #define PHOSPHOR_BLOOM_TRIAD_SIZE_MODE _PHOSPHOR_BLOOM_TRIADS_LARGER_THAN_3_PIXELS // [0 - 4] + #endif +#endif + +// Here's a helpful chart: +// MaxTriadSize BlurSize MinTriadCountsByResolution +// 3.0 9.0 480/640/960/1920 triads at 1080p/1440p/2160p/4320p, 4:3 aspect +// 6.0 17.0 240/320/480/960 triads at 1080p/1440p/2160p/4320p, 4:3 aspect +// 9.0 25.0 160/213/320/640 triads at 1080p/1440p/2160p/4320p, 4:3 aspect +// 12.0 31.0 120/160/240/480 triads at 1080p/1440p/2160p/4320p, 4:3 aspect +// 18.0 43.0 80/107/160/320 triads at 1080p/1440p/2160p/4320p, 4:3 aspect + +/////////////////////////////// USER PARAMETERS ////////////////////////////// + +// Note: Many of these static parameters are overridden by runtime shader +// parameters when those are enabled. However, many others are static codepath +// options that were cleaner or more convert to code as static constants. + +// GAMMA: +static const float crt_gamma_static = 2.5; // range [1, 5] +static const float lcd_gamma_static = 2.2; // range [1, 5] + +// LEVELS MANAGEMENT: +// Control the final multiplicative image contrast: +static const float levels_contrast_static = 1.0; // range [0, 4) +// We auto-dim to avoid clipping between passes and restore brightness +// later. Control the dim factor here: Lower values clip less but crush +// blacks more (static only for now). +static const float levels_autodim_temp = 0.5; // range (0, 1] default is 0.5 but that was unnecessarily dark for me, so I set it to 1.0 + +// HALATION/DIFFUSION/BLOOM: +// Halation weight: How much energy should be lost to electrons bounding +// around under the CRT glass and exciting random phosphors? +static const float halation_weight_static = 0.0; // range [0, 1] +// Refractive diffusion weight: How much light should spread/diffuse from +// refracting through the CRT glass? +static const float diffusion_weight_static = 0.075; // range [0, 1] +// Underestimate brightness: Bright areas bloom more, but we can base the +// bloom brightpass on a lower brightness to sharpen phosphors, or a higher +// brightness to soften them. Low values clip, but >= 0.8 looks okay. +static const float bloom_underestimate_levels_static = 0.8; // range [0, 5] +// Blur all colors more than necessary for a softer phosphor bloom? +static const float bloom_excess_static = 0.0; // range [0, 1] +// The BLOOM_APPROX pass approximates a phosphor blur early on with a small +// blurred resize of the input (convergence offsets are applied as well). +// There are three filter options (static option only for now): +// 0.) Bilinear resize: A fast, close approximation to a 4x4 resize +// if min_allowed_viewport_triads and the BLOOM_APPROX resolution are sane +// and gaussian_beam_max_sigma is low. +// 1.) 3x3 resize blur: Medium speed, soft/smeared from bilinear blurring, +// always uses a static sigma regardless of gaussian_beam_max_sigma or +// mask_num_triads_across. +// 2.) True 4x4 Gaussian resize: Slowest, technically correct. +// These options are more pronounced for the fast, unbloomed shader version. +#ifndef RADEON_FIX + #define RADEON_FIX 0 +#endif + +#if !RADEON_FIX + static const float bloom_approx_filter_static = 2.0; +#else + static const float bloom_approx_filter_static = 1.0; +#endif + +// ELECTRON BEAM SCANLINE DISTRIBUTION: +// How many scanlines should contribute light to each pixel? Using more +// scanlines is slower (especially for a generalized Gaussian) but less +// distorted with larger beam sigmas (especially for a pure Gaussian). The +// max_beam_sigma at which the closest unused weight is guaranteed < +// 1.0/255.0 (for a 3x antialiased pure Gaussian) is: +// 2 scanlines: max_beam_sigma = 0.2089; distortions begin ~0.34; 141.7 FPS pure, 131.9 FPS generalized +// 3 scanlines, max_beam_sigma = 0.3879; distortions begin ~0.52; 137.5 FPS pure; 123.8 FPS generalized +// 4 scanlines, max_beam_sigma = 0.5723; distortions begin ~0.70; 134.7 FPS pure; 117.2 FPS generalized +// 5 scanlines, max_beam_sigma = 0.7591; distortions begin ~0.89; 131.6 FPS pure; 112.1 FPS generalized +// 6 scanlines, max_beam_sigma = 0.9483; distortions begin ~1.08; 127.9 FPS pure; 105.6 FPS generalized +static const float beam_num_scanlines = 3.0; // range [2, 6] +// A generalized Gaussian beam varies shape with color too, now just width. +// It's slower but more flexible (static option only for now). +static const bool beam_generalized_gaussian = true; +// What kind of scanline antialiasing do you want? +// 0: Sample weights at 1x; 1: Sample weights at 3x; 2: Compute an integral +// Integrals are slow (especially for generalized Gaussians) and rarely any +// better than 3x antialiasing (static option only for now). +static const float beam_antialias_level = 1.0; // range [0, 2] +// Min/max standard deviations for scanline beams: Higher values widen and +// soften scanlines. Depending on other options, low min sigmas can alias. +static const float gaussian_beam_min_sigma_static = 0.02; // range (0, 1] +static const float gaussian_beam_max_sigma_static = 0.3; // range (0, 1] +// Beam width varies as a function of color: A power function (0) is more +// configurable, but a spherical function (1) gives the widest beam +// variability without aliasing (static option only for now). +static const float beam_spot_shape_function = 0.0; +// Spot shape power: Powers <= 1 give smoother spot shapes but lower +// sharpness. Powers >= 1.0 are awful unless mix/max sigmas are close. +static const float gaussian_beam_spot_power_static = 1.0/3.0; // range (0, 16] +// Generalized Gaussian max shape parameters: Higher values give flatter +// scanline plateaus and steeper dropoffs, simultaneously widening and +// sharpening scanlines at the cost of aliasing. 2.0 is pure Gaussian, and +// values > ~40.0 cause artifacts with integrals. +static const float gaussian_beam_min_shape_static = 2.0; // range [2, 32] +static const float gaussian_beam_max_shape_static = 4.0; // range [2, 32] +// Generalized Gaussian shape power: Affects how quickly the distribution +// changes shape from Gaussian to steep/plateaued as color increases from 0 +// to 1.0. Higher powers appear softer for most colors, and lower powers +// appear sharper for most colors. +static const float gaussian_beam_shape_power_static = 1.0/4.0; // range (0, 16] +// What filter should be used to sample scanlines horizontally? +// 0: Quilez (fast), 1: Gaussian (configurable), 2: Lanczos2 (sharp) +static const float beam_horiz_filter_static = 0.0; +// Standard deviation for horizontal Gaussian resampling: +static const float beam_horiz_sigma_static = 0.35; // range (0, 2/3] +// Do horizontal scanline sampling in linear RGB (correct light mixing), +// gamma-encoded RGB (darker, hard spot shape, may better match bandwidth- +// limiting circuitry in some CRT's), or a weighted avg.? +static const float beam_horiz_linear_rgb_weight_static = 1.0; // range [0, 1] +// Simulate scanline misconvergence? This needs 3x horizontal texture +// samples and 3x texture samples of BLOOM_APPROX and HALATION_BLUR in +// later passes (static option only for now). +static const bool beam_misconvergence = true; +// Convergence offsets in x/y directions for R/G/B scanline beams in units +// of scanlines. Positive offsets go right/down; ranges [-2, 2] +static const float2 convergence_offsets_r_static = float2(0.1, 0.2); +static const float2 convergence_offsets_g_static = float2(0.3, 0.4); +static const float2 convergence_offsets_b_static = float2(0.5, 0.6); +// Detect interlacing (static option only for now)? +static const bool interlace_detect = true; +// Assume 1080-line sources are interlaced? +static const bool interlace_1080i_static = false; +// For interlaced sources, assume TFF (top-field first) or BFF order? +// (Whether this matters depends on the nature of the interlaced input.) +static const bool interlace_back_field_first_static = false; + +// ANTIALIASING: +// What AA level do you want for curvature/overscan/subpixels? Options: +// 0x (none), 1x (sample subpixels), 4x, 5x, 6x, 7x, 8x, 12x, 16x, 20x, 24x +// (Static option only for now) +#ifndef antialias_level + #define antialias_level 0.0 +#endif +// static const float aa_level = 12.0; // range [0, 24] +// static const float aa_level = 0.0; // range [0, 24] +// What antialiasing filter do you want (static option only)? Options: +// 0: Box (separable), 1: Box (cylindrical), +// 2: Tent (separable), 3: Tent (cylindrical), +// 4: Gaussian (separable), 5: Gaussian (cylindrical), +// 6: Cubic* (separable), 7: Cubic* (cylindrical, poor) +// 8: Lanczos Sinc (separable), 9: Lanczos Jinc (cylindrical, poor) +// * = Especially slow with _RUNTIME_ANTIALIAS_WEIGHTS +#ifndef antialias_filter + #define antialias_filter 6 +#endif +static const float aa_filter = antialias_filter; // range [0, 9] +// Flip the sample grid on odd/even frames (static option only for now)? +#ifndef antialias_temporal + #define antialias_temporal false +#endif +static const bool aa_temporal = antialias_temporal; +// Use RGB subpixel offsets for antialiasing? The pixel is at green, and +// the blue offset is the negative r offset; range [0, 0.5] +static const float2 aa_subpixel_r_offset_static = float2(-1.0/3.0, 0.0);//float2(0.0); +// Cubics: See http://www.imagemagick.org/Usage/filter/#mitchell +// 1.) "Keys cubics" with B = 1 - 2C are considered the highest quality. +// 2.) C = 0.5 (default) is Catmull-Rom; higher C's apply sharpening. +// 3.) C = 1.0/3.0 is the Mitchell-Netravali filter. +// 4.) C = 0.0 is a soft spline filter. +static const float aa_cubic_c_static = 0.5; // range [0, 4] +// Standard deviation for Gaussian antialiasing: Try 0.5/aa_pixel_diameter. +static const float aa_gauss_sigma_static = 0.5; // range [0.0625, 1.0] + +// PHOSPHOR MASK: +// Mask type: 0 = aperture grille, 1 = slot mask, 2 = shadow mask +// 3 = lowres grille, 4 = lowres slot, 5 = lowres shadow +static const float mask_type_static = 4.0; // range [0, 5] +// We can sample the mask three ways. Pick 2/3 from: Pretty/Fast/Flexible. +// 0.) Sinc-resize to the desired dot pitch manually (pretty/slow/flexible). +// This requires _PHOSPHOR_MASK_MANUALLY_RESIZE to be #defined. +// 1.) Hardware-resize to the desired dot pitch (ugly/fast/flexible). This +// is halfway decent with LUT mipmapping but atrocious without it. +// 2.) Tile it without resizing at a 1:1 texel:pixel ratio for flat coords +// (pretty/fast/inflexible). Each input LUT has a fixed dot pitch. +// This mode reuses the same masks, so triads will be enormous unless +// you change the mask LUT filenames in your .cgp file. +static const float mask_sample_mode_static = 0.0; // range [0, 2] +// Prefer setting the triad size (0.0) or number on the screen (1.0)? +// If _RUNTIME_PHOSPHOR_BLOOM_SIGMA isn't #defined, the specified triad size +// will always be used to calculate the full bloom sigma statically. +static const float mask_size_param_static = 0.0; // range [0, 1] +// Specify the phosphor triad size, in pixels. Each tile (usually with 8 +// triads) will be rounded to the nearest integer tile size and clamped to +// obey minimum size constraints (imposed to reduce downsize taps) and +// maximum size constraints (imposed to have a sane MASK_RESIZE FBO size). +// To increase the size limit, double the viewport-relative scales for the +// two MASK_RESIZE passes in crt-royale.cgp and user-cgp-contants.h. +// range [1, mask_texture_small_size/mask_triads_per_tile] +static const float mask_triad_width_static = 24.0 / 8.0; +// If mask_size_param is 1.0/true, we'll go by this instead (the +// final size will be rounded and constrained as above); default 480.0 +static const float mask_num_triads_across_static = 480.0; +// How many lobes should the sinc/Lanczos resizer use? More lobes require +// more samples and avoid moire a bit better, but some is unavoidable +// depending on the destination size (static option for now). +static const float mask_sinc_lobes = 3.0; // range [2, 4] +// The mask is resized using a variable number of taps in each dimension, +// but some Cg profiles always fetch a constant number of taps no matter +// what (no dynamic branching). We can limit the maximum number of taps if +// we statically limit the minimum phosphor triad size. Larger values are +// faster, but the limit IS enforced (static option only, forever); +// range [1, mask_texture_small_size/mask_triads_per_tile] +// TODO: Make this 1.0 and compensate with smarter sampling! +static const float mask_min_allowed_triad_size = 2.0; + +// GEOMETRY: +// Geometry mode: +// 0: Off (default), 1: Spherical mapping (like cgwg's), +// 2: Alt. spherical mapping (more bulbous), 3: Cylindrical/Trinitron +static const float geom_mode_static = 0.0; // range [0, 3] +// Radius of curvature: Measured in units of your viewport's diagonal size. +static const float geom_radius_static = 2.0; // range [1/(2*pi), 1024] +// View dist is the distance from the player to their physical screen, in +// units of the viewport's diagonal size. It controls the field of view. +static const float geom_view_dist_static = 2.0; // range [0.5, 1024] +// Tilt angle in radians (clockwise around up and right vectors): +static const float2 geom_tilt_angle_static = float2(0.0, 0.0); // range [-pi, pi] +// Aspect ratio: When the true viewport size is unknown, this value is used +// to help convert between the phosphor triad size and count, along with +// the mask_resize_viewport_scale constant from user-cgp-constants.h. Set +// this equal to Retroarch's display aspect ratio (DAR) for best results; +// range [1, geom_max_aspect_ratio from user-cgp-constants.h]; +// default (256/224)*(54/47) = 1.313069909 (see below) +static const float geom_aspect_ratio_static = 1.313069909; +// Before getting into overscan, here's some general aspect ratio info: +// - DAR = display aspect ratio = SAR * PAR; as in your Retroarch setting +// - SAR = storage aspect ratio = DAR / PAR; square pixel emulator frame AR +// - PAR = pixel aspect ratio = DAR / SAR; holds regardless of cropping +// Geometry processing has to "undo" the screen-space 2D DAR to calculate +// 3D view vectors, then reapplies the aspect ratio to the simulated CRT in +// uv-space. To ensure the source SAR is intended for a ~4:3 DAR, either: +// a.) Enable Retroarch's "Crop Overscan" +// b.) Readd horizontal padding: Set overscan to e.g. N*(1.0, 240.0/224.0) +// Real consoles use horizontal black padding in the signal, but emulators +// often crop this without cropping the vertical padding; a 256x224 [S]NES +// frame (8:7 SAR) is intended for a ~4:3 DAR, but a 256x240 frame is not. +// The correct [S]NES PAR is 54:47, found by blargg and NewRisingSun: +// http://board.zsnes.com/phpBB3/viewtopic.php?f=22&t=11928&start=50 +// http://forums.nesdev.com/viewtopic.php?p=24815#p24815 +// For flat output, it's okay to set DAR = [existing] SAR * [correct] PAR +// without doing a. or b., but horizontal image borders will be tighter +// than vertical ones, messing up curvature and overscan. Fixing the +// padding first corrects this. +// Overscan: Amount to "zoom in" before cropping. You can zoom uniformly +// or adjust x/y independently to e.g. readd horizontal padding, as noted +// above: Values < 1.0 zoom out; range (0, inf) +static const float2 geom_overscan_static = float2(1.0, 1.0);// * 1.005 * (1.0, 240/224.0) +// Compute a proper pixel-space to texture-space matrix even without ddx()/ +// ddy()? This is ~8.5% slower but improves antialiasing/subpixel filtering +// with strong curvature (static option only for now). +static const bool geom_force_correct_tangent_matrix = true; + +// BORDERS: +// Rounded border size in texture uv coords: +static const float border_size_static = 0.015; // range [0, 0.5] +// Border darkness: Moderate values darken the border smoothly, and high +// values make the image very dark just inside the border: +static const float border_darkness_static = 2.0; // range [0, inf) +// Border compression: High numbers compress border transitions, narrowing +// the dark border area. +static const float border_compress_static = 2.5; // range [1, inf) + +// TODO: Nuke this +#define mask_size_xy float2(512, 512) + +#endif // _USER_SETTINGS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/bloom.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/bloom.fxh new file mode 100644 index 000000000..8cc4832f5 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/bloom.fxh @@ -0,0 +1,149 @@ +#ifndef _BLOOM_H +#define _BLOOM_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + +#include "../lib/user-settings.fxh" +#include "../lib/derived-settings-and-constants.fxh" +#include "../lib/bind-shader-params.fxh" +#include "../lib/gamma-management.fxh" +#include "../lib/downsampling-functions.fxh" +#include "../lib/blur-functions.fxh" +#include "../lib/bloom-functions.fxh" + +#include "shared-objects.fxh" + + +void approximateBloomVertPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target +) { + const float2 delta_uv = blur_radius * float2(0.0, rcp(TEX_BEAMCONVERGENCE_HEIGHT)); + + color = float4(opaque_linear_downsample( + samplerBeamConvergence, texcoord, + uint((bloomapprox_downsizing_factor - 1)/2), + delta_uv + ), 1); +} + +void approximateBloomHorizPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target +) { + const float2 delta_uv = blur_radius * float2(rcp(TEX_BEAMCONVERGENCE_WIDTH), 0.0); + + color = float4(opaque_linear_downsample( + samplerBloomApproxVert, texcoord, + uint((bloomapprox_downsizing_factor - 1)/2), + delta_uv + ), 1); +} + + +void bloomHorizontalVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float bloom_sigma_runtime : TEXCOORD1 +) { + PostProcessVS(id, position, texcoord); + + bloom_sigma_runtime = get_min_sigma_to_blur_triad(calc_triad_size().x, bloom_diff_thresh_); +} + +void bloomHorizontalPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + in float bloom_sigma_runtime : TEXCOORD1, + + out float4 color : SV_Target +) { + const float2 bloom_dxdy = float2(rcp(TEX_BLOOMVERTICAL_WIDTH), 0); + + // Blur the vertically blurred brightpass horizontally by 9/17/25/43x: + const float bloom_sigma = get_final_bloom_sigma(bloom_sigma_runtime); + const float3 blurred_brightpass = tex2DblurNfast(samplerBloomVertical, + texcoord, bloom_dxdy, bloom_sigma, get_intermediate_gamma()); + + // Sample the masked scanlines. Alpha contains the auto-dim factor: + const float3 intensity_dim = tex2D_linearize(samplerMaskedScanlines, texcoord, get_intermediate_gamma()).rgb; + const float auto_dim_factor = levels_autodim_temp; + const float undim_factor = 1.0/auto_dim_factor; + + // Calculate the mask dimpass, add it to the blurred brightpass, and + // undim (from scanline auto-dim) and amplify (from mask dim) the result: + const float mask_amplify = get_mask_amplify(); + const float3 brightpass = tex2D_linearize(samplerBrightpass, texcoord, get_intermediate_gamma()).rgb; + const float3 dimpass = intensity_dim - brightpass; + const float3 phosphor_bloom = (dimpass + blurred_brightpass) * + mask_amplify * undim_factor * levels_contrast; + + // Sample the halation texture, and let some light bleed into refractive + // diffusion. Conceptually this occurs before the phosphor bloom, but + // adding it in earlier passes causes black crush in the diffusion colors. + const float3 raw_diffusion_color = tex2D_linearize(samplerBlurHorizontal, texcoord, get_intermediate_gamma()).rgb; + const float3 raw_halation_color = dot(raw_diffusion_color, float3(1, 1, 1)) / 3.0; + const float3 diffusion_color = levels_contrast * lerp(raw_diffusion_color, raw_halation_color, halation_weight); + const float3 final_bloom = lerp(phosphor_bloom, diffusion_color, diffusion_weight); + + // Encode and output the bloomed image: + color = encode_output(float4(final_bloom, 1.0), get_intermediate_gamma()); +} + + +void bloomVerticalVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float bloom_sigma_runtime : TEXCOORD1 +) { + PostProcessVS(id, position, texcoord); + + bloom_sigma_runtime = get_min_sigma_to_blur_triad(calc_triad_size().x, bloom_diff_thresh_); +} + +void bloomVerticalPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + in float bloom_sigma_runtime : TEXCOORD1, + + out float4 color : SV_Target +) { + const float2 bloom_dxdy = float2(0, rcp(TEX_BLOOMVERTICAL_HEIGHT)); + + // Blur the brightpass horizontally with a 9/17/25/43x blur: + const float bloom_sigma = get_final_bloom_sigma(bloom_sigma_runtime); + const float3 color3 = tex2DblurNfast(samplerBrightpass, texcoord, + bloom_dxdy, bloom_sigma, get_intermediate_gamma()); + + // Encode and output the blurred image: + color = encode_output(float4(color3, 1.0), get_intermediate_gamma()); +} + +#endif // _BLOOM_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/blurring.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/blurring.fxh new file mode 100644 index 000000000..0695eace8 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/blurring.fxh @@ -0,0 +1,131 @@ +#ifndef _BLURRING_H +#define _BLURRING_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2014 TroggleMonkey +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +#include "../lib/gamma-management.fxh" +#include "../lib/blur-functions.fxh" + +#include "shared-objects.fxh" + + +void blurHorizontalVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float2 blur_dxdy : TEXCOORD1 +) { + PostProcessVS(id, position, texcoord); + + // Get the uv sample distance between output pixels. Blurs are not generic + // Gaussian resizers, and correct blurs require: + // 1.) OutputSize == InputSize * 2^m, where m is an integer <= 0. + // 2.) mipmap_inputN = "true" for this pass in the preset if m != 0 + // 3.) filter_linearN = "true" except for 1x scale nearest neighbor blurs + // Gaussian resizers would upsize using the distance between input texels + // (not output pixels), but we avoid this and consistently blur at the + // destination size. Otherwise, combining statically calculated weights + // with bilinear sample exploitation would result in terrible artifacts. + static const float2 output_size = TEX_BLURHORIZONTAL_SIZE; + static const float2 dxdy = 1.0 / output_size; + // This blur is vertical-only, so zero out the horizontal offset: + blur_dxdy = float2(dxdy.x, 0.0); +} + +void blurHorizontalPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + in float2 blur_dxdy : TEXCOORD1, + + out float4 color : SV_Target +) { + static const float3 blur_color = tex2Dblur9fast(samplerBlurVertical, texcoord, blur_dxdy, get_intermediate_gamma()); + // Encode and output the blurred image: + // color = encode_output(float4(blur_color, 1.0), 1.0); + color = encode_output(float4(blur_color, 1.0), get_intermediate_gamma()); +} + + +void blurVerticalVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float2 blur_dxdy : TEXCOORD1 +) { + PostProcessVS(id, position, texcoord); + + // Get the uv sample distance between output pixels. Blurs are not generic + // Gaussian resizers, and correct blurs require: + // 1.) OutputSize == InputSize * 2^m, where m is an integer <= 0. + // 2.) mipmap_inputN = "true" for this pass in the preset if m != 0 + // 3.) filter_linearN = "true" except for 1x scale nearest neighbor blurs + // Gaussian resizers would upsize using the distance between input texels + // (not output pixels), but we avoid this and consistently blur at the + // destination size. Otherwise, combining statically calculated weights + // with bilinear sample exploitation would result in terrible artifacts. + static const float2 output_size = TEX_BLURVERTICAL_SIZE; + static const float2 dxdy = 1.0 / output_size; + // This blur is vertical-only, so zero out the horizontal offset: + blur_dxdy = float2(0.0, dxdy.y); +} + +void blurVerticalPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + in float2 blur_dxdy : TEXCOORD1, + + out float4 color : SV_Target +) { + static const float3 blur_color = tex2Dblur9fast(samplerBloomApproxHoriz, texcoord, blur_dxdy, get_intermediate_gamma()); + // Encode and output the blurred image: + // color = encode_output(float4(blur_color, 1.0), 1.0); + color = encode_output(float4(blur_color, 1.0), get_intermediate_gamma()); +} + +#endif // _BLURRING_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/brightpass.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/brightpass.fxh new file mode 100644 index 000000000..fd5ad8fb8 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/brightpass.fxh @@ -0,0 +1,90 @@ +#ifndef _BRIGHTPASS_H +#define _BRIGHTPASS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +#include "../lib/user-settings.fxh" +#include "../lib/derived-settings-and-constants.fxh" +#include "../lib/bind-shader-params.fxh" +#include "../lib/gamma-management.fxh" +#include "../lib/phosphor-mask-calculations.fxh" +#include "../lib/scanline-functions.fxh" +#include "../lib/bloom-functions.fxh" +#include "../lib/blur-functions.fxh" + + +void brightpassVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float bloom_sigma_runtime : TEXCOORD1 +) { + PostProcessVS(id, position, texcoord); + + bloom_sigma_runtime = get_min_sigma_to_blur_triad(calc_triad_size().x, bloom_diff_thresh_); +} + +void brightpassPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + in float bloom_sigma_runtime : TEXCOORD1, + + out float4 color : SV_Target +) { + // Sample the masked scanlines: + const float3 intensity_dim = tex2D_linearize(samplerMaskedScanlines, texcoord, get_intermediate_gamma()).rgb; + // Get the full intensity, including auto-undimming, and mask compensation: + const float mask_amplify = get_mask_amplify(); + const float3 intensity = intensity_dim * rcp(levels_autodim_temp) * mask_amplify * levels_contrast; + + // Sample BLOOM_APPROX to estimate what a straight blur of masked scanlines + // would look like, so we can estimate how much energy we'll receive from + // blooming neighbors: + const float3 phosphor_blur_approx = levels_contrast * tex2D_linearize(samplerBloomApproxHoriz, texcoord, get_intermediate_gamma()).rgb; + + // Compute the blur weight for the center texel and the maximum energy we + // expect to receive from neighbors: + const float bloom_sigma = get_final_bloom_sigma(bloom_sigma_runtime); + const float center_weight = get_center_weight(bloom_sigma); + const float3 max_area_contribution_approx = + max(float3(0.0, 0.0, 0.0), phosphor_blur_approx - center_weight * intensity); + // Assume neighbors will blur 100% of their intensity (blur_ratio = 1.0), + // because it actually gets better results (on top of being very simple), + // but adjust all intensities for the user's desired underestimate factor: + const float3 area_contrib_underestimate = bloom_underestimate_levels * max_area_contribution_approx; + const float3 intensity_underestimate = bloom_underestimate_levels * intensity; + // Calculate the blur_ratio, the ratio of intensity we want to blur: + const float3 blur_ratio_temp = + ((float3(1.0, 1.0, 1.0) - area_contrib_underestimate) / + intensity_underestimate - float3(1.0, 1.0, 1.0)) / (center_weight - 1.0); + const float3 blur_ratio = saturate(blur_ratio_temp); + // Calculate the brightpass based on the auto-dimmed, unamplified, masked + // scanlines, encode if necessary, and return! + const float3 brightpass = intensity_dim * + lerp(blur_ratio, float3(1.0, 1.0, 1.0), bloom_excess); + + color = encode_output(float4(brightpass, 1.0), get_intermediate_gamma()); +} + +#endif // _BRIGHTPASS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/content-box.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/content-box.fxh new file mode 100644 index 000000000..ccde83258 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/content-box.fxh @@ -0,0 +1,221 @@ +#ifndef _CONTENT_BOX_H +#define _CONTENT_BOX_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2020 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +#include "shared-objects.fxh" + + +void contentCropVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0 +) { + #if _DX9_ACTIVE + texcoord.x = (id == 1 || id == 3) ? content_right : content_left; + texcoord.y = (id > 1) ? content_lower : content_upper; + + position.x = (id == 1 || id == 3) ? 1 : -1; + position.y = (id > 1) ? -1 : 1; + position.zw = 1; + #else + texcoord.x = (id & 1) ? content_right : content_left; + texcoord.y = (id & 2) ? content_lower : content_upper; + + position.x = (id & 1) ? 1 : -1; + position.y = (id & 2) ? -1 : 1; + position.zw = 1; + #endif +} + +#if USE_VERTEX_UNCROPPING +/* + * Using the vertex shader for uncropping can save about 0.1ms in some apps. + * However, some apps like SNES9X w/ DX9 don't trigger a refresh of the entire screen, + * which in turn causes the ReShade UI to "stick around" after it's closed. + * + * The slower algorithm forces the entire screen to refresh, which forces the + * area outside the content box to be black. I assume most users will prefer + * the results of the slower algorithm and won't notice the 0.1ms. Users who + * need that 0.1ms can use a preprocessor def to recover that time. + */ + void contentUncropVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0 + ) { + #if _DX9_ACTIVE + texcoord.x = id == 1 || id == 3; + texcoord.y = id < 2; + + position.x = (id == 1 || id == 3) ? content_scale.x : -content_scale.x; + position.y = (id > 1) ? content_scale.y : -content_scale.y; + position.zw = 1; + #else + texcoord.x = id & 1; + texcoord.y = !(id & 2); + + position.x = (id & 1) ? content_scale.x : -content_scale.x; + position.y = (id & 2) ? content_scale.y : -content_scale.y; + position.zw = 1; + #endif + } + + void uncropContentPixelShader( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target + ) { + color = tex2D(samplerGeometry, texcoord); + } +#else + void contentUncropVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0 + ) { + // TODO: There's probably a better way to code this. + // I'll figure it out later. + #if _DX9_ACTIVE + texcoord.x = id == 1 || id == 3; + texcoord.y = id < 2; + + position.x = (id == 1 || id == 3) ? 1 : -1; + position.y = (id > 1) ? 1 : -1; + position.zw = 1; + #else + texcoord.x = id & 1; + texcoord.y = !(id & 2); + + position.x = (id & 1) ? 1 : -1; + position.y = (id & 2) ? 1 : -1; + position.zw = 1; + #endif + } + + void uncropContentPixelShader( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target + ) { + const bool is_in_boundary = float( + texcoord.x >= content_left && texcoord.x <= content_right && + texcoord.y >= content_upper && texcoord.y <= content_lower + ); + const float2 texcoord_uncropped = ((texcoord - content_offset) * buffer_size + 0) / content_size; + + const float4 raw_color = tex2D(samplerGeometry, texcoord_uncropped); + color = float4(is_in_boundary * raw_color.rgb, raw_color.a); + } +#endif + + +#if CONTENT_BOX_VISIBLE + #ifndef CONTENT_BOX_INSCRIBED + #define CONTENT_BOX_INSCRIBED 1 + #endif + + #ifndef CONTENT_BOX_THICKNESS + #define CONTENT_BOX_THICKNESS 5 + #endif + + #ifndef CONTENT_BOX_COLOR_R + #define CONTENT_BOX_COLOR_R 1.0 + #endif + + #ifndef CONTENT_BOX_COLOR_G + #define CONTENT_BOX_COLOR_G 0.0 + #endif + + #ifndef CONTENT_BOX_COLOR_B + #define CONTENT_BOX_COLOR_B 0.0 + #endif + + static const float vert_line_thickness = float(CONTENT_BOX_THICKNESS) / BUFFER_WIDTH; + static const float horiz_line_thickness = float(CONTENT_BOX_THICKNESS) / BUFFER_HEIGHT; + + #if CONTENT_BOX_INSCRIBED + // Set the outer borders to the edge of the content + static const float left_line_1 = content_left; + static const float left_line_2 = left_line_1 + vert_line_thickness; + static const float right_line_2 = content_right; + static const float right_line_1 = right_line_2 - vert_line_thickness; + + static const float upper_line_1 = content_upper; + static const float upper_line_2 = upper_line_1 + horiz_line_thickness; + static const float lower_line_2 = content_lower; + static const float lower_line_1 = lower_line_2 - horiz_line_thickness; + #else + // Set the inner borders to the edge of the content + static const float left_line_2 = content_left; + static const float left_line_1 = left_line_2 - vert_line_thickness; + static const float right_line_1 = content_right; + static const float right_line_2 = right_line_1 + vert_line_thickness; + + static const float upper_line_2 = content_upper; + static const float upper_line_1 = upper_line_2 - horiz_line_thickness; + static const float lower_line_1 = content_lower; + static const float lower_line_2 = lower_line_1 + horiz_line_thickness; + #endif + + + static const float4 box_color = float4( + CONTENT_BOX_COLOR_R, + CONTENT_BOX_COLOR_G, + CONTENT_BOX_COLOR_B, + 1.0 + ); + + void contentBoxPixelShader( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target + ) { + + const bool is_inside_outerbound = ( + texcoord.x >= left_line_1 && texcoord.x <= right_line_2 && + texcoord.y >= upper_line_1 && texcoord.y <= lower_line_2 + ); + const bool is_outside_innerbound = ( + texcoord.x <= left_line_2 || texcoord.x >= right_line_1 || + texcoord.y <= upper_line_2 || texcoord.y >= lower_line_1 + ); + + if (is_inside_outerbound && is_outside_innerbound) { + color = box_color; + } + else { + color = tex2D(ReShade::BackBuffer, texcoord); + } + } + + +#endif // CONTENT_BOX_VISIBLE +#endif // _CONTENT_BOX_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/deinterlace.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/deinterlace.fxh new file mode 100644 index 000000000..15878c5df --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/deinterlace.fxh @@ -0,0 +1,137 @@ +#ifndef _DEINTERLACE_H +#define _DEINTERLACE_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2020 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +#include "../lib/user-settings.fxh" +#include "../lib/derived-settings-and-constants.fxh" +#include "../lib/bind-shader-params.fxh" +#include "../lib/gamma-management.fxh" +#include "../lib/scanline-functions.fxh" + + + +void freezeFrameVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0 +) { + float use_deinterlacing_tex = enable_interlacing && ( + scanline_deinterlacing_mode == 2 || scanline_deinterlacing_mode == 3 + ); + + texcoord.x = (id == 2) ? use_deinterlacing_tex*2.0 : 0.0; + texcoord.y = (id == 1) ? 2.0 : 0.0; + position = float4(texcoord * float2(2, -2) + float2(-1, 1), 0, 1); +} + +void freezeFramePS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target +) { + color = tex2D(samplerBeamConvergence, texcoord); +} + + +void deinterlaceVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float2 v_step : TEXCOORD1 +) { + freezeFrameVS(id, position, texcoord); + + v_step = float2(0.0, scanline_thickness * rcp(TEX_FREEZEFRAME_HEIGHT)); +} + + +void deinterlacePS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + in float2 v_step : TEXCOORD1, + + out float4 color : SV_Target +) { + // float2 scanline_offset_norm; + // float triangle_wave_freq; + // bool field_parity; + // bool wrong_field; + // calc_wrong_field(texcoord, scanline_offset_norm, triangle_wave_freq, field_parity, wrong_field); + + float2 rotated_coord = lerp(texcoord.yx, texcoord, geom_rotation_mode == 0 || geom_rotation_mode == 2); + float scale = lerp(CONTENT_WIDTH, CONTENT_HEIGHT, geom_rotation_mode == 0 || geom_rotation_mode == 2); + + InterpolationFieldData interpolation_data = calc_interpolation_field_data(rotated_coord, scale); + + // TODO: add scanline_parity to calc_wrong_field() + + // Weaving + // Sample texcoord from this frame and the previous frame + // If we're in the correct field, use the current sample + // If we're in the wrong field, average the current and prev samples + // In this case, we're probably averaging a color with 0 and producing a brightness of 0.5. + [branch] + if (enable_interlacing && scanline_deinterlacing_mode == 2) { + // const float cur_scanline_idx = get_curr_scanline_idx(texcoord.y, content_size.y); + // const float wrong_field = curr_line_is_wrong_field(cur_scanline_idx); + + const float4 cur_line_color = tex2D_nograd(samplerBeamConvergence, texcoord); + const float4 cur_line_prev_color = tex2D_nograd(samplerFreezeFrame, texcoord); + + const float4 avg_color = (cur_line_color + cur_line_prev_color) / 2.0; + + // Multiply by 1.5, so each pair of scanlines has total brightness 2 + const float4 raw_out_color = lerp(1.5*cur_line_color, avg_color, interpolation_data.wrong_field); + color = encode_output(raw_out_color, deinterlacing_blend_gamma); + } + // Blended Weaving + // Sample texcoord from this frame + // From the previous frame, sample the current scanline's sibling + // Do this by shifting up or down by a line + // If we're in the correct field, use the current sample + // If we're in the wrong field, average the current and prev samples + // In this case, we're averaging two fully illuminated colors + else if (enable_interlacing && scanline_deinterlacing_mode == 3) { + const float2 raw_offset = lerp(1, -1, interpolation_data.scanline_parity) * v_step; + const float2 curr_offset = lerp(0, raw_offset, interpolation_data.wrong_field); + const float2 prev_offset = lerp(raw_offset, 0, interpolation_data.wrong_field); + + const float4 cur_line_color = tex2D_nograd(samplerBeamConvergence, texcoord + curr_offset); + const float4 prev_line_color = tex2D_nograd(samplerFreezeFrame, texcoord + prev_offset); + + const float4 avg_color = (cur_line_color + prev_line_color) / 2.0; + const float4 raw_out_color = lerp(cur_line_color, avg_color, interpolation_data.wrong_field); + color = encode_output(raw_out_color, deinterlacing_blend_gamma); + } + // No temporal blending + else { + color = tex2D_nograd(samplerBeamConvergence, texcoord); + } +} + +#endif // _DEINTERLACE_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/electron-beams.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/electron-beams.fxh new file mode 100644 index 000000000..a0bec77c0 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/electron-beams.fxh @@ -0,0 +1,347 @@ +#ifndef _ELECTRON_BEAMS_H +#define _ELECTRON_BEAMS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +#include "../lib/bind-shader-params.fxh" +#include "../lib/gamma-management.fxh" +#include "../lib/scanline-functions.fxh" + +#include "content-box.fxh" +#include "shared-objects.fxh" + + +void calculateBeamDistsVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0 +) { + const float compute_mask_factor = frame_count % 60 == 0 || overlay_active > 0; + + texcoord.x = (id == 2) ? compute_mask_factor*2.0 : 0.0; + texcoord.y = (id == 1) ? 2.0 : 0.0; + position = float4(texcoord * float2(2, -2) + float2(-1, 1), 0, 1); +} + + +void calculateBeamDistsPS( + in float4 position : SV_Position, + in float2 texcoord : TEXCOORD0, + + out float4 beam_strength : SV_Target +) { + InterpolationFieldData interpolation_data = precalc_interpolation_field_data(texcoord); + + // We have to subtract off the texcoord offset to make sure we're using domain [0, 1] + const float color_corrected = texcoord.x - 1.0 / TEX_BEAMDIST_WIDTH; + + // Digital shape + // Beam will be perfectly rectangular + [branch] + if (beam_shape_mode == 0) { + // Double the intensity when interlacing to maintain the same apparent brightness + const float interlacing_brightness_factor = 1 + float( + enable_interlacing && + (scanline_deinterlacing_mode != 2) && + (scanline_deinterlacing_mode != 3) + ); + const float raw_beam_strength = (1 - interpolation_data.scanline_parity * enable_interlacing) * interlacing_brightness_factor * levels_autodim_temp; + + beam_strength = float4(color_corrected * raw_beam_strength, 0, 0, 1); + } + // Linear shape + // Beam intensity will drop off linarly with distance from center + // Works better than gaussian with narrow scanlines (about 1-6 pixels wide) + // Will only consider contribution from nearest scanline + else if (beam_shape_mode == 1) { + const float beam_dist_y = triangle_wave(texcoord.y, interpolation_data.triangle_wave_freq); + + const bool scanline_is_wider_than_1 = scanline_thickness > 1; + const bool deinterlacing_mode_requires_boost = ( + enable_interlacing && + (scanline_deinterlacing_mode != 2) && + (scanline_deinterlacing_mode != 3) + ); + + const float interlacing_brightness_factor = (1 + scanline_is_wider_than_1) * (1 + deinterlacing_mode_requires_boost); + // const float raw_beam_strength = (1 - beam_dist_y) * (1 - interpolation_data.scanline_parity * enable_interlacing) * interlacing_brightness_factor * levels_autodim_temp; + // const float raw_beam_strength = (1 - beam_dist_y); + const float raw_beam_strength = saturate(-beam_dist_y * rcp(linear_beam_thickness) + 1); + const float adj_beam_strength = raw_beam_strength * (1 - interpolation_data.scanline_parity * enable_interlacing) * interlacing_brightness_factor * levels_autodim_temp; + + beam_strength = float4(color_corrected * adj_beam_strength, 0, 0, 1); + } + // Gaussian Shape + // Beam will be a distorted Gaussian, dependent on color brightness and hyperparameters + // Will only consider contribution from nearest scanline + else if (beam_shape_mode == 2) { + // Calculate {sigma, shape}_range outside of scanline_contrib so it's only + // done once per pixel (not 6 times) with runtime params. Don't reuse the + // vertex shader calculations, so static versions can be constant-folded. + const float sigma_range = max(gaussian_beam_max_sigma, gaussian_beam_min_sigma) - gaussian_beam_min_sigma; + const float shape_range = max(gaussian_beam_max_shape, gaussian_beam_min_shape) - gaussian_beam_min_shape; + + const float beam_dist_factor = 1 + float(enable_interlacing); + const float freq_adj = interpolation_data.triangle_wave_freq * rcp(beam_dist_factor); + // The conditional 0.25*f offset ensures the interlaced scanlines align with the non-interlaced ones as in the other beam shapes + const float frame_offset = enable_interlacing * (!interpolation_data.field_parity * 0.5 + 0.25) * rcp(freq_adj); + const float beam_dist_y = triangle_wave((texcoord.y - frame_offset), freq_adj) * rcp(linear_beam_thickness); + + const float interlacing_brightness_factor = 1 + float( + !enable_interlacing && + (scanline_thickness > 1) + ) + float( + enable_interlacing && + (scanline_deinterlacing_mode != 2) && + (scanline_deinterlacing_mode != 3) + ); + const float raw_beam_strength = get_gaussian_beam_strength( + beam_dist_y, color_corrected, + sigma_range, shape_range + ) * interlacing_brightness_factor * levels_autodim_temp; + + beam_strength = float4(raw_beam_strength, 0, 0, 1); + } + // Gaussian Shape + // Beam will be a distorted Gaussian, dependent on color brightness and hyperparameters + // Will consider contributions from current scanline and two neighboring in-field scanlines + else { + // Calculate {sigma, shape}_range outside of scanline_contrib so it's only + // done once per pixel (not 6 times) with runtime params. Don't reuse the + // vertex shader calculations, so static versions can be constant-folded. + const float sigma_range = max(gaussian_beam_max_sigma, gaussian_beam_min_sigma) - gaussian_beam_min_sigma; + const float shape_range = max(gaussian_beam_max_shape, gaussian_beam_min_shape) - gaussian_beam_min_shape; + + const float beam_dist_factor = (1 + float(enable_interlacing)); + const float freq_adj = interpolation_data.triangle_wave_freq * rcp(beam_dist_factor); + // The conditional 0.25*f offset ensures the interlaced scanlines align with the non-interlaced ones as in the other beam shapes + const float frame_offset = enable_interlacing * (!interpolation_data.field_parity * 0.5 + 0.25) * rcp(freq_adj); + const float curr_beam_dist_y = triangle_wave(texcoord.y - frame_offset, freq_adj) * rcp(linear_beam_thickness); + const float upper_beam_dist_y = (sawtooth_incr_wave(texcoord.y - frame_offset, freq_adj)*2 + 1) * rcp(linear_beam_thickness); + const float lower_beam_dist_y = 4 * rcp(linear_beam_thickness) - upper_beam_dist_y; + + const float upper_beam_strength = get_gaussian_beam_strength( + upper_beam_dist_y, color_corrected, + sigma_range, shape_range + ); + const float curr_beam_strength = get_gaussian_beam_strength( + curr_beam_dist_y, color_corrected, + sigma_range, shape_range + ); + const float lower_beam_strength = get_gaussian_beam_strength( + lower_beam_dist_y, color_corrected, + sigma_range, shape_range + ); + + const float interlacing_brightness_factor = 1 + float( + !enable_interlacing && + (scanline_thickness > 1) + ) + float( + enable_interlacing && + (scanline_deinterlacing_mode != 2) && + (scanline_deinterlacing_mode != 3) + ); + const float3 raw_beam_strength = float3(curr_beam_strength, upper_beam_strength, lower_beam_strength) * interlacing_brightness_factor * levels_autodim_temp; + + beam_strength = float4(raw_beam_strength, 1); + } +} + + +void simulateEletronBeamsVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float4 runtime_bin_shapes : TEXCOORD1 +) { + #if ENABLE_PREBLUR + PostProcessVS(id, position, texcoord); + #else + // texcoord.x = (id == 0 || id == 2) ? content_left : content_right; + // texcoord.y = (id < 2) ? content_lower : content_upper; + // position.x = (id == 0 || id == 2) ? -1 : 1; + // position.y = (id < 2) ? -1 : 1; + // position.zw = 1; + contentCropVS(id, position, texcoord); + #endif + + bool screen_is_landscape = geom_rotation_mode == 0 || geom_rotation_mode == 2; + + // Mode 0: size of pixel in [0, 1] = pixel_dims / viewport_size + // Mode 1: size of pixel in [0, 1] = viewport_size / grid_dims + // float2 runtime_pixel_size = (pixel_grid_mode == 0) ? pixel_size * rcp(content_size) : rcp(pixel_grid_resolution); + float2 runtime_pixel_size = rcp(content_size); + float2 runtime_scanline_shape = lerp( + float2(scanline_thickness, 1), + float2(1, scanline_thickness), + screen_is_landscape + ) * rcp(content_size); + + runtime_bin_shapes = float4(runtime_pixel_size, runtime_scanline_shape); +} + +void simulateEletronBeamsPS( + in float4 position : SV_Position, + in float2 texcoord : TEXCOORD0, + in float4 runtime_bin_shapes : TEXCOORD1, + + out float4 color : SV_Target +) { + bool screen_is_landscape = geom_rotation_mode == 0 || geom_rotation_mode == 2; + float2 rotated_coord = lerp(texcoord.yx, texcoord, screen_is_landscape); + float scale = lerp(CONTENT_WIDTH, CONTENT_HEIGHT, screen_is_landscape); + + // InterpolationFieldData interpolation_data = precalc_interpolation_field_data(rotated_coord); + + // // We have to subtract off the texcoord offset to make sure we're using domain [0, 1] + // const float color_corrected = rotated_coord.x - 1.0 / scale; + + + InterpolationFieldData interpolation_data = calc_interpolation_field_data(rotated_coord, scale); + const float ypos = (rotated_coord.y * interpolation_data.triangle_wave_freq + interpolation_data.field_parity) * 0.5; + + float2 texcoord_scanlined = round_coord(texcoord, 0, runtime_bin_shapes.zw); + + // Sample from the neighboring scanline when in the wrong field + [branch] + if (interpolation_data.wrong_field && screen_is_landscape) { + const float coord_moved_up = texcoord_scanlined.y <= texcoord.y; + const float direction = lerp(-1, 1, coord_moved_up); + texcoord_scanlined.y += direction * scanline_thickness * rcp(content_size.y); + } + else if (interpolation_data.wrong_field) { + const float coord_moved_up = texcoord_scanlined.x <= texcoord.x; + const float direction = lerp(-1, 1, coord_moved_up); + texcoord_scanlined.x += direction * scanline_thickness * rcp(content_size.x); + } + + // Now we apply pixellation and cropping + // float2 texcoord_pixellated = round_coord( + // texcoord_scanlined, + // pixel_grid_offset * rcp(content_size), + // runtime_bin_shapes.xy + // ); + float2 texcoord_pixellated = texcoord_scanlined; + + const float2 texcoord_uncropped = texcoord_pixellated; + #if ENABLE_PREBLUR + // If the pre-blur pass ran, then it's already handled cropping. + // const float2 texcoord_uncropped = texcoord_pixellated; + #define source_sampler samplerPreblurHoriz + #else + // const float2 texcoord_uncropped = texcoord_pixellated * content_scale + content_offset; + #define source_sampler ReShade::BackBuffer + #endif + + [branch] + if (beam_shape_mode < 3) { + const float4 scanline_color = tex2Dlod_linearize( + source_sampler, + texcoord_uncropped, + get_input_gamma() + ); + + const float beam_strength_r = tex2D_nograd(samplerBeamDist, float2(scanline_color.r, ypos)).x; + const float beam_strength_g = tex2D_nograd(samplerBeamDist, float2(scanline_color.g, ypos)).x; + const float beam_strength_b = tex2D_nograd(samplerBeamDist, float2(scanline_color.b, ypos)).x; + const float4 beam_strength = float4(beam_strength_r, beam_strength_g, beam_strength_b, 1); + + color = beam_strength; + } + else { + const float2 offset = float2(0, scanline_thickness) * (1 + enable_interlacing) * rcp(content_size); + + const float4 curr_scanline_color = tex2Dlod_linearize( + source_sampler, + texcoord_uncropped, + get_input_gamma() + ); + const float4 upper_scanline_color = tex2Dlod_linearize( + source_sampler, + texcoord_uncropped - offset, + get_input_gamma() + ); + const float4 lower_scanline_color = tex2Dlod_linearize( + source_sampler, + texcoord_uncropped + offset, + get_input_gamma() + ); + + const float curr_beam_strength_r = tex2D_nograd(samplerBeamDist, float2(curr_scanline_color.r, ypos)).x; + const float curr_beam_strength_g = tex2D_nograd(samplerBeamDist, float2(curr_scanline_color.g, ypos)).x; + const float curr_beam_strength_b = tex2D_nograd(samplerBeamDist, float2(curr_scanline_color.b, ypos)).x; + + const float upper_beam_strength_r = tex2D_nograd(samplerBeamDist, float2(upper_scanline_color.r, ypos)).y; + const float upper_beam_strength_g = tex2D_nograd(samplerBeamDist, float2(upper_scanline_color.g, ypos)).y; + const float upper_beam_strength_b = tex2D_nograd(samplerBeamDist, float2(upper_scanline_color.b, ypos)).y; + + const float lower_beam_strength_r = tex2D_nograd(samplerBeamDist, float2(lower_scanline_color.r, ypos)).z; + const float lower_beam_strength_g = tex2D_nograd(samplerBeamDist, float2(lower_scanline_color.g, ypos)).z; + const float lower_beam_strength_b = tex2D_nograd(samplerBeamDist, float2(lower_scanline_color.b, ypos)).z; + + color = float4( + curr_beam_strength_r + upper_beam_strength_r + lower_beam_strength_r, + curr_beam_strength_g + upper_beam_strength_g + lower_beam_strength_g, + curr_beam_strength_b + upper_beam_strength_b + lower_beam_strength_b, + 1 + ); + } +} + +void beamConvergenceVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float run_convergence : TEXCOORD1 +) { + PostProcessVS(id, position, texcoord); + const uint3 x_flag = convergence_offset_x != 0; + const uint3 y_flag = convergence_offset_y != 0; + run_convergence = dot(x_flag, 1) + dot(y_flag, 1); +} + +void beamConvergencePS( + in float4 position : SV_Position, + in float2 texcoord : TEXCOORD0, + in float run_convergence : TEXCOORD1, + + out float4 color : SV_TARGET +) { + // [branch] + if (!run_convergence) { + color = tex2D(samplerElectronBeams, texcoord - float2(0, scanline_offset * rcp(content_size.y))); + } + else { + const float3 offset_sample = sample_rgb_scanline( + samplerElectronBeams, texcoord - float2(0, scanline_offset * rcp(content_size.y)), + TEX_ELECTRONBEAMS_SIZE, rcp(TEX_ELECTRONBEAMS_SIZE) + ); + + color = float4(offset_sample, 1); + } +} + +#endif // _ELECTRON_BEAMS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/geometry-aa-last-pass.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/geometry-aa-last-pass.fxh new file mode 100644 index 000000000..bd6a7dca4 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/geometry-aa-last-pass.fxh @@ -0,0 +1,220 @@ +#ifndef _GEOMETRY_AA_LAST_PASS_H +#define _GEOMETRY_AA_LAST_PASS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale: A full-featured CRT shader, with cheese. +// Copyright (C) 2014 TroggleMonkey +// +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +#include "../lib/user-settings.fxh" +#include "../lib/derived-settings-and-constants.fxh" +#include "../lib/bind-shader-params.fxh" +#include "../lib/gamma-management.fxh" +#include "../lib/tex2Dantialias.fxh" +#include "../lib/geometry-functions.fxh" + +// Disabled in the ReShade port because I don't know a good way to make these +// static AND global AND defined with sin(), cos(), or pow(). + +// #if !_RUNTIME_GEOMETRY_TILT +// // Create a local-to-global rotation matrix for the CRT's coordinate frame +// // and its global-to-local inverse. See the vertex shader for details. +// // It's faster to compute these statically if possible. +// static const float2 sin_tilt = sin(geom_tilt_angle_static); +// static const float2 cos_tilt = cos(geom_tilt_angle_static); +// static const float3x3 geom_local_to_global_static = float3x3( +// cos_tilt.x, sin_tilt.y*sin_tilt.x, cos_tilt.y*sin_tilt.x, +// 0.0, cos_tilt.y, -sin_tilt.y, +// -sin_tilt.x, sin_tilt.y*cos_tilt.x, cos_tilt.y*cos_tilt.x); +// static const float3x3 geom_global_to_local_static = float3x3( +// cos_tilt.x, 0.0, -sin_tilt.x, +// sin_tilt.y*sin_tilt.x, cos_tilt.y, sin_tilt.y*cos_tilt.x, +// cos_tilt.y*sin_tilt.x, -sin_tilt.y, cos_tilt.y*cos_tilt.x); +// #endif + +float2x2 mul_scale(float2 scale, float2x2 mtrx) +{ + float4 temp_matrix = float4(mtrx[0][0], mtrx[0][1], mtrx[1][0], mtrx[1][1]) * scale.xxyy; + return float2x2(temp_matrix.x, temp_matrix.y, temp_matrix.z, temp_matrix.w); +} + + +void geometryVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float2 output_size_inv : TEXCOORD1, + out float4 geom_aspect_and_overscan : TEXCOORD2, + out float3 eye_pos_local : TEXCOORD3, + out float3 global_to_local_row0 : TEXCOORD4, + out float3 global_to_local_row1 : TEXCOORD5, + out float3 global_to_local_row2 : TEXCOORD6 +) { + PostProcessVS(id, position, texcoord); + + output_size_inv = 1.0 / content_size; + + // Get aspect/overscan vectors from scalar parameters (likely uniforms): + const float viewport_aspect_ratio = output_size_inv.y / output_size_inv.x; + const float2 geom_aspect = get_aspect_vector(viewport_aspect_ratio); + const float2 geom_overscan = get_geom_overscan_vector(); + geom_aspect_and_overscan = float4(geom_aspect, geom_overscan); + + #if _RUNTIME_GEOMETRY_TILT + // Create a local-to-global rotation matrix for the CRT's coordinate + // frame and its global-to-local inverse. Rotate around the x axis + // first (pitch) and then the y axis (yaw) with yucky Euler angles. + // Positive angles go clockwise around the right-vec and up-vec. + // Runtime shader parameters prevent us from computing these globally, + // but we can still combine the pitch/yaw matrices by hand to cut a + // few instructions. Note that cg matrices fill row1 first, then row2, + // etc. (row-major order). + const float2 geom_tilt_angle = get_geom_tilt_angle_vector(); + const float2 sin_tilt = sin(geom_tilt_angle); + const float2 cos_tilt = cos(geom_tilt_angle); + // Conceptual breakdown: + static const float3x3 rot_x_matrix = float3x3( + 1.0, 0.0, 0.0, + 0.0, cos_tilt.y, -sin_tilt.y, + 0.0, sin_tilt.y, cos_tilt.y); + static const float3x3 rot_y_matrix = float3x3( + cos_tilt.x, 0.0, sin_tilt.x, + 0.0, 1.0, 0.0, + -sin_tilt.x, 0.0, cos_tilt.x); + static const float3x3 local_to_global = + mul(rot_y_matrix, rot_x_matrix); +/* static const float3x3 global_to_local = + transpose(local_to_global); + const float3x3 local_to_global = float3x3( + cos_tilt.x, sin_tilt.y*sin_tilt.x, cos_tilt.y*sin_tilt.x, + 0.0, cos_tilt.y, sin_tilt.y, + sin_tilt.x, sin_tilt.y*cos_tilt.x, cos_tilt.y*cos_tilt.x); +*/ // This is a pure rotation, so transpose = inverse: + const float3x3 global_to_local = transpose(local_to_global); + // Decompose the matrix into 3 float3's for output: + global_to_local_row0 = float3(global_to_local[0][0], global_to_local[0][1], global_to_local[0][2]);//._m00_m01_m02); + global_to_local_row1 = float3(global_to_local[1][0], global_to_local[1][1], global_to_local[1][2]);//._m10_m11_m12); + global_to_local_row2 = float3(global_to_local[2][0], global_to_local[2][1], global_to_local[2][2]);//._m20_m21_m22); + #else + static const float3x3 global_to_local = geom_global_to_local_static; + static const float3x3 local_to_global = geom_local_to_global_static; + #endif + + // Get an optimal eye position based on geom_view_dist, viewport_aspect, + // and CRT radius/rotation: + #if _RUNTIME_GEOMETRY_MODE + const float geom_mode = geom_mode_runtime; + #else + static const float geom_mode = geom_mode_static; + #endif + const float3 eye_pos_global = get_ideal_global_eye_pos(local_to_global, geom_aspect, geom_mode); + eye_pos_local = mul(global_to_local, eye_pos_global); +} + +void geometryPS( + in float4 position : SV_Position, + in float2 texcoord : TEXCOORD0, + in float2 output_size_inv : TEXCOORD1, + in float4 geom_aspect_and_overscan : TEXCOORD2, + in float3 eye_pos_local : TEXCOORD3, + in float3 global_to_local_row0 : TEXCOORD4, + in float3 global_to_local_row1 : TEXCOORD5, + in float3 global_to_local_row2 : TEXCOORD6, + + out float4 color : SV_Target +) { + // Localize some parameters: + const float2 geom_aspect = geom_aspect_and_overscan.xy; + const float2 geom_overscan = geom_aspect_and_overscan.zw; + #if _RUNTIME_GEOMETRY_TILT + const float3x3 global_to_local = float3x3(global_to_local_row0, + global_to_local_row1, global_to_local_row2); + #else + static const float3x3 global_to_local = geom_global_to_local_static; + #endif + #if _RUNTIME_GEOMETRY_MODE + const float geom_mode = geom_mode_runtime; + #else + static const float geom_mode = geom_mode_static; + #endif + + // Get flat and curved texture coords for the current fragment point sample + // and a pixel_to_tangent_video_uv matrix for transforming pixel offsets: + // video_uv = relative position in video frame, mapped to [0.0, 1.0] range + // tex_uv = relative position in padded texture, mapped to [0.0, 1.0] range + const float2 flat_video_uv = texcoord; + float2x2 pixel_to_video_uv; + float2 video_uv_no_geom_overscan; + if(geom_mode > 0.5) + { + video_uv_no_geom_overscan = + get_curved_video_uv_coords_and_tangent_matrix(flat_video_uv, + eye_pos_local, output_size_inv, geom_aspect, + geom_mode, global_to_local, pixel_to_video_uv); + } + else + { + video_uv_no_geom_overscan = flat_video_uv; + pixel_to_video_uv = float2x2( + output_size_inv.x, 0.0, 0.0, output_size_inv.y); + } + // Correct for overscan here (not in curvature code): + const float2 video_uv = + (video_uv_no_geom_overscan - float2(0.5, 0.5))/geom_overscan + float2(0.5, 0.5); + const float2 tex_uv = video_uv; + + // Get a matrix transforming pixel vectors to tex_uv vectors: + const float2x2 pixel_to_tex_uv = + mul_scale(1.0 / geom_overscan, pixel_to_video_uv); + + // Sample! Skip antialiasing if antialias_level < 0.5 or both of these hold: + // 1.) Geometry/curvature isn't used + // 2.) Overscan == float2(1.0, 1.0) + // Skipping AA is sharper, but it's only faster with dynamic branches. + const float2 abs_aa_r_offset = abs(get_aa_subpixel_r_offset()); + // this next check seems to always return true, even when it shouldn't so disabling it for now + const bool need_subpixel_aa = false;//abs_aa_r_offset.x + abs_aa_r_offset.y > 0.0; + float3 raw_color; + + if(antialias_level > 0.5 && (geom_mode > 0.5 || any(bool2((geom_overscan.x != 1.0), (geom_overscan.y != 1.0))))) + { + // Sample the input with antialiasing (due to sharp phosphors, etc.): + raw_color = tex2Daa(samplerBloomHorizontal, tex_uv, pixel_to_tex_uv, float(frame_count), get_intermediate_gamma()); + } + else if(antialias_level > 0.5 && need_subpixel_aa) + { + // Sample at each subpixel location: + raw_color = tex2Daa_subpixel_weights_only( + samplerBloomHorizontal, tex_uv, pixel_to_tex_uv, get_intermediate_gamma()); + } + else + { + raw_color = tex2D_linearize(samplerBloomHorizontal, tex_uv, get_intermediate_gamma()).rgb; + } + + // Dim borders and output the final result: + const float border_dim_factor = get_border_dim_factor(video_uv, geom_aspect); + const float3 final_color = raw_color * border_dim_factor; + + color = encode_output(float4(final_color, 1.0), get_output_gamma()); +} + +#endif // _GEOMETRY_AA_LAST_PASS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/input-blurring.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/input-blurring.fxh new file mode 100644 index 000000000..6b4444004 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/input-blurring.fxh @@ -0,0 +1,74 @@ +#ifndef _INPUT_BLURRING_H +#define _INPUT_BLURRING_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2022 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +// Theoretically this could go in blurring.fxh +// But that file has a bunch of GPL stuff in it. +// Keeping it separate makes it easier to communicate that this portion is +// available under the MIT license. + +#include "../lib/downsampling-functions.fxh" + +#include "content-box.fxh" +#include "shared-objects.fxh" + + +void preblurVertPS( + in const float4 pos : SV_Position, + in const float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target +) { + const float2 texcoord_uncropped = texcoord; + + const float2 max_delta_uv = float2(0.0, rcp(content_size.y)) * preblur_effect_radius; + const float2 delta_uv = max_delta_uv * rcp(max(preblur_sampling_radius.y, 1)); + + color = float4(opaque_linear_downsample( + ReShade::BackBuffer, + texcoord_uncropped, + preblur_sampling_radius.y, + delta_uv + ), 1); +} + +void preblurHorizPS( + in const float4 pos : SV_Position, + in const float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target +) { + const float2 max_delta_uv = float2(rcp(content_size.x), 0.0) * preblur_effect_radius; + const float2 delta_uv = max_delta_uv * rcp(max(preblur_sampling_radius.x, 1)); + + color = float4(opaque_linear_downsample( + samplerPreblurVert, + texcoord, + preblur_sampling_radius.x, + delta_uv + ), 1); +} + +#endif // _INPUT_BLURRING_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/phosphor-mask.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/phosphor-mask.fxh new file mode 100644 index 000000000..9a2e1649b --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/phosphor-mask.fxh @@ -0,0 +1,211 @@ +#ifndef _PHOSPHOR_MASK_H +#define _PHOSPHOR_MASK_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2022 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +#include "../lib/bind-shader-params.fxh" +#include "../lib/phosphor-mask-calculations.fxh" + +#include "shared-objects.fxh" + + +// Split into 64 segments that overlap a little bit +static const float num_segments = 64; +static const float segment_offset = 0.015625; // 1/64 +static const float segment_width = 0.0234375; // 1/128 + +void generatePhosphorMaskVS( + in uint id : SV_VertexID, + + out float4 position : SV_Position, + out float2 texcoord : TEXCOORD0, + out float2 viewport_frequency_factor: TEXCOORD1, + out float2 mask_pq_x : TEXCOORD2, + out float2 mask_pq_y : TEXCOORD3 +) { + const float screen_segment_idx = frame_count % num_segments; + const float left_coord = lerp(segment_offset * screen_segment_idx, 0, overlay_active > 0); + const float right_coord = lerp(left_coord + segment_width, 1, overlay_active > 0); + const float pos_center = 2 * (left_coord + 0.5 * segment_width - 0.5); + const float pos_left = lerp(pos_center - segment_width, -1, overlay_active > 0); + const float pos_right = lerp(pos_center + segment_width, 1, overlay_active > 0); + + #if _DX9_ACTIVE + texcoord.x = (id == 1 || id == 3) ? right_coord : left_coord; + texcoord.y = (id > 1) ? 1 : 0; + + position.x = (id == 1 || id == 3) ? pos_right : pos_left; + position.y = (id > 1) ? -1 : 1; + position.zw = 1; + #else + texcoord.x = (id & 1) ? right_coord : left_coord; + texcoord.y = (id & 2) ? 1 : 0; + + position.x = (id & 1) ? pos_right : pos_left; + position.y = (id & 2) ? -1 : 1; + position.zw = 1; + #endif + + viewport_frequency_factor = calc_phosphor_viewport_frequency_factor(); + + // We don't alter these based on screen rotation because they're independent of screen dimensions. + float edge_norm_tx; + float edge_norm_ty; + [flatten] + switch (mask_type) { + case 0: + edge_norm_tx = grille_edge_norm_t; + break; + case 1: + edge_norm_tx = slot_edge_norm_tx; + edge_norm_ty = slot_edge_norm_ty; + break; + case 2: + edge_norm_tx = shadow_edge_norm_tx; + edge_norm_ty = shadow_edge_norm_ty; + break; + case 3: + edge_norm_tx = smallgrille_edge_norm_t; + break; + case 4: + edge_norm_tx = smallslot_edge_norm_tx; + edge_norm_ty = smallslot_edge_norm_ty; + break; + default: + edge_norm_tx = smallshadow_edge_norm_tx; + edge_norm_ty = smallshadow_edge_norm_ty; + break; + } + + const float2 thickness_scaled = linearize_phosphor_thickness_param(phosphor_thickness); + const float mask_p_x = exp(-calculate_phosphor_p_value(edge_norm_tx, thickness_scaled.x, phosphor_sharpness.x)); + const float mask_p_y = exp(-calculate_phosphor_p_value(edge_norm_ty, thickness_scaled.y, phosphor_sharpness.y)); + mask_pq_x = float2(mask_p_x, phosphor_sharpness.x); + mask_pq_y = float2(mask_p_y, phosphor_sharpness.y); +} + +void generatePhosphorMaskPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + in float2 viewport_frequency_factor: TEXCOORD1, + in float2 mask_pq_x : TEXCOORD2, + in float2 mask_pq_y : TEXCOORD3, + + out float4 color : SV_Target +) { + [branch] + if (geom_rotation_mode == 1 || geom_rotation_mode == 3) { + texcoord = texcoord.yx; + viewport_frequency_factor = viewport_frequency_factor.yx; + } + + float3 phosphor_color; + [branch] + if (mask_type == 0) { + phosphor_color = get_phosphor_intensity_grille( + texcoord, + viewport_frequency_factor, + mask_pq_x + ); + } + else if (mask_type == 1) { + phosphor_color = get_phosphor_intensity_slot( + texcoord, + viewport_frequency_factor, + mask_pq_x, + mask_pq_y + ); + } + else if (mask_type == 2) { + phosphor_color = get_phosphor_intensity_shadow( + texcoord, + viewport_frequency_factor, + float2(mask_pq_x.y, mask_pq_y.y) + ); + } + else if (mask_type == 3) { + phosphor_color = get_phosphor_intensity_grille_small( + texcoord, + viewport_frequency_factor, + mask_pq_x + ); + } + else if (mask_type == 4) { + phosphor_color = get_phosphor_intensity_slot_small( + texcoord, + viewport_frequency_factor, + mask_pq_x, + mask_pq_y + ); + } + else { + phosphor_color = get_phosphor_intensity_shadow_small( + texcoord, + viewport_frequency_factor, + mask_pq_x, + mask_pq_y + ); + } + + color = float4(phosphor_color, 1.0); +} + + +void applyComputedPhosphorMaskPS( + in float4 pos : SV_Position, + in float2 texcoord : TEXCOORD0, + + out float4 color : SV_Target +) { + bool use_deinterlacing_tex = enable_interlacing && ( + scanline_deinterlacing_mode == 2 || scanline_deinterlacing_mode == 3 + ); + + float3 scanline_color_dim; + [branch] + if (use_deinterlacing_tex) scanline_color_dim = tex2D(samplerDeinterlace, texcoord).rgb; + else scanline_color_dim = tex2D(samplerBeamConvergence, texcoord).rgb; + + const float3 phosphor_color = tex2D(samplerPhosphorMask, texcoord).rgb; + + // Sample the halation texture (auto-dim to match the scanlines), and + // account for both horizontal and vertical convergence offsets, given + // in units of texels horizontally and same-field scanlines vertically: + const float3 halation_color = tex2D_linearize(samplerBlurHorizontal, texcoord, get_intermediate_gamma()).rgb; + + // Apply halation: Halation models electrons flying around under the glass + // and hitting the wrong phosphors (of any color). It desaturates, so + // average the halation electrons to a scalar. Reduce the local scanline + // intensity accordingly to conserve energy. + const float halation_intensity_dim_scalar = dot(halation_color, float3(1, 1, 1)) / 3.0; + const float3 halation_intensity_dim = halation_intensity_dim_scalar; + const float3 electron_intensity_dim = lerp(scanline_color_dim, halation_intensity_dim, halation_weight); + + // Apply the phosphor mask: + const float3 phosphor_emission_dim = electron_intensity_dim * phosphor_color; + + color = float4(phosphor_emission_dim, 1.0); +} + +#endif // _PHOSPHOR_MASK_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/shaders/shared-objects.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/shared-objects.fxh new file mode 100644 index 000000000..0ef24ff54 --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/shaders/shared-objects.fxh @@ -0,0 +1,370 @@ +#ifndef _SHARED_OBJECTS_H +#define _SHARED_OBJECTS_H + +///////////////////////////// GPL LICENSE NOTICE ///////////////////////////// + +// crt-royale-reshade: A port of TroggleMonkey's crt-royale from libretro to ReShade. +// Copyright (C) 2020 Alex Gunter +// +// This program is free software; you can redistribute it and/or modify it +// under the terms of the GNU General Public License as published by the Free +// Software Foundation; either version 2 of the License, or any later version. +// +// This program is distributed in the hope that it will be useful, but WITHOUT +// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +// FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for +// more details. +// +// You should have received a copy of the GNU General Public License along with +// this program; if not, write to the Free Software Foundation, Inc., 59 Temple +// Place, Suite 330, Boston, MA 02111-1307 USA + + +#include "../lib/helper-functions-and-macros.fxh" +#include "../lib/derived-settings-and-constants.fxh" +#include "../lib/bind-shader-params.fxh" + + +// Yes, the WIDTH/HEIGHT/SIZE defines are kinda weird. +// Yes, we have to have them or something similar. This is for D3D11 which +// returns (0, 0) when you call tex2Dsize() on the pass's render target. + + +// Pass 0 Buffer (cropPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is in interlacingPass +// electronBeamPass -> beamConvergencePass +// deinterlacePass -> phosphorMaskPass +// brightpassPass -> bloomHorizontalPass +// #define TEX_CROP_WIDTH content_size.x +// #define TEX_CROP_HEIGHT content_size.y +// #define TEX_CROP_SIZE int2(TEX_CROP_WIDTH, TEX_CROP_HEIGHT) +// texture2D texCrop { +// Width = TEX_CROP_WIDTH; +// Height = TEX_CROP_HEIGHT; + +// Format = RGBA16; +// }; +// sampler2D samplerCrop { Texture = texCrop; }; + + +// Pass 1 Buffer (interlacingPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is in electronBeamPass +// beamConvergencPass -> freezeFramePass +// phosphorMaskPass -> bloomHorizontalPass +// #define TEX_INTERLACED_WIDTH content_size.x +// #define TEX_INTERLACED_HEIGHT content_size.y +// #define TEX_INTERLACED_SIZE int2(TEX_INTERLACED_WIDTH, TEX_INTERLACED_HEIGHT) +// texture2D texInterlaced { +// Width = TEX_INTERLACED_WIDTH; +// Height = TEX_INTERLACED_HEIGHT; + +// Format = RGBA16; +// }; +// sampler2D samplerInterlaced { Texture = texInterlaced; }; + +// Pass 2 Buffer (electronBeamPass) +// Last usage is in beamConvergencePass + + +#define TEX_PREBLUR_VERT_WIDTH content_size.x +#define TEX_PREBLUR_VERT_HEIGHT content_size.y +static const uint2 TEX_PREBLUR_SIZE = uint2(TEX_PREBLUR_VERT_WIDTH, TEX_PREBLUR_VERT_HEIGHT); +texture2D texPreblurVert < pooled = true; > { + Width = TEX_PREBLUR_VERT_WIDTH; + Height = TEX_PREBLUR_VERT_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerPreblurVert { Texture = texPreblurVert; }; + +#define TEX_PREBLUR_HORIZ_WIDTH content_size.x +#define TEX_PREBLUR_HORIZ_HEIGHT content_size.y +static const uint2 TEX_PREBLUR_SIZE = uint2(TEX_PREBLUR_HORIZ_WIDTH, TEX_PREBLUR_HORIZ_HEIGHT); +texture2D texPreblurHoriz < pooled = true; > { + Width = TEX_PREBLUR_HORIZ_WIDTH; + Height = TEX_PREBLUR_HORIZ_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerPreblurHoriz { Texture = texPreblurHoriz; }; + + +#define TEX_BEAMDIST_WIDTH num_beamdist_color_samples +#define TEX_BEAMDIST_HEIGHT num_beamdist_dist_samples +#define TEX_BEAMDIST_SIZE int2(TEX_BEAMDIST_WIDTH, TEX_BEAMDIST_HEIGHT) +texture2D texBeamDist < pooled = false; > { + Width = TEX_BEAMDIST_WIDTH; + Height = TEX_BEAMDIST_HEIGHT; + + + Format = RGB10A2; +}; +sampler2D samplerBeamDist { + Texture = texBeamDist; + AddressV = WRAP; +}; + + +// Pass 2 Buffer (electronBeamPass) +// Last usage is in beamConvergencePass +#define TEX_ELECTRONBEAMS_WIDTH content_size.x +#define TEX_ELECTRONBEAMS_HEIGHT content_size.y +#define TEX_ELECTRONBEAMS_SIZE int2(TEX_ELECTRONBEAMS_WIDTH, TEX_ELECTRONBEAMS_HEIGHT) +texture2D texElectronBeams < pooled = true; > { + Width = TEX_ELECTRONBEAMS_WIDTH; + Height = TEX_ELECTRONBEAMS_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerElectronBeams { + Texture = texElectronBeams; + + AddressU = BORDER; + AddressV = BORDER; +}; +// #define texElectronBeams texCrop +// #define samplerElectronBeams samplerCrop + + +// Pass 3 Buffer (beamConvergencPass) +// Last usage is freezeFramePass +#define TEX_BEAMCONVERGENCE_WIDTH content_size.x +#define TEX_BEAMCONVERGENCE_HEIGHT content_size.y +#define TEX_BEAMCONVERGENCE_SIZE int2(TEX_BEAMCONVERGENCE_WIDTH, TEX_BEAMCONVERGENCE_HEIGHT) +texture2D texBeamConvergence < pooled = true; > { + Width = TEX_BEAMCONVERGENCE_WIDTH; + Height = TEX_BEAMCONVERGENCE_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerBeamConvergence { Texture = texBeamConvergence; }; +// #define texBeamConvergence texInterlaced +// #define samplerBeamConvergence samplerInterlaced + + +/* +// Pass 4 Buffer (bloomApproxPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is in brightpassPass +#define TEX_BLOOMAPPROX_WIDTH 320 +#define TEX_BLOOMAPPROX_HEIGHT 240 +#define TEX_BLOOMAPPROX_SIZE int2(TEX_BLOOMAPPROX_WIDTH, TEX_BLOOMAPPROX_HEIGHT) +texture2D texBloomApprox { + Width = TEX_BLOOMAPPROX_WIDTH; + Height = TEX_BLOOMAPPROX_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerBloomApprox { Texture = texBloomApprox; }; +*/ + +// Pass 4a Buffer (bloomApproxVerticalPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is in brightpassPass +#define TEX_BLOOMAPPROXVERT_WIDTH content_size.x +// #define TEX_BLOOMAPPROXVERT_HEIGHT 240 +#define TEX_BLOOMAPPROXVERT_HEIGHT int(content_size.y / bloomapprox_downsizing_factor) +#define TEX_BLOOMAPPROXVERT_SIZE int2(TEX_BLOOMAPPROXVERT_WIDTH, TEX_BLOOMAPPROXVERT_HEIGHT) +texture2D texBloomApproxVert < pooled = true; > { + Width = TEX_BLOOMAPPROXVERT_WIDTH; + Height = TEX_BLOOMAPPROXVERT_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerBloomApproxVert { Texture = texBloomApproxVert; }; + +// Pass 4b Buffer (bloomApproxHorizontalPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is in brightpassPass +// #define TEX_BLOOMAPPROXHORIZ_WIDTH 320 +// #define TEX_BLOOMAPPROXHORIZ_HEIGHT 240 +#define TEX_BLOOMAPPROXHORIZ_WIDTH int(content_size.x / bloomapprox_downsizing_factor) +#define TEX_BLOOMAPPROXHORIZ_HEIGHT TEX_BLOOMAPPROXVERT_HEIGHT +#define TEX_BLOOMAPPROXHORIZ_SIZE int2(TEX_BLOOMAPPROXHORIZ_WIDTH, TEX_BLOOMAPPROXHORIZ_HEIGHT) +texture2D texBloomApproxHoriz < pooled = true; > { + Width = TEX_BLOOMAPPROXHORIZ_WIDTH; + Height = TEX_BLOOMAPPROXHORIZ_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerBloomApproxHoriz { Texture = texBloomApproxHoriz; }; + +// Pass 5 Buffer (blurVerticalPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is blurHorizontalPass +#define TEX_BLURVERTICAL_WIDTH TEX_BLOOMAPPROXHORIZ_WIDTH +#define TEX_BLURVERTICAL_HEIGHT TEX_BLOOMAPPROXHORIZ_HEIGHT +#define TEX_BLURVERTICAL_SIZE int2(TEX_BLURVERTICAL_WIDTH, TEX_BLURVERTICAL_HEIGHT) +texture2D texBlurVertical < pooled = true; > { + Width = TEX_BLURVERTICAL_WIDTH; + Height = TEX_BLURVERTICAL_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerBlurVertical { Texture = texBlurVertical; }; + + +// Pass 6 Buffer (blurHorizontalPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is bloomHorizontalPass +#define TEX_BLURHORIZONTAL_WIDTH TEX_BLOOMAPPROXHORIZ_WIDTH +#define TEX_BLURHORIZONTAL_HEIGHT TEX_BLOOMAPPROXHORIZ_HEIGHT +#define TEX_BLURHORIZONTAL_SIZE int2(TEX_BLURHORIZONTAL_WIDTH, TEX_BLURHORIZONTAL_HEIGHT) +texture2D texBlurHorizontal < pooled = true; > { + Width = TEX_BLURHORIZONTAL_WIDTH; + Height = TEX_BLURHORIZONTAL_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerBlurHorizontal { Texture = texBlurHorizontal; }; + + +// Pass 7 (deinterlacePass) +// Last usage is phosphorMaskPass +#define TEX_DEINTERLACE_WIDTH content_size.x +#define TEX_DEINTERLACE_HEIGHT content_size.y +#define TEX_DEINTERLACE_SIZE int2(TEX_DEINTERLACE_WIDTH, TEX_DEINTERLACE_HEIGHT) +#if _DX9_ACTIVE == 0 + texture2D texDeinterlace < pooled = true; > { + Width = TEX_DEINTERLACE_WIDTH; + Height = TEX_DEINTERLACE_HEIGHT; + + Format = RGBA16; + }; + sampler2D samplerDeinterlace { Texture = texDeinterlace; }; +#else + #define texDeinterlace texElectronBeams + #define samplerDeinterlace samplerElectronBeams +#endif + +// Pass 8 (freezeFramePass) +// Do not condition this on __RENDERER__. It will not work if another +// pass corrupts it. +#define TEX_FREEZEFRAME_WIDTH content_size.x +#define TEX_FREEZEFRAME_HEIGHT content_size.y +#define TEX_FREEZEFRAME_SIZE int2(TEX_FREEZEFRAME_WIDTH, TEX_FREEZEFRAME_HEIGHT +texture2D texFreezeFrame < pooled = false; > { + Width = TEX_FREEZEFRAME_WIDTH; + Height = TEX_FREEZEFRAME_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerFreezeFrame { Texture = texFreezeFrame; }; + + +// Pass 10 Mask Texture (phosphorMaskResizeHorizontalPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +#define TEX_PHOSPHORMASK_WIDTH content_size.x +#define TEX_PHOSPHORMASK_HEIGHT content_size.y +#define TEX_PHOSPHORMASKL_SIZE int2(TEX_PHOSPHORMASK_WIDTH, TEX_PHOSPHORMASK_HEIGHT) +texture2D texPhosphorMask < pooled = false; > { + Width = TEX_PHOSPHORMASK_WIDTH; + Height = TEX_PHOSPHORMASK_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerPhosphorMask { Texture = texPhosphorMask; }; + + +// Pass 11 Buffer (phosphorMaskPass) +// Last usage is bloomHorizontalPass +#define TEX_MASKEDSCANLINES_WIDTH content_size.x +#define TEX_MASKEDSCANLINES_HEIGHT content_size.y +#define TEX_MASKEDSCANLINES_SIZE int2(TEX_MASKEDSCANLINES_WIDTH, TEX_MASKEDSCANLINES_HEIGHT) + +#if _DX9_ACTIVE == 0 + texture2D texMaskedScanlines < pooled = true; > { + Width = TEX_MASKEDSCANLINES_WIDTH; + Height = TEX_MASKEDSCANLINES_HEIGHT; + + Format = RGBA16; + }; + sampler2D samplerMaskedScanlines { Texture = texMaskedScanlines; }; +#else + #define texMaskedScanlines texBeamConvergence + #define samplerMaskedScanlines samplerBeamConvergence +#endif + + +// Pass 12 Buffer (brightpassPass) +// Last usage is bloomHorizontalPass +#define TEX_BRIGHTPASS_WIDTH content_size.x +#define TEX_BRIGHTPASS_HEIGHT content_size.y +#define TEX_BRIGHTPASS_SIZE int2(TEX_BRIGHTPASS_WIDTH, TEX_BRIGHTPASS_HEIGHT) + +#if _DX9_ACTIVE == 0 + texture2D texBrightpass < pooled = true; > { + Width = TEX_BRIGHTPASS_WIDTH; + Height = TEX_BRIGHTPASS_HEIGHT; + + Format = RGBA16; + }; + sampler2D samplerBrightpass { Texture = texBrightpass; }; +#else + #define texBrightpass texElectronBeams + #define samplerBrightpass samplerElectronBeams +#endif + + +// Pass 13 Buffer (bloomVerticalPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is bloomHorizontalPass +#define TEX_BLOOMVERTICAL_WIDTH content_size.x +#define TEX_BLOOMVERTICAL_HEIGHT content_size.y +#define TEX_BLOOMVERTICAL_SIZE int2(TEX_BLOOMVERTICAL_WIDTH, TEX_BLOOMVERTICAL_HEIGHT) +texture2D texBloomVertical < pooled = true; > { + Width = TEX_BLOOMVERTICAL_WIDTH; + Height = TEX_BLOOMVERTICAL_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerBloomVertical { Texture = texBloomVertical; }; + + +// Pass 14 Buffer (bloomHorizontalPass) +// Cannot be conditioned on __RENDERER__ b/c there are no +// available buffers of the same size +// Last usage is geometryPass +#define TEX_BLOOMHORIZONTAL_WIDTH content_size.x +#define TEX_BLOOMHORIZONTAL_HEIGHT content_size.y +#define TEX_BLOOMHORIZONTAL_SIZE int2(TEX_BLOOMHORIZONTAL_WIDTH, TEX_BLOOMHORIZONTAL_HEIGHT) +texture2D texBloomHorizontal < pooled = true; > { + Width = TEX_BLOOMHORIZONTAL_WIDTH; + Height = TEX_BLOOMHORIZONTAL_HEIGHT; + + Format = RGBA16; +}; +sampler2D samplerBloomHorizontal { Texture = texBloomHorizontal; }; + + +// Pass 15 Buffer (geometryPass) +// Last usage is uncropPass +#define TEX_GEOMETRY_WIDTH content_size.x +#define TEX_GEOMETRY_HEIGHT content_size.y +#define TEX_GEOMETRY_SIZE int2(TEX_GEOMETRY_WIDTH, TEX_GEOMETRY_HEIGHT) + +#if _DX9_ACTIVE == 0 + texture2D texGeometry < pooled = true; > { + Width = TEX_GEOMETRY_WIDTH; + Height = TEX_GEOMETRY_HEIGHT; + + Format = RGBA16; + }; + sampler2D samplerGeometry { Texture = texGeometry; }; +#else + #define texGeometry texElectronBeams + #define samplerGeometry samplerElectronBeams +#endif + +#endif // _SHARED_OBJECTS_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/Shaders/crt-royale/version-number.fxh b/data/resources/shaders/reshade/Shaders/crt-royale/version-number.fxh new file mode 100644 index 000000000..79d17d2fd --- /dev/null +++ b/data/resources/shaders/reshade/Shaders/crt-royale/version-number.fxh @@ -0,0 +1,44 @@ +#ifndef _VERSION_NUMBER_H +#define _VERSION_NUMBER_H + +///////////////////////////////// MIT LICENSE //////////////////////////////// + +// Copyright (C) 2022 Alex Gunter +// +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to +// deal in the Software without restriction, including without limitation the +// rights to use, copy, modify, merge, publish, distribute, sublicense, and/or +// sell copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +// IN THE SOFTWARE. + + +#define MAJOR_VERSION 2 +#define MINOR_VERSION 1 +#define PATCH_VERSION 0 + +// Yes, both sibling preprocessor functions are necessary. +// Don't "simplify" this, or the substitution won't work. +#define BUILD_DOT_VERSION_(mav, miv, pav) #mav "." #miv "." #pav +#define BUILD_DOT_VERSION(mav, miv, pav) BUILD_DOT_VERSION_(mav, miv, pav) +#define DOT_VERSION_STR BUILD_DOT_VERSION(MAJOR_VERSION, MINOR_VERSION, PATCH_VERSION) + +// Again, yes, both sibling preprocessor functions are necessary. +// Don't "simplify" this, or the substitution won't work. +#define BUILD_UNDERSCORE_VERSION_(prefix, mav, miv, pav) prefix ## _ ## mav ## _ ## miv ## _ ## pav +#define BUILD_UNDERSCORE_VERSION(p, mav, miv, pav) BUILD_UNDERSCORE_VERSION_(p, mav, miv, pav) +#define APPEND_VERSION_SUFFIX(prefix) BUILD_UNDERSCORE_VERSION(prefix, MAJOR_VERSION, MINOR_VERSION, PATCH_VERSION) + + +#endif // _VERSION_NUMBER_H \ No newline at end of file diff --git a/data/resources/shaders/reshade/source.txt b/data/resources/shaders/reshade/source.txt index f97a2f2f6..eb18a52a9 100644 --- a/data/resources/shaders/reshade/source.txt +++ b/data/resources/shaders/reshade/source.txt @@ -1,2 +1,3 @@ https://github.com/crosire/reshade-shaders https://github.com/Matsilagi/RSRetroArch/ +https://github.com/akgunter/crt-royale-reshade diff --git a/data/resources/thirdparty.html b/data/resources/thirdparty.html index 46755cf9d..f1021ea18 100644 --- a/data/resources/thirdparty.html +++ b/data/resources/thirdparty.html @@ -3,7 +3,7 @@ 
 DuckStation PS1 Emulator
-Copyright (C) 2019-2023 Connor McLaughlin <stenzek@gmail.com>
+Copyright (C) 2019-2024 Connor McLaughlin <stenzek@gmail.com>
 
 This program is free software: you can redistribute it and/or modify
 it under the terms of the GNU General Public License as published by
@@ -1758,7 +1758,11 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 SOFTWARE.
-Some shaders provided with the application are sourced from https://github.com/Matsilagi/RSRetroArch/.
+Some shaders provided with the application are sourced from: + License details are included in the relevant shader source files, under resources\shaders\reshade.