ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

GS: Remove reduced-depth-range hack

This commit is contained in:
Stenzek 2024-03-29 16:54:50 +10:00 committed by Connor McLaughlin
parent 7ddf6386f1
commit 11ee0a8613
11 changed files with 20 additions and 86 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
bin/resources/shaders/opengl/convert.glsl
View File

@ -80,11 +80,7 @@ void ps_convert_rgba8_16bits()
void ps_convert_float32_32bits()
{
// Convert a GL_FLOAT32 depth texture into a 32 bits UINT texture
#if HAS_CLIP_CONTROL
SV_Target1 = uint(exp2(32.0f) * sample_c().r);
#else
SV_Target1 = uint(exp2(24.0f) * sample_c().r);
#endif
}
#endif
@ -92,11 +88,7 @@ void ps_convert_float32_32bits()
void ps_convert_float32_rgba8()
{
// Convert a GL_FLOAT32 depth texture into a RGBA color texture
#if HAS_CLIP_CONTROL
uint d = uint(sample_c().r * exp2(32.0f));
#else
uint d = uint(sample_c().r * exp2(24.0f));
#endif
SV_Target0 = vec4(uvec4((d & 0xFFu), ((d >> 8) & 0xFFu), ((d >> 16) & 0xFFu), (d >> 24))) / vec4(255.0);
}
#endif
@ -105,11 +97,7 @@ void ps_convert_float32_rgba8()
void ps_convert_float16_rgb5a1()
{
// Convert a GL_FLOAT32 (only 16 lsb) depth into a RGB5A1 color texture
#if HAS_CLIP_CONTROL
uint d = uint(sample_c().r * exp2(32.0f));
#else
uint d = uint(sample_c().r * exp2(24.0f));
#endif
SV_Target0 = vec4(uvec4(d << 3, d >> 2, d >> 7, d >> 8) & uvec4(0xf8, 0xf8, 0xf8, 0x80)) / 255.0f;
}
#endif
@ -117,41 +105,25 @@ void ps_convert_float16_rgb5a1()
float rgba8_to_depth32(vec4 unorm)
{
uvec4 c = uvec4(unorm * vec4(255.5f));
#if HAS_CLIP_CONTROL
return float(c.r | (c.g << 8) | (c.b << 16) | (c.a << 24)) * exp2(-32.0f);
#else
return float(c.r | (c.g << 8) | (c.b << 16) | (c.a << 24)) * exp2(-24.0f);
#endif
}
float rgba8_to_depth24(vec4 unorm)
{
uvec3 c = uvec3(unorm.rgb * vec3(255.5f));
#if HAS_CLIP_CONTROL
return float(c.r | (c.g << 8) | (c.b << 16)) * exp2(-32.0f);
#else
return float(c.r | (c.g << 8) | (c.b << 16)) * exp2(-24.0f);
#endif
}
float rgba8_to_depth16(vec4 unorm)
{
uvec2 c = uvec2(unorm.rg * vec2(255.5f));
#if HAS_CLIP_CONTROL
return float(c.r | (c.g << 8)) * exp2(-32.0f);
#else
return float(c.r | (c.g << 8)) * exp2(-24.0f);
#endif
}
float rgb5a1_to_depth16(vec4 unorm)
{
uvec4 c = uvec4(unorm * vec4(255.5f));
#if HAS_CLIP_CONTROL
return float(((c.r & 0xF8u) >> 3) | ((c.g & 0xF8u) << 2) | ((c.b & 0xF8u) << 7) | ((c.a & 0x80u) << 8)) * exp2(-32.0f);
#else
return float(((c.r & 0xF8u) >> 3) | ((c.g & 0xF8u) << 2) | ((c.b & 0xF8u) << 7) | ((c.a & 0x80u) << 8)) * exp2(-24.0f);
#endif
}
#ifdef ps_convert_rgba8_float32

16
bin/resources/shaders/opengl/tfx_fs.glsl
View File

@ -317,11 +317,7 @@ mat4 sample_4p(uvec4 u)
int fetch_raw_depth()
{
#if HAS_CLIP_CONTROL
float multiplier = exp2(32.0f);
#else
float multiplier = exp2(24.0f);
#endif
#if PS_TEX_IS_FB == 1
return int(fetch_rt().r * multiplier);
@ -423,21 +419,13 @@ vec4 sample_depth(vec2 st)
#elif PS_DEPTH_FMT == 1
// Based on ps_convert_float32_rgba8 of convert
// Convert a GL_FLOAT32 depth texture into a RGBA color texture
#if HAS_CLIP_CONTROL
uint d = uint(fetch_c(uv).r * exp2(32.0f));
#else
uint d = uint(fetch_c(uv).r * exp2(24.0f));
#endif
uint d = uint(fetch_c(uv).r * exp2(32.0f));
t = vec4(uvec4((d & 0xFFu), ((d >> 8) & 0xFFu), ((d >> 16) & 0xFFu), (d >> 24)));
#elif PS_DEPTH_FMT == 2
// Based on ps_convert_float16_rgb5a1 of convert
// Convert a GL_FLOAT32 (only 16 lsb) depth into a RGB5A1 color texture
#if HAS_CLIP_CONTROL
uint d = uint(fetch_c(uv).r * exp2(32.0f));
#else
uint d = uint(fetch_c(uv).r * exp2(24.0f));
#endif
uint d = uint(fetch_c(uv).r * exp2(32.0f));
t = vec4(uvec4((d & 0x1Fu), ((d >> 5) & 0x1Fu), ((d >> 10) & 0x1Fu), (d >> 15) & 0x01u)) * vec4(8.0f, 8.0f, 8.0f, 128.0f);
#elif PS_DEPTH_FMT == 3

27
bin/resources/shaders/opengl/tfx_vgs.glsl
View File

@ -70,22 +70,10 @@ void vs_main()
// example: ceil(afterseveralvertextransformations(y = 133)) => 134 => line 133 stays empty
// input granularity is 1/16 pixel, anything smaller than that won't step drawing up/left by one pixel
// example: 133.0625 (133 + 1/16) should start from line 134, ceil(133.0625 - 0.05) still above 133
vec4 p;
p.xy = vec2(i_p) - vec2(0.05f, 0.05f);
p.xy = p.xy * VertexScale - VertexOffset;
p.w = 1.0f;
#if HAS_CLIP_CONTROL
p.z = float(z) * exp_min32;
#else
// GLES doesn't support ARB_clip_control, so remap it to -1..1. We also reduce the range from 32 bits
// to 24 bits, which means some games with very large depth ranges will not render correctly. But,
// for most, it's okay, and really, the best we can do.
p.z = min(float(z) * exp2(-23.0f), 2.0f) - 1.0f;
#endif
gl_Position = p;
gl_Position.xy = vec2(i_p) - vec2(0.05f, 0.05f);
gl_Position.xy = gl_Position.xy * VertexScale - VertexOffset;
gl_Position.z = float(z) * exp_min32;
gl_Position.w = 1.0f;
texture_coord();
@ -144,13 +132,8 @@ ProcessedVertex load_vertex(uint index)
uint z = min(i_z, MaxDepth);
vtx.p.xy = vec2(i_p) - vec2(0.05f, 0.05f);
vtx.p.xy = vtx.p.xy * VertexScale - VertexOffset;
vtx.p.w = 1.0f;
#if HAS_CLIP_CONTROL
vtx.p.z = float(z) * exp_min32;
#else
vtx.p.z = min(float(z) * exp2(-23.0f), 2.0f) - 1.0f;
#endif
vtx.p.w = 1.0f;
vec2 uv = vec2(i_uv) - TextureOffset;
vec2 st = i_st - TextureOffset;

1
pcsx2/GS/Renderers/Common/GSDevice.h
View File

@ -728,7 +728,6 @@ public:
bool dxt_textures : 1; ///< Supports DXTn texture compression, i.e. S3TC and BC1-3.
bool bptc_textures : 1; ///< Supports BC6/7 texture compression.
bool framebuffer_fetch : 1; ///< Can sample from the framebuffer without texture barriers.
bool clip_control : 1; ///< Can use 0..1 depth range instead of -1..1.
bool stencil_buffer : 1; ///< Supports stencil buffer, and can use for DATE.
bool cas_sharpening : 1; ///< Supports sufficient functionality for contrast adaptive sharpening.
bool test_and_sample_depth: 1; ///< Supports concurrently binding the depth-stencil buffer for sampling and depth testing.

1
pcsx2/GS/Renderers/DX11/GSDevice11.cpp
View File

@ -52,7 +52,6 @@ GSDevice11::GSDevice11()
m_features.bptc_textures = false;
m_features.framebuffer_fetch = false;
m_features.stencil_buffer = true;
m_features.clip_control = true;
m_features.cas_sharpening = true;
m_features.test_and_sample_depth = false;
}

1
pcsx2/GS/Renderers/DX12/GSDevice12.cpp
View File

@ -1195,7 +1195,6 @@ bool GSDevice12::CheckFeatures()
m_features.point_expand = false;
m_features.line_expand = false;
m_features.framebuffer_fetch = false;
m_features.clip_control = true;
m_features.stencil_buffer = true;
m_features.cas_sharpening = true;
m_features.test_and_sample_depth = false;

6
pcsx2/GS/Renderers/HW/GSRendererHW.cpp
View File

@ -3393,7 +3393,7 @@ void GSRendererHW::EmulateZbuffer(const GSTextureCache::Target* ds)
}
else if (!m_cached_ctx.ZBUF.ZMSK)
{
m_conf.cb_ps.TA_MaxDepth_Af.z = static_cast<float>(max_z) * (g_gs_device->Features().clip_control ? 0x1p-32f : 0x1p-24f);
m_conf.cb_ps.TA_MaxDepth_Af.z = static_cast<float>(max_z) * 0x1p-32f;
m_conf.ps.zclamp = 1;
}
}
@ -3950,7 +3950,7 @@ void GSRendererHW::EmulateBlending(int rt_alpha_min, int rt_alpha_max, bool& DAT
const bool blend_non_recursive = !!(blend_flag & BLEND_NO_REC);
// BLEND MIX selection, use a mix of hw/sw blending
const bool blend_mix1 = !!(blend_flag & BLEND_MIX1) && !(m_conf.ps.blend_b == m_conf.ps.blend_d && (alpha_c0_high_min_one || alpha_c2_high_one));
const bool blend_mix1 = !!(blend_flag & BLEND_MIX1);
const bool blend_mix2 = !!(blend_flag & BLEND_MIX2);
const bool blend_mix3 = !!(blend_flag & BLEND_MIX3);
bool blend_mix = (blend_mix1 || blend_mix2 || blend_mix3) && COLCLAMP.CLAMP;
@ -6395,7 +6395,7 @@ bool GSRendererHW::TryTargetClear(GSTextureCache::Target* rt, GSTextureCache::Ta
{
const u32 max_z = 0xFFFFFFFF >> (GSLocalMemory::m_psm[m_cached_ctx.ZBUF.PSM].fmt * 8);
const u32 z = std::min(max_z, m_vertex.buff[1].XYZ.Z);
const float d = static_cast<float>(z) * (g_gs_device->Features().clip_control ? 0x1p-32f : 0x1p-24f);
const float d = static_cast<float>(z) * 0x1p-32f;
GL_INS("TryTargetClear(): DS at %x <= %f", ds->m_TEX0.TBP0, d);
g_gs_device->ClearDepth(ds->m_texture, d);
ds->m_alpha_max = z >> 24;

4
pcsx2/GS/Renderers/HW/GSTextureCache.cpp
View File

@ -2795,7 +2795,7 @@ void GSTextureCache::ScaleTargetForDisplay(Target* t, const GIFRegTEX0& dispfb,
float GSTextureCache::ConvertColorToDepth(u32 c, ShaderConvert convert)
{
const float mult = std::exp2(g_gs_device->Features().clip_control ? -32.0f : -24.0f);
const float mult = std::exp2(-32.0f);
switch (convert)
{
case ShaderConvert::RGB5A1_TO_FLOAT16:
@ -2817,7 +2817,7 @@ float GSTextureCache::ConvertColorToDepth(u32 c, ShaderConvert convert)
u32 GSTextureCache::ConvertDepthToColor(float d, ShaderConvert convert)
{
const float mult = std::exp2(g_gs_device->Features().clip_control ? 32.0f : 24.0f);
const float mult = std::exp2(32.0f);
switch (convert)
{
case ShaderConvert::FLOAT16_TO_RGB5A1:

1
pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm
View File

@ -897,7 +897,6 @@ bool GSDeviceMTL::Create()
m_features.dxt_textures = true;
m_features.bptc_textures = true;
m_features.framebuffer_fetch = m_dev.features.framebuffer_fetch && !GSConfig.DisableFramebufferFetch;
m_features.clip_control = true;
m_features.stencil_buffer = true;
m_features.cas_sharpening = true;
m_features.test_and_sample_depth = true;

18
pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp
View File

@ -517,8 +517,7 @@ bool GSDeviceOGL::Create()
// This extension allow FS depth to range from -1 to 1. So
// gl_position.z could range from [0, 1]
// Change depth convention
if (m_features.clip_control)
glClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE);
glClipControl(GL_LOWER_LEFT, GL_ZERO_TO_ONE);
// ****************************************************************
// HW renderer shader
@ -671,6 +670,12 @@ bool GSDeviceOGL::CheckFeatures(bool& buggy_pbo)
"GS", "GL_ARB_copy_image is not supported, this is required for the OpenGL renderer.");
return false;
}
if (!GLAD_GL_VERSION_4_5 && !GLAD_GL_ARB_clip_control)
{
Host::ReportFormattedErrorAsync(
"GS", "GL_ARB_clip_control is not supported, this is required for the OpenGL renderer.");
return false;
}
if (!GLAD_GL_ARB_viewport_array)
{
@ -733,10 +738,6 @@ bool GSDeviceOGL::CheckFeatures(bool& buggy_pbo)
m_features.bptc_textures =
GLAD_GL_VERSION_4_2 || GLAD_GL_ARB_texture_compression_bptc || GLAD_GL_EXT_texture_compression_bptc;
m_features.prefer_new_textures = false;
m_features.clip_control = GLAD_GL_ARB_clip_control;
if (!m_features.clip_control)
Host::AddOSDMessage(
"GL_ARB_clip_control is not supported, this will cause rendering issues.", Host::OSD_ERROR_DURATION);
m_features.stencil_buffer = true;
m_features.test_and_sample_depth = m_features.texture_barrier;
@ -1281,11 +1282,6 @@ std::string GSDeviceOGL::GenGlslHeader(const std::string_view& entry, GLenum typ
else
header += "#define HAS_FRAMEBUFFER_FETCH 0\n";
if (m_features.clip_control)
header += "#define HAS_CLIP_CONTROL 1\n";
else
header += "#define HAS_CLIP_CONTROL 0\n";
// Allow to puts several shader in 1 files
switch (type)
{

3
pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp
View File

@ -2669,7 +2669,6 @@ bool GSDeviceVK::CheckFeatures()
m_features.prefer_new_textures = true;
m_features.provoking_vertex_last = m_optional_extensions.vk_ext_provoking_vertex;
m_features.clip_control = true;
m_features.vs_expand = !GSConfig.DisableVertexShaderExpand;
if (!m_features.texture_barrier)
@ -2699,7 +2698,7 @@ bool GSDeviceVK::CheckFeatures()
m_features.line_expand =
(m_device_features.wideLines && limits.lineWidthRange[0] <= f_upscale && limits.lineWidthRange[1] >= f_upscale);
DevCon.WriteLn("Optional features:%s%s%s%s%s%s", m_features.primitive_id ? " primitive_id" : "",
DevCon.WriteLn("Optional features:%s%s%s%s%s", m_features.primitive_id ? " primitive_id" : "",
m_features.texture_barrier ? " texture_barrier" : "", m_features.framebuffer_fetch ? " framebuffer_fetch" : "",
m_features.provoking_vertex_last ? " provoking_vertex_last" : "", m_features.vs_expand ? " vs_expand" : "");
DevCon.WriteLn("Using %s for point expansion and %s for line expansion.",