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

get the OpenGL renderer going. 

sorta.

(also make the blackmagic_II branch obsolete in the process)
This commit is contained in:
Arisotura 2020-05-25 18:25:50 +02:00
parent 10f9eda58a
commit 36f4cdbbbf
13 changed files with 1241 additions and 42 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/CMakeLists.txt
View File

@ -16,6 +16,8 @@ add_library(core STATIC
FIFO.h
GBACart.cpp
GPU.cpp
GPU_OpenGL.cpp
GPU_OpenGL_shaders.h
GPU2D.cpp
GPU3D.cpp
GPU3D_OpenGL.cpp

2
src/GPU.cpp
View File

@ -985,6 +985,8 @@ void StartScanline(u32 line)
GPU2D_A->VBlank();
GPU2D_B->VBlank();
GPU3D::VBlank();
if (Accelerated) GLCompositor::RenderFrame();
}
else if (VCount == 144)
{

14
src/GPU.h
View File

@ -422,6 +422,20 @@ void SetDispStat(u32 cpu, u16 val);
void SetVCount(u16 val);
namespace GLCompositor
{
bool Init();
void DeInit();
void Reset();
void UpdateDisplaySettings();
void RenderFrame();
void BindOutputTexture();
}
}
#endif

2
src/GPU2D.cpp
View File

@ -728,6 +728,8 @@ u32 GPU2D::ColorComposite(int i, u32 val1, u32 val2)
case 3: return ColorBrightnessDown(val1, EVY);
case 4: return ColorBlend5(val1, val2);
}
return val1;
}

2
src/GPU3D.cpp
View File

@ -616,7 +616,7 @@ int InitRenderer(bool hasGL)
if (!GLRenderer::Init())
renderer = 0;
}
printf("renderer: %d\n", renderer);
if (renderer == 0) SoftRenderer::Init();
Renderer = renderer;

31
src/GPU3D_OpenGL.cpp
View File

@ -29,6 +29,8 @@ namespace GPU3D
namespace GLRenderer
{
using namespace OpenGL;
// GL version requirements
// * texelFetch: 3.0 (GLSL 1.30) (3.2/1.50 for MS)
// * UBO: 3.1
@ -142,7 +144,7 @@ bool BuildRenderShader(u32 flags, const char* vs, const char* fs)
strcpy(&fsbuf[headerlen], kRenderFSCommon);
strcpy(&fsbuf[headerlen + fsclen], fs);
bool ret = OpenGL_BuildShaderProgram(vsbuf, fsbuf, RenderShader[flags], shadername);
bool ret = OpenGL::BuildShaderProgram(vsbuf, fsbuf, RenderShader[flags], shadername);
delete[] vsbuf;
delete[] fsbuf;
@ -158,7 +160,7 @@ bool BuildRenderShader(u32 flags, const char* vs, const char* fs)
glBindFragDataLocation(prog, 0, "oColor");
glBindFragDataLocation(prog, 1, "oAttr");
if (!OpenGL_LinkShaderProgram(RenderShader[flags]))
if (!OpenGL::LinkShaderProgram(RenderShader[flags]))
return false;
GLint uni_id = glGetUniformBlockIndex(prog, "uConfig");
@ -202,14 +204,14 @@ bool Init()
glClearDepth(1.0);
if (!OpenGL_BuildShaderProgram(kClearVS, kClearFS, ClearShaderPlain, "ClearShader"))
if (!OpenGL::BuildShaderProgram(kClearVS, kClearFS, ClearShaderPlain, "ClearShader"))
return false;
glBindAttribLocation(ClearShaderPlain[2], 0, "vPosition");
glBindFragDataLocation(ClearShaderPlain[2], 0, "oColor");
glBindFragDataLocation(ClearShaderPlain[2], 1, "oAttr");
if (!OpenGL_LinkShaderProgram(ClearShaderPlain))
if (!OpenGL::LinkShaderProgram(ClearShaderPlain))
return false;
ClearUniformLoc[0] = glGetUniformLocation(ClearShaderPlain[2], "uColor");
@ -237,15 +239,15 @@ bool Init()
kRenderVS_W, kRenderFS_WSM)) return false;
if (!OpenGL_BuildShaderProgram(kFinalPassVS, kFinalPassEdgeFS, FinalPassEdgeShader, "FinalPassEdgeShader"))
if (!OpenGL::BuildShaderProgram(kFinalPassVS, kFinalPassEdgeFS, FinalPassEdgeShader, "FinalPassEdgeShader"))
return false;
if (!OpenGL_BuildShaderProgram(kFinalPassVS, kFinalPassFogFS, FinalPassFogShader, "FinalPassFogShader"))
if (!OpenGL::BuildShaderProgram(kFinalPassVS, kFinalPassFogFS, FinalPassFogShader, "FinalPassFogShader"))
return false;
glBindAttribLocation(FinalPassEdgeShader[2], 0, "vPosition");
glBindFragDataLocation(FinalPassEdgeShader[2], 0, "oColor");
if (!OpenGL_LinkShaderProgram(FinalPassEdgeShader))
if (!OpenGL::LinkShaderProgram(FinalPassEdgeShader))
return false;
uni_id = glGetUniformBlockIndex(FinalPassEdgeShader[2], "uConfig");
@ -261,7 +263,7 @@ bool Init()
glBindAttribLocation(FinalPassFogShader[2], 0, "vPosition");
glBindFragDataLocation(FinalPassFogShader[2], 0, "oColor");
if (!OpenGL_LinkShaderProgram(FinalPassFogShader))
if (!OpenGL::LinkShaderProgram(FinalPassFogShader))
return false;
uni_id = glGetUniformBlockIndex(FinalPassFogShader[2], "uConfig");
@ -371,11 +373,19 @@ bool Init()
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB5_A1, 1024, 48, 0, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV, NULL);
if (!GPU::GLCompositor::Init())
{
// TODO: clean up things? fail more gracefully??
return false;
}
return true;
}
void DeInit()
{
GPU::GLCompositor::DeInit();
glDeleteTextures(1, &TexMemID);
glDeleteTextures(1, &TexPalMemID);
@ -392,12 +402,13 @@ void DeInit()
for (int i = 0; i < 16; i++)
{
if (!RenderShader[i][2]) continue;
OpenGL_DeleteShaderProgram(RenderShader[i]);
OpenGL::DeleteShaderProgram(RenderShader[i]);
}
}
void Reset()
{
GPU::GLCompositor::Reset();
}
void UpdateDisplaySettings()
@ -480,6 +491,8 @@ void UpdateDisplaySettings()
//glLineWidth(scale);
//glLineWidth(1.5);
GPU::GLCompositor::UpdateDisplaySettings();
}

207
src/GPU_OpenGL.cpp Normal file
View File

@ -0,0 +1,207 @@
/*
Copyright 2016-2020 Arisotura
This file is part of melonDS.
melonDS 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 3 of the License, or (at your option)
any later version.
melonDS 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 melonDS. If not, see http://www.gnu.org/licenses/.
*/
#include <stdio.h>
#include <string.h>
#include "NDS.h"
#include "GPU.h"
#include "Config.h"
#include "OpenGLSupport.h"
#include "GPU_OpenGL_shaders.h"
namespace GPU
{
namespace GLCompositor
{
using namespace OpenGL;
int Scale;
int ScreenH, ScreenW;
GLuint CompShader[1][3];
GLuint CompScaleLoc[1];
GLuint CompVertexBufferID;
GLuint CompVertexArrayID;
float CompVertices[2 * 3*2 * 2]; // position
GLuint CompScreenInputTex;
GLuint CompScreenOutputTex;
GLuint CompScreenOutputFB;
bool Init()
{
if (!OpenGL::BuildShaderProgram(kCompositorVS, kCompositorFS_Nearest, CompShader[0], "CompositorShader"))
//if (!OpenGL::BuildShaderProgram(kCompositorVS, kCompositorFS_Linear, CompShader[0], "CompositorShader"))
//if (!OpenGL::BuildShaderProgram(kCompositorVS_xBRZ, kCompositorFS_xBRZ, CompShader[0], "CompositorShader"))
return false;
for (int i = 0; i < 1; i++)
{
GLint uni_id;
glBindAttribLocation(CompShader[i][2], 0, "vPosition");
glBindFragDataLocation(CompShader[i][2], 0, "oColor");
if (!OpenGL::LinkShaderProgram(CompShader[i]))
return false;
CompScaleLoc[i] = glGetUniformLocation(CompShader[i][2], "u3DScale");
glUseProgram(CompShader[i][2]);
uni_id = glGetUniformLocation(CompShader[i][2], "ScreenTex");
glUniform1i(uni_id, 0);
uni_id = glGetUniformLocation(CompShader[i][2], "_3DTex");
glUniform1i(uni_id, 1);
}
#define SETVERTEX(i, x, y) \
CompVertices[2*(i) + 0] = x; \
CompVertices[2*(i) + 1] = y;
// top screen
SETVERTEX(0, -1, 1);
SETVERTEX(1, 1, 0);
SETVERTEX(2, 1, 1);
SETVERTEX(3, -1, 1);
SETVERTEX(4, -1, 0);
SETVERTEX(5, 1, 0);
// bottom screen
SETVERTEX(6, -1, 0);
SETVERTEX(7, 1, -1);
SETVERTEX(8, 1, 0);
SETVERTEX(9, -1, 0);
SETVERTEX(10, -1, -1);
SETVERTEX(11, 1, -1);
#undef SETVERTEX
glGenBuffers(1, &CompVertexBufferID);
glBindBuffer(GL_ARRAY_BUFFER, CompVertexBufferID);
glBufferData(GL_ARRAY_BUFFER, sizeof(CompVertices), CompVertices, GL_STATIC_DRAW);
glGenVertexArrays(1, &CompVertexArrayID);
glBindVertexArray(CompVertexArrayID);
glEnableVertexAttribArray(0); // position
glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 2*4, (void*)(0));
glGenFramebuffers(1, &CompScreenOutputFB);
glGenTextures(1, &CompScreenInputTex);
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, CompScreenInputTex);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA8UI, 256*3 + 1, 192*2, 0, GL_RGBA_INTEGER, GL_UNSIGNED_BYTE, NULL);
glGenTextures(1, &CompScreenOutputTex);
glBindTexture(GL_TEXTURE_2D, CompScreenOutputTex);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
return true;
}
void DeInit()
{
glDeleteFramebuffers(1, &CompScreenOutputFB);
glDeleteTextures(1, &CompScreenInputTex);
glDeleteTextures(1, &CompScreenOutputTex);
glDeleteVertexArrays(1, &CompVertexArrayID);
glDeleteBuffers(1, &CompVertexBufferID);
for (int i = 0; i < 1; i++)
OpenGL::DeleteShaderProgram(CompShader[i]);
}
void Reset()
{
}
void UpdateDisplaySettings()
{
int scale = Config::GL_ScaleFactor;
Scale = scale;
ScreenW = 256 * scale;
ScreenH = 384 * scale;
glBindTexture(GL_TEXTURE_2D, CompScreenOutputTex);
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, ScreenW, ScreenH, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
GLenum fbassign[] = {GL_COLOR_ATTACHMENT0};
glBindFramebuffer(GL_FRAMEBUFFER, CompScreenOutputFB);
glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, CompScreenOutputTex, 0);
glDrawBuffers(1, fbassign);
}
void RenderFrame()
{
glBindFramebuffer(GL_FRAMEBUFFER, CompScreenOutputFB);
glDisable(GL_DEPTH_TEST);
glDisable(GL_STENCIL_TEST);
glDisable(GL_BLEND);
glColorMaski(0, GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
glViewport(0, 0, ScreenW, ScreenH);
// TODO: select more shaders (filtering, etc)
OpenGL::UseShaderProgram(CompShader[0]);
glUniform1ui(CompScaleLoc[0], Scale);
//if (RunningSomething)
{
int frontbuf = GPU::FrontBuffer;
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, CompScreenInputTex);
if (GPU::Framebuffer[frontbuf][0] && GPU::Framebuffer[frontbuf][1])
{
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 256*3 + 1, 192, GL_RGBA_INTEGER,
GL_UNSIGNED_BYTE, GPU::Framebuffer[frontbuf][0]);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 192, 256*3 + 1, 192, GL_RGBA_INTEGER,
GL_UNSIGNED_BYTE, GPU::Framebuffer[frontbuf][1]);
}
glActiveTexture(GL_TEXTURE1);
GPU3D::GLRenderer::SetupAccelFrame();
glBindBuffer(GL_ARRAY_BUFFER, CompVertexBufferID);
glBindVertexArray(CompVertexArrayID);
glDrawArrays(GL_TRIANGLES, 0, 4*3);
}
}
void BindOutputTexture()
{
glBindTexture(GL_TEXTURE_2D, CompScreenOutputTex);
}
}
}

867
src/GPU_OpenGL_shaders.h Normal file
View File

@ -0,0 +1,867 @@
/*
Copyright 2016-2020 Arisotura
This file is part of melonDS.
melonDS 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 3 of the License, or (at your option)
any later version.
melonDS 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 melonDS. If not, see http://www.gnu.org/licenses/.
*/
#ifndef GPU_OPENGL_SHADERS_H
#define GPU_OPENGL_SHADERS_H
const char* kCompositorVS = R"(#version 140
in vec2 vPosition;
smooth out vec2 fTexcoord;
void main()
{
vec4 fpos;
fpos.xy = vPosition;
fpos.z = 0.0;
fpos.w = 1.0;
gl_Position = fpos;
fTexcoord = (vPosition + vec2(1.0, 1.0)) * (vec2(256.0, 384.0) / 2.0);
}
)";
const char* kCompositorFS_Nearest = R"(#version 140
uniform uint u3DScale;
uniform usampler2D ScreenTex;
uniform sampler2D _3DTex;
smooth in vec2 fTexcoord;
out vec4 oColor;
void main()
{
ivec4 pixel = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord), 0));
ivec4 mbright = ivec4(texelFetch(ScreenTex, ivec2(256*3, int(fTexcoord.y)), 0));
int dispmode = mbright.b & 0x3;
if (dispmode == 1)
{
ivec4 val1 = pixel;
ivec4 val2 = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(256,0), 0));
ivec4 val3 = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(512,0), 0));
int compmode = val3.a & 0xF;
int eva, evb, evy;
if (compmode == 4)
{
// 3D on top, blending
float xpos = val3.r + fract(fTexcoord.x);
float ypos = mod(fTexcoord.y, 192);
ivec4 _3dpix = ivec4(texelFetch(_3DTex, ivec2(vec2(xpos, ypos)*u3DScale), 0).bgra
* vec4(63,63,63,31));
if (_3dpix.a > 0)
{
eva = (_3dpix.a & 0x1F) + 1;
evb = 32 - eva;
val1 = ((_3dpix * eva) + (val1 * evb)) >> 5;
if (eva <= 16) val1 += ivec4(1,1,1,0);
val1 = min(val1, 0x3F);
}
else
val1 = val2;
}
else if (compmode == 1)
{
// 3D on bottom, blending
float xpos = val3.r + fract(fTexcoord.x);
float ypos = mod(fTexcoord.y, 192);
ivec4 _3dpix = ivec4(texelFetch(_3DTex, ivec2(vec2(xpos, ypos)*u3DScale), 0).bgra
* vec4(63,63,63,31));
if (_3dpix.a > 0)
{
eva = val3.g;
evb = val3.b;
val1 = ((val1 * eva) + (_3dpix * evb)) >> 4;
val1 = min(val1, 0x3F);
}
else
val1 = val2;
}
else if (compmode <= 3)
{
// 3D on top, normal/fade
float xpos = val3.r + fract(fTexcoord.x);
float ypos = mod(fTexcoord.y, 192);
ivec4 _3dpix = ivec4(texelFetch(_3DTex, ivec2(vec2(xpos, ypos)*u3DScale), 0).bgra
* vec4(63,63,63,31));
if (_3dpix.a > 0)
{
evy = val3.g;
val1 = _3dpix;
if (compmode == 2) val1 += ((ivec4(0x3F,0x3F,0x3F,0) - val1) * evy) >> 4;
else if (compmode == 3) val1 -= (val1 * evy) >> 4;
}
else
val1 = val2;
}
pixel = val1;
}
if (dispmode != 0)
{
int brightmode = mbright.g >> 6;
if (brightmode == 1)
{
// up
int evy = mbright.r & 0x1F;
if (evy > 16) evy = 16;
pixel += ((ivec4(0x3F,0x3F,0x3F,0) - pixel) * evy) >> 4;
}
else if (brightmode == 2)
{
// down
int evy = mbright.r & 0x1F;
if (evy > 16) evy = 16;
pixel -= (pixel * evy) >> 4;
}
}
pixel.rgb <<= 2;
pixel.rgb |= (pixel.rgb >> 6);
// TODO: filters
oColor = vec4(vec3(pixel.rgb) / 255.0, 1.0);
}
)";
const char* kCompositorFS_Linear = R"(#version 140
uniform uint u3DScale;
uniform usampler2D ScreenTex;
uniform sampler2D _3DTex;
smooth in vec2 fTexcoord;
out vec4 oColor;
ivec4 Get3DPixel(vec2 pos)
{
return ivec4(texelFetch(_3DTex, ivec2(pos*u3DScale), 0).bgra
* vec4(63,63,63,31));
}
ivec4 GetFullPixel(ivec4 val1, ivec4 val2, ivec4 val3, ivec4 _3dpix)
{
int compmode = val3.a & 0xF;
int eva, evb, evy;
if (compmode == 4)
{
// 3D on top, blending
if (_3dpix.a > 0)
{
eva = (_3dpix.a & 0x1F) + 1;
evb = 32 - eva;
val1 = ((_3dpix * eva) + (val1 * evb)) >> 5;
if (eva <= 16) val1 += ivec4(1,1,1,0);
val1 = min(val1, 0x3F);
}
else
val1 = val2;
}
else if (compmode == 1)
{
// 3D on bottom, blending
if (_3dpix.a > 0)
{
eva = val3.g;
evb = val3.b;
val1 = ((val1 * eva) + (_3dpix * evb)) >> 4;
val1 = min(val1, 0x3F);
}
else
val1 = val2;
}
else if (compmode <= 3)
{
// 3D on top, normal/fade
if (_3dpix.a > 0)
{
evy = val3.g;
val1 = _3dpix;
if (compmode == 2) val1 += ((ivec4(0x3F,0x3F,0x3F,0) - val1) * evy) >> 4;
else if (compmode == 3) val1 -= (val1 * evy) >> 4;
}
else
val1 = val2;
}
return val1;
}
ivec4 imix(ivec4 a, ivec4 b, float x)
{
return ivec4(vec4(a)*(1-x) + vec4(b)*x);
}
void main()
{
ivec4 pixel = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord), 0));
ivec4 mbright = ivec4(texelFetch(ScreenTex, ivec2(256*3, int(fTexcoord.y)), 0));
int dispmode = mbright.b & 0x3;
if (dispmode == 1)
{
ivec4 val1 = pixel;
ivec4 val2 = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(256,0), 0));
ivec4 val3 = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(512,0), 0));
float xfract = fract(fTexcoord.x);
float yfract = fract(fTexcoord.y);
float xpos = val3.r + xfract;
float ypos = mod(fTexcoord.y, 192);
ivec4 _3dpix = Get3DPixel(vec2(xpos,ypos));
ivec4 p00 = GetFullPixel(val1, val2, val3, _3dpix);
int xdisp = 1 - int(step(255, fTexcoord.x));
int ydisp = 1 - int(step(191, ypos));
ivec4 p01 = GetFullPixel(ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(xdisp+0 ,0), 0)),
ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(xdisp+256,0), 0)),
ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(xdisp+512,0), 0)),
_3dpix);
ivec4 p10 = GetFullPixel(ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(0+0 ,ydisp), 0)),
ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(0+256,ydisp), 0)),
ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(0+512,ydisp), 0)),
_3dpix);
ivec4 p11 = GetFullPixel(ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(xdisp+0 ,ydisp), 0)),
ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(xdisp+256,ydisp), 0)),
ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(xdisp+512,ydisp), 0)),
_3dpix);
ivec4 pa = imix(p00, p01, xfract);
ivec4 pb = imix(p10, p11, xfract);
pixel = imix(pa, pb, yfract);
}
if (dispmode != 0)
{
int brightmode = mbright.g >> 6;
if (brightmode == 1)
{
// up
int evy = mbright.r & 0x1F;
if (evy > 16) evy = 16;
pixel += ((ivec4(0x3F,0x3F,0x3F,0) - pixel) * evy) >> 4;
}
else if (brightmode == 2)
{
// down
int evy = mbright.r & 0x1F;
if (evy > 16) evy = 16;
pixel -= (pixel * evy) >> 4;
}
}
pixel.rgb <<= 2;
pixel.rgb |= (pixel.rgb >> 6);
// TODO: filters
oColor = vec4(vec3(pixel.rgb) / 255.0, 1.0);
}
)";
// HUGE TEST ZONE ARRLGD
const char* kCompositorVS_xBRZ = R"(#version 140
#define BLEND_NONE 0
#define BLEND_NORMAL 1
#define BLEND_DOMINANT 2
#define LUMINANCE_WEIGHT 1.0
#define EQUAL_COLOR_TOLERANCE 30.0/255.0
#define STEEP_DIRECTION_THRESHOLD 2.2
#define DOMINANT_DIRECTION_THRESHOLD 3.6
#if __VERSION__ >= 130
#define COMPAT_VARYING out
#define COMPAT_ATTRIBUTE in
#define COMPAT_TEXTURE texture
#else
#define COMPAT_VARYING varying
#define COMPAT_ATTRIBUTE attribute
#define COMPAT_TEXTURE texture2D
#endif
#ifdef GL_ES
#define COMPAT_PRECISION mediump
#else
#define COMPAT_PRECISION
#endif
COMPAT_ATTRIBUTE vec2 vPosition;
COMPAT_VARYING vec4 TEX0;
COMPAT_VARYING vec4 t1;
COMPAT_VARYING vec4 t2;
COMPAT_VARYING vec4 t3;
COMPAT_VARYING vec4 t4;
COMPAT_VARYING vec4 t5;
COMPAT_VARYING vec4 t6;
COMPAT_VARYING vec4 t7;
uniform COMPAT_PRECISION int FrameDirection;
uniform COMPAT_PRECISION int FrameCount;
uniform COMPAT_PRECISION vec2 OutputSize;
uniform COMPAT_PRECISION vec2 TextureSize;
uniform COMPAT_PRECISION vec2 InputSize;
// vertex compatibility #defines
#define vTexCoord TEX0.xy
#define SourceSize vec4(TextureSize, 1.0 / TextureSize) //either TextureSize or InputSize
#define outsize vec4(OutputSize, 1.0 / OutputSize)
void main()
{
vec4 fpos;
fpos.xy = vPosition;
fpos.z = 0.0;
fpos.w = 1.0;
gl_Position = fpos;
vec2 TexCoord = (vPosition + vec2(1.0, 1.0)) * (vec2(256.0, 384.0) / 2.0);
//gl_Position = MVPMatrix * VertexCoord;
//COL0 = COLOR;
TEX0.xy = TexCoord.xy;
vec2 ps = vec2(1,1);//vec2(SourceSize.z, SourceSize.w);
float dx = ps.x;
float dy = ps.y;
// A1 B1 C1
// A0 A B C C4
// D0 D E F F4
// G0 G H I I4
// G5 H5 I5
t1 = vTexCoord.xxxy + vec4( -dx, 0.0, dx,-2.0*dy); // A1 B1 C1
t2 = vTexCoord.xxxy + vec4( -dx, 0.0, dx, -dy); // A B C
t3 = vTexCoord.xxxy + vec4( -dx, 0.0, dx, 0.0); // D E F
t4 = vTexCoord.xxxy + vec4( -dx, 0.0, dx, dy); // G H I
t5 = vTexCoord.xxxy + vec4( -dx, 0.0, dx, 2.0*dy); // G5 H5 I5
t6 = vTexCoord.xyyy + vec4(-2.0*dx,-dy, 0.0, dy); // A0 D0 G0
t7 = vTexCoord.xyyy + vec4( 2.0*dx,-dy, 0.0, dy); // C4 F4 I4
}
)";
const char* kCompositorFS_xBRZ = R"(#version 140
#define BLEND_NONE 0
#define BLEND_NORMAL 1
#define BLEND_DOMINANT 2
#define LUMINANCE_WEIGHT 1.0
#define EQUAL_COLOR_TOLERANCE 30.0/255.0
#define STEEP_DIRECTION_THRESHOLD 2.2
#define DOMINANT_DIRECTION_THRESHOLD 3.6
#if __VERSION__ >= 130
#define COMPAT_VARYING in
//#define COMPAT_TEXTURE texture
#define FragColor oColor
#else
#define COMPAT_VARYING varying
#define FragColor gl_FragColor
//#define COMPAT_TEXTURE texture2D
#endif
#ifdef GL_ES
#ifdef GL_FRAGMENT_PRECISION_HIGH
precision highp float;
#else
precision mediump float;
#endif
#define COMPAT_PRECISION mediump
#else
#define COMPAT_PRECISION
#endif
uniform uint u3DScale;
uniform usampler2D ScreenTex;
uniform sampler2D _3DTex;
smooth in vec2 fTexcoord;
out vec4 oColor;
//uniform COMPAT_PRECISION vec2 OutputSize;
//uniform COMPAT_PRECISION vec2 TextureSize;
#define TextureSize vec2(256,384)
//uniform COMPAT_PRECISION vec2 InputSize;
//uniform sampler2D Texture;
#define Texture 1312
COMPAT_VARYING vec4 TEX0;
COMPAT_VARYING vec4 t1;
COMPAT_VARYING vec4 t2;
COMPAT_VARYING vec4 t3;
COMPAT_VARYING vec4 t4;
COMPAT_VARYING vec4 t5;
COMPAT_VARYING vec4 t6;
COMPAT_VARYING vec4 t7;
// fragment compatibility #defines
#define Source Texture
#define vTexCoord TEX0.xy
#define SourceSize vec4(TextureSize, 1.0 / TextureSize) //either TextureSize or InputSize
#define outsize vec4(OutputSize, 1.0 / OutputSize)
const float one_sixth = 1.0 / 6.0;
const float two_sixth = 2.0 / 6.0;
const float four_sixth = 4.0 / 6.0;
const float five_sixth = 5.0 / 6.0;
vec4 Get2DPixel(vec2 texcoord, int level)
{
ivec4 pixel = ivec4(texelFetch(ScreenTex, ivec2(texcoord) + ivec2(level*256,0), 0));
return vec4(pixel) / vec4(63.0, 63.0, 63.0, 31.0);
}
ivec4 Get3DPixel(vec2 pos)
{
return ivec4(texelFetch(_3DTex, ivec2(pos*u3DScale), 0).bgra
* vec4(63,63,63,31));
}
float reduce(const vec3 color)
{
return dot(color, vec3(65536.0, 256.0, 1.0));
}
float DistYCbCr(const vec3 pixA, const vec3 pixB)
{
const vec3 w = vec3(0.2627, 0.6780, 0.0593);
const float scaleB = 0.5 / (1.0 - w.b);
const float scaleR = 0.5 / (1.0 - w.r);
vec3 diff = pixA - pixB;
float Y = dot(diff, w);
float Cb = scaleB * (diff.b - Y);
float Cr = scaleR * (diff.r - Y);
return sqrt( ((LUMINANCE_WEIGHT * Y) * (LUMINANCE_WEIGHT * Y)) + (Cb * Cb) + (Cr * Cr) );
}
bool IsPixEqual(const vec3 pixA, const vec3 pixB)
{
return (DistYCbCr(pixA, pixB) < EQUAL_COLOR_TOLERANCE);
}
bool IsBlendingNeeded(const ivec4 blend)
{
return any(notEqual(blend, ivec4(BLEND_NONE)));
}
//---------------------------------------
// Input Pixel Mapping: --|21|22|23|--
// 19|06|07|08|09
// 18|05|00|01|10
// 17|04|03|02|11
// --|15|14|13|--
//
// Output Pixel Mapping: 20|21|22|23|24|25
// 19|06|07|08|09|26
// 18|05|00|01|10|27
// 17|04|03|02|11|28
// 16|15|14|13|12|29
// 35|34|33|32|31|30
ivec4 GetFiltered2DPixel(int level)
{
vec2 f = fract(vTexCoord.xy);// * SourceSize.xy);
//---------------------------------------
// Input Pixel Mapping: 20|21|22|23|24
// 19|06|07|08|09
// 18|05|00|01|10
// 17|04|03|02|11
// 16|15|14|13|12
vec3 src[25];
src[21] = Get2DPixel(t1.xw, level).rgb;
src[22] = Get2DPixel(t1.yw, level).rgb;
src[23] = Get2DPixel(t1.zw, level).rgb;
src[ 6] = Get2DPixel(t2.xw, level).rgb;
src[ 7] = Get2DPixel(t2.yw, level).rgb;
src[ 8] = Get2DPixel(t2.zw, level).rgb;
src[ 5] = Get2DPixel(t3.xw, level).rgb;
src[ 0] = Get2DPixel(t3.yw, level).rgb;
src[ 1] = Get2DPixel(t3.zw, level).rgb;
src[ 4] = Get2DPixel(t4.xw, level).rgb;
src[ 3] = Get2DPixel(t4.yw, level).rgb;
src[ 2] = Get2DPixel(t4.zw, level).rgb;
src[15] = Get2DPixel(t5.xw, level).rgb;
src[14] = Get2DPixel(t5.yw, level).rgb;
src[13] = Get2DPixel(t5.zw, level).rgb;
src[19] = Get2DPixel(t6.xy, level).rgb;
src[18] = Get2DPixel(t6.xz, level).rgb;
src[17] = Get2DPixel(t6.xw, level).rgb;
src[ 9] = Get2DPixel(t7.xy, level).rgb;
src[10] = Get2DPixel(t7.xz, level).rgb;
src[11] = Get2DPixel(t7.xw, level).rgb;
float v[9];
v[0] = reduce(src[0]);
v[1] = reduce(src[1]);
v[2] = reduce(src[2]);
v[3] = reduce(src[3]);
v[4] = reduce(src[4]);
v[5] = reduce(src[5]);
v[6] = reduce(src[6]);
v[7] = reduce(src[7]);
v[8] = reduce(src[8]);
ivec4 blendResult = ivec4(BLEND_NONE);
// Preprocess corners
// Pixel Tap Mapping: --|--|--|--|--
// --|--|07|08|--
// --|05|00|01|10
// --|04|03|02|11
// --|--|14|13|--
// Corner (1, 1)
if ( ((v[0] == v[1] && v[3] == v[2]) || (v[0] == v[3] && v[1] == v[2])) == false)
{
float dist_03_01 = DistYCbCr(src[ 4], src[ 0]) + DistYCbCr(src[ 0], src[ 8]) + DistYCbCr(src[14], src[ 2]) + DistYCbCr(src[ 2], src[10]) + (4.0 * DistYCbCr(src[ 3], src[ 1]));
float dist_00_02 = DistYCbCr(src[ 5], src[ 3]) + DistYCbCr(src[ 3], src[13]) + DistYCbCr(src[ 7], src[ 1]) + DistYCbCr(src[ 1], src[11]) + (4.0 * DistYCbCr(src[ 0], src[ 2]));
bool dominantGradient = (DOMINANT_DIRECTION_THRESHOLD * dist_03_01) < dist_00_02;
blendResult[2] = ((dist_03_01 < dist_00_02) && (v[0] != v[1]) && (v[0] != v[3])) ? ((dominantGradient) ? BLEND_DOMINANT : BLEND_NORMAL) : BLEND_NONE;
}
// Pixel Tap Mapping: --|--|--|--|--
// --|06|07|--|--
// 18|05|00|01|--
// 17|04|03|02|--
// --|15|14|--|--
// Corner (0, 1)
if ( ((v[5] == v[0] && v[4] == v[3]) || (v[5] == v[4] && v[0] == v[3])) == false)
{
float dist_04_00 = DistYCbCr(src[17], src[ 5]) + DistYCbCr(src[ 5], src[ 7]) + DistYCbCr(src[15], src[ 3]) + DistYCbCr(src[ 3], src[ 1]) + (4.0 * DistYCbCr(src[ 4], src[ 0]));
float dist_05_03 = DistYCbCr(src[18], src[ 4]) + DistYCbCr(src[ 4], src[14]) + DistYCbCr(src[ 6], src[ 0]) + DistYCbCr(src[ 0], src[ 2]) + (4.0 * DistYCbCr(src[ 5], src[ 3]));
bool dominantGradient = (DOMINANT_DIRECTION_THRESHOLD * dist_05_03) < dist_04_00;
blendResult[3] = ((dist_04_00 > dist_05_03) && (v[0] != v[5]) && (v[0] != v[3])) ? ((dominantGradient) ? BLEND_DOMINANT : BLEND_NORMAL) : BLEND_NONE;
}
// Pixel Tap Mapping: --|--|22|23|--
// --|06|07|08|09
// --|05|00|01|10
// --|--|03|02|--
// --|--|--|--|--
// Corner (1, 0)
if ( ((v[7] == v[8] && v[0] == v[1]) || (v[7] == v[0] && v[8] == v[1])) == false)
{
float dist_00_08 = DistYCbCr(src[ 5], src[ 7]) + DistYCbCr(src[ 7], src[23]) + DistYCbCr(src[ 3], src[ 1]) + DistYCbCr(src[ 1], src[ 9]) + (4.0 * DistYCbCr(src[ 0], src[ 8]));
float dist_07_01 = DistYCbCr(src[ 6], src[ 0]) + DistYCbCr(src[ 0], src[ 2]) + DistYCbCr(src[22], src[ 8]) + DistYCbCr(src[ 8], src[10]) + (4.0 * DistYCbCr(src[ 7], src[ 1]));
bool dominantGradient = (DOMINANT_DIRECTION_THRESHOLD * dist_07_01) < dist_00_08;
blendResult[1] = ((dist_00_08 > dist_07_01) && (v[0] != v[7]) && (v[0] != v[1])) ? ((dominantGradient) ? BLEND_DOMINANT : BLEND_NORMAL) : BLEND_NONE;
}
// Pixel Tap Mapping: --|21|22|--|--
// 19|06|07|08|--
// 18|05|00|01|--
// --|04|03|--|--
// --|--|--|--|--
// Corner (0, 0)
if ( ((v[6] == v[7] && v[5] == v[0]) || (v[6] == v[5] && v[7] == v[0])) == false)
{
float dist_05_07 = DistYCbCr(src[18], src[ 6]) + DistYCbCr(src[ 6], src[22]) + DistYCbCr(src[ 4], src[ 0]) + DistYCbCr(src[ 0], src[ 8]) + (4.0 * DistYCbCr(src[ 5], src[ 7]));
float dist_06_00 = DistYCbCr(src[19], src[ 5]) + DistYCbCr(src[ 5], src[ 3]) + DistYCbCr(src[21], src[ 7]) + DistYCbCr(src[ 7], src[ 1]) + (4.0 * DistYCbCr(src[ 6], src[ 0]));
bool dominantGradient = (DOMINANT_DIRECTION_THRESHOLD * dist_05_07) < dist_06_00;
blendResult[0] = ((dist_05_07 < dist_06_00) && (v[0] != v[5]) && (v[0] != v[7])) ? ((dominantGradient) ? BLEND_DOMINANT : BLEND_NORMAL) : BLEND_NONE;
}
vec3 dst[16];
dst[ 0] = src[0];
dst[ 1] = src[0];
dst[ 2] = src[0];
dst[ 3] = src[0];
dst[ 4] = src[0];
dst[ 5] = src[0];
dst[ 6] = src[0];
dst[ 7] = src[0];
dst[ 8] = src[0];
dst[ 9] = src[0];
dst[10] = src[0];
dst[11] = src[0];
dst[12] = src[0];
dst[13] = src[0];
dst[14] = src[0];
dst[15] = src[0];
// Scale pixel
if (IsBlendingNeeded(blendResult) == true)
{
float dist_01_04 = DistYCbCr(src[1], src[4]);
float dist_03_08 = DistYCbCr(src[3], src[8]);
bool haveShallowLine = (STEEP_DIRECTION_THRESHOLD * dist_01_04 <= dist_03_08) && (v[0] != v[4]) && (v[5] != v[4]);
bool haveSteepLine = (STEEP_DIRECTION_THRESHOLD * dist_03_08 <= dist_01_04) && (v[0] != v[8]) && (v[7] != v[8]);
bool needBlend = (blendResult[2] != BLEND_NONE);
bool doLineBlend = ( blendResult[2] >= BLEND_DOMINANT ||
((blendResult[1] != BLEND_NONE && !IsPixEqual(src[0], src[4])) ||
(blendResult[3] != BLEND_NONE && !IsPixEqual(src[0], src[8])) ||
(IsPixEqual(src[4], src[3]) && IsPixEqual(src[3], src[2]) && IsPixEqual(src[2], src[1]) && IsPixEqual(src[1], src[8]) && IsPixEqual(src[0], src[2]) == false) ) == false );
vec3 blendPix = ( DistYCbCr(src[0], src[1]) <= DistYCbCr(src[0], src[3]) ) ? src[1] : src[3];
dst[ 2] = mix(dst[ 2], blendPix, (needBlend && doLineBlend) ? ((haveShallowLine) ? ((haveSteepLine) ? 1.0/3.0 : 0.25) : ((haveSteepLine) ? 0.25 : 0.00)) : 0.00);
dst[ 9] = mix(dst[ 9], blendPix, (needBlend && doLineBlend && haveSteepLine) ? 0.25 : 0.00);
dst[10] = mix(dst[10], blendPix, (needBlend && doLineBlend && haveSteepLine) ? 0.75 : 0.00);
dst[11] = mix(dst[11], blendPix, (needBlend) ? ((doLineBlend) ? ((haveSteepLine) ? 1.00 : ((haveShallowLine) ? 0.75 : 0.50)) : 0.08677704501) : 0.00);
dst[12] = mix(dst[12], blendPix, (needBlend) ? ((doLineBlend) ? 1.00 : 0.6848532563) : 0.00);
dst[13] = mix(dst[13], blendPix, (needBlend) ? ((doLineBlend) ? ((haveShallowLine) ? 1.00 : ((haveSteepLine) ? 0.75 : 0.50)) : 0.08677704501) : 0.00);
dst[14] = mix(dst[14], blendPix, (needBlend && doLineBlend && haveShallowLine) ? 0.75 : 0.00);
dst[15] = mix(dst[15], blendPix, (needBlend && doLineBlend && haveShallowLine) ? 0.25 : 0.00);
dist_01_04 = DistYCbCr(src[7], src[2]);
dist_03_08 = DistYCbCr(src[1], src[6]);
haveShallowLine = (STEEP_DIRECTION_THRESHOLD * dist_01_04 <= dist_03_08) && (v[0] != v[2]) && (v[3] != v[2]);
haveSteepLine = (STEEP_DIRECTION_THRESHOLD * dist_03_08 <= dist_01_04) && (v[0] != v[6]) && (v[5] != v[6]);
needBlend = (blendResult[1] != BLEND_NONE);
doLineBlend = ( blendResult[1] >= BLEND_DOMINANT ||
!((blendResult[0] != BLEND_NONE && !IsPixEqual(src[0], src[2])) ||
(blendResult[2] != BLEND_NONE && !IsPixEqual(src[0], src[6])) ||
(IsPixEqual(src[2], src[1]) && IsPixEqual(src[1], src[8]) && IsPixEqual(src[8], src[7]) && IsPixEqual(src[7], src[6]) && !IsPixEqual(src[0], src[8])) ) );
blendPix = ( DistYCbCr(src[0], src[7]) <= DistYCbCr(src[0], src[1]) ) ? src[7] : src[1];
dst[ 1] = mix(dst[ 1], blendPix, (needBlend && doLineBlend) ? ((haveShallowLine) ? ((haveSteepLine) ? 1.0/3.0 : 0.25) : ((haveSteepLine) ? 0.25 : 0.00)) : 0.00);
dst[ 6] = mix(dst[ 6], blendPix, (needBlend && doLineBlend && haveSteepLine) ? 0.25 : 0.00);
dst[ 7] = mix(dst[ 7], blendPix, (needBlend && doLineBlend && haveSteepLine) ? 0.75 : 0.00);
dst[ 8] = mix(dst[ 8], blendPix, (needBlend) ? ((doLineBlend) ? ((haveSteepLine) ? 1.00 : ((haveShallowLine) ? 0.75 : 0.50)) : 0.08677704501) : 0.00);
dst[ 9] = mix(dst[ 9], blendPix, (needBlend) ? ((doLineBlend) ? 1.00 : 0.6848532563) : 0.00);
dst[10] = mix(dst[10], blendPix, (needBlend) ? ((doLineBlend) ? ((haveShallowLine) ? 1.00 : ((haveSteepLine) ? 0.75 : 0.50)) : 0.08677704501) : 0.00);
dst[11] = mix(dst[11], blendPix, (needBlend && doLineBlend && haveShallowLine) ? 0.75 : 0.00);
dst[12] = mix(dst[12], blendPix, (needBlend && doLineBlend && haveShallowLine) ? 0.25 : 0.00);
dist_01_04 = DistYCbCr(src[5], src[8]);
dist_03_08 = DistYCbCr(src[7], src[4]);
haveShallowLine = (STEEP_DIRECTION_THRESHOLD * dist_01_04 <= dist_03_08) && (v[0] != v[8]) && (v[1] != v[8]);
haveSteepLine = (STEEP_DIRECTION_THRESHOLD * dist_03_08 <= dist_01_04) && (v[0] != v[4]) && (v[3] != v[4]);
needBlend = (blendResult[0] != BLEND_NONE);
doLineBlend = ( blendResult[0] >= BLEND_DOMINANT ||
!((blendResult[3] != BLEND_NONE && !IsPixEqual(src[0], src[8])) ||
(blendResult[1] != BLEND_NONE && !IsPixEqual(src[0], src[4])) ||
(IsPixEqual(src[8], src[7]) && IsPixEqual(src[7], src[6]) && IsPixEqual(src[6], src[5]) && IsPixEqual(src[5], src[4]) && !IsPixEqual(src[0], src[6])) ) );
blendPix = ( DistYCbCr(src[0], src[5]) <= DistYCbCr(src[0], src[7]) ) ? src[5] : src[7];
dst[ 0] = mix(dst[ 0], blendPix, (needBlend && doLineBlend) ? ((haveShallowLine) ? ((haveSteepLine) ? 1.0/3.0 : 0.25) : ((haveSteepLine) ? 0.25 : 0.00)) : 0.00);
dst[15] = mix(dst[15], blendPix, (needBlend && doLineBlend && haveSteepLine) ? 0.25 : 0.00);
dst[ 4] = mix(dst[ 4], blendPix, (needBlend && doLineBlend && haveSteepLine) ? 0.75 : 0.00);
dst[ 5] = mix(dst[ 5], blendPix, (needBlend) ? ((doLineBlend) ? ((haveSteepLine) ? 1.00 : ((haveShallowLine) ? 0.75 : 0.50)) : 0.08677704501) : 0.00);
dst[ 6] = mix(dst[ 6], blendPix, (needBlend) ? ((doLineBlend) ? 1.00 : 0.6848532563) : 0.00);
dst[ 7] = mix(dst[ 7], blendPix, (needBlend) ? ((doLineBlend) ? ((haveShallowLine) ? 1.00 : ((haveSteepLine) ? 0.75 : 0.50)) : 0.08677704501) : 0.00);
dst[ 8] = mix(dst[ 8], blendPix, (needBlend && doLineBlend && haveShallowLine) ? 0.75 : 0.00);
dst[ 9] = mix(dst[ 9], blendPix, (needBlend && doLineBlend && haveShallowLine) ? 0.25 : 0.00);
dist_01_04 = DistYCbCr(src[3], src[6]);
dist_03_08 = DistYCbCr(src[5], src[2]);
haveShallowLine = (STEEP_DIRECTION_THRESHOLD * dist_01_04 <= dist_03_08) && (v[0] != v[6]) && (v[7] != v[6]);
haveSteepLine = (STEEP_DIRECTION_THRESHOLD * dist_03_08 <= dist_01_04) && (v[0] != v[2]) && (v[1] != v[2]);
needBlend = (blendResult[3] != BLEND_NONE);
doLineBlend = ( blendResult[3] >= BLEND_DOMINANT ||
!((blendResult[2] != BLEND_NONE && !IsPixEqual(src[0], src[6])) ||
(blendResult[0] != BLEND_NONE && !IsPixEqual(src[0], src[2])) ||
(IsPixEqual(src[6], src[5]) && IsPixEqual(src[5], src[4]) && IsPixEqual(src[4], src[3]) && IsPixEqual(src[3], src[2]) && !IsPixEqual(src[0], src[4])) ) );
blendPix = ( DistYCbCr(src[0], src[3]) <= DistYCbCr(src[0], src[5]) ) ? src[3] : src[5];
dst[ 3] = mix(dst[ 3], blendPix, (needBlend && doLineBlend) ? ((haveShallowLine) ? ((haveSteepLine) ? 1.0/3.0 : 0.25) : ((haveSteepLine) ? 0.25 : 0.00)) : 0.00);
dst[12] = mix(dst[12], blendPix, (needBlend && doLineBlend && haveSteepLine) ? 0.25 : 0.00);
dst[13] = mix(dst[13], blendPix, (needBlend && doLineBlend && haveSteepLine) ? 0.75 : 0.00);
dst[14] = mix(dst[14], blendPix, (needBlend) ? ((doLineBlend) ? ((haveSteepLine) ? 1.00 : ((haveShallowLine) ? 0.75 : 0.50)) : 0.08677704501) : 0.00);
dst[15] = mix(dst[15], blendPix, (needBlend) ? ((doLineBlend) ? 1.00 : 0.6848532563) : 0.00);
dst[ 4] = mix(dst[ 4], blendPix, (needBlend) ? ((doLineBlend) ? ((haveShallowLine) ? 1.00 : ((haveSteepLine) ? 0.75 : 0.50)) : 0.08677704501) : 0.00);
dst[ 5] = mix(dst[ 5], blendPix, (needBlend && doLineBlend && haveShallowLine) ? 0.75 : 0.00);
dst[ 6] = mix(dst[ 6], blendPix, (needBlend && doLineBlend && haveShallowLine) ? 0.25 : 0.00);
}
vec3 res = mix( mix( mix( mix(dst[ 6], dst[ 7], step(0.25, f.x)), mix(dst[ 8], dst[ 9], step(0.75, f.x)), step(0.50, f.x)),
mix( mix(dst[ 5], dst[ 0], step(0.25, f.x)), mix(dst[ 1], dst[10], step(0.75, f.x)), step(0.50, f.x)), step(0.25, f.y)),
mix( mix( mix(dst[ 4], dst[ 3], step(0.25, f.x)), mix(dst[ 2], dst[11], step(0.75, f.x)), step(0.50, f.x)),
mix( mix(dst[15], dst[14], step(0.25, f.x)), mix(dst[13], dst[12], step(0.75, f.x)), step(0.50, f.x)), step(0.75, f.y)),
step(0.50, f.y));
return ivec4(res * vec3(63,63,63), 0);
}
void main()
{
vec2 fTexcoord = vTexCoord.xy;
ivec4 pixel;// = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord), 0));
ivec4 mbright = ivec4(texelFetch(ScreenTex, ivec2(256*3, int(fTexcoord.y)), 0));
int dispmode = mbright.b & 0x3;
if (dispmode == 1)
{
ivec4 val1;// = pixel;
//ivec4 val2 = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(256,0), 0));
ivec4 val3 = ivec4(texelFetch(ScreenTex, ivec2(fTexcoord) + ivec2(512,0), 0));
int compmode = val3.a & 0xF;
int eva, evb, evy;
float xpos = val3.r + fract(fTexcoord.x);
float ypos = mod(fTexcoord.y, 192);
ivec4 _3dpix = Get3DPixel(vec2(xpos, ypos));
if (compmode == 4)
{
// 3D on top, blending
if (_3dpix.a > 0)
{
eva = (_3dpix.a & 0x1F) + 1;
if (eva == 32)
{
val1 = _3dpix;
}
else
{
evb = 32 - eva;
val1 = GetFiltered2DPixel(0);
val1 = ((_3dpix * eva) + (val1 * evb)) >> 5;
if (eva <= 16) val1 += ivec4(1,1,1,0);
val1 = min(val1, 0x3F);
}
}
else
val1 = GetFiltered2DPixel(1);
}
else if (compmode == 1)
{
// 3D on bottom, blending
if (_3dpix.a > 0)
{
eva = val3.g;
evb = val3.b;
val1 = GetFiltered2DPixel(0);
val1 = ((val1 * eva) + (_3dpix * evb)) >> 4;
val1 = min(val1, 0x3F);
}
else
val1 = GetFiltered2DPixel(1);
}
else if (compmode <= 3)
{
// 3D on top, normal/fade
if (_3dpix.a > 0)
{
evy = val3.g;
val1 = _3dpix;
if (compmode == 2) val1 += ((ivec4(0x3F,0x3F,0x3F,0) - val1) * evy) >> 4;
else if (compmode == 3) val1 -= (val1 * evy) >> 4;
}
else
val1 = GetFiltered2DPixel(1);
}
else
val1 = GetFiltered2DPixel(0);
pixel = val1;
}
else
{
pixel = GetFiltered2DPixel(0);
}
if (dispmode != 0)
{
int brightmode = mbright.g >> 6;
if (brightmode == 1)
{
// up
int evy = mbright.r & 0x1F;
if (evy > 16) evy = 16;
pixel += ((ivec4(0x3F,0x3F,0x3F,0) - pixel) * evy) >> 4;
}
else if (brightmode == 2)
{
// down
int evy = mbright.r & 0x1F;
if (evy > 16) evy = 16;
pixel -= (pixel * evy) >> 4;
}
}
pixel.rgb <<= 2;
pixel.rgb |= (pixel.rgb >> 6);
FragColor = vec4(vec3(pixel.rgb) / 255.0, 1.0);
}
)";
#endif // GPU_OPENGL_SHADERS_H

14
src/OpenGLSupport.cpp
View File

@ -19,18 +19,20 @@
#include "OpenGLSupport.h"
namespace OpenGL
{
DO_PROCLIST(DECLPROC);
bool OpenGL_Init()
bool Init()
{
DO_PROCLIST(LOADPROC);
return true;
}
bool OpenGL_BuildShaderProgram(const char* vs, const char* fs, GLuint* ids, const char* name)
bool BuildShaderProgram(const char* vs, const char* fs, GLuint* ids, const char* name)
{
int len;
int res;
@ -89,7 +91,7 @@ bool OpenGL_BuildShaderProgram(const char* vs, const char* fs, GLuint* ids, cons
return true;
}
bool OpenGL_LinkShaderProgram(GLuint* ids)
bool LinkShaderProgram(GLuint* ids)
{
int res;
@ -115,14 +117,16 @@ bool OpenGL_LinkShaderProgram(GLuint* ids)
return true;
}
void OpenGL_DeleteShaderProgram(GLuint* ids)
void DeleteShaderProgram(GLuint* ids)
{
glDeleteShader(ids[0]);
glDeleteShader(ids[1]);
glDeleteProgram(ids[2]);
}
void OpenGL_UseShaderProgram(GLuint* ids)
void UseShaderProgram(GLuint* ids)
{
glUseProgram(ids[2]);
}
}

32
src/OpenGLSupport.h
View File

@ -21,6 +21,8 @@
#include <stdio.h>
#include <string.h>
// TODO: different includes for each platform
#include <GL/gl.h>
#include <GL/glext.h>
@ -45,23 +47,11 @@
// if you need more OpenGL functions, add them to the macronator here
// TODO: handle conditionally loading certain functions for different GL versions
#ifndef __WIN32__
#define DO_PROCLIST_1_3(func)
#else
#define DO_PROCLIST_1_3(func) \
func(GLACTIVETEXTURE, glActiveTexture); \
func(GLBLENDCOLOR, glBlendColor); \
#endif
#define DO_PROCLIST(func) \
DO_PROCLIST_1_3(func) \
func(GLACTIVETEXTURE, glActiveTexture); \
func(GLBLENDCOLOR, glBlendColor); \
\
func(GLGENFRAMEBUFFERS, glGenFramebuffers); \
func(GLDELETEFRAMEBUFFERS, glDeleteFramebuffers); \
@ -122,14 +112,18 @@
func(GLGETSTRINGI, glGetStringi); \
namespace OpenGL
{
DO_PROCLIST(DECLPROC_EXT);
bool Init();
bool OpenGL_Init();
bool BuildShaderProgram(const char* vs, const char* fs, GLuint* ids, const char* name);
bool LinkShaderProgram(GLuint* ids);
void DeleteShaderProgram(GLuint* ids);
void UseShaderProgram(GLuint* ids);
bool OpenGL_BuildShaderProgram(const char* vs, const char* fs, GLuint* ids, const char* name);
bool OpenGL_LinkShaderProgram(GLuint* ids);
void OpenGL_DeleteShaderProgram(GLuint* ids);
void OpenGL_UseShaderProgram(GLuint* ids);
}
#endif // OPENGLSUPPORT_H

3
src/frontend/qt_sdl/Platform.cpp
View File

@ -56,6 +56,7 @@
char* EmuDirectory;
void emuStop();
void* oglGetProcAddress(const char* proc);
namespace Platform
@ -418,7 +419,7 @@ void Semaphore_Post(void* sema)
void* GL_GetProcAddress(const char* proc)
{
return NULL;//uiGLGetProcAddress(proc);
return oglGetProcAddress(proc);
}

96
src/frontend/qt_sdl/main.cpp
View File

@ -45,6 +45,7 @@
#include "NDS.h"
#include "GBACart.h"
#include "OpenGLSupport.h"
#include "GPU.h"
#include "SPU.h"
#include "Wifi.h"
@ -258,6 +259,54 @@ EmuThread::EmuThread(QObject* parent) : QThread(parent)
connect(this, SIGNAL(windowEmuPause()), mainWindow->actPause, SLOT(trigger()));
connect(this, SIGNAL(windowEmuReset()), mainWindow->actReset, SLOT(trigger()));
connect(this, SIGNAL(screenLayoutChange()), mainWindow->panel, SLOT(onScreenLayoutChanged()));
initOpenGL();
}
void EmuThread::initOpenGL()
{
QOpenGLContext* windowctx = mainWindow->getOGLContext();
QSurfaceFormat format = windowctx->format();
oglSurface = new QOffscreenSurface();
oglSurface->setFormat(format);
oglSurface->create();
if (!oglSurface->isValid())
{
// TODO handle this!
printf("oglSurface shat itself :(\n");
delete oglSurface;
return;
}
oglContext = new QOpenGLContext();//oglSurface);
oglContext->setFormat(oglSurface->format());
oglContext->setShareContext(windowctx);
if (!oglContext->create())
{
// TODO handle this!
printf("oglContext shat itself :(\n");
delete oglContext;
delete oglSurface;
return;
}
oglContext->moveToThread(this);
}
void deinitOpenGL()
{
// TODO!!
}
void* oglGetProcAddress(const char* proc)
{
return emuThread->oglGetProcAddress(proc);
}
void* EmuThread::oglGetProcAddress(const char* proc)
{
return (void*)oglContext->getProcAddress(proc);
}
void EmuThread::run()
@ -279,7 +328,11 @@ void EmuThread::run()
}
else*/
{
GPU3D::InitRenderer(false);
//GPU3D::InitRenderer(false);
bool res = oglContext->makeCurrent(oglSurface);
printf("good? %d\n", res);
OpenGL::Init();
GPU3D::InitRenderer(res);
}
Input::Init();
@ -755,6 +808,11 @@ void ScreenPanelGL::initializeGL()
{
initializeOpenGLFunctions();
const GLubyte* renderer = glGetString(GL_RENDERER); // get renderer string
const GLubyte* version = glGetString(GL_VERSION); // version as a string
printf("OpenGL: renderer: %s\n", renderer);
printf("OpenGL: version: %s\n", version);
glClearColor(0, 0, 0, 1);
screenShader = new QOpenGLShaderProgram(this);
@ -828,14 +886,24 @@ void ScreenPanelGL::paintGL()
int frontbuf = GPU::FrontBuffer;
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, screenTexture);
if (GPU::Framebuffer[frontbuf][0] && GPU::Framebuffer[frontbuf][1])
if (true)
{
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 256, 192, GL_RGBA,
GL_UNSIGNED_BYTE, GPU::Framebuffer[frontbuf][0]);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 192, 256, 192, GL_RGBA,
GL_UNSIGNED_BYTE, GPU::Framebuffer[frontbuf][1]);
// hardware-accelerated render
GPU::GLCompositor::BindOutputTexture();
}
else
{
// regular render
glBindTexture(GL_TEXTURE_2D, screenTexture);
if (GPU::Framebuffer[frontbuf][0] && GPU::Framebuffer[frontbuf][1])
{
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 256, 192, GL_RGBA,
GL_UNSIGNED_BYTE, GPU::Framebuffer[frontbuf][0]);
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 192, 256, 192, GL_RGBA,
GL_UNSIGNED_BYTE, GPU::Framebuffer[frontbuf][1]);
}
}
GLint filter = Config::ScreenFilter ? GL_LINEAR : GL_NEAREST;
@ -1141,6 +1209,13 @@ MainWindow::~MainWindow()
{
}
QOpenGLContext* MainWindow::getOGLContext()
{
// TODO: check whether we can actually pull this!
QOpenGLWidget* glpanel = (QOpenGLWidget*)panel;
return glpanel->context();
}
void MainWindow::resizeEvent(QResizeEvent* event)
{
int w = event->size().width();
@ -1755,6 +1830,13 @@ int main(int argc, char** argv)
}
#endif
QSurfaceFormat format;
format.setDepthBufferSize(24);
format.setStencilBufferSize(8);
format.setVersion(3, 2);
format.setProfile(QSurfaceFormat::CoreProfile);
QSurfaceFormat::setDefaultFormat(format);
audioSync = SDL_CreateCond();
audioSyncLock = SDL_CreateMutex();

11
src/frontend/qt_sdl/main.h
View File

@ -25,6 +25,7 @@
#include <QImage>
#include <QActionGroup>
#include <QOffscreenSurface>
#include <QOpenGLWidget>
#include <QOpenGLContext>
#include <QOpenGLFunctions>
@ -40,6 +41,11 @@ class EmuThread : public QThread
public:
explicit EmuThread(QObject* parent = nullptr);
void initOpenGL();
void deinitOpenGL();
void* oglGetProcAddress(const char* proc);
void changeWindowTitle(char* title);
// to be called from the UI thread
@ -67,6 +73,9 @@ private:
volatile int EmuStatus;
int PrevEmuStatus;
int EmuRunning;
QOffscreenSurface* oglSurface;
QOpenGLContext* oglContext;
};
@ -161,6 +170,8 @@ public:
explicit MainWindow(QWidget* parent = nullptr);
~MainWindow();
QOpenGLContext* getOGLContext();
protected:
void resizeEvent(QResizeEvent* event) override;