Compute shader renderer (#2041)

* nothing works yet

* don't double buffer 3D framebuffers for the GL Renderer
looks like leftovers from when 3D+2D composition was done in the frontend

* oops

* it works!

* implement display capture for compute renderer
it's actually just all stolen from the regular OpenGL renderer

* fix bad indirect call

* handle cleanup properly

* add hires rendering to the compute shader renderer

* fix UB
also misc changes to use more unsigned multiplication
also fix framebuffer resize

* correct edge filling behaviour when AA is disabled

* fix full color textures

* fix edge marking (polygon id is 6-bit not 5)
also make the code a bit nicer

* take all edge cases into account for XMin/XMax calculation

* use hires coordinate again

* stop using fixed size buffers based on scale factor in shaders
this makes shader compile times tolerable on Wintel
- beginning of the shader cache
- increase size of tile idx in workdesc to 20 bits

* apparently & is not defined on bvec4
why does this even compile on Intel and Nvidia?

* put the texture cache into it's own file

* add compute shader renderer properly to the GUI
also add option to toggle using high resolution vertex coordinates

* unbind sampler object in compute shader renderer

* fix GetRangedBitMask for 64 bit aligned 64 bits
pretty embarassing

* convert NonStupidBitfield.h back to LF only new lines

* actually adapt to latest changes

* fix stupid merge

* actually make compute shader renderer work with newest changes

* show progress on shader compilation

* remove merge leftover

src/CMakeLists.txt

@@ -35,6 +35,8 @@ add_library(core STATIC
     GPU2D_Soft.cpp
     GPU3D.cpp
     GPU3D_Soft.cpp
+    GPU3D_Texcache.cpp
+    GPU3D_Texcache.h
     melonDLDI.h
     NDS.cpp
     NDSCart.cpp
@@ -78,6 +80,9 @@ if (ENABLE_OGLRENDERER)
         GPU_OpenGL.cpp
         GPU_OpenGL_shaders.h
         GPU3D_OpenGL.cpp
+        GPU3D_Compute.cpp
+        GPU3D_TexcacheOpenGL.cpp
+        GPU3D_TexcacheOpenGL.h
         GPU3D_OpenGL_shaders.h
         OpenGLSupport.cpp)
 

src/DMA.cpp

@@ -21,6 +21,7 @@
 #include "DSi.h"
 #include "DMA.h"
 #include "GPU.h"
+#include "GPU3D.h"
 #include "DMA_Timings.h"
 #include "Platform.h"
 

src/DSi_NDMA.cpp

@@ -22,6 +22,7 @@
 #include "DSi_NDMA.h"
 #include "GPU.h"
 #include "DSi_AES.h"
+#include "GPU3D.h"
 
 namespace melonDS
 {

src/GPU.cpp

@@ -23,7 +23,7 @@
 #include "ARMJIT.h"
 
 #include "GPU2D_Soft.h"
-#include "GPU3D_Soft.h"
+#include "GPU3D.h"
 
 namespace melonDS
 {

src/GPU2D.cpp

@@ -20,6 +20,7 @@
 #include <string.h>
 #include "NDS.h"
 #include "GPU.h"
+#include "GPU3D.h"
 
 namespace melonDS
 {

src/GPU2D_Soft.cpp

@@ -18,7 +18,7 @@
 
 #include "GPU2D_Soft.h"
 #include "GPU.h"
-#include "GPU3D_OpenGL.h"
+#include "GPU3D.h"
 
 namespace melonDS
 {

src/GPU3D.cpp

@@ -24,6 +24,7 @@
 #include "FIFO.h"
 #include "GPU3D_Soft.h"
 #include "Platform.h"
+#include "GPU3D.h"
 
 namespace melonDS
 {

src/GPU3D.h

@@ -349,7 +349,14 @@ public:
     virtual void RestartFrame(GPU& gpu) {};
     virtual u32* GetLine(int line) = 0;
     virtual void Blit(const GPU& gpu) {};
+
+    virtual void SetupAccelFrame() {}
     virtual void PrepareCaptureFrame() {}
+    virtual void BindOutputTexture(int buffer) {}
+
+    virtual bool NeedsShaderCompile() { return false; }
+    virtual void ShaderCompileStep(int& current, int& count) {}
+
 protected:
     Renderer3D(bool Accelerated);
 };

src/GPU3D_Compute.h

@@ -0,0 +1,242 @@
+/*
+    Copyright 2016-2022 melonDS team
+
+    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 GPU3D_COMPUTE
+#define GPU3D_COMPUTE
+
+#include <memory>
+
+#include "types.h"
+
+#include "GPU3D.h"
+
+#include "OpenGLSupport.h"
+#include "GPU_OpenGL.h"
+
+#include "GPU3D_TexcacheOpenGL.h"
+
+#include "NonStupidBitfield.h"
+
+namespace melonDS
+{
+
+class ComputeRenderer : public Renderer3D
+{
+public:
+    static std::unique_ptr<ComputeRenderer> New();
+    ~ComputeRenderer() override;
+
+    void Reset(GPU& gpu) override;
+
+    void SetRenderSettings(int scale, bool highResolutionCoordinates);
+
+    void VCount144(GPU& gpu) override;
+
+    void RenderFrame(GPU& gpu) override;
+    void RestartFrame(GPU& gpu) override;
+    u32* GetLine(int line) override;
+
+    void SetupAccelFrame() override;
+    void PrepareCaptureFrame() override;
+
+    void BindOutputTexture(int buffer) override;
+
+    void Blit(const GPU& gpu) override;
+    void Stop(const GPU& gpu) override;
+
+    bool NeedsShaderCompile() { return ShaderStepIdx != 33; }
+    void ShaderCompileStep(int& current, int& count) override;
+private:
+    ComputeRenderer(GLCompositor&& compositor);
+
+    GLuint ShaderInterpXSpans[2];
+    GLuint ShaderBinCombined;
+    GLuint ShaderDepthBlend[2];
+    GLuint ShaderRasteriseNoTexture[2];
+    GLuint ShaderRasteriseNoTextureToon[2];
+    GLuint ShaderRasteriseNoTextureHighlight[2];
+    GLuint ShaderRasteriseUseTextureDecal[2];
+    GLuint ShaderRasteriseUseTextureModulate[2];
+    GLuint ShaderRasteriseUseTextureToon[2];
+    GLuint ShaderRasteriseUseTextureHighlight[2];
+    GLuint ShaderRasteriseShadowMask[2];
+    GLuint ShaderClearCoarseBinMask;
+    GLuint ShaderClearIndirectWorkCount;
+    GLuint ShaderCalculateWorkListOffset;
+    GLuint ShaderSortWork;
+    GLuint ShaderFinalPass[8];
+
+    GLuint YSpanIndicesTextureMemory;
+    GLuint YSpanIndicesTexture;
+    GLuint YSpanSetupMemory;
+    GLuint XSpanSetupMemory;
+    GLuint BinResultMemory;
+    GLuint RenderPolygonMemory;
+    GLuint WorkDescMemory;
+
+    enum
+    {
+        tilememoryLayer_Color,
+        tilememoryLayer_Depth,
+        tilememoryLayer_Attr,
+        tilememoryLayer_Num,
+    };
+
+    GLuint TileMemory[tilememoryLayer_Num];
+    GLuint FinalTileMemory;
+
+    u32 DummyLine[256] = {};
+
+    struct SpanSetupY
+    {
+        // Attributes
+        s32 Z0, Z1, W0, W1;
+        s32 ColorR0, ColorG0, ColorB0;
+        s32 ColorR1, ColorG1, ColorB1;
+        s32 TexcoordU0, TexcoordV0;
+        s32 TexcoordU1, TexcoordV1;
+
+        // Interpolator
+        s32 I0, I1;
+        s32 Linear;
+        s32 IRecip;
+        s32 W0n, W0d, W1d;
+
+        // Slope
+        s32 Increment;
+
+        s32 X0, X1, Y0, Y1;
+        s32 XMin, XMax;
+        s32 DxInitial;
+
+        s32 XCovIncr;
+        u32 IsDummy;
+    };
+    struct SpanSetupX
+    {
+        s32 X0, X1;
+
+        s32 EdgeLenL, EdgeLenR, EdgeCovL, EdgeCovR;
+
+        s32 XRecip;
+
+        u32 Flags;
+
+        s32 Z0, Z1, W0, W1;
+        s32 ColorR0, ColorG0, ColorB0;
+        s32 ColorR1, ColorG1, ColorB1;
+        s32 TexcoordU0, TexcoordV0;
+        s32 TexcoordU1, TexcoordV1;
+
+        s32 CovLInitial, CovRInitial;
+    };
+    struct SetupIndices
+    {
+        u16 PolyIdx, SpanIdxL, SpanIdxR, Y;
+    };
+    struct RenderPolygon
+    {
+        u32 FirstXSpan;
+        s32 YTop, YBot;
+
+        s32 XMin, XMax;
+        s32 XMinY, XMaxY;
+
+        u32 Variant;
+        u32 Attr;
+
+        float TextureLayer;
+    };
+
+    static constexpr int TileSize = 8;
+    static constexpr int CoarseTileCountX = 8;
+    static constexpr int CoarseTileCountY = 4;
+    static constexpr int CoarseTileW = CoarseTileCountX * TileSize;
+    static constexpr int CoarseTileH = CoarseTileCountY * TileSize;
+
+    static constexpr int BinStride = 2048/32;
+    static constexpr int CoarseBinStride = BinStride/32;
+
+    static constexpr int MaxVariants = 256;
+
+    static constexpr int UniformIdxCurVariant = 0;
+    static constexpr int UniformIdxTextureSize = 1;
+
+    static constexpr int MaxFullscreenLayers = 16;
+
+    struct BinResultHeader
+    {
+        u32 VariantWorkCount[MaxVariants*4];
+        u32 SortedWorkOffset[MaxVariants];
+
+        u32 SortWorkWorkCount[4];
+    };
+
+    static const int MaxYSpanSetups = 6144*2;
+    std::vector<SetupIndices> YSpanIndices;
+    SpanSetupY YSpanSetups[MaxYSpanSetups];
+    RenderPolygon RenderPolygons[2048];
+
+    TexcacheOpenGL Texcache;
+
+    struct MetaUniform
+    {
+        u32 NumPolygons;
+        u32 NumVariants;
+
+        u32 AlphaRef;
+        u32 DispCnt;
+
+        u32 ToonTable[4*34];
+
+        u32 ClearColor, ClearDepth, ClearAttr;
+
+        u32 FogOffset, FogShift, FogColor;
+    };
+    GLuint MetaUniformMemory;
+
+    GLuint Samplers[9];
+
+    GLuint Framebuffer = 0;
+    GLuint LowResFramebuffer;
+    GLuint PixelBuffer;
+
+    u32 FramebufferCPU[256*192];
+
+    int ScreenWidth, ScreenHeight;
+    int TilesPerLine, TileLines;
+    int ScaleFactor = -1;
+    int MaxWorkTiles;
+    bool HiresCoordinates;
+
+    GLCompositor CurGLCompositor;
+
+    int ShaderStepIdx = 0;
+
+    void DeleteShaders();
+
+    void SetupAttrs(SpanSetupY* span, Polygon* poly, int from, int to);
+    void SetupYSpan(RenderPolygon* rp, SpanSetupY* span, Polygon* poly, int from, int to, int side, s32 positions[10][2]);
+    void SetupYSpanDummy(RenderPolygon* rp, SpanSetupY* span, Polygon* poly, int vertex, int side, s32 positions[10][2]);
+
+    bool CompileShader(GLuint& shader, const std::string& source, const std::initializer_list<const char*>& defines);
+};
+
+}
+
+#endif

src/GPU3D_OpenGL.cpp

@@ -28,46 +28,32 @@
 namespace melonDS
 {
 
-bool GLRenderer::BuildRenderShader(u32 flags, const char* vs, const char* fs)
+bool GLRenderer::BuildRenderShader(u32 flags, const std::string& vs, const std::string& fs)
 {
     char shadername[32];
     snprintf(shadername, sizeof(shadername), "RenderShader%02X", flags);
 
     int headerlen = strlen(kShaderHeader);
 
-    int vslen = strlen(vs);
+    std::string vsbuf;
-    int vsclen = strlen(kRenderVSCommon);
+    vsbuf += kShaderHeader;
-    char* vsbuf = new char[headerlen + vsclen + vslen + 1];
+    vsbuf += kRenderVSCommon;
-    strcpy(&vsbuf[0], kShaderHeader);
+    vsbuf += vs;
-    strcpy(&vsbuf[headerlen], kRenderVSCommon);
-    strcpy(&vsbuf[headerlen + vsclen], vs);
 
-    int fslen = strlen(fs);
+    std::string fsbuf;
-    int fsclen = strlen(kRenderFSCommon);
+    fsbuf += kShaderHeader;
-    char* fsbuf = new char[headerlen + fsclen + fslen + 1];
+    fsbuf += kRenderFSCommon;
-    strcpy(&fsbuf[0], kShaderHeader);
+    fsbuf += fs;
-    strcpy(&fsbuf[headerlen], kRenderFSCommon);
-    strcpy(&fsbuf[headerlen + fsclen], fs);
 
-    bool ret = OpenGL::BuildShaderProgram(vsbuf, fsbuf, RenderShader[flags], shadername);
+    GLuint prog;
-
+    bool ret = OpenGL::CompileVertexFragmentProgram(prog,
-    delete[] vsbuf;
+        vsbuf, fsbuf,
-    delete[] fsbuf;
+        shadername,
+        {{"vPosition", 0}, {"vColor", 1}, {"vTexcoord", 2}, {"vPolygonAttr", 3}},
+        {{"oColor", 0}, {"oAttr", 1}});
 
     if (!ret) return false;
 
-    GLuint prog = RenderShader[flags][2];
-
-    glBindAttribLocation(prog, 0, "vPosition");
-    glBindAttribLocation(prog, 1, "vColor");
-    glBindAttribLocation(prog, 2, "vTexcoord");
-    glBindAttribLocation(prog, 3, "vPolygonAttr");
-    glBindFragDataLocation(prog, 0, "oColor");
-    glBindFragDataLocation(prog, 1, "oAttr");
-
-    if (!OpenGL::LinkShaderProgram(RenderShader[flags]))
-        return false;
-
     GLint uni_id = glGetUniformBlockIndex(prog, "uConfig");
     glUniformBlockBinding(prog, uni_id, 0);
 
@@ -78,13 +64,15 @@ bool GLRenderer::BuildRenderShader(u32 flags, const char* vs, const char* fs)
     uni_id = glGetUniformLocation(prog, "TexPalMem");
     glUniform1i(uni_id, 1);
 
+    RenderShader[flags] = prog;
+
     return true;
 }
 
 void GLRenderer::UseRenderShader(u32 flags)
 {
     if (CurShaderID == flags) return;
-    glUseProgram(RenderShader[flags][2]);
+    glUseProgram(RenderShader[flags]);
     CurShaderID = flags;
 }
 
@@ -125,21 +113,17 @@ std::unique_ptr<GLRenderer> GLRenderer::New() noexcept
     glDepthRange(0, 1);
     glClearDepth(1.0);
 
-
+    if (!OpenGL::CompileVertexFragmentProgram(result->ClearShaderPlain,
-    if (!OpenGL::BuildShaderProgram(kClearVS, kClearFS, result->ClearShaderPlain, "ClearShader"))
+            kClearVS, kClearFS,
+            "ClearShader",
+            {{"vPosition", 0}},
+            {{"oColor", 0}, {"oAttr", 1}}))
         return nullptr;
 
-    glBindAttribLocation(result->ClearShaderPlain[2], 0, "vPosition");
+    result->ClearUniformLoc[0] = glGetUniformLocation(result->ClearShaderPlain, "uColor");
-    glBindFragDataLocation(result->ClearShaderPlain[2], 0, "oColor");
+    result->ClearUniformLoc[1] = glGetUniformLocation(result->ClearShaderPlain, "uDepth");
-    glBindFragDataLocation(result->ClearShaderPlain[2], 1, "oAttr");
+    result->ClearUniformLoc[2] = glGetUniformLocation(result->ClearShaderPlain, "uOpaquePolyID");
-
+    result->ClearUniformLoc[3] = glGetUniformLocation(result->ClearShaderPlain, "uFogFlag");
-    if (!OpenGL::LinkShaderProgram(result->ClearShaderPlain))
-        return nullptr;
-
-    result->ClearUniformLoc[0] = glGetUniformLocation(result->ClearShaderPlain[2], "uColor");
-    result->ClearUniformLoc[1] = glGetUniformLocation(result->ClearShaderPlain[2], "uDepth");
-    result->ClearUniformLoc[2] = glGetUniformLocation(result->ClearShaderPlain[2], "uOpaquePolyID");
-    result->ClearUniformLoc[3] = glGetUniformLocation(result->ClearShaderPlain[2], "uFogFlag");
 
     memset(result->RenderShader, 0, sizeof(RenderShader));
 
@@ -167,42 +151,35 @@ std::unique_ptr<GLRenderer> GLRenderer::New() noexcept
     if (!result->BuildRenderShader(RenderFlag_ShadowMask | RenderFlag_WBuffer, kRenderVS_W, kRenderFS_WSM))
         return nullptr;
 
-    if (!OpenGL::BuildShaderProgram(kFinalPassVS, kFinalPassEdgeFS, result->FinalPassEdgeShader, "FinalPassEdgeShader"))
+    if (!OpenGL::CompileVertexFragmentProgram(result->FinalPassEdgeShader,
+            kFinalPassVS, kFinalPassEdgeFS,
+            "FinalPassEdgeShader",
+            {{"vPosition", 0}},
+            {{"oColor", 0}}))
+        return nullptr;
+    if (!OpenGL::CompileVertexFragmentProgram(result->FinalPassFogShader,
+            kFinalPassVS, kFinalPassFogFS,
+            "FinalPassFogShader",
+            {{"vPosition", 0}},
+            {{"oColor", 0}}))
         return nullptr;
 
-    if (!OpenGL::BuildShaderProgram(kFinalPassVS, kFinalPassFogFS, result->FinalPassFogShader, "FinalPassFogShader"))
+    GLuint uni_id = glGetUniformBlockIndex(result->FinalPassEdgeShader, "uConfig");
-        return nullptr;
+    glUniformBlockBinding(result->FinalPassEdgeShader, uni_id, 0);
 
-    glBindAttribLocation(result->FinalPassEdgeShader[2], 0, "vPosition");
+    glUseProgram(result->FinalPassEdgeShader);
-    glBindFragDataLocation(result->FinalPassEdgeShader[2], 0, "oColor");
+    uni_id = glGetUniformLocation(result->FinalPassEdgeShader, "DepthBuffer");
-
-    if (!OpenGL::LinkShaderProgram(result->FinalPassEdgeShader))
-        return nullptr;
-
-    GLint uni_id = glGetUniformBlockIndex(result->FinalPassEdgeShader[2], "uConfig");
-    glUniformBlockBinding(result->FinalPassEdgeShader[2], uni_id, 0);
-
-    glUseProgram(result->FinalPassEdgeShader[2]);
-
-    uni_id = glGetUniformLocation(result->FinalPassEdgeShader[2], "DepthBuffer");
     glUniform1i(uni_id, 0);
-    uni_id = glGetUniformLocation(result->FinalPassEdgeShader[2], "AttrBuffer");
+    uni_id = glGetUniformLocation(result->FinalPassEdgeShader, "AttrBuffer");
     glUniform1i(uni_id, 1);
 
-    glBindAttribLocation(result->FinalPassFogShader[2], 0, "vPosition");
+    uni_id = glGetUniformBlockIndex(result->FinalPassFogShader, "uConfig");
-    glBindFragDataLocation(result->FinalPassFogShader[2], 0, "oColor");
+    glUniformBlockBinding(result->FinalPassFogShader, uni_id, 0);
 
-    if (!OpenGL::LinkShaderProgram(result->FinalPassFogShader))
+    glUseProgram(result->FinalPassFogShader);
-        return nullptr;
+    uni_id = glGetUniformLocation(result->FinalPassFogShader, "DepthBuffer");
-
-    uni_id = glGetUniformBlockIndex(result->FinalPassFogShader[2], "uConfig");
-    glUniformBlockBinding(result->FinalPassFogShader[2], uni_id, 0);
-
-    glUseProgram(result->FinalPassFogShader[2]);
-
-    uni_id = glGetUniformLocation(result->FinalPassFogShader[2], "DepthBuffer");
     glUniform1i(uni_id, 0);
-    uni_id = glGetUniformLocation(result->FinalPassFogShader[2], "AttrBuffer");
+    uni_id = glGetUniformLocation(result->FinalPassFogShader, "AttrBuffer");
     glUniform1i(uni_id, 1);
 
 
@@ -255,29 +232,26 @@ std::unique_ptr<GLRenderer> GLRenderer::New() noexcept
     glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, result->IndexBufferID);
     glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(IndexBuffer), nullptr, GL_DYNAMIC_DRAW);
 
-    glGenFramebuffers(4, &result->FramebufferID[0]);
+    glGenFramebuffers(1, &result->MainFramebuffer);
-    glBindFramebuffer(GL_FRAMEBUFFER, result->FramebufferID[0]);
-
-    glGenTextures(8, &result->FramebufferTex[0]);
-    result->FrontBuffer = 0;
 
     // color buffers
-    SetupDefaultTexParams(result->FramebufferTex[0]);
+    glGenTextures(1, &result->ColorBufferTex);
-    SetupDefaultTexParams(result->FramebufferTex[1]);
+    SetupDefaultTexParams(result->ColorBufferTex);
 
     // depth/stencil buffer
-    SetupDefaultTexParams(result->FramebufferTex[4]);
+    glGenTextures(1, &result->DepthBufferTex);
-    SetupDefaultTexParams(result->FramebufferTex[6]);
+    SetupDefaultTexParams(result->DepthBufferTex);
 
     // attribute buffer
     // R: opaque polyID (for edgemarking)
     // G: edge flag
     // B: fog flag
-    SetupDefaultTexParams(result->FramebufferTex[5]);
+    glGenTextures(1, &result->AttrBufferTex);
-    SetupDefaultTexParams(result->FramebufferTex[7]);
+    SetupDefaultTexParams(result->AttrBufferTex);
 
     // downscale framebuffer for display capture (always 256x192)
-    SetupDefaultTexParams(result->FramebufferTex[3]);
+    glGenTextures(1, &result->DownScaleBufferTex);
+    SetupDefaultTexParams(result->DownScaleBufferTex);
     glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, 256, 192, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
 
     glEnable(GL_BLEND);
@@ -315,8 +289,12 @@ GLRenderer::~GLRenderer()
     glDeleteTextures(1, &TexMemID);
     glDeleteTextures(1, &TexPalMemID);
 
-    glDeleteFramebuffers(4, &FramebufferID[0]);
+    glDeleteFramebuffers(1, &MainFramebuffer);
-    glDeleteTextures(8, &FramebufferTex[0]);
+    glDeleteFramebuffers(1, &DownscaleFramebuffer);
+    glDeleteTextures(1, &ColorBufferTex);
+    glDeleteTextures(1, &DepthBufferTex);
+    glDeleteTextures(1, &AttrBufferTex);
+    glDeleteTextures(1, &DownScaleBufferTex);
 
     glDeleteVertexArrays(1, &VertexArrayID);
     glDeleteBuffers(1, &VertexBufferID);
@@ -327,8 +305,8 @@ GLRenderer::~GLRenderer()
 
     for (int i = 0; i < 16; i++)
     {
-        if (!RenderShader[i][2]) continue;
+        if (!RenderShader[i]) continue;
-        OpenGL::DeleteShaderProgram(RenderShader[i]);
+        glDeleteProgram(RenderShader[i]);
     }
 }
 
@@ -361,40 +339,25 @@ void GLRenderer::SetRenderSettings(bool betterpolygons, int scale) noexcept
     ScreenW = 256 * scale;
     ScreenH = 192 * scale;
 
-    glBindTexture(GL_TEXTURE_2D, FramebufferTex[0]);
+    glBindTexture(GL_TEXTURE_2D, ColorBufferTex);
-    glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, ScreenW, ScreenH, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
-    glBindTexture(GL_TEXTURE_2D, FramebufferTex[1]);
     glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, ScreenW, ScreenH, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);
 
-    glBindTexture(GL_TEXTURE_2D, FramebufferTex[4]);
+    glBindTexture(GL_TEXTURE_2D, DepthBufferTex);
     glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH24_STENCIL8, ScreenW, ScreenH, 0, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8, NULL);
-    glBindTexture(GL_TEXTURE_2D, FramebufferTex[5]);
+    glBindTexture(GL_TEXTURE_2D, AttrBufferTex);
     glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB, ScreenW, ScreenH, 0, GL_RGB, GL_UNSIGNED_BYTE, NULL);
 
-    glBindTexture(GL_TEXTURE_2D, FramebufferTex[6]);
+    glBindFramebuffer(GL_FRAMEBUFFER, DownscaleFramebuffer);
-    glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH24_STENCIL8, ScreenW, ScreenH, 0, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8, NULL);
+    glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, DownScaleBufferTex, 0);
-    glBindTexture(GL_TEXTURE_2D, FramebufferTex[7]);
-    glTexImage2D(GL_TEXTURE_2D, 0, GL_RGB, ScreenW, ScreenH, 0, GL_RGB, GL_UNSIGNED_BYTE, NULL);
-
-    glBindFramebuffer(GL_FRAMEBUFFER, FramebufferID[3]);
-    glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, FramebufferTex[3], 0);
 
     GLenum fbassign[2] = {GL_COLOR_ATTACHMENT0, GL_COLOR_ATTACHMENT1};
 
-    glBindFramebuffer(GL_FRAMEBUFFER, FramebufferID[0]);
+    glBindFramebuffer(GL_FRAMEBUFFER, MainFramebuffer);
-    glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, FramebufferTex[0], 0);
+    glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, ColorBufferTex, 0);
-    glFramebufferTexture(GL_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, FramebufferTex[4], 0);
+    glFramebufferTexture(GL_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, DepthBufferTex, 0);
-    glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT1, FramebufferTex[5], 0);
+    glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT1, AttrBufferTex, 0);
     glDrawBuffers(2, fbassign);
 
-    glBindFramebuffer(GL_FRAMEBUFFER, FramebufferID[1]);
-    glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, FramebufferTex[1], 0);
-    glFramebufferTexture(GL_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, FramebufferTex[6], 0);
-    glFramebufferTexture(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT1, FramebufferTex[7], 0);
-    glDrawBuffers(2, fbassign);
-
-    glBindFramebuffer(GL_FRAMEBUFFER, FramebufferID[0]);
-
     glBindBuffer(GL_PIXEL_PACK_BUFFER, PixelbufferID);
     glBufferData(GL_PIXEL_PACK_BUFFER, 256*192*4, NULL, GL_DYNAMIC_READ);
 
@@ -1103,9 +1066,9 @@ void GLRenderer::RenderSceneChunk(const GPU3D& gpu3d, int y, int h)
         glStencilMask(0);
 
         glActiveTexture(GL_TEXTURE0);
-        glBindTexture(GL_TEXTURE_2D, FramebufferTex[FrontBuffer ? 6 : 4]);
+        glBindTexture(GL_TEXTURE_2D, DepthBufferTex);
         glActiveTexture(GL_TEXTURE1);
-        glBindTexture(GL_TEXTURE_2D, FramebufferTex[FrontBuffer ? 7 : 5]);
+        glBindTexture(GL_TEXTURE_2D, AttrBufferTex);
 
         glBindBuffer(GL_ARRAY_BUFFER, ClearVertexBufferID);
         glBindVertexArray(ClearVertexArrayID);
@@ -1115,7 +1078,7 @@ void GLRenderer::RenderSceneChunk(const GPU3D& gpu3d, int y, int h)
             // edge marking
             // TODO: depth/polyid values at screen edges
 
-            glUseProgram(FinalPassEdgeShader[2]);
+            glUseProgram(FinalPassEdgeShader);
 
             glBlendFuncSeparate(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA, GL_ZERO, GL_ONE);
 
@@ -1126,7 +1089,7 @@ void GLRenderer::RenderSceneChunk(const GPU3D& gpu3d, int y, int h)
         {
             // fog
 
-            glUseProgram(FinalPassFogShader[2]);
+            glUseProgram(FinalPassFogShader);
 
             if (gpu3d.RenderDispCnt & (1<<6))
                 glBlendFuncSeparate(GL_ZERO, GL_ONE, GL_CONSTANT_COLOR, GL_ONE_MINUS_SRC_ALPHA);
@@ -1154,7 +1117,7 @@ void GLRenderer::RenderFrame(GPU& gpu)
     CurShaderID = -1;
 
     glBindFramebuffer(GL_READ_FRAMEBUFFER, 0);
-    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, FramebufferID[FrontBuffer]);
+    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, MainFramebuffer);
 
     ShaderConfig.uScreenSize[0] = ScreenW;
     ShaderConfig.uScreenSize[1] = ScreenH;
@@ -1260,7 +1223,7 @@ void GLRenderer::RenderFrame(GPU& gpu)
     // TODO: check whether 'clear polygon ID' affects translucent polyID
     // (for example when alpha is 1..30)
     {
-        glUseProgram(ClearShaderPlain[2]);
+        glUseProgram(ClearShaderPlain);
         glDepthFunc(GL_ALWAYS);
 
         u32 r = gpu.GPU3D.RenderClearAttr1 & 0x1F;
@@ -1320,8 +1283,6 @@ void GLRenderer::RenderFrame(GPU& gpu)
 
         RenderSceneChunk(gpu.GPU3D, 0, 192);
     }
-
-    FrontBuffer = FrontBuffer ? 0 : 1;
 }
 
 void GLRenderer::Stop(const GPU& gpu)
@@ -1331,16 +1292,14 @@ void GLRenderer::Stop(const GPU& gpu)
 
 void GLRenderer::PrepareCaptureFrame()
 {
-    // TODO: make sure this picks the right buffer when doing antialiasing
+    glBindFramebuffer(GL_READ_FRAMEBUFFER, MainFramebuffer);
-    int original_fb = FrontBuffer^1;
-
-    glBindFramebuffer(GL_READ_FRAMEBUFFER, FramebufferID[original_fb]);
     glReadBuffer(GL_COLOR_ATTACHMENT0);
-    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, FramebufferID[3]);
+    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, DownscaleFramebuffer);
     glDrawBuffer(GL_COLOR_ATTACHMENT0);
     glBlitFramebuffer(0, 0, ScreenW, ScreenH, 0, 0, 256, 192, GL_COLOR_BUFFER_BIT, GL_NEAREST);
 
-    glBindFramebuffer(GL_READ_FRAMEBUFFER, FramebufferID[3]);
+    glBindBuffer(GL_PIXEL_PACK_BUFFER, PixelbufferID);
+    glBindFramebuffer(GL_READ_FRAMEBUFFER, DownscaleFramebuffer);
     glReadPixels(0, 0, 256, 192, GL_BGRA, GL_UNSIGNED_BYTE, NULL);
 }
 
@@ -1349,12 +1308,18 @@ void GLRenderer::Blit(const GPU& gpu)
     CurGLCompositor.RenderFrame(gpu, *this);
 }
 
+void GLRenderer::BindOutputTexture(int buffer)
+{
+    CurGLCompositor.BindOutputTexture(buffer);
+}
+
 u32* GLRenderer::GetLine(int line)
 {
     int stride = 256;
 
     if (line == 0)
     {
+        glBindBuffer(GL_PIXEL_PACK_BUFFER, PixelbufferID);
         u8* data = (u8*)glMapBuffer(GL_PIXEL_PACK_BUFFER, GL_READ_ONLY);
         if (data) memcpy(&Framebuffer[stride*0], data, 4*stride*192);
         glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
@@ -1374,7 +1339,7 @@ u32* GLRenderer::GetLine(int line)
 
 void GLRenderer::SetupAccelFrame()
 {
-    glBindTexture(GL_TEXTURE_2D, FramebufferTex[FrontBuffer]);
+    glBindTexture(GL_TEXTURE_2D, ColorBufferTex);
 }
 
 }

src/GPU3D_OpenGL.h

@@ -44,12 +44,11 @@ public:
     void Stop(const GPU& gpu) override;
     u32* GetLine(int line) override;
 
-    void SetupAccelFrame();
+    void SetupAccelFrame() override;
     void PrepareCaptureFrame() override;
     void Blit(const GPU& gpu) override;
 
-    [[nodiscard]] const GLCompositor& GetCompositor() const noexcept { return CurGLCompositor; }
+    void BindOutputTexture(int buffer) override;
-    GLCompositor& GetCompositor() noexcept { return CurGLCompositor; }
 
     static std::unique_ptr<GLRenderer> New() noexcept;
 private:
@@ -77,7 +76,7 @@ private:
     GLCompositor CurGLCompositor;
     RendererPolygon PolygonList[2048] {};
 
-    bool BuildRenderShader(u32 flags, const char* vs, const char* fs);
+    bool BuildRenderShader(u32 flags, const std::string& vs, const std::string& fs);
     void UseRenderShader(u32 flags);
     void SetupPolygon(RendererPolygon* rp, Polygon* polygon) const;
     u32* SetupVertex(const Polygon* poly, int vid, const Vertex* vtx, u32 vtxattr, u32* vptr) const;
@@ -96,13 +95,13 @@ private:
     };
 
 
-    GLuint ClearShaderPlain[3] {};
+    GLuint ClearShaderPlain {};
 
-    GLuint RenderShader[16][3] {};
+    GLuint RenderShader[16] {};
     GLuint CurShaderID = -1;
 
-    GLuint FinalPassEdgeShader[3] {};
+    GLuint FinalPassEdgeShader {};
-    GLuint FinalPassFogShader[3] {};
+    GLuint FinalPassFogShader {};
 
     // std140 compliant structure
     struct
@@ -155,12 +154,12 @@ private:
     bool BetterPolygons {};
     int ScreenW {}, ScreenH {};
 
-    GLuint FramebufferTex[8] {};
+    GLuint ColorBufferTex {}, DepthBufferTex {}, AttrBufferTex {};
-    int FrontBuffer {};
+    GLuint DownScaleBufferTex {};
-    GLuint FramebufferID[4] {}, PixelbufferID {};
+    GLuint PixelbufferID {};
+
+    GLuint MainFramebuffer {}, DownscaleFramebuffer {};
     u32 Framebuffer[256*192] {};
-
-
 };
 }
 #endif

src/GPU3D_Soft.cpp

@@ -95,8 +95,8 @@ void SoftRenderer::EnableRenderThread()
     }
 }
 
-SoftRenderer::SoftRenderer(bool threaded) noexcept
+SoftRenderer::SoftRenderer() noexcept
-    : Renderer3D(false), Threaded(threaded)
+    : Renderer3D(false)
 {
     Sema_RenderStart = Platform::Semaphore_Create();
     Sema_RenderDone = Platform::Semaphore_Create();

src/GPU3D_Soft.h

@@ -29,7 +29,7 @@ namespace melonDS
 class SoftRenderer : public Renderer3D
 {
 public:
-    SoftRenderer(bool threaded = false) noexcept;
+    SoftRenderer() noexcept;
     ~SoftRenderer() override;
     void Reset(GPU& gpu) override;
 
@@ -504,7 +504,7 @@ private:
 
     // threading
 
-    bool Threaded;
+    bool Threaded = false;
     Platform::Thread* RenderThread;
     std::atomic_bool RenderThreadRunning;
     std::atomic_bool RenderThreadRendering;

src/GPU3D_Texcache.cpp

@@ -0,0 +1,269 @@
+#include "GPU3D_Texcache.h"
+
+namespace melonDS
+{
+
+inline u16 ColorAvg(u16 color0, u16 color1)
+{
+    u32 r0 = color0 & 0x001F;
+    u32 g0 = color0 & 0x03E0;
+    u32 b0 = color0 & 0x7C00;
+    u32 r1 = color1 & 0x001F;
+    u32 g1 = color1 & 0x03E0;
+    u32 b1 = color1 & 0x7C00;
+
+    u32 r = (r0 + r1) >> 1;
+    u32 g = ((g0 + g1) >> 1) & 0x03E0;
+    u32 b = ((b0 + b1) >> 1) & 0x7C00;
+
+    return r | g | b;
+}
+
+inline u16 Color5of3(u16 color0, u16 color1)
+{
+    u32 r0 = color0 & 0x001F;
+    u32 g0 = color0 & 0x03E0;
+    u32 b0 = color0 & 0x7C00;
+    u32 r1 = color1 & 0x001F;
+    u32 g1 = color1 & 0x03E0;
+    u32 b1 = color1 & 0x7C00;
+
+    u32 r = (r0*5 + r1*3) >> 3;
+    u32 g = ((g0*5 + g1*3) >> 3) & 0x03E0;
+    u32 b = ((b0*5 + b1*3) >> 3) & 0x7C00;
+
+    return r | g | b;
+}
+
+inline u16 Color3of5(u16 color0, u16 color1)
+{
+    u32 r0 = color0 & 0x001F;
+    u32 g0 = color0 & 0x03E0;
+    u32 b0 = color0 & 0x7C00;
+    u32 r1 = color1 & 0x001F;
+    u32 g1 = color1 & 0x03E0;
+    u32 b1 = color1 & 0x7C00;
+
+    u32 r = (r0*3 + r1*5) >> 3;
+    u32 g = ((g0*3 + g1*5) >> 3) & 0x03E0;
+    u32 b = ((b0*3 + b1*5) >> 3) & 0x7C00;
+
+    return r | g | b;
+}
+
+inline u32 ConvertRGB5ToRGB8(u16 val)
+{
+    return (((u32)val & 0x1F) << 3)
+        | (((u32)val & 0x3E0) << 6)
+        | (((u32)val & 0x7C00) << 9);
+}
+inline u32 ConvertRGB5ToBGR8(u16 val)
+{
+    return (((u32)val & 0x1F) << 9)
+        | (((u32)val & 0x3E0) << 6)
+        | (((u32)val & 0x7C00) << 3);
+}
+inline u32 ConvertRGB5ToRGB6(u16 val)
+{
+    u8 r = (val & 0x1F) << 1;
+    u8 g = (val & 0x3E0) >> 4;
+    u8 b = (val & 0x7C00) >> 9;
+    if (r) r++;
+    if (g) g++;
+    if (b) b++;
+    return (u32)r | ((u32)g << 8) | ((u32)b << 16);
+}
+
+template <int outputFmt>
+void ConvertBitmapTexture(u32 width, u32 height, u32* output, u8* texData)
+{
+    for (u32 i = 0; i < width*height; i++)
+    {
+        u16 value = *(u16*)&texData[i * 2];
+
+        switch (outputFmt)
+        {
+        case outputFmt_RGB6A5:
+            output[i] = ConvertRGB5ToRGB6(value) | (value & 0x8000 ? 0x1F000000 : 0);
+            break;
+        case outputFmt_RGBA8:
+            output[i] = ConvertRGB5ToRGB8(value) | (value & 0x8000 ? 0xFF000000 : 0);
+            break;
+        case outputFmt_BGRA8:
+            output[i] = ConvertRGB5ToBGR8(value) | (value & 0x8000 ? 0xFF000000 : 0);
+            break;
+        }
+    }
+}
+
+template void ConvertBitmapTexture<outputFmt_RGB6A5>(u32 width, u32 height, u32* output, u8* texData);
+
+template <int outputFmt>
+void ConvertCompressedTexture(u32 width, u32 height, u32* output, u8* texData, u8* texAuxData, u16* palData)
+{
+    // we process a whole block at the time
+    for (int y = 0; y < height / 4; y++)
+    {
+        for (int x = 0; x < width / 4; x++)
+        {
+            u32 data = ((u32*)texData)[x + y * (width / 4)];
+            u16 auxData = ((u16*)texAuxData)[x + y * (width / 4)];
+
+            u32 paletteOffset = auxData & 0x3FFF;
+            u16 color0 = palData[paletteOffset*2] | 0x8000;
+            u16 color1 = palData[paletteOffset*2+1] | 0x8000;
+            u16 color2, color3;
+
+            switch ((auxData >> 14) & 0x3)
+            {
+            case 0:
+                color2 = palData[paletteOffset*2+2] | 0x8000;
+                color3 = 0;
+                break;
+            case 1:
+                {
+                    u32 r0 = color0 & 0x001F;
+                    u32 g0 = color0 & 0x03E0;
+                    u32 b0 = color0 & 0x7C00;
+                    u32 r1 = color1 & 0x001F;
+                    u32 g1 = color1 & 0x03E0;
+                    u32 b1 = color1 & 0x7C00;
+
+                    u32 r = (r0 + r1) >> 1;
+                    u32 g = ((g0 + g1) >> 1) & 0x03E0;
+                    u32 b = ((b0 + b1) >> 1) & 0x7C00;
+                    color2 = r | g | b | 0x8000;
+                }
+                color3 = 0;
+                break;
+            case 2:
+                color2 = palData[paletteOffset*2+2] | 0x8000;
+                color3 = palData[paletteOffset*2+3] | 0x8000;
+                break;
+            case 3:
+                {
+                    u32 r0 = color0 & 0x001F;
+                    u32 g0 = color0 & 0x03E0;
+                    u32 b0 = color0 & 0x7C00;
+                    u32 r1 = color1 & 0x001F;
+                    u32 g1 = color1 & 0x03E0;
+                    u32 b1 = color1 & 0x7C00;
+
+                    u32 r = (r0*5 + r1*3) >> 3;
+                    u32 g = ((g0*5 + g1*3) >> 3) & 0x03E0;
+                    u32 b = ((b0*5 + b1*3) >> 3) & 0x7C00;
+
+                    color2 = r | g | b | 0x8000;
+                }
+                {
+                    u32 r0 = color0 & 0x001F;
+                    u32 g0 = color0 & 0x03E0;
+                    u32 b0 = color0 & 0x7C00;
+                    u32 r1 = color1 & 0x001F;
+                    u32 g1 = color1 & 0x03E0;
+                    u32 b1 = color1 & 0x7C00;
+
+                    u32 r = (r0*3 + r1*5) >> 3;
+                    u32 g = ((g0*3 + g1*5) >> 3) & 0x03E0;
+                    u32 b = ((b0*3 + b1*5) >> 3) & 0x7C00;
+
+                    color3 = r | g | b | 0x8000;
+                }
+                break;
+            }
+
+            // in 2020 our default data types are big enough to be used as lookup tables...
+            u64 packed = color0 | ((u64)color1 << 16) | ((u64)color2 << 32) | ((u64)color3 << 48);
+
+            for (int j = 0; j < 4; j++)
+            {
+                for (int i = 0; i < 4; i++)
+                {
+                    u16 color = (packed >> 16 * (data >> 2 * (i + j * 4))) & 0xFFFF;
+                    u32 res;
+                    switch (outputFmt)
+                    {
+                    case outputFmt_RGB6A5: res = ConvertRGB5ToRGB6(color)
+                        | ((color & 0x8000) ? 0x1F000000 : 0); break;
+                    case outputFmt_RGBA8: res = ConvertRGB5ToRGB8(color)
+                        | ((color & 0x8000) ? 0xFF000000 : 0); break;
+                    case outputFmt_BGRA8: res = ConvertRGB5ToBGR8(color)
+                        | ((color & 0x8000) ? 0xFF000000 : 0); break;
+                    }
+                    output[x * 4 + i + (y * 4 + j) * width] = res;
+                }
+            }
+        }
+    }
+}
+
+template void ConvertCompressedTexture<outputFmt_RGB6A5>(u32, u32, u32*, u8*, u8*, u16*);
+
+template <int outputFmt, int X, int Y>
+void ConvertAXIYTexture(u32 width, u32 height, u32* output, u8* texData, u16* palData)
+{
+    for (int y = 0; y < height; y++)
+    {
+        for (int x = 0; x < width; x++)
+        {
+            u8 val = texData[x + y * width];
+
+            u32 idx = val & ((1 << Y) - 1);
+
+            u16 color = palData[idx];
+            u32 alpha = (val >> Y) & ((1 << X) - 1);
+            if (X != 5)
+                alpha = alpha * 4 + alpha / 2;
+
+            u32 res;
+            switch (outputFmt)
+            {
+            case outputFmt_RGB6A5: res = ConvertRGB5ToRGB6(color) | alpha << 24; break;
+            // make sure full alpha == 255
+            case outputFmt_RGBA8: res = ConvertRGB5ToRGB8(color) | (alpha << 27 | (alpha & 0x1C) << 22); break;
+            case outputFmt_BGRA8: res = ConvertRGB5ToBGR8(color) | (alpha << 27 | (alpha & 0x1C) << 22); break;
+            }
+            output[x + y * width] = res;
+        }
+    }
+}
+
+template void ConvertAXIYTexture<outputFmt_RGB6A5, 5, 3>(u32, u32, u32*, u8*, u16*);
+template void ConvertAXIYTexture<outputFmt_RGB6A5, 3, 5>(u32, u32, u32*, u8*, u16*);
+
+template <int outputFmt, int colorBits>
+void ConvertNColorsTexture(u32 width, u32 height, u32* output, u8* texData, u16* palData, bool color0Transparent)
+{
+    for (int y = 0; y < height; y++)
+    {
+        for (int x = 0; x < width / (8 / colorBits); x++)
+        {
+            u8 val = texData[x + y * (width / (8 / colorBits))];
+
+            for (int i = 0; i < 8 / colorBits; i++)
+            {
+                u32 index = (val >> (i * colorBits)) & ((1 << colorBits) - 1);
+                u16 color = palData[index];
+
+                bool transparent = color0Transparent && index == 0;
+                u32 res;
+                switch (outputFmt)
+                {
+                case outputFmt_RGB6A5: res = ConvertRGB5ToRGB6(color)
+                    | (transparent ? 0 : 0x1F000000); break;
+                case outputFmt_RGBA8: res = ConvertRGB5ToRGB8(color)
+                    | (transparent ? 0 : 0xFF000000); break;
+                case outputFmt_BGRA8: res = ConvertRGB5ToBGR8(color)
+                    | (transparent ? 0 : 0xFF000000); break;
+                }
+                output[x * (8 / colorBits) + y * width + i] = res;
+            }
+        }
+    }
+}
+
+template void ConvertNColorsTexture<outputFmt_RGB6A5, 2>(u32, u32, u32*, u8*, u16*, bool);
+template void ConvertNColorsTexture<outputFmt_RGB6A5, 4>(u32, u32, u32*, u8*, u16*, bool);
+template void ConvertNColorsTexture<outputFmt_RGB6A5, 8>(u32, u32, u32*, u8*, u16*, bool);
+
+}

src/GPU3D_Texcache.h

@@ -0,0 +1,310 @@
+#ifndef GPU3D_TEXCACHE
+#define GPU3D_TEXCACHE
+
+#include "types.h"
+#include "GPU.h"
+
+#include <assert.h>
+#include <unordered_map>
+#include <vector>
+
+#define XXH_STATIC_LINKING_ONLY
+#include "xxhash/xxhash.h"
+
+namespace melonDS
+{
+
+inline u32 TextureWidth(u32 texparam)
+{
+    return 8 << ((texparam >> 20) & 0x7);
+}
+
+inline u32 TextureHeight(u32 texparam)
+{
+    return 8 << ((texparam >> 23) & 0x7);
+}
+
+enum
+{
+    outputFmt_RGB6A5,
+    outputFmt_RGBA8,
+    outputFmt_BGRA8
+};
+
+template <int outputFmt>
+void ConvertBitmapTexture(u32 width, u32 height, u32* output, u8* texData);
+template <int outputFmt>
+void ConvertCompressedTexture(u32 width, u32 height, u32* output, u8* texData, u8* texAuxData, u16* palData);
+template <int outputFmt, int X, int Y>
+void ConvertAXIYTexture(u32 width, u32 height, u32* output, u8* texData, u16* palData);
+template <int outputFmt, int colorBits>
+void ConvertNColorsTexture(u32 width, u32 height, u32* output, u8* texData, u16* palData, bool color0Transparent);
+
+template <typename TexLoaderT, typename TexHandleT>
+class Texcache
+{
+public:
+    Texcache(const TexLoaderT& texloader)
+        : TexLoader(texloader) // probably better if this would be a move constructor???
+    {}
+
+    bool Update(GPU& gpu)
+    {
+        auto textureDirty = gpu.VRAMDirty_Texture.DeriveState(gpu.VRAMMap_Texture, gpu);
+        auto texPalDirty = gpu.VRAMDirty_TexPal.DeriveState(gpu.VRAMMap_TexPal, gpu);
+
+        bool textureChanged = gpu.MakeVRAMFlat_TextureCoherent(textureDirty);
+        bool texPalChanged = gpu.MakeVRAMFlat_TexPalCoherent(texPalDirty);
+
+        if (textureChanged || texPalChanged)
+        {
+            //printf("check invalidation %d\n", TexCache.size());
+            for (auto it = Cache.begin(); it != Cache.end();)
+            {
+                TexCacheEntry& entry = it->second;
+                if (textureChanged)
+                {
+                    for (u32 i = 0; i < 2; i++)
+                    {
+                        u32 startBit = entry.TextureRAMStart[i] / VRAMDirtyGranularity;
+                        u32 bitsCount = ((entry.TextureRAMStart[i] + entry.TextureRAMSize[i] + VRAMDirtyGranularity - 1) / VRAMDirtyGranularity) - startBit;
+
+                        u32 startEntry = startBit >> 6;
+                        u64 entriesCount = ((startBit + bitsCount + 0x3F) >> 6) - startEntry;
+                        for (u32 j = startEntry; j < startEntry + entriesCount; j++)
+                        {
+                            if (GetRangedBitMask(j, startBit, bitsCount) & textureDirty.Data[j])
+                            {
+                                u64 newTexHash = XXH3_64bits(&gpu.VRAMFlat_Texture[entry.TextureRAMStart[i]], entry.TextureRAMSize[i]);
+
+                                if (newTexHash != entry.TextureHash[i])
+                                    goto invalidate;
+                            }
+                        }
+                    }
+                }
+
+                if (texPalChanged && entry.TexPalSize > 0)
+                {
+                    u32 startBit = entry.TexPalStart / VRAMDirtyGranularity;
+                    u32 bitsCount = ((entry.TexPalStart + entry.TexPalSize + VRAMDirtyGranularity - 1) / VRAMDirtyGranularity) - startBit;
+
+                    u32 startEntry = startBit >> 6;
+                    u64 entriesCount = ((startBit + bitsCount + 0x3F) >> 6) - startEntry;
+                    for (u32 j = startEntry; j < startEntry + entriesCount; j++)
+                    {
+                        if (GetRangedBitMask(j, startBit, bitsCount) & texPalDirty.Data[j])
+                        {
+                            u64 newPalHash = XXH3_64bits(&gpu.VRAMFlat_TexPal[entry.TexPalStart], entry.TexPalSize);
+                            if (newPalHash != entry.TexPalHash)
+                                goto invalidate;
+                        }
+                    }
+                }
+
+                it++;
+                continue;
+            invalidate:
+                FreeTextures[entry.WidthLog2][entry.HeightLog2].push_back(entry.Texture);
+
+                //printf("invalidating texture %d\n", entry.ImageDescriptor);
+
+                it = Cache.erase(it);
+            }
+
+            return true;
+        }
+
+        return false;
+    }
+
+    void GetTexture(GPU& gpu, u32 texParam, u32 palBase, TexHandleT& textureHandle, u32& layer, u32*& helper)
+    {
+        // remove sampling and texcoord gen params
+        texParam &= ~0xC00F0000;
+
+        u32 fmt = (texParam >> 26) & 0x7;
+        u64 key = texParam;
+        if (fmt != 7)
+        {
+            key |= (u64)palBase << 32;
+            if (fmt == 5)
+                key &= ~((u64)1 << 29);
+        }
+        //printf("%" PRIx64 " %" PRIx32 " %" PRIx32 "\n", key, texParam, palBase);
+
+        assert(fmt != 0 && "no texture is not a texture format!");
+
+        auto it = Cache.find(key);
+
+        if (it != Cache.end())
+        {
+            textureHandle = it->second.Texture.TextureID;
+            layer = it->second.Texture.Layer;
+            helper = &it->second.LastVariant;
+            return;
+        }
+
+        u32 widthLog2 = (texParam >> 20) & 0x7;
+        u32 heightLog2 = (texParam >> 23) & 0x7;
+        u32 width = 8 << widthLog2;
+        u32 height = 8 << heightLog2;
+
+        u32 addr = (texParam & 0xFFFF) * 8;
+
+        TexCacheEntry entry = {0};
+
+        entry.TextureRAMStart[0] = addr;
+        entry.WidthLog2 = widthLog2;
+        entry.HeightLog2 = heightLog2;
+
+        // apparently a new texture
+        if (fmt == 7)
+        {
+            entry.TextureRAMSize[0] = width*height*2;
+
+            ConvertBitmapTexture<outputFmt_RGB6A5>(width, height, DecodingBuffer, &gpu.VRAMFlat_Texture[addr]);
+        }
+        else if (fmt == 5)
+        {
+            u8* texData = &gpu.VRAMFlat_Texture[addr];
+            u32 slot1addr = 0x20000 + ((addr & 0x1FFFC) >> 1);
+            if (addr >= 0x40000)
+                slot1addr += 0x10000;
+            u8* texAuxData = &gpu.VRAMFlat_Texture[slot1addr];
+
+            u16* palData = (u16*)(gpu.VRAMFlat_TexPal + palBase*16);
+
+            entry.TextureRAMSize[0] = width*height/16*4;
+            entry.TextureRAMStart[1] = slot1addr;
+            entry.TextureRAMSize[1] = width*height/16*2;
+            entry.TexPalStart = palBase*16;
+            entry.TexPalSize = 0x10000;
+
+            ConvertCompressedTexture<outputFmt_RGB6A5>(width, height, DecodingBuffer, texData, texAuxData, palData);
+        }
+        else
+        {
+            u32 texSize, palAddr = palBase*16, numPalEntries;
+            switch (fmt)
+            {
+            case 1: texSize = width*height; numPalEntries = 32; break;
+            case 6: texSize = width*height; numPalEntries = 8; break;
+            case 2: texSize = width*height/4; numPalEntries = 4; palAddr >>= 1; break;
+            case 3: texSize = width*height/2; numPalEntries = 16; break;
+            case 4: texSize = width*height; numPalEntries = 256; break;
+            }
+
+            palAddr &= 0x1FFFF;
+
+            /*printf("creating texture | fmt: %d | %dx%d | %08x | %08x\n", fmt, width, height, addr, palAddr);
+            svcSleepThread(1000*1000);*/
+
+            entry.TextureRAMSize[0] = texSize;
+            entry.TexPalStart = palAddr;
+            entry.TexPalSize = numPalEntries*2;
+
+            u8* texData = &gpu.VRAMFlat_Texture[addr];
+            u16* palData = (u16*)(gpu.VRAMFlat_TexPal + palAddr);
+
+            //assert(entry.TexPalStart+entry.TexPalSize <= 128*1024*1024);
+
+            bool color0Transparent = texParam & (1 << 29);
+
+            switch (fmt)
+            {
+            case 1: ConvertAXIYTexture<outputFmt_RGB6A5, 3, 5>(width, height, DecodingBuffer, texData, palData); break;
+            case 6: ConvertAXIYTexture<outputFmt_RGB6A5, 5, 3>(width, height, DecodingBuffer, texData, palData); break;
+            case 2: ConvertNColorsTexture<outputFmt_RGB6A5, 2>(width, height, DecodingBuffer, texData, palData, color0Transparent); break;
+            case 3: ConvertNColorsTexture<outputFmt_RGB6A5, 4>(width, height, DecodingBuffer, texData, palData, color0Transparent); break;
+            case 4: ConvertNColorsTexture<outputFmt_RGB6A5, 8>(width, height, DecodingBuffer, texData, palData, color0Transparent); break;
+            }
+        }
+
+        for (int i = 0; i < 2; i++)
+        {
+            if (entry.TextureRAMSize[i])
+                entry.TextureHash[i] = XXH3_64bits(&gpu.VRAMFlat_Texture[entry.TextureRAMStart[i]], entry.TextureRAMSize[i]);
+        }
+        if (entry.TexPalSize)
+            entry.TexPalHash = XXH3_64bits(&gpu.VRAMFlat_TexPal[entry.TexPalStart], entry.TexPalSize);
+
+        auto& texArrays = TexArrays[widthLog2][heightLog2];
+        auto& freeTextures = FreeTextures[widthLog2][heightLog2];
+
+        if (freeTextures.size() == 0)
+        {
+            texArrays.resize(texArrays.size()+1);
+            TexHandleT& array = texArrays[texArrays.size()-1];
+
+            u32 layers = std::min<u32>((8*1024*1024) / (width*height*4), 64);
+
+            // allocate new array texture
+            //printf("allocating new layer set for %d %d %d %d\n", width, height, texArrays.size()-1, array.ImageDescriptor);
+            array = TexLoader.GenerateTexture(width, height, layers);
+
+            for (u32 i = 0; i < layers; i++)
+            {
+                freeTextures.push_back(TexArrayEntry{array, i});
+            }
+        }
+
+        TexArrayEntry storagePlace = freeTextures[freeTextures.size()-1];
+        freeTextures.pop_back();
+
+        entry.Texture = storagePlace;
+
+        TexLoader.UploadTexture(storagePlace.TextureID, width, height, storagePlace.Layer, DecodingBuffer);
+        //printf("using storage place %d %d | %d %d (%d)\n", width, height, storagePlace.TexArrayIdx, storagePlace.LayerIdx, array.ImageDescriptor);
+
+        textureHandle = storagePlace.TextureID;
+        layer = storagePlace.Layer;
+        helper = &Cache.emplace(std::make_pair(key, entry)).first->second.LastVariant;
+    }
+
+    void Reset()
+    {
+        for (u32 i = 0; i < 8; i++)
+        {
+            for (u32 j = 0; j < 8; j++)
+            {
+                for (u32 k = 0; k < TexArrays[i][j].size(); k++)
+                    TexLoader.DeleteTexture(TexArrays[i][j][k]);
+                TexArrays[i][j].clear();
+                FreeTextures[i][j].clear();
+            }
+        }
+        Cache.clear();
+    }
+private:
+    struct TexArrayEntry
+    {
+        TexHandleT TextureID;
+        u32 Layer;
+    };
+
+    struct TexCacheEntry
+    {
+        u32 LastVariant; // very cheap way to make variant lookup faster
+
+        u32 TextureRAMStart[2], TextureRAMSize[2];
+        u32 TexPalStart, TexPalSize;
+        u8 WidthLog2, HeightLog2;
+        TexArrayEntry Texture;
+
+        u64 TextureHash[2];
+        u64 TexPalHash;
+    };
+    std::unordered_map<u64, TexCacheEntry> Cache;
+
+    TexLoaderT TexLoader;
+
+    std::vector<TexArrayEntry> FreeTextures[8][8];
+    std::vector<TexHandleT> TexArrays[8][8];
+
+    u32 DecodingBuffer[1024*1024];
+};
+
+}
+
+#endif

src/GPU3D_TexcacheOpenGL.cpp

@@ -0,0 +1,29 @@
+#include "GPU3D_TexcacheOpenGL.h"
+
+namespace melonDS
+{
+
+GLuint TexcacheOpenGLLoader::GenerateTexture(u32 width, u32 height, u32 layers)
+{
+    GLuint texarray;
+    glGenTextures(1, &texarray);
+    glBindTexture(GL_TEXTURE_2D_ARRAY, texarray);
+    glTexStorage3D(GL_TEXTURE_2D_ARRAY, 1, GL_RGBA8UI, width, height, layers);
+    return texarray;
+}
+
+void TexcacheOpenGLLoader::UploadTexture(GLuint handle, u32 width, u32 height, u32 layer, void* data)
+{
+    glBindTexture(GL_TEXTURE_2D_ARRAY, handle);
+    glTexSubImage3D(GL_TEXTURE_2D_ARRAY,
+        0, 0, 0, layer,
+        width, height, 1,
+        GL_RGBA_INTEGER, GL_UNSIGNED_BYTE, data);
+}
+
+void TexcacheOpenGLLoader::DeleteTexture(GLuint handle)
+{
+    glDeleteTextures(1, &handle);
+}
+
+}

src/GPU3D_TexcacheOpenGL.h

@@ -0,0 +1,25 @@
+#ifndef GPU3D_TEXCACHEOPENGL
+#define GPU3D_TEXCACHEOPENGL
+
+#include "GPU3D_Texcache.h"
+#include "OpenGLSupport.h"
+
+namespace melonDS
+{
+
+template <typename, typename>
+class Texcache;
+
+class TexcacheOpenGLLoader
+{
+public:
+    GLuint GenerateTexture(u32 width, u32 height, u32 layers);
+    void UploadTexture(GLuint handle, u32 width, u32 height, u32 layer, void* data);
+    void DeleteTexture(GLuint handle);
+};
+
+using TexcacheOpenGL = Texcache<TexcacheOpenGLLoader, GLuint>;
+
+}
+
+#endif

src/GPU_OpenGL.cpp

@@ -36,32 +36,27 @@ using namespace OpenGL;
 std::optional<GLCompositor> GLCompositor::New() noexcept
 {
     assert(glBindAttribLocation != nullptr);
+    GLuint CompShader {};
 
-    std::array<GLuint, 3> CompShader {};
+    if (!OpenGL::CompileVertexFragmentProgram(CompShader,
-    if (!OpenGL::BuildShaderProgram(kCompositorVS, kCompositorFS_Nearest, &CompShader[0], "CompositorShader"))
+            kCompositorVS, kCompositorFS_Nearest, 
-        return std::nullopt;
+            "CompositorShader",
-
+            {{"vPosition", 0}, {"vTexcoord", 1}},
-    glBindAttribLocation(CompShader[2], 0, "vPosition");
+            {{"oColor", 0}}))
-    glBindAttribLocation(CompShader[2], 1, "vTexcoord");
-    glBindFragDataLocation(CompShader[2], 0, "oColor");
-
-    if (!OpenGL::LinkShaderProgram(CompShader.data()))
-        // OpenGL::LinkShaderProgram already deletes the shader program object
-        // if linking the shaders together failed.
         return std::nullopt;
 
     return { GLCompositor(CompShader) };
 }
 
-GLCompositor::GLCompositor(std::array<GLuint, 3> compShader) noexcept : CompShader(compShader)
+GLCompositor::GLCompositor(GLuint compShader) noexcept : CompShader(compShader)
 {
-    CompScaleLoc = glGetUniformLocation(CompShader[2], "u3DScale");
+    CompScaleLoc = glGetUniformLocation(CompShader, "u3DScale");
-    Comp3DXPosLoc = glGetUniformLocation(CompShader[2], "u3DXPos");
+    Comp3DXPosLoc = glGetUniformLocation(CompShader, "u3DXPos");
 
-    glUseProgram(CompShader[2]);
+    glUseProgram(CompShader);
-    GLuint screenTextureUniform = glGetUniformLocation(CompShader[2], "ScreenTex");
+    GLuint screenTextureUniform = glGetUniformLocation(CompShader, "ScreenTex");
     glUniform1i(screenTextureUniform, 0);
-    GLuint _3dTextureUniform = glGetUniformLocation(CompShader[2], "_3DTex");
+    GLuint _3dTextureUniform = glGetUniformLocation(CompShader, "_3DTex");
     glUniform1i(_3dTextureUniform, 1);
 
     // all this mess is to prevent bleeding
@@ -136,7 +131,7 @@ GLCompositor::~GLCompositor()
     glDeleteVertexArrays(1, &CompVertexArrayID);
     glDeleteBuffers(1, &CompVertexBufferID);
 
-    OpenGL::DeleteShaderProgram(CompShader.data());
+    glDeleteProgram(CompShader);
 }
 
 
@@ -174,7 +169,7 @@ GLCompositor& GLCompositor::operator=(GLCompositor&& other) noexcept
         CompVertices = other.CompVertices;
 
         // Clean up these resources before overwriting them
-        OpenGL::DeleteShaderProgram(CompShader.data());
+        glDeleteProgram(CompShader);
         CompShader = other.CompShader;
 
         glDeleteBuffers(1, &CompVertexBufferID);
@@ -244,11 +239,11 @@ void GLCompositor::Stop(const GPU& gpu) noexcept
     glBindFramebuffer(GL_FRAMEBUFFER, 0);
 }
 
-void GLCompositor::RenderFrame(const GPU& gpu, GLRenderer& renderer) noexcept
+void GLCompositor::RenderFrame(const GPU& gpu, Renderer3D& renderer) noexcept
 {
-    int frontbuf = gpu.FrontBuffer;
+    int backbuf = gpu.FrontBuffer ^ 1;
     glBindFramebuffer(GL_READ_FRAMEBUFFER, 0);
-    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, CompScreenOutputFB[frontbuf]);
+    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, CompScreenOutputFB[backbuf]);
 
     glDisable(GL_DEPTH_TEST);
     glDisable(GL_STENCIL_TEST);
@@ -260,7 +255,7 @@ void GLCompositor::RenderFrame(const GPU& gpu, GLRenderer& renderer) noexcept
     glClear(GL_COLOR_BUFFER_BIT);
 
     // TODO: select more shaders (filtering, etc)
-    OpenGL::UseShaderProgram(CompShader.data());
+    glUseProgram(CompShader);
     glUniform1ui(CompScaleLoc, Scale);
 
     // TODO: support setting this midframe, if ever needed
@@ -269,12 +264,12 @@ void GLCompositor::RenderFrame(const GPU& gpu, GLRenderer& renderer) noexcept
     glActiveTexture(GL_TEXTURE0);
     glBindTexture(GL_TEXTURE_2D, CompScreenInputTex);
 
-    if (gpu.Framebuffer[frontbuf][0] && gpu.Framebuffer[frontbuf][1])
+    if (gpu.Framebuffer[backbuf][0] && gpu.Framebuffer[backbuf][1])
     {
         glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, 256*3 + 1, 192, GL_RGBA_INTEGER,
-                        GL_UNSIGNED_BYTE, gpu.Framebuffer[frontbuf][0].get());
+                        GL_UNSIGNED_BYTE, gpu.Framebuffer[backbuf][0].get());
         glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 192, 256*3 + 1, 192, GL_RGBA_INTEGER,
-                        GL_UNSIGNED_BYTE, gpu.Framebuffer[frontbuf][1].get());
+                        GL_UNSIGNED_BYTE, gpu.Framebuffer[backbuf][1].get());
     }
 
     glActiveTexture(GL_TEXTURE1);

src/GPU_OpenGL.h

@@ -28,6 +28,7 @@ namespace melonDS
 class GPU;
 struct RenderSettings;
 class GLRenderer;
+class Renderer3D;
 class GLCompositor
 {
 public:
@@ -42,14 +43,14 @@ public:
     [[nodiscard]] int GetScaleFactor() const noexcept { return Scale; }
 
     void Stop(const GPU& gpu) noexcept;
-    void RenderFrame(const GPU& gpu, GLRenderer& renderer) noexcept;
+    void RenderFrame(const GPU& gpu, Renderer3D& renderer) noexcept;
     void BindOutputTexture(int buf);
 private:
-    GLCompositor(std::array<GLuint, 3> CompShader) noexcept;
+    GLCompositor(GLuint CompShader) noexcept;
     int Scale = 0;
     int ScreenH = 0, ScreenW = 0;
 
-    std::array<GLuint, 3> CompShader {};
+    GLuint CompShader {};
     GLuint CompScaleLoc = 0;
     GLuint Comp3DXPosLoc = 0;
 

src/NonStupidBitfield.h

@@ -26,11 +26,38 @@
 #include <initializer_list>
 #include <algorithm>
 
+namespace melonDS
+{
+
+inline u64 GetRangedBitMask(u32 idx, u32 startBit, u32 bitsCount)
+{
+    u32 startEntry = startBit >> 6;
+    u64 entriesCount = ((startBit + bitsCount + 0x3F) >> 6) - startEntry;
+
+    if (entriesCount > 1)
+    {
+        if (idx == startEntry)
+            return 0xFFFFFFFFFFFFFFFF << (startBit & 0x3F);
+        if (((startBit + bitsCount) & 0x3F) && idx == startEntry + entriesCount - 1)
+            return ~(0xFFFFFFFFFFFFFFFF << ((startBit + bitsCount) & 0x3F));
+
+        return 0xFFFFFFFFFFFFFFFF;
+    }
+    else if (idx == startEntry)
+    {
+        return bitsCount == 64
+            ? 0xFFFFFFFFFFFFFFFF
+            : ((1ULL << bitsCount) - 1) << (startBit & 0x3F);
+    }
+    else
+    {
+        return 0;
+    }
+}
+
 // like std::bitset but less stupid and optimised for 
 // our use case (keeping track of memory invalidations)
 
-namespace melonDS
-{
 template <u32 Size>
 struct NonStupidBitField
 {
@@ -166,6 +193,11 @@ struct NonStupidBitField
         return Ref{*this, idx};
     }
 
+    bool operator[](u32 idx) const
+    {
+        return Data[idx >> 6] & (1ULL << (idx & 0x3F));
+    }
+
     void SetRange(u32 startBit, u32 bitsCount)
     {
         u32 startEntry = startBit >> 6;
@@ -187,6 +219,26 @@ struct NonStupidBitField
         }
     }
 
+    int Min() const
+    {
+        for (int i = 0; i < DataLength; i++)
+        {
+            if (Data[i])
+                return i * 64 + __builtin_ctzll(Data[i]);
+        }
+        return -1;
+    }
+
+    int Max() const
+    {
+        for (int i = DataLength - 1; i >= 0; i--)
+        {
+            if (Data[i])
+                return i * 64 + (63 - __builtin_clzll(Data[i]));
+        }
+        return -1;
+    }
+
     NonStupidBitField& operator|=(const NonStupidBitField<Size>& other)
     {
         for (u32 i = 0; i < DataLength; i++)
@@ -195,6 +247,7 @@ struct NonStupidBitField
         }
         return *this;
     }
+
     NonStupidBitField& operator&=(const NonStupidBitField<Size>& other)
     {
         for (u32 i = 0; i < DataLength; i++)
@@ -203,6 +256,20 @@ struct NonStupidBitField
         }
         return *this;
     }
+
+    operator bool() const
+    {
+        for (int i = 0; i < DataLength - 1; i++)
+        {
+            if (Data[i])
+                return true;
+        }
+        if (Data[DataLength-1] & ((Size&0x3F) ? ~(0xFFFFFFFFFFFFFFFF << (Size&0x3F)) : 0xFFFFFFFFFFFFFFFF))
+        {
+            return true;
+        }
+        return false;
+    }
 };
 
 }

src/OpenGLSupport.cpp

@@ -18,6 +18,14 @@
 
 #include "OpenGLSupport.h"
 
+#include <unordered_map>
+#include <vector>
+
+#include <assert.h>
+
+#define XXH_STATIC_LINKING_ONLY
+#include "xxhash/xxhash.h"
+
 namespace melonDS
 {
 
@@ -27,9 +35,158 @@ using Platform::LogLevel;
 namespace OpenGL
 {
 
-bool BuildShaderProgram(const char* vs, const char* fs, GLuint* ids, const char* name)
+struct ShaderCacheEntry
+{
+    u32 Length;
+    u8* Data;
+    u32 BinaryFormat;
+
+    ShaderCacheEntry(u8* data, u32 length, u32 binaryFmt)
+        : Length(length), Data(data), BinaryFormat(binaryFmt)
+    {
+        assert(data != nullptr);
+    }
+
+    ShaderCacheEntry(const ShaderCacheEntry&) = delete;
+    ShaderCacheEntry(ShaderCacheEntry&& other)
+    {
+        Data = other.Data;
+        Length = other.Length;
+        BinaryFormat = other.BinaryFormat;
+
+        other.Data = nullptr;
+        other.Length = 0;
+        other.BinaryFormat = 0;
+    }
+
+    ~ShaderCacheEntry()
+    {
+        if (Data) // check whether it was moved
+            delete[] Data;
+    }
+};
+
+std::unordered_map<u64, ShaderCacheEntry> ShaderCache;
+std::vector<u64> NewShaders;
+
+constexpr u32 ShaderCacheMagic = 0x11CAC4E1;
+constexpr u32 ShaderCacheVersion = 1;
+
+void LoadShaderCache()
+{
+    // for now the shader cache only contains only compute shaders
+    // because they take the longest to compile
+    Platform::FileHandle* file = Platform::OpenLocalFile("shadercache", Platform::FileMode::Read);
+    if (file == nullptr)
+    {
+        Log(LogLevel::Error, "Could not find shader cache\n");
+        return;
+    }
+
+    u32 magic, version, numPrograms;
+    if (Platform::FileRead(&magic, 4, 1, file) != 1 || magic != ShaderCacheMagic)
+    {
+        Log(LogLevel::Error, "Shader cache file has invalid magic\n");
+        goto fileInvalid;
+    }
+
+    if (Platform::FileRead(&version, 4, 1, file) != 1 || version != ShaderCacheVersion)
+    {
+        Log(LogLevel::Error, "Shader cache file has bad version\n");
+        goto fileInvalid;
+    }
+
+    if (Platform::FileRead(&numPrograms, 4, 1, file) != 1)
+    {
+        Log(LogLevel::Error, "Shader cache file invalid program count\n");
+        goto fileInvalid;
+    }
+
+    // not the best approach, because once changes pile up
+    // we read and overwrite the old files
+    for (u32 i = 0; i < numPrograms; i++)
+    {
+        int error = 3;
+
+        u32 length, binaryFormat;
+        u64 sourceHash;
+        error -= Platform::FileRead(&sourceHash, 8, 1, file);
+        error -= Platform::FileRead(&length, 4, 1, file);
+        error -= Platform::FileRead(&binaryFormat, 4, 1, file);
+
+        if (error != 0)
+        {
+            Log(LogLevel::Error, "Invalid shader cache entry\n");
+            goto fileInvalid;
+        }
+
+        u8* data = new u8[length];
+        if (Platform::FileRead(data, length, 1, file) != 1)
+        {
+            Log(LogLevel::Error, "Could not read shader cache entry data\n");
+            delete[] data;
+            goto fileInvalid;
+        }
+
+        ShaderCache.erase(sourceHash);
+        ShaderCache.emplace(sourceHash, ShaderCacheEntry(data, length, binaryFormat));
+    }
+
+fileInvalid:
+    Platform::CloseFile(file);
+}
+
+void SaveShaderCache()
+{
+    Platform::FileHandle* file = Platform::OpenLocalFile("shadercache", Platform::FileMode::ReadWrite);
+
+    if (file == nullptr)
+    {
+        Log(LogLevel::Error, "Could not open or create shader cache file\n");
+        return;
+    }
+
+    int written = 3;
+    u32 magic = ShaderCacheMagic, version = ShaderCacheVersion, numPrograms = ShaderCache.size();
+    written -= Platform::FileWrite(&magic, 4, 1, file);
+    written -= Platform::FileWrite(&version, 4, 1, file);
+    written -= Platform::FileWrite(&numPrograms, 4, 1, file);
+
+    if (written != 0)
+    {
+        Log(LogLevel::Error, "Could not write shader cache header\n");
+        goto writeError;
+    }
+
+    Platform::FileSeek(file, 0, Platform::FileSeekOrigin::End);
+
+    printf("new shaders %d\n", NewShaders.size());
+
+    for (u64 newShader : NewShaders)
+    {
+        int error = 4;
+        auto it = ShaderCache.find(newShader);
+
+        error -= Platform::FileWrite(&it->first, 8, 1, file);
+        error -= Platform::FileWrite(&it->second.Length, 4, 1, file);
+        error -= Platform::FileWrite(&it->second.BinaryFormat, 4, 1, file);
+        error -= Platform::FileWrite(it->second.Data, it->second.Length, 1, file);
+
+        if (error != 0)
+        {
+            Log(LogLevel::Error, "Could not insert new shader cache entry\n");
+            goto writeError;
+        }
+    }
+
+writeError:
+    Platform::CloseFile(file);
+
+    NewShaders.clear();
+}
+
+bool CompilerShader(GLuint& id, const std::string& source, const std::string& name, const std::string& type)
 {
-    int len;
     int res;
 
     if (!glCreateShader)
@@ -38,61 +195,32 @@ bool BuildShaderProgram(const char* vs, const char* fs, GLuint* ids, const char*
         return false;
     }
 
-    ids[0] = glCreateShader(GL_VERTEX_SHADER);
+    const char* sourceC = source.c_str();
-    len = strlen(vs);
+    int len = source.length();
-    glShaderSource(ids[0], 1, &vs, &len);
+    glShaderSource(id, 1, &sourceC, &len);
-    glCompileShader(ids[0]);
 
-    glGetShaderiv(ids[0], GL_COMPILE_STATUS, &res);
+    glCompileShader(id);
+
+    glGetShaderiv(id, GL_COMPILE_STATUS, &res);
     if (res != GL_TRUE)
     {
-        glGetShaderiv(ids[0], GL_INFO_LOG_LENGTH, &res);
+        glGetShaderiv(id, GL_INFO_LOG_LENGTH, &res);
         if (res < 1) res = 1024;
         char* log = new char[res+1];
-        glGetShaderInfoLog(ids[0], res+1, NULL, log);
+        glGetShaderInfoLog(id, res+1, NULL, log);
-        Log(LogLevel::Error, "OpenGL: failed to compile vertex shader %s: %s\n", name, log);
+        Log(LogLevel::Error, "OpenGL: failed to compile %s shader %s: %s\n", type.c_str(), name.c_str(), log);
-        Log(LogLevel::Debug, "shader source:\n--\n%s\n--\n", vs);
+        Log(LogLevel::Debug, "shader source:\n--\n%s\n--\n", source.c_str());
         delete[] log;
 
-        glDeleteShader(ids[0]);
+        glDeleteShader(id);
 
         return false;
     }
 
-    ids[1] = glCreateShader(GL_FRAGMENT_SHADER);
-    len = strlen(fs);
-    glShaderSource(ids[1], 1, &fs, &len);
-    glCompileShader(ids[1]);
-
-    glGetShaderiv(ids[1], GL_COMPILE_STATUS, &res);
-    if (res != GL_TRUE)
-    {
-        glGetShaderiv(ids[1], GL_INFO_LOG_LENGTH, &res);
-        if (res < 1) res = 1024;
-        char* log = new char[res+1];
-        glGetShaderInfoLog(ids[1], res+1, NULL, log);
-        Log(LogLevel::Error, "OpenGL: failed to compile fragment shader %s: %s\n", name, log);
-        //printf("shader source:\n--\n%s\n--\n", fs);
-        delete[] log;
-
-        Platform::FileHandle* logf = Platform::OpenFile("shaderfail.log", Platform::FileMode::WriteText);
-        Platform::FileWrite(fs, len+1, 1, logf);
-        Platform::CloseFile(logf);
-
-        glDeleteShader(ids[0]);
-        glDeleteShader(ids[1]);
-
-        return false;
-    }
-
-    ids[2] = glCreateProgram();
-    glAttachShader(ids[2], ids[0]);
-    glAttachShader(ids[2], ids[1]);
-
     return true;
 }
 
-bool LinkShaderProgram(GLuint* ids)
+bool LinkProgram(GLuint& result, GLuint* ids, int numIds)
 {
     int res;
 
@@ -102,46 +230,132 @@ bool LinkShaderProgram(GLuint* ids)
         return false;
     }
 
-    glLinkProgram(ids[2]);
+    for (int i = 0; i < numIds; i++)
+    {
+        glAttachShader(result, ids[i]);
+    }
 
-    glDetachShader(ids[2], ids[0]);
+    glLinkProgram(result);
-    glDetachShader(ids[2], ids[1]);
 
-    glDeleteShader(ids[0]);
+    for (int i = 0; i < numIds; i++)
-    glDeleteShader(ids[1]);
+        glDetachShader(result, ids[i]);
 
-    glGetProgramiv(ids[2], GL_LINK_STATUS, &res);
+    glGetProgramiv(result, GL_LINK_STATUS, &res);
     if (res != GL_TRUE)
     {
-        glGetProgramiv(ids[2], GL_INFO_LOG_LENGTH, &res);
+        glGetProgramiv(result, GL_INFO_LOG_LENGTH, &res);
         if (res < 1) res = 1024;
         char* log = new char[res+1];
-        glGetProgramInfoLog(ids[2], res+1, NULL, log);
+        glGetProgramInfoLog(result, res+1, NULL, log);
         Log(LogLevel::Error, "OpenGL: failed to link shader program: %s\n", log);
         delete[] log;
 
-        glDeleteProgram(ids[2]);
-
         return false;
     }
 
     return true;
 }
 
-void DeleteShaderProgram(GLuint* ids)
+bool CompileComputeProgram(GLuint& result, const std::string& source, const std::string& name)
 {
-    if (glDeleteProgram)
+    result = glCreateProgram();
-    { // If OpenGL isn't loaded, then there's no shader program to delete
-        glDeleteProgram(ids[2]);
-    }
-}
 
-void UseShaderProgram(GLuint* ids)
+    /*u64 sourceHash = XXH64(source.data(), source.size(), 0);
+    auto it = ShaderCache.find(sourceHash);
+    if (it != ShaderCache.end())
     {
-    if (glUseProgram)
+        glProgramBinary(result, it->second.BinaryFormat, it->second.Data, it->second.Length);
-    { // If OpenGL isn't loaded, then there's no shader program to use
+
-        glUseProgram(ids[2]);
+        GLint linkStatus;
+        glGetProgramiv(result, GL_LINK_STATUS, &linkStatus);
+        if (linkStatus == GL_TRUE)
+        {
+            Log(LogLevel::Info, "Restored shader %s from cache\n", name.c_str());
+            return true;
         }
+        else
+        {
+        }
+    }*/
+    Log(LogLevel::Error, "Shader %s from cache was rejected\n", name.c_str());
+
+    GLuint shader;
+    bool linkingSucess = false;
+
+    if (!glCreateShader || !glDeleteShader)
+        goto error;
+
+    shader = glCreateShader(GL_COMPUTE_SHADER);
+
+    if (!CompilerShader(shader, source, name, "compute"))
+        goto error;
+
+    linkingSucess = LinkProgram(result, &shader, 1);
+
+error:
+    glDeleteShader(shader);
+
+    if (!linkingSucess)
+    {
+        glDeleteProgram(result);
+    }
+    /*else
+    {
+        GLint length;
+        GLenum format;
+        glGetProgramiv(result, GL_PROGRAM_BINARY_LENGTH, &length);
+
+        u8* buffer = new u8[length];
+        glGetProgramBinary(result, length, nullptr, &format, buffer);
+
+        ShaderCache.emplace(sourceHash, ShaderCacheEntry(buffer, length, format));
+        NewShaders.push_back(sourceHash);
+    }*/
+
+    return linkingSucess;
+}
+
+bool CompileVertexFragmentProgram(GLuint& result,
+    const std::string& vs, const std::string& fs,
+    const std::string& name,
+    const std::initializer_list<AttributeTarget>& vertexInAttrs,
+    const std::initializer_list<AttributeTarget>& fragmentOutAttrs)
+{
+    GLuint shaders[2] =
+    {
+        glCreateShader(GL_VERTEX_SHADER),
+        glCreateShader(GL_FRAGMENT_SHADER)
+    };
+    result = glCreateProgram();
+
+    bool linkingSucess = false;
+
+    if (!CompilerShader(shaders[0], vs, name, "vertex"))
+        goto error;
+
+    if (!CompilerShader(shaders[1], fs, name, "fragment"))
+        goto error;
+
+
+    for (const AttributeTarget& target : vertexInAttrs)
+    {
+        glBindAttribLocation(result, target.Location, target.Name);
+    }
+    for (const AttributeTarget& target : fragmentOutAttrs)
+    {
+        glBindFragDataLocation(result, target.Location, target.Name);
+    }
+
+    linkingSucess = LinkProgram(result, shaders, 2);
+
+error:
+    glDeleteShader(shaders[1]);
+    glDeleteShader(shaders[0]);
+
+    if (!linkingSucess)
+        glDeleteProgram(result);
+
+    return linkingSucess;
 }
 
 }

src/OpenGLSupport.h

@@ -28,10 +28,23 @@
 namespace melonDS::OpenGL
 {
 
-bool BuildShaderProgram(const char* vs, const char* fs, GLuint* ids, const char* name);
+void LoadShaderCache();
-bool LinkShaderProgram(GLuint* ids);
+void SaveShaderCache();
-void DeleteShaderProgram(GLuint* ids);
+
-void UseShaderProgram(GLuint* ids);
+struct AttributeTarget
+{
+    const char* Name;
+    u32 Location;
+};
+
+
+bool CompileVertexFragmentProgram(GLuint& result,
+    const std::string& vs, const std::string& fs,
+    const std::string& name,
+    const std::initializer_list<AttributeTarget>& vertexInAttrs,
+    const std::initializer_list<AttributeTarget>& fragmentOutAttrs);
+
+bool CompileComputeProgram(GLuint& result, const std::string& source, const std::string& name);
 
 }
 

src/frontend/qt_sdl/Config.cpp

@@ -22,6 +22,7 @@
 #include <inttypes.h>
 #include "Platform.h"
 #include "Config.h"
+#include "GPU.h"
 
 
 namespace Config
@@ -59,6 +60,7 @@ bool Threaded3D;
 
 int GL_ScaleFactor;
 bool GL_BetterPolygons;
+bool GL_HiresCoordinates;
 
 bool LimitFPS;
 int MaxFPS;
@@ -246,11 +248,12 @@ ConfigEntry ConfigFile[] =
     {"ScreenVSync",         1, &ScreenVSync,         false, false},
     {"ScreenVSyncInterval", 0, &ScreenVSyncInterval, 1, false},
 
-    {"3DRenderer", 0, &_3DRenderer, 0, false},
+    {"3DRenderer", 0, &_3DRenderer, renderer3D_Software, false},
     {"Threaded3D", 1, &Threaded3D, true, false},
 
     {"GL_ScaleFactor", 0, &GL_ScaleFactor, 1, false},
     {"GL_BetterPolygons", 1, &GL_BetterPolygons, false, false},
+    {"GL_HiresCoordinates", 1, &GL_HiresCoordinates, true, false},
 
     {"LimitFPS", 1, &LimitFPS, true, false},
     {"MaxFPS", 0, &MaxFPS, 1000, false},

src/frontend/qt_sdl/Config.h

@@ -51,6 +51,16 @@ enum
     micInputType_MAX,
 };
 
+enum
+{
+    renderer3D_Software = 0,
+#ifdef OGLRENDERER_ENABLED
+    renderer3D_OpenGL,
+    renderer3D_OpenGLCompute,
+#endif
+    renderer3D_Max,
+};
+
 namespace Config
 {
 
@@ -103,6 +113,7 @@ extern bool Threaded3D;
 
 extern int GL_ScaleFactor;
 extern bool GL_BetterPolygons;
+extern bool GL_HiresCoordinates;
 
 extern bool LimitFPS;
 extern int MaxFPS;

src/frontend/qt_sdl/EmuThread.cpp

@@ -52,10 +52,12 @@
 #include "DSi_I2C.h"
 #include "GPU3D_Soft.h"
 #include "GPU3D_OpenGL.h"
+#include "GPU3D_Compute.h"
 
 #include "Savestate.h"
 
 #include "ROMManager.h"
+#include "EmuThread.h"
 //#include "ArchiveUtil.h"
 //#include "CameraManager.h"
 
@@ -95,8 +97,7 @@ EmuThread::EmuThread(QObject* parent) : QThread(parent)
 
 std::unique_ptr<NDS> EmuThread::CreateConsole(
     std::unique_ptr<melonDS::NDSCart::CartCommon> &&ndscart,
-    std::unique_ptr<melonDS::GBACart::CartCommon>&& gbacart
+    std::unique_ptr<melonDS::GBACart::CartCommon> &&gbacart) noexcept
-) noexcept
 {
     auto arm7bios = ROMManager::LoadARM7BIOS();
     if (!arm7bios)
@@ -326,21 +327,12 @@ void EmuThread::run()
         videoRenderer = 0;
     }
 
-    if (videoRenderer == 0)
+    updateRenderer();
-    { // If we're using the software renderer...
-        NDS->GPU.SetRenderer3D(std::make_unique<SoftRenderer>(Config::Threaded3D != 0));
-    }
-    else
-    {
-        auto glrenderer =  melonDS::GLRenderer::New();
-        glrenderer->SetRenderSettings(Config::GL_BetterPolygons, Config::GL_ScaleFactor);
-        NDS->GPU.SetRenderer3D(std::move(glrenderer));
-    }
 
     Input::Init();
 
     u32 nframes = 0;
-    double perfCountsSec = 1.0 / SDL_GetPerformanceFrequency();
+    perfCountsSec = 1.0 / SDL_GetPerformanceFrequency();
     double lastTime = SDL_GetPerformanceCounter() * perfCountsSec;
     double frameLimitError = 0.0;
     double lastMeasureTime = lastTime;
@@ -451,20 +443,9 @@ void EmuThread::run()
                     videoRenderer = 0;
                 }
 
-                videoRenderer = screenGL ? Config::_3DRenderer : 0;
+                updateRenderer();
 
                 videoSettingsDirty = false;
-
-                if (videoRenderer == 0)
-                { // If we're using the software renderer...
-                    NDS->GPU.SetRenderer3D(std::make_unique<SoftRenderer>(Config::Threaded3D != 0));
-                }
-                else
-                {
-                    auto glrenderer =  melonDS::GLRenderer::New();
-                    glrenderer->SetRenderSettings(Config::GL_BetterPolygons, Config::GL_ScaleFactor);
-                    NDS->GPU.SetRenderer3D(std::move(glrenderer));
-                }
             }
 
             // process input and hotkeys
@@ -512,7 +493,16 @@ void EmuThread::run()
 
 
             // emulate
-            u32 nlines = NDS->RunFrame();
+            u32 nlines;
+            if (NDS->GPU.GetRenderer3D().NeedsShaderCompile())
+            {
+                compileShaders();
+                nlines = 0;
+            }
+            else
+            {
+                nlines = NDS->RunFrame();
+            }
 
             if (ROMManager::NDSSave)
                 ROMManager::NDSSave->CheckFlush();
@@ -750,3 +740,53 @@ bool EmuThread::emuIsActive()
 {
     return (RunningSomething == 1);
 }
+
+void EmuThread::updateRenderer()
+{
+    if (videoRenderer != lastVideoRenderer)
+    {
+        printf("creating renderer %d\n", videoRenderer);
+        switch (videoRenderer)
+        {
+        case renderer3D_Software:
+            NDS->GPU.SetRenderer3D(std::make_unique<SoftRenderer>());
+            break;
+        case renderer3D_OpenGL:
+                NDS->GPU.SetRenderer3D(GLRenderer::New());
+            break;
+        case renderer3D_OpenGLCompute:
+            NDS->GPU.SetRenderer3D(ComputeRenderer::New());
+            break;
+        default: __builtin_unreachable();
+        }
+    }
+    lastVideoRenderer = videoRenderer;
+
+    switch (videoRenderer)
+    {
+    case renderer3D_Software:
+        static_cast<SoftRenderer&>(NDS->GPU.GetRenderer3D()).SetThreaded(Config::Threaded3D, NDS->GPU);
+        break;
+    case renderer3D_OpenGL:
+        static_cast<GLRenderer&>(NDS->GPU.GetRenderer3D()).SetRenderSettings(Config::GL_BetterPolygons, Config::GL_ScaleFactor);
+        break;
+    case renderer3D_OpenGLCompute:
+        static_cast<ComputeRenderer&>(NDS->GPU.GetRenderer3D()).SetRenderSettings(Config::GL_ScaleFactor, Config::GL_HiresCoordinates);
+        break;
+    default: __builtin_unreachable();
+    }
+}
+
+void EmuThread::compileShaders()
+{
+    int currentShader, shadersCount;
+    u64 startTime = SDL_GetPerformanceCounter();
+    // kind of hacky to look at the wallclock, though it is easier than
+    // than disabling vsync
+    do
+    {
+        NDS->GPU.GetRenderer3D().ShaderCompileStep(currentShader, shadersCount);
+    } while (NDS->GPU.GetRenderer3D().NeedsShaderCompile() &&
+        (SDL_GetPerformanceCounter() - startTime) * perfCountsSec < 1.0 / 6.0);
+    mainWindow->osdAddMessage(0, "Compiling shader %d/%d", currentShader+1, shadersCount);
+}

src/frontend/qt_sdl/EmuThread.h

@@ -94,6 +94,9 @@ signals:
     void syncVolumeLevel();
 
 private:
+    void updateRenderer();
+    void compileShaders();
+
     std::unique_ptr<melonDS::NDS> CreateConsole(
         std::unique_ptr<melonDS::NDSCart::CartCommon>&& ndscart,
         std::unique_ptr<melonDS::GBACart::CartCommon>&& gbacart
@@ -127,8 +130,9 @@ private:
 
     int autoScreenSizing;
 
-    int videoRenderer;
+    int lastVideoRenderer = -1;
-    bool videoSettingsDirty;
+
+    double perfCountsSec;
 };
 
 #endif // EMUTHREAD_H

src/frontend/qt_sdl/Screen.cpp

@@ -709,19 +709,17 @@ void ScreenPanelGL::initOpenGL()
 
     glContext->MakeCurrent();
 
-    OpenGL::BuildShaderProgram(kScreenVS, kScreenFS, screenShaderProgram, "ScreenShader");
+    OpenGL::CompileVertexFragmentProgram(screenShaderProgram,
-    GLuint pid = screenShaderProgram[2];
+        kScreenVS, kScreenFS,
-    glBindAttribLocation(pid, 0, "vPosition");
+        "ScreenShader",
-    glBindAttribLocation(pid, 1, "vTexcoord");
+        {{"vPosition", 0}, {"vTexcoord", 1}},
-    glBindFragDataLocation(pid, 0, "oColor");
+        {{"oColor", 0}});
 
-    OpenGL::LinkShaderProgram(screenShaderProgram);
+    glUseProgram(screenShaderProgram);
+    glUniform1i(glGetUniformLocation(screenShaderProgram, "ScreenTex"), 0);
 
-    glUseProgram(pid);
+    screenShaderScreenSizeULoc = glGetUniformLocation(screenShaderProgram, "uScreenSize");
-    glUniform1i(glGetUniformLocation(pid, "ScreenTex"), 0);
+    screenShaderTransformULoc = glGetUniformLocation(screenShaderProgram, "uTransform");
-
-    screenShaderScreenSizeULoc = glGetUniformLocation(pid, "uScreenSize");
-    screenShaderTransformULoc = glGetUniformLocation(pid, "uTransform");
 
     // to prevent bleeding between both parts of the screen
     // with bilinear filtering enabled
@@ -769,21 +767,19 @@ void ScreenPanelGL::initOpenGL()
     memset(zeroData, 0, sizeof(zeroData));
     glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 192, 256, 2, GL_RGBA, GL_UNSIGNED_BYTE, zeroData);
 
+    OpenGL::CompileVertexFragmentProgram(osdShader,
+        kScreenVS_OSD, kScreenFS_OSD,
+        "OSDShader",
+        {{"vPosition", 0}},
+        {{"oColor", 0}});
 
-    OpenGL::BuildShaderProgram(kScreenVS_OSD, kScreenFS_OSD, osdShader, "OSDShader");
+    glUseProgram(osdShader);
+    glUniform1i(glGetUniformLocation(osdShader, "OSDTex"), 0);
 
-    pid = osdShader[2];
+    osdScreenSizeULoc = glGetUniformLocation(osdShader, "uScreenSize");
-    glBindAttribLocation(pid, 0, "vPosition");
+    osdPosULoc = glGetUniformLocation(osdShader, "uOSDPos");
-    glBindFragDataLocation(pid, 0, "oColor");
+    osdSizeULoc = glGetUniformLocation(osdShader, "uOSDSize");
-
+    osdScaleFactorULoc = glGetUniformLocation(osdShader, "uScaleFactor");
-    OpenGL::LinkShaderProgram(osdShader);
-    glUseProgram(pid);
-    glUniform1i(glGetUniformLocation(pid, "OSDTex"), 0);
-
-    osdScreenSizeULoc = glGetUniformLocation(pid, "uScreenSize");
-    osdPosULoc = glGetUniformLocation(pid, "uOSDPos");
-    osdSizeULoc = glGetUniformLocation(pid, "uOSDSize");
-    osdScaleFactorULoc = glGetUniformLocation(pid, "uScaleFactor");
 
     const float osdvertices[6*2] =
     {
@@ -818,8 +814,7 @@ void ScreenPanelGL::deinitOpenGL()
     glDeleteVertexArrays(1, &screenVertexArray);
     glDeleteBuffers(1, &screenVertexBuffer);
 
-    OpenGL::DeleteShaderProgram(screenShaderProgram);
+    glDeleteProgram(screenShaderProgram);
-
 
     for (const auto& [key, tex] : osdTextures)
     {
@@ -830,8 +825,7 @@ void ScreenPanelGL::deinitOpenGL()
     glDeleteVertexArrays(1, &osdVertexArray);
     glDeleteBuffers(1, &osdVertexBuffer);
 
-    OpenGL::DeleteShaderProgram(osdShader);
+    glDeleteProgram(osdShader);
-
 
     glContext->DoneCurrent();
 
@@ -885,7 +879,7 @@ void ScreenPanelGL::drawScreenGL()
 
     glViewport(0, 0, w, h);
 
-    glUseProgram(screenShaderProgram[2]);
+    glUseProgram(screenShaderProgram);
     glUniform2f(screenShaderScreenSizeULoc, w / factor, h / factor);
 
     int frontbuf = emuThread->FrontBuffer;
@@ -895,7 +889,7 @@ void ScreenPanelGL::drawScreenGL()
     if (emuThread->NDS->GPU.GetRenderer3D().Accelerated)
     {
         // hardware-accelerated render
-        static_cast<GLRenderer&>(emuThread->NDS->GPU.GetRenderer3D()).GetCompositor().BindOutputTexture(frontbuf);
+        emuThread->NDS->GPU.GetRenderer3D().BindOutputTexture(frontbuf);
     }
     else
 #endif
@@ -936,7 +930,7 @@ void ScreenPanelGL::drawScreenGL()
 
         u32 y = kOSDMargin;
 
-        glUseProgram(osdShader[2]);
+        glUseProgram(osdShader);
 
         glUniform2f(osdScreenSizeULoc, w, h);
         glUniform1f(osdScaleFactorULoc, factor);

src/frontend/qt_sdl/Screen.h

@@ -172,7 +172,7 @@ private:
 
     GLuint screenVertexBuffer, screenVertexArray;
     GLuint screenTexture;
-    GLuint screenShaderProgram[3];
+    GLuint screenShaderProgram;
     GLuint screenShaderTransformULoc, screenShaderScreenSizeULoc;
 
     QMutex screenSettingsLock;
@@ -181,7 +181,7 @@ private:
 
     int lastScreenWidth = -1, lastScreenHeight = -1;
 
-    GLuint osdShader[3];
+    GLuint osdShader;
     GLint osdScreenSizeULoc, osdPosULoc, osdSizeULoc;
     GLfloat osdScaleFactorULoc;
     GLuint osdVertexArray;

src/frontend/qt_sdl/VideoSettingsDialog.cpp

@@ -23,6 +23,7 @@
 #include "types.h"
 #include "Platform.h"
 #include "Config.h"
+#include "GPU.h"
 
 #include "VideoSettingsDialog.h"
 #include "ui_VideoSettingsDialog.h"
@@ -30,11 +31,20 @@
 
 inline bool UsesGL()
 {
-    return (Config::ScreenUseGL != 0) || (Config::_3DRenderer != 0);
+    return (Config::ScreenUseGL != 0) || (Config::_3DRenderer != renderer3D_Software);
 }
 
 VideoSettingsDialog* VideoSettingsDialog::currentDlg = nullptr;
 
+void VideoSettingsDialog::setEnabled()
+{
+    bool softwareRenderer = Config::_3DRenderer == renderer3D_Software;
+    ui->cbGLDisplay->setEnabled(softwareRenderer);
+    ui->cbSoftwareThreaded->setEnabled(softwareRenderer);
+    ui->cbxGLResolution->setEnabled(!softwareRenderer);
+    ui->cbBetterPolygons->setEnabled(Config::_3DRenderer == renderer3D_OpenGL);
+    ui->cbxComputeHiResCoords->setEnabled(Config::_3DRenderer == renderer3D_OpenGLCompute);
+}
 
 VideoSettingsDialog::VideoSettingsDialog(QWidget* parent) : QDialog(parent), ui(new Ui::VideoSettingsDialog)
 {
@@ -48,10 +58,12 @@ VideoSettingsDialog::VideoSettingsDialog(QWidget* parent) : QDialog(parent), ui(
     oldSoftThreaded = Config::Threaded3D;
     oldGLScale = Config::GL_ScaleFactor;
     oldGLBetterPolygons = Config::GL_BetterPolygons;
+    oldHiresCoordinates = Config::GL_HiresCoordinates;
 
     grp3DRenderer = new QButtonGroup(this);
-    grp3DRenderer->addButton(ui->rb3DSoftware, 0);
+    grp3DRenderer->addButton(ui->rb3DSoftware, renderer3D_Software);
-    grp3DRenderer->addButton(ui->rb3DOpenGL,   1);
+    grp3DRenderer->addButton(ui->rb3DOpenGL,   renderer3D_OpenGL);
+    grp3DRenderer->addButton(ui->rb3DCompute,  renderer3D_OpenGLCompute);
 #if QT_VERSION < QT_VERSION_CHECK(5, 15, 0)
     connect(grp3DRenderer, SIGNAL(buttonClicked(int)), this, SLOT(onChange3DRenderer(int)));
 #else
@@ -75,25 +87,13 @@ VideoSettingsDialog::VideoSettingsDialog(QWidget* parent) : QDialog(parent), ui(
     ui->cbxGLResolution->setCurrentIndex(Config::GL_ScaleFactor-1);
 
     ui->cbBetterPolygons->setChecked(Config::GL_BetterPolygons != 0);
+    ui->cbxComputeHiResCoords->setChecked(Config::GL_HiresCoordinates != 0);
 
     if (!Config::ScreenVSync)
         ui->sbVSyncInterval->setEnabled(false);
     setVsyncControlEnable(UsesGL());
 
-    if (Config::_3DRenderer == 0)
+    setEnabled();
-    {
-        ui->cbGLDisplay->setEnabled(true);
-        ui->cbSoftwareThreaded->setEnabled(true);
-        ui->cbxGLResolution->setEnabled(false);
-        ui->cbBetterPolygons->setEnabled(false);
-    }
-    else
-    {
-        ui->cbGLDisplay->setEnabled(false);
-        ui->cbSoftwareThreaded->setEnabled(false);
-        ui->cbxGLResolution->setEnabled(true);
-        ui->cbBetterPolygons->setEnabled(true);
-    }
 }
 
 VideoSettingsDialog::~VideoSettingsDialog()
@@ -119,6 +119,7 @@ void VideoSettingsDialog::on_VideoSettingsDialog_rejected()
     Config::Threaded3D = oldSoftThreaded;
     Config::GL_ScaleFactor = oldGLScale;
     Config::GL_BetterPolygons = oldGLBetterPolygons;
+    Config::GL_HiresCoordinates = oldHiresCoordinates;
 
     emit updateVideoSettings(old_gl != UsesGL());
 
@@ -133,31 +134,18 @@ void VideoSettingsDialog::setVsyncControlEnable(bool hasOGL)
 
 void VideoSettingsDialog::onChange3DRenderer(int renderer)
 {
-    bool old_gl = (Config::ScreenUseGL != 0) || (Config::_3DRenderer != 0);
+    bool old_gl = UsesGL();
 
     Config::_3DRenderer = renderer;
 
-    if (renderer == 0)
+    setEnabled();
-    {
-        ui->cbGLDisplay->setEnabled(true);
-        ui->cbSoftwareThreaded->setEnabled(true);
-        ui->cbxGLResolution->setEnabled(false);
-        ui->cbBetterPolygons->setEnabled(false);
-    }
-    else
-    {
-        ui->cbGLDisplay->setEnabled(false);
-        ui->cbSoftwareThreaded->setEnabled(false);
-        ui->cbxGLResolution->setEnabled(true);
-        ui->cbBetterPolygons->setEnabled(true);
-    }
 
     emit updateVideoSettings(old_gl != UsesGL());
 }
 
 void VideoSettingsDialog::on_cbGLDisplay_stateChanged(int state)
 {
-    bool old_gl = (Config::ScreenUseGL != 0) || (Config::_3DRenderer != 0);
+    bool old_gl = UsesGL();
 
     Config::ScreenUseGL = (state != 0);
 
@@ -205,3 +193,10 @@ void VideoSettingsDialog::on_cbBetterPolygons_stateChanged(int state)
 
     emit updateVideoSettings(false);
 }
+
+void VideoSettingsDialog::on_cbxComputeHiResCoords_stateChanged(int state)
+{
+    Config::GL_HiresCoordinates = (state != 0);
+
+    emit updateVideoSettings(false);
+}

src/frontend/qt_sdl/VideoSettingsDialog.h

@@ -65,10 +65,12 @@ private slots:
 
     void on_cbxGLResolution_currentIndexChanged(int idx);
     void on_cbBetterPolygons_stateChanged(int state);
+    void on_cbxComputeHiResCoords_stateChanged(int state);
 
     void on_cbSoftwareThreaded_stateChanged(int state);
 private:
     void setVsyncControlEnable(bool hasOGL);
+    void setEnabled();
 
     Ui::VideoSettingsDialog* ui;
 
@@ -81,6 +83,7 @@ private:
     int oldSoftThreaded;
     int oldGLScale;
     int oldGLBetterPolygons;
+    int oldHiresCoordinates;
 };
 
 #endif // VIDEOSETTINGSDIALOG_H

src/frontend/qt_sdl/VideoSettingsDialog.ui

@@ -6,7 +6,7 @@
    <rect>
     <x>0</x>
     <y>0</y>
-    <width>408</width>
+    <width>427</width>
     <height>262</height>
    </rect>
   </property>
@@ -24,7 +24,7 @@
     <enum>QLayout::SetFixedSize</enum>
    </property>
    <property name="horizontalSpacing">
-    <number>-1</number>
+    <number>6</number>
    </property>
    <item row="1" column="1">
     <widget class="QGroupBox" name="groupBox_3">
@@ -39,13 +39,6 @@
         </property>
        </widget>
       </item>
-      <item row="1" column="0">
-       <widget class="QComboBox" name="cbxGLResolution">
-        <property name="whatsThis">
-         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;The resolution at which the 3D graphics will be rendered. Higher resolutions improve graphics quality when the main window is enlarged, but may also cause glitches.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
-        </property>
-       </widget>
-      </item>
       <item row="2" column="0">
        <widget class="QCheckBox" name="cbBetterPolygons">
         <property name="whatsThis">
@@ -56,6 +49,20 @@
         </property>
        </widget>
       </item>
+      <item row="1" column="0">
+       <widget class="QComboBox" name="cbxGLResolution">
+        <property name="whatsThis">
+         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;The resolution at which the 3D graphics will be rendered. Higher resolutions improve graphics quality when the main window is enlarged, but may also cause glitches.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
+        </property>
+       </widget>
+      </item>
+      <item row="3" column="0">
+       <widget class="QCheckBox" name="cbxComputeHiResCoords">
+        <property name="text">
+         <string>Use high resolution coordinates</string>
+        </property>
+       </widget>
+      </item>
      </layout>
     </widget>
    </item>
@@ -94,23 +101,7 @@
       <string>Display settings</string>
      </property>
      <layout class="QGridLayout" name="gridLayout_2">
-      <item row="6" column="0">
+      <item row="7" column="1">
-       <widget class="QLabel" name="label_2">
-        <property name="sizePolicy">
-         <sizepolicy hsizetype="Fixed" vsizetype="Fixed">
-          <horstretch>0</horstretch>
-          <verstretch>0</verstretch>
-         </sizepolicy>
-        </property>
-        <property name="whatsThis">
-         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;The interval at which to synchronize to the monitor's refresh rate. Set to 1 for a 60Hz monitor, 2 for 120Hz, ...&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
-        </property>
-        <property name="text">
-         <string>VSync interval:</string>
-        </property>
-       </widget>
-      </item>
-      <item row="6" column="1">
        <widget class="QSpinBox" name="sbVSyncInterval">
         <property name="whatsThis">
          <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;The interval at which to synchronize to the monitor's refresh rate. Set to 1 for a 60Hz monitor, 2 for 120Hz, ...&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
@@ -123,7 +114,7 @@
         </property>
        </widget>
       </item>
-      <item row="4" column="0" colspan="2">
+      <item row="5" column="0" colspan="2">
        <widget class="QCheckBox" name="cbGLDisplay">
         <property name="whatsThis">
          <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;Use OpenGL to draw the DS screens to the main window. May result in better frame pacing. Mandatory when using the OpenGL 3D renderer.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
@@ -133,17 +124,7 @@
         </property>
        </widget>
       </item>
-      <item row="5" column="0" colspan="2">
+      <item row="4" column="0" colspan="2">
-       <widget class="QCheckBox" name="cbVSync">
-        <property name="whatsThis">
-         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;When using OpenGL, synchronize the video output to your monitor's refresh rate.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
-        </property>
-        <property name="text">
-         <string>VSync</string>
-        </property>
-       </widget>
-      </item>
-      <item row="3" column="0" colspan="2">
        <spacer name="verticalSpacer">
         <property name="orientation">
          <enum>Qt::Vertical</enum>
@@ -159,13 +140,39 @@
         </property>
        </spacer>
       </item>
+      <item row="6" column="0" colspan="2">
+       <widget class="QCheckBox" name="cbVSync">
+        <property name="whatsThis">
+         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;When using OpenGL, synchronize the video output to your monitor's refresh rate.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
+        </property>
+        <property name="text">
+         <string>VSync</string>
+        </property>
+       </widget>
+      </item>
+      <item row="7" column="0">
+       <widget class="QLabel" name="label_2">
+        <property name="sizePolicy">
+         <sizepolicy hsizetype="Fixed" vsizetype="Fixed">
+          <horstretch>0</horstretch>
+          <verstretch>0</verstretch>
+         </sizepolicy>
+        </property>
+        <property name="whatsThis">
+         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;The interval at which to synchronize to the monitor's refresh rate. Set to 1 for a 60Hz monitor, 2 for 120Hz, ...&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
+        </property>
+        <property name="text">
+         <string>VSync interval:</string>
+        </property>
+       </widget>
+      </item>
       <item row="2" column="0" colspan="2">
        <widget class="QRadioButton" name="rb3DOpenGL">
         <property name="whatsThis">
          <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;The OpenGL renderer may be faster than software and supports graphical enhancements, but is more prone to glitches.&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
         </property>
         <property name="text">
-         <string>OpenGL</string>
+         <string>OpenGL (Classic)</string>
         </property>
        </widget>
       </item>
@@ -186,6 +193,13 @@
         </property>
        </widget>
       </item>
+      <item row="3" column="0">
+       <widget class="QRadioButton" name="rb3DCompute">
+        <property name="text">
+         <string>OpenGL (Compute shader)</string>
+        </property>
+       </widget>
+      </item>
      </layout>
     </widget>
    </item>

src/frontend/qt_sdl/Window.cpp

@@ -2048,6 +2048,7 @@ void MainWindow::onUpdateVideoSettings(bool glchange)
         connect(emuThread, SIGNAL(windowUpdate()), panel, SLOT(repaint()));
     }
 
+    printf("update video settings\n");
     videoSettingsDirty = true;
 
     if (glchange)

src/frontend/qt_sdl/main.cpp

@@ -175,10 +175,6 @@ bool camStarted[2];
 //extern int AspectRatiosNum;
 
 
-
-
-
-
 static bool FileExtensionInList(const QString& filename, const QStringList& extensions, Qt::CaseSensitivity cs = Qt::CaseInsensitive)
 {
     return std::any_of(extensions.cbegin(), extensions.cend(), [&](const auto& ext) {
@@ -339,10 +335,10 @@ int main(int argc, char** argv)
 
     if (!Config::Load()) QMessageBox::critical(NULL, "melonDS", "Unable to write to config.\nPlease check the write permissions of the folder you placed melonDS in.");
 
-#define SANITIZE(var, min, max)  { var = std::clamp(var, min, max); }
+#define SANITIZE(var, min, max)  { var = std::clamp<int>(var, min, max); }
     SANITIZE(Config::ConsoleType, 0, 1);
 #ifdef OGLRENDERER_ENABLED
-    SANITIZE(Config::_3DRenderer, 0, 1); // 0 is the software renderer, 1 is the OpenGL renderer
+    SANITIZE(Config::_3DRenderer, 0, renderer3D_Max);
 #else
     SANITIZE(Config::_3DRenderer, 0, 0);
 #endif