File melonds_PR2065.patch of Package melonds
From 57d20751fcd1ff52fe9f8890d565c65c5b6abc13 Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Sun, 9 Jun 2024 18:47:43 +0200
Subject: [PATCH 1/9] Fix glMemoryBarrier flags
---
src/GPU3D_Compute.cpp | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index da2559507..027d68066 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -932,23 +932,23 @@ void ComputeRenderer::RenderFrame(GPU& gpu)
glBindImageTexture(0, YSpanIndicesTexture, 0, GL_FALSE, 0, GL_READ_ONLY, GL_RGBA16UI);
glUseProgram(ShaderInterpXSpans[wbuffer]);
glDispatchCompute((numSetupIndices + 31) / 32, 1, 1);
- glMemoryBarrier(GL_SHADER_STORAGE_BUFFER);
+ glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
// bin polygons
glUseProgram(ShaderBinCombined);
glDispatchCompute(((gpu.GPU3D.RenderNumPolygons + 31) / 32), ScreenWidth/CoarseTileW, ScreenHeight/CoarseTileH);
- glMemoryBarrier(GL_SHADER_STORAGE_BUFFER);
+ glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
// calculate list offsets
glUseProgram(ShaderCalculateWorkListOffset);
glDispatchCompute((numVariants + 31) / 32, 1, 1);
- glMemoryBarrier(GL_SHADER_STORAGE_BUFFER);
+ glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
// sort shader work
glUseProgram(ShaderSortWork);
glBindBuffer(GL_DISPATCH_INDIRECT_BUFFER, BinResultMemory);
glDispatchComputeIndirect(offsetof(BinResultHeader, SortWorkWorkCount));
- glMemoryBarrier(GL_SHADER_STORAGE_BUFFER);
+ glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
glActiveTexture(GL_TEXTURE0);
From 528501149e21e380898dd77a4c527087f0d8651e Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Sun, 9 Jun 2024 20:07:26 +0200
Subject: [PATCH 2/9] Scale TileSize with internal resolution
---
src/GPU3D_Compute.cpp | 7 +++++++
src/GPU3D_Compute.h | 6 +++---
src/GPU3D_Compute_shaders.h | 1 -
3 files changed, 10 insertions(+), 4 deletions(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index 027d68066..6d6115fe1 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -19,6 +19,7 @@
#include "GPU3D_Compute.h"
#include <assert.h>
+#include <algorithm>
#include "OpenGLSupport.h"
@@ -50,6 +51,8 @@ bool ComputeRenderer::CompileShader(GLuint& shader, const std::string& source, c
shaderSource += std::to_string(ScreenHeight);
shaderSource += "\n#define MaxWorkTiles ";
shaderSource += std::to_string(MaxWorkTiles);
+ shaderSource += "\n#define TileSize ";
+ shaderSource += std::to_string(TileSize);
shaderSource += ComputeRendererShaders::Common;
shaderSource += source;
@@ -310,6 +313,10 @@ void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinate
ScreenWidth = 256 * ScaleFactor;
ScreenHeight = 192 * ScaleFactor;
+ TileSize = std::min(8 * (1 << (ScaleFactor / 5)), 32);
+ CoarseTileW = CoarseTileCountX * TileSize;
+ CoarseTileH = CoarseTileCountY * TileSize;
+
TilesPerLine = ScreenWidth/TileSize;
TileLines = ScreenHeight/TileSize;
diff --git a/src/GPU3D_Compute.h b/src/GPU3D_Compute.h
index 7544c09e0..1e225948b 100644
--- a/src/GPU3D_Compute.h
+++ b/src/GPU3D_Compute.h
@@ -163,11 +163,11 @@ class ComputeRenderer : public Renderer3D
float TextureLayer;
};
- static constexpr int TileSize = 8;
+ int TileSize;
static constexpr int CoarseTileCountX = 8;
static constexpr int CoarseTileCountY = 4;
- static constexpr int CoarseTileW = CoarseTileCountX * TileSize;
- static constexpr int CoarseTileH = CoarseTileCountY * TileSize;
+ int CoarseTileW;
+ int CoarseTileH;
static constexpr int BinStride = 2048/32;
static constexpr int CoarseBinStride = BinStride/32;
diff --git a/src/GPU3D_Compute_shaders.h b/src/GPU3D_Compute_shaders.h
index 572f9ad66..0ad57dbd1 100644
--- a/src/GPU3D_Compute_shaders.h
+++ b/src/GPU3D_Compute_shaders.h
@@ -339,7 +339,6 @@ const uint ResultAttrStart = ResultDepthStart+ScreenWidth*ScreenHeight*2;
const char* Common = R"(
-#define TileSize 8
const int CoarseTileCountX = 8;
const int CoarseTileCountY = 4;
const int CoarseTileW = (CoarseTileCountX * TileSize);
From 078b4c43f4af165279e0fd235abb3cb241b2b20b Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Sun, 9 Jun 2024 21:46:44 +0200
Subject: [PATCH 3/9] Clean up tile size calc
---
src/GPU3D_Compute.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index 6d6115fe1..8e0ea4f36 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -313,7 +313,7 @@ void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinate
ScreenWidth = 256 * ScaleFactor;
ScreenHeight = 192 * ScaleFactor;
- TileSize = std::min(8 * (1 << (ScaleFactor / 5)), 32);
+ TileSize = std::min(8 << (ScaleFactor / 5), 32);
CoarseTileW = CoarseTileCountX * TileSize;
CoarseTileH = CoarseTileCountY * TileSize;
From 7433388beb6662673b021a1c4b0ceba9513bd156 Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Mon, 10 Jun 2024 20:15:05 +0200
Subject: [PATCH 4/9] Use a different CoarseTileCountY with size 32 tiles
---
src/GPU3D_Compute.cpp | 3 +++
src/GPU3D_Compute.h | 2 +-
src/GPU3D_Compute_shaders.h | 1 -
3 files changed, 4 insertions(+), 2 deletions(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index 8e0ea4f36..294cd5180 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -53,6 +53,8 @@ bool ComputeRenderer::CompileShader(GLuint& shader, const std::string& source, c
shaderSource += std::to_string(MaxWorkTiles);
shaderSource += "\n#define TileSize ";
shaderSource += std::to_string(TileSize);
+ shaderSource += "\nconst int CoarseTileCountY = ";
+ shaderSource += std::to_string(CoarseTileCountY) + ";";
shaderSource += ComputeRendererShaders::Common;
shaderSource += source;
@@ -314,6 +316,7 @@ void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinate
ScreenHeight = 192 * ScaleFactor;
TileSize = std::min(8 << (ScaleFactor / 5), 32);
+ CoarseTileCountY = TileSize < 32 ? 4 : 6;
CoarseTileW = CoarseTileCountX * TileSize;
CoarseTileH = CoarseTileCountY * TileSize;
diff --git a/src/GPU3D_Compute.h b/src/GPU3D_Compute.h
index 1e225948b..4f944e42f 100644
--- a/src/GPU3D_Compute.h
+++ b/src/GPU3D_Compute.h
@@ -165,7 +165,7 @@ class ComputeRenderer : public Renderer3D
int TileSize;
static constexpr int CoarseTileCountX = 8;
- static constexpr int CoarseTileCountY = 4;
+ int CoarseTileCountY;
int CoarseTileW;
int CoarseTileH;
diff --git a/src/GPU3D_Compute_shaders.h b/src/GPU3D_Compute_shaders.h
index 0ad57dbd1..9fb4aae76 100644
--- a/src/GPU3D_Compute_shaders.h
+++ b/src/GPU3D_Compute_shaders.h
@@ -340,7 +340,6 @@ const uint ResultAttrStart = ResultDepthStart+ScreenWidth*ScreenHeight*2;
const char* Common = R"(
const int CoarseTileCountX = 8;
-const int CoarseTileCountY = 4;
const int CoarseTileW = (CoarseTileCountX * TileSize);
const int CoarseTileH = (CoarseTileCountY * TileSize);
From d24079e693befa45e0ea1d89d70f48adf25614a8 Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Tue, 11 Jun 2024 15:56:24 +0200
Subject: [PATCH 5/9] Better tile scaling calc
---
src/GPU3D_Compute.cpp | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index 294cd5180..9d5a23785 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -302,6 +302,8 @@ void ComputeRenderer::Reset(GPU& gpu)
void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinates)
{
+ unsigned char TileScale;
+
CurGLCompositor.SetScaleFactor(scale);
if (ScaleFactor != -1)
@@ -315,7 +317,13 @@ void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinate
ScreenWidth = 256 * ScaleFactor;
ScreenHeight = 192 * ScaleFactor;
- TileSize = std::min(8 << (ScaleFactor / 5), 32);
+ //Starting at 4.5x we want to double TileSize every time scale doubles
+ TileScale = 2 * ScaleFactor / 9;
+ TileScale &= ~(TileScale >> 1);
+ TileScale <<= 1;
+ TileScale += TileScale == 0;
+
+ TileSize = std::min(8 * TileScale, 32);
CoarseTileCountY = TileSize < 32 ? 4 : 6;
CoarseTileW = CoarseTileCountX * TileSize;
CoarseTileH = CoarseTileCountY * TileSize;
From 57b5e16e6dfbeb0e3d53e32f2a32e654e2fb3203 Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Tue, 11 Jun 2024 17:22:36 +0200
Subject: [PATCH 6/9] Actually good way to get the MS bit
---
src/GPU3D_Compute.cpp | 9 +++++++--
src/Utils.h | 14 ++++++++++++++
2 files changed, 21 insertions(+), 2 deletions(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index 9d5a23785..c22a66b96 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -21,6 +21,8 @@
#include <assert.h>
#include <algorithm>
+#include "Utils.h"
+
#include "OpenGLSupport.h"
#include "GPU3D_Compute_shaders.h"
@@ -302,7 +304,7 @@ void ComputeRenderer::Reset(GPU& gpu)
void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinates)
{
- unsigned char TileScale;
+ u8 TileScale;
CurGLCompositor.SetScaleFactor(scale);
@@ -319,9 +321,12 @@ void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinate
//Starting at 4.5x we want to double TileSize every time scale doubles
TileScale = 2 * ScaleFactor / 9;
- TileScale &= ~(TileScale >> 1);
+ TileScale = GetMSBit(TileScale);
TileScale <<= 1;
TileScale += TileScale == 0;
+
+ std::printf("Scale: %d\n", ScaleFactor);
+ std::printf("TileScale: %d\n", TileScale);
TileSize = std::min(8 * TileScale, 32);
CoarseTileCountY = TileSize < 32 ? 4 : 6;
diff --git a/src/Utils.h b/src/Utils.h
index 63be217b8..e6444c51f 100644
--- a/src/Utils.h
+++ b/src/Utils.h
@@ -38,6 +38,20 @@ std::pair<std::unique_ptr<u8[]>, u32> PadToPowerOf2(const u8* data, u32 len) noe
std::unique_ptr<u8[]> CopyToUnique(const u8* data, u32 len) noexcept;
+template <typename T>
+T GetMSBit(T val)
+{
+ val |= (val >> 1);
+ val |= (val >> 2);
+ val |= (val >> 4);
+
+ if constexpr(sizeof(val) > 1) val |= (val >> 8);
+ if constexpr(sizeof(val) > 2) val |= (val >> 16);
+ if constexpr(sizeof(val) > 4) val |= (val >> 32);
+
+ return val - (val >> 1);
+}
+
}
#endif // MELONDS_UTILS_H
From f097cc3b4e93bcd9b5db73d6be68d21dcfb85955 Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Thu, 13 Jun 2024 00:30:33 +0200
Subject: [PATCH 7/9] Tie BinCombined's local_size_x to the coarse tiles' area
---
src/GPU3D_Compute.cpp | 5 ++++-
src/GPU3D_Compute.h | 1 +
src/GPU3D_Compute_shaders.h | 2 +-
3 files changed, 6 insertions(+), 2 deletions(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index c22a66b96..ecf485440 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -57,6 +57,8 @@ bool ComputeRenderer::CompileShader(GLuint& shader, const std::string& source, c
shaderSource += std::to_string(TileSize);
shaderSource += "\nconst int CoarseTileCountY = ";
shaderSource += std::to_string(CoarseTileCountY) + ";";
+ shaderSource += "\n#define CoarseTileArea ";
+ shaderSource += std::to_string(CoarseTileArea);
shaderSource += ComputeRendererShaders::Common;
shaderSource += source;
@@ -330,6 +332,7 @@ void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinate
TileSize = std::min(8 * TileScale, 32);
CoarseTileCountY = TileSize < 32 ? 4 : 6;
+ CoarseTileArea = CoarseTileCountX * CoarseTileCountY;
CoarseTileW = CoarseTileCountX * TileSize;
CoarseTileH = CoarseTileCountY * TileSize;
@@ -959,7 +962,7 @@ void ComputeRenderer::RenderFrame(GPU& gpu)
// bin polygons
glUseProgram(ShaderBinCombined);
- glDispatchCompute(((gpu.GPU3D.RenderNumPolygons + 31) / 32), ScreenWidth/CoarseTileW, ScreenHeight/CoarseTileH);
+ glDispatchCompute(((gpu.GPU3D.RenderNumPolygons + CoarseTileArea - 1) / CoarseTileArea), ScreenWidth/CoarseTileW, ScreenHeight/CoarseTileH);
glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
// calculate list offsets
diff --git a/src/GPU3D_Compute.h b/src/GPU3D_Compute.h
index 4f944e42f..6a5fd499c 100644
--- a/src/GPU3D_Compute.h
+++ b/src/GPU3D_Compute.h
@@ -166,6 +166,7 @@ class ComputeRenderer : public Renderer3D
int TileSize;
static constexpr int CoarseTileCountX = 8;
int CoarseTileCountY;
+ int CoarseTileArea;
int CoarseTileW;
int CoarseTileH;
diff --git a/src/GPU3D_Compute_shaders.h b/src/GPU3D_Compute_shaders.h
index 9fb4aae76..9b3190295 100644
--- a/src/GPU3D_Compute_shaders.h
+++ b/src/GPU3D_Compute_shaders.h
@@ -862,7 +862,7 @@ const std::string BinCombined =
XSpanSetupBuffer +
WorkDescBuffer + R"(
-layout (local_size_x = 32) in;
+layout (local_size_x = CoarseTileArea) in;
bool BinPolygon(Polygon polygon, ivec2 topLeft, ivec2 botRight)
{
From 9c87d9998f1ad793b7ea442aed8407851d65b750 Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Thu, 13 Jun 2024 00:46:05 +0200
Subject: [PATCH 8/9] Work count X is unrelated to the local size here
---
src/GPU3D_Compute.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index ecf485440..93aac5ce3 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -962,7 +962,7 @@ void ComputeRenderer::RenderFrame(GPU& gpu)
// bin polygons
glUseProgram(ShaderBinCombined);
- glDispatchCompute(((gpu.GPU3D.RenderNumPolygons + CoarseTileArea - 1) / CoarseTileArea), ScreenWidth/CoarseTileW, ScreenHeight/CoarseTileH);
+ glDispatchCompute(((gpu.GPU3D.RenderNumPolygons + 31) / 32), ScreenWidth/CoarseTileW, ScreenHeight/CoarseTileH);
glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
// calculate list offsets
From 6f8ce9fe897fab0277754209122df4181fb4a177 Mon Sep 17 00:00:00 2001
From: FireNX70 <firenx70@gmail.com>
Date: Thu, 13 Jun 2024 07:57:49 +0200
Subject: [PATCH 9/9] Adjust ClearCoarseBinMask's local size according to
TileSize
---
src/GPU3D_Compute.cpp | 5 ++++-
src/GPU3D_Compute.h | 1 +
src/GPU3D_Compute_shaders.h | 2 +-
3 files changed, 6 insertions(+), 2 deletions(-)
diff --git a/src/GPU3D_Compute.cpp b/src/GPU3D_Compute.cpp
index 93aac5ce3..16a3d80a8 100644
--- a/src/GPU3D_Compute.cpp
+++ b/src/GPU3D_Compute.cpp
@@ -59,6 +59,8 @@ bool ComputeRenderer::CompileShader(GLuint& shader, const std::string& source, c
shaderSource += std::to_string(CoarseTileCountY) + ";";
shaderSource += "\n#define CoarseTileArea ";
shaderSource += std::to_string(CoarseTileArea);
+ shaderSource += "\n#define ClearCoarseBinMaskLocalSize ";
+ shaderSource += std::to_string(ClearCoarseBinMaskLocalSize);
shaderSource += ComputeRendererShaders::Common;
shaderSource += source;
@@ -332,6 +334,7 @@ void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinate
TileSize = std::min(8 * TileScale, 32);
CoarseTileCountY = TileSize < 32 ? 4 : 6;
+ ClearCoarseBinMaskLocalSize = TileSize < 32 ? 64 : 48;
CoarseTileArea = CoarseTileCountX * CoarseTileCountY;
CoarseTileW = CoarseTileCountX * TileSize;
CoarseTileH = CoarseTileCountY * TileSize;
@@ -944,7 +947,7 @@ void ComputeRenderer::RenderFrame(GPU& gpu)
glBindBufferBase(GL_UNIFORM_BUFFER, 0, MetaUniformMemory);
glUseProgram(ShaderClearCoarseBinMask);
- glDispatchCompute(TilesPerLine*TileLines/32, 1, 1);
+ glDispatchCompute(TilesPerLine*TileLines/ClearCoarseBinMaskLocalSize, 1, 1);
bool wbuffer = false;
if (numYSpans > 0)
diff --git a/src/GPU3D_Compute.h b/src/GPU3D_Compute.h
index 6a5fd499c..30766ec7a 100644
--- a/src/GPU3D_Compute.h
+++ b/src/GPU3D_Compute.h
@@ -169,6 +169,7 @@ class ComputeRenderer : public Renderer3D
int CoarseTileArea;
int CoarseTileW;
int CoarseTileH;
+ int ClearCoarseBinMaskLocalSize;
static constexpr int BinStride = 2048/32;
static constexpr int CoarseBinStride = BinStride/32;
diff --git a/src/GPU3D_Compute_shaders.h b/src/GPU3D_Compute_shaders.h
index 9b3190295..556346109 100644
--- a/src/GPU3D_Compute_shaders.h
+++ b/src/GPU3D_Compute_shaders.h
@@ -846,7 +846,7 @@ void main()
const std::string ClearCoarseBinMask =
BinningBuffer + R"(
-layout (local_size_x = 32) in;
+layout (local_size_x = ClearCoarseBinMaskLocalSize) in;
void main()
{
