piwalker/melonDS
1
0
Fork 0
mirror of https://github.com/melonDS-emu/melonDS.git synced 2025-07-09 15:29:43 -06:00
Code Issues Packages Projects Releases Wiki Activity

Tie BinCombined's local_size_x to the coarse tiles' area

Browse Source
This commit is contained in:
FireNX70
2024-06-13 00:30:33 +02:00
parent 57b5e16e6d
commit f097cc3b4e
3 changed files with 6 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/GPU3D_Compute.cpp
View File

@ -57,6 +57,8 @@ bool ComputeRenderer::CompileShader(GLuint& shader, const std::string& source, c
shaderSource += std::to_string(TileSize); shaderSource += std::to_string(TileSize);
shaderSource += "\nconst int CoarseTileCountY = "; shaderSource += "\nconst int CoarseTileCountY = ";
shaderSource += std::to_string(CoarseTileCountY) + ";"; shaderSource += std::to_string(CoarseTileCountY) + ";";
shaderSource += "\n#define CoarseTileArea ";
shaderSource += std::to_string(CoarseTileArea);
shaderSource += ComputeRendererShaders::Common; shaderSource += ComputeRendererShaders::Common;
shaderSource += source; shaderSource += source;
@ -330,6 +332,7 @@ void ComputeRenderer::SetRenderSettings(int scale, bool highResolutionCoordinate
TileSize = std::min(8 * TileScale, 32); TileSize = std::min(8 * TileScale, 32);
CoarseTileCountY = TileSize < 32 ? 4 : 6; CoarseTileCountY = TileSize < 32 ? 4 : 6;
CoarseTileArea = CoarseTileCountX * CoarseTileCountY;
CoarseTileW = CoarseTileCountX * TileSize; CoarseTileW = CoarseTileCountX * TileSize;
CoarseTileH = CoarseTileCountY * TileSize; CoarseTileH = CoarseTileCountY * TileSize;
@ -959,7 +962,7 @@ void ComputeRenderer::RenderFrame(GPU& gpu)
// bin polygons // bin polygons
glUseProgram(ShaderBinCombined); 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); glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
// calculate list offsets // calculate list offsets

1
src/GPU3D_Compute.h
View File

@ -166,6 +166,7 @@ private:
int TileSize; int TileSize;
static constexpr int CoarseTileCountX = 8; static constexpr int CoarseTileCountX = 8;
int CoarseTileCountY; int CoarseTileCountY;
int CoarseTileArea;
int CoarseTileW; int CoarseTileW;
int CoarseTileH; int CoarseTileH;

2
src/GPU3D_Compute_shaders.h
View File

@ -862,7 +862,7 @@ const std::string BinCombined =
XSpanSetupBuffer + XSpanSetupBuffer +
WorkDescBuffer + R"( WorkDescBuffer + R"(
layout (local_size_x = 32) in; layout (local_size_x = CoarseTileArea) in;
bool BinPolygon(Polygon polygon, ivec2 topLeft, ivec2 botRight) bool BinPolygon(Polygon polygon, ivec2 topLeft, ivec2 botRight)
{ {