piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-21 13:20:27 -06:00
Code Issues Packages Projects Releases Wiki Activity

this is a pure optimization commit:

Browse Source 
return to the  old values in constants in pixelshader, the old values give mi 3 or 4 fps more,(maybe some kind of compiler optimization) in some games and with the current algorithm i notice no difference with this values, please report any problem.
optimizes SSAA to make it a little faster and, the quality should be the same but with a little speedup.
change the way the frame is processed depending is xfb is enabled or not to make this a little faster.
please test and report any problem

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5820 8ced0084-cf51-0410-be5f-012b33b47a6e
This commit is contained in:
Rodolfo Osvaldo Bogado
2010-07-02 17:09:53 +00:00
parent 6cacc52ebf
commit 12a676c273
14 changed files with 312 additions and 243 deletions
Download Patch File Download Diff File Expand all files Collapse all files

84
Source/Plugins/Plugin_VideoDX9/Src/PixelShaderCache.cpp
View File

@ -46,24 +46,26 @@ static std::set<u32> unique_shaders;
static float lastPSconstants[C_COLORMATRIX+16][4];
static LPDIRECT3DPIXELSHADER9 s_ColorMatrixProgram[3];
static LPDIRECT3DPIXELSHADER9 s_ColorCopyProgram[3];
static LPDIRECT3DPIXELSHADER9 s_DepthMatrixProgram[3];
#define MAX_SSAA_SHADERS 3
static LPDIRECT3DPIXELSHADER9 s_ColorMatrixProgram[MAX_SSAA_SHADERS];
static LPDIRECT3DPIXELSHADER9 s_ColorCopyProgram[MAX_SSAA_SHADERS];
static LPDIRECT3DPIXELSHADER9 s_DepthMatrixProgram[MAX_SSAA_SHADERS];
static LPDIRECT3DPIXELSHADER9 s_ClearProgram = 0;
LPDIRECT3DPIXELSHADER9 PixelShaderCache::GetColorMatrixProgram(int SSAAMode)
{
return s_ColorMatrixProgram[SSAAMode % 3];
return s_ColorMatrixProgram[SSAAMode % MAX_SSAA_SHADERS];
}
LPDIRECT3DPIXELSHADER9 PixelShaderCache::GetDepthMatrixProgram(int SSAAMode)
{
return s_DepthMatrixProgram[SSAAMode % 3];
return s_DepthMatrixProgram[SSAAMode % MAX_SSAA_SHADERS];
}
LPDIRECT3DPIXELSHADER9 PixelShaderCache::GetColorCopyProgram(int SSAAMode)
{
return s_ColorCopyProgram[SSAAMode % 3];
return s_ColorCopyProgram[SSAAMode % MAX_SSAA_SHADERS];
}
LPDIRECT3DPIXELSHADER9 PixelShaderCache::GetClearProgram()
@ -134,34 +136,32 @@ void PixelShaderCache::Init()
"in float2 uv0 : TEXCOORD0){\n"
"ocol0 = tex2D(samp0,uv0);\n"
"}\n");
s_ColorCopyProgram[0] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
s_ColorCopyProgram[0] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
//4 Samples
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
//1 Samples SSAA
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
"void main(\n"
"out float4 ocol0 : COLOR0,\n"
"in float4 uv0 : TEXCOORD0,\n"
"in float4 uv1 : TEXCOORD1,\n"
"in float4 uv2 : TEXCOORD2,\n"
"in float4 uv3 : TEXCOORD3,\n"
"in float4 uv4 : TEXCOORD4){\n"
"ocol0 = (tex2D(samp0,uv1.xy) + tex2D(samp0,uv2.xy) + tex2D(samp0,uv3.xy) + tex2D(samp0,uv4.xy))*0.25f;\n"
"in float4 uv1 : TEXCOORD1){\n"
"ocol0 = tex2D(samp0,uv0.xy);\n"
"}\n");
s_ColorCopyProgram[1] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
//9 Samples
//4 Samples SSAA
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
"void main(\n"
"out float4 ocol0 : COLOR0,\n"
"in float4 uv0 : TEXCOORD0,\n"
"in float4 uv1 : TEXCOORD1,\n"
"in float4 uv2 : TEXCOORD2,\n"
"in float4 uv3 : TEXCOORD3,\n"
"in float4 uv4 : TEXCOORD4){\n"
"ocol0 = (tex2D(samp0,uv1.xy) + tex2D(samp0,uv1.wz) + tex2D(samp0,uv2.xy) + tex2D(samp0,uv2.wz) + tex2D(samp0,uv3.xy) + tex2D(samp0,uv3.wz) + tex2D(samp0,uv4.xy) + tex2D(samp0,uv4.wz) + tex2D(samp0,uv0.xy))/9.0f;\n"
"in float4 uv3 : TEXCOORD3){\n"
"ocol0 = (tex2D(samp0,uv1.xy) + tex2D(samp0,uv1.wz) + tex2D(samp0,uv2.xy) + tex2D(samp0,uv2.wz))*0.25;\n"
"}\n");
s_ColorCopyProgram[2] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
//Color conversion Programs
//1 sample
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
@ -172,25 +172,21 @@ void PixelShaderCache::Init()
"float4 texcol = tex2D(samp0,uv0);\n"
"ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
"}\n",C_COLORMATRIX);
s_ColorMatrixProgram[0] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
s_ColorMatrixProgram[0] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
//4 samples
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
//1 samples SSAA
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
"uniform float4 cColMatrix[5] : register(c%d);\n"
"void main(\n"
"out float4 ocol0 : COLOR0,\n"
"in float4 uv0 : TEXCOORD0,\n"
"in float4 uv1 : TEXCOORD1,\n"
"in float4 uv2 : TEXCOORD2,\n"
"in float4 uv3 : TEXCOORD3,\n"
"in float4 uv4 : TEXCOORD4,\n"
"in float4 uv5 : TEXCOORD5){\n"
"float4 texcol = (tex2D(samp0,float2(clamp(uv1.x,uv5.x,uv5.z),clamp(uv1.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv2.x,uv5.x,uv5.z),clamp(uv2.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv3.x,uv5.x,uv5.z),clamp(uv3.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv4.x,uv5.x,uv5.z),clamp(uv4.y,uv5.y,uv5.w))))*0.25f;\n"
"in float4 uv1 : TEXCOORD1){\n"
"float4 texcol = tex2D(samp0,uv0.xy);\n"
"ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
"}\n",C_COLORMATRIX);
s_ColorMatrixProgram[1] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
//9 samples
//4 samples SSAA
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
"uniform float4 cColMatrix[5] : register(c%d);\n"
"void main(\n"
@ -198,10 +194,8 @@ void PixelShaderCache::Init()
"in float4 uv0 : TEXCOORD0,\n"
"in float4 uv1 : TEXCOORD1,\n"
"in float4 uv2 : TEXCOORD2,\n"
"in float4 uv3 : TEXCOORD3,\n"
"in float4 uv4 : TEXCOORD4,\n"
"in float4 uv5 : TEXCOORD5){\n"
"float4 texcol = (tex2D(samp0,float2(clamp(uv1.x,uv5.x,uv5.z),clamp(uv1.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv1.w,uv5.x,uv5.z),clamp(uv1.z,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv2.x,uv5.x,uv5.z),clamp(uv2.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv2.w,uv5.x,uv5.z),clamp(uv2.z,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv3.x,uv5.x,uv5.z),clamp(uv3.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv3.w,uv5.x,uv5.z),clamp(uv3.z,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv4.x,uv5.x,uv5.z),clamp(uv4.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv4.w,uv5.x,uv5.z),clamp(uv4.z,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv0.x,uv5.x,uv5.z),clamp(uv0.y,uv5.y,uv5.w))))/9;\n"
"in float4 uv3 : TEXCOORD3){\n"
"float4 texcol = (tex2D(samp0,uv1.xy) + tex2D(samp0,uv1.wz) + tex2D(samp0,uv2.xy) + tex2D(samp0,uv2.wz))*0.25f;\n"
"ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
"}\n",C_COLORMATRIX);
s_ColorMatrixProgram[2] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
@ -216,29 +210,25 @@ void PixelShaderCache::Init()
"float4 texcol = tex2D(samp0,uv0);\n"
"float4 EncodedDepth = frac((texcol.r * (16777215.0f/16777216.0f)) * float4(1.0f,255.0f,255.0f*255.0f,255.0f*255.0f*255.0f));\n"
"texcol = float4((EncodedDepth.rgb * (16777216.0f/16777215.0f)),1.0f);\n"
"ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
"ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
"}\n",C_COLORMATRIX);
s_DepthMatrixProgram[0] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
s_DepthMatrixProgram[0] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
//4 sample
//1 sample SSAA
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
"uniform float4 cColMatrix[5] : register(c%d);\n"
"void main(\n"
"out float4 ocol0 : COLOR0,\n"
"in float4 uv0 : TEXCOORD0,\n"
"in float4 uv1 : TEXCOORD1,\n"
"in float4 uv2 : TEXCOORD2,\n"
"in float4 uv3 : TEXCOORD3,\n"
"in float4 uv4 : TEXCOORD4,\n"
"in float4 uv5 : TEXCOORD5){\n"
"float4 texcol = (tex2D(samp0,float2(clamp(uv1.x,uv5.x,uv5.z),clamp(uv1.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv2.x,uv5.x,uv5.z),clamp(uv2.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv3.x,uv5.x,uv5.z),clamp(uv3.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv4.x,uv5.x,uv5.z),clamp(uv4.y,uv5.y,uv5.w))))*0.25f;\n"
"in float4 uv1 : TEXCOORD1){\n"
"float4 texcol = tex2D(samp0,uv0.xy);\n"
"float4 EncodedDepth = frac((texcol.r * (16777215.0f/16777216.0f)) * float4(1.0f,255.0f,255.0f*255.0f,255.0f*255.0f*255.0f));\n"
"texcol = float4((EncodedDepth.rgb * (16777216.0f/16777215.0f)),1.0f);\n"
"ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
"}\n",C_COLORMATRIX);
s_DepthMatrixProgram[1] = D3D::CompileAndCreatePixelShader(pprog, (int)strlen(pprog));
//9 sample
//4 sample SSAA
sprintf(pprog, "uniform sampler samp0 : register(s0);\n"
"uniform float4 cColMatrix[5] : register(c%d);\n"
"void main(\n"
@ -246,10 +236,8 @@ void PixelShaderCache::Init()
"in float4 uv0 : TEXCOORD0,\n"
"in float4 uv1 : TEXCOORD1,\n"
"in float4 uv2 : TEXCOORD2,\n"
"in float4 uv3 : TEXCOORD3,\n"
"in float4 uv4 : TEXCOORD4,\n"
"in float4 uv5 : TEXCOORD5){\n"
"float4 texcol = (tex2D(samp0,float2(clamp(uv1.x,uv5.x,uv5.z),clamp(uv1.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv1.w,uv5.x,uv5.z),clamp(uv1.z,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv2.x,uv5.x,uv5.z),clamp(uv2.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv2.w,uv5.x,uv5.z),clamp(uv2.z,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv3.x,uv5.x,uv5.z),clamp(uv3.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv3.w,uv5.x,uv5.z),clamp(uv3.z,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv4.x,uv5.x,uv5.z),clamp(uv4.y,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv4.w,uv5.x,uv5.z),clamp(uv4.z,uv5.y,uv5.w))) + tex2D(samp0,float2(clamp(uv0.x,uv5.x,uv5.z),clamp(uv0.y,uv5.y,uv5.w))))/9;\n"
"in float4 uv3 : TEXCOORD3){\n"
"float4 texcol = (tex2D(samp0,uv1.xy) + tex2D(samp0,uv1.wz) + tex2D(samp0,uv2.xy) + tex2D(samp0,uv2.wz))*0.25f;\n"
"float4 EncodedDepth = frac((texcol.r * (16777215.0f/16777216.0f)) * float4(1.0f,255.0f,255.0f*255.0f,255.0f*255.0f*255.0f));\n"
"texcol = float4((EncodedDepth.rgb * (16777216.0f/16777215.0f)),1.0f);\n"
"ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
@ -285,7 +273,7 @@ void PixelShaderCache::Clear()
void PixelShaderCache::Shutdown()
{
for(int i = 0;i<3;i++)
for(int i = 0;i < MAX_SSAA_SHADERS; i++)
{
if (s_ColorMatrixProgram[i]) s_ColorMatrixProgram[i]->Release();
s_ColorMatrixProgram[i] = NULL;