first try of primitive restart index generator

Convert all quads+triangles into trangle_strip and uses primitive restart to split them.
Speed up triangle_strip, but slows down all others primitive formats.
Only implemented in ogl.
This commit is contained in:
degasus 2013-03-29 14:27:33 +01:00
parent c743e75d92
commit ca8554e7d1
8 changed files with 102 additions and 33 deletions

View File

@ -18,6 +18,7 @@
#include <cstddef>
#include "Common.h"
#include "VideoConfig.h"
#include "IndexGenerator.h"
/*
@ -90,6 +91,8 @@ __forceinline void IndexGenerator::WriteTriangle(u32 index1, u32 index2, u32 ind
*Tptr++ = index1;
*Tptr++ = index2;
*Tptr++ = index3;
if(g_Config.backend_info.bSupportsPrimitiveRestart)
*Tptr++ = 65535;
++numT;
}
@ -105,15 +108,25 @@ void IndexGenerator::AddList(u32 const numVerts)
void IndexGenerator::AddStrip(u32 const numVerts)
{
bool wind = false;
for (u32 i = 2; i < numVerts; ++i)
{
WriteTriangle(
index + i - 2,
index + i - !wind,
index + i - wind);
if(g_Config.backend_info.bSupportsPrimitiveRestart) {
for (u32 i = 0; i < numVerts; ++i)
{
*Tptr++ = index + i;
}
*Tptr++ = 65535;
numT += numVerts - 2;
wind ^= true;
} else {
bool wind = false;
for (u32 i = 2; i < numVerts; ++i)
{
WriteTriangle(
index + i - 2,
index + i - !wind,
index + i - wind);
wind ^= true;
}
}
}
@ -130,8 +143,17 @@ void IndexGenerator::AddQuads(u32 numVerts)
auto const numQuads = numVerts / 4;
for (u32 i = 0; i != numQuads; ++i)
{
WriteTriangle(index + i * 4, index + i * 4 + 1, index + i * 4 + 2);
WriteTriangle(index + i * 4, index + i * 4 + 2, index + i * 4 + 3);
if(g_Config.backend_info.bSupportsPrimitiveRestart) {
*Tptr++ = index + i * 4 + 0;
*Tptr++ = index + i * 4 + 1;
*Tptr++ = index + i * 4 + 3;
*Tptr++ = index + i * 4 + 2;
*Tptr++ = 65535;
numT += 2;
} else {
WriteTriangle(index + i * 4, index + i * 4 + 1, index + i * 4 + 2);
WriteTriangle(index + i * 4, index + i * 4 + 2, index + i * 4 + 3);
}
}
}

View File

@ -659,6 +659,10 @@ void VertexLoader::SetVAT(u32 _group0, u32 _group1, u32 _group2)
m_VtxAttr.texCoord[7].Elements = vat.g2.Tex7CoordElements;
m_VtxAttr.texCoord[7].Format = vat.g2.Tex7CoordFormat;
m_VtxAttr.texCoord[7].Frac = vat.g2.Tex7Frac;
if(!m_VtxAttr.ByteDequant) {
ERROR_LOG(VIDEO, "ByteDequant is set to zero");
}
};
void VertexLoader::AppendToString(std::string *dest) const

View File

@ -74,27 +74,53 @@ bool VertexManager::IsFlushed() const
u32 VertexManager::GetRemainingIndices(int primitive)
{
switch (primitive)
{
case GX_DRAW_QUADS:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 6 * 4;
case GX_DRAW_TRIANGLES:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen());
case GX_DRAW_TRIANGLE_STRIP:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 3 + 2;
case GX_DRAW_TRIANGLE_FAN:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 3 + 2;
case GX_DRAW_LINES:
return (MAXIBUFFERSIZE - IndexGenerator::GetLineindexLen());
case GX_DRAW_LINE_STRIP:
return (MAXIBUFFERSIZE - IndexGenerator::GetLineindexLen()) / 2 + 1;
if(g_Config.backend_info.bSupportsPrimitiveRestart) {
switch (primitive)
{
case GX_DRAW_QUADS:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 5 * 4;
case GX_DRAW_TRIANGLES:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 4 * 3;
case GX_DRAW_TRIANGLE_STRIP:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 1 - 1;
case GX_DRAW_TRIANGLE_FAN:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 4 + 2;
case GX_DRAW_POINTS:
return (MAXIBUFFERSIZE - IndexGenerator::GetPointindexLen());
case GX_DRAW_LINES:
return (MAXIBUFFERSIZE - IndexGenerator::GetLineindexLen());
case GX_DRAW_LINE_STRIP:
return (MAXIBUFFERSIZE - IndexGenerator::GetLineindexLen()) / 2 + 1;
default:
return 0;
case GX_DRAW_POINTS:
return (MAXIBUFFERSIZE - IndexGenerator::GetPointindexLen());
default:
return 0;
}
} else {
switch (primitive)
{
case GX_DRAW_QUADS:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 6 * 4;
case GX_DRAW_TRIANGLES:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen());
case GX_DRAW_TRIANGLE_STRIP:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 3 + 2;
case GX_DRAW_TRIANGLE_FAN:
return (MAXIBUFFERSIZE - IndexGenerator::GetTriangleindexLen()) / 3 + 2;
case GX_DRAW_LINES:
return (MAXIBUFFERSIZE - IndexGenerator::GetLineindexLen());
case GX_DRAW_LINE_STRIP:
return (MAXIBUFFERSIZE - IndexGenerator::GetLineindexLen()) / 2 + 1;
case GX_DRAW_POINTS:
return (MAXIBUFFERSIZE - IndexGenerator::GetPointindexLen());
default:
return 0;
}
}
}

View File

@ -153,6 +153,7 @@ struct VideoConfig
bool bSupportsDualSourceBlend; // only supported by D3D11 and OpenGL
bool bSupportsFormatReinterpretation;
bool bSupportsPixelLighting;
bool bSupportsPrimitiveRestart;
bool bSupportsGLSLUBO; // needed by pixelShaderGen, so must stay in videoCommon
} backend_info;

View File

@ -96,6 +96,7 @@ void InitBackendInfo()
g_Config.backend_info.bSupportsDualSourceBlend = true;
g_Config.backend_info.bSupportsFormatReinterpretation = true;
g_Config.backend_info.bSupportsPixelLighting = true;
g_Config.backend_info.bSupportsPrimitiveRestart = false; // TODO: dx11 does support it, but it isn't implemented
IDXGIFactory* factory;
IDXGIAdapter* ad;

View File

@ -101,6 +101,7 @@ void InitBackendInfo()
g_Config.backend_info.bUseRGBATextures = false;
g_Config.backend_info.bUseMinimalMipCount = true;
g_Config.backend_info.bSupports3DVision = true;
g_Config.backend_info.bSupportsPrimitiveRestart = false; // TODO: figure out if it does
OSVERSIONINFO info;
ZeroMemory(&info, sizeof(OSVERSIONINFO));
info.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);

View File

@ -293,6 +293,7 @@ Renderer::Renderer()
g_Config.backend_info.bSupportsDualSourceBlend = GLEW_ARB_blend_func_extended;
g_Config.backend_info.bSupportsGLSLUBO = GLEW_ARB_uniform_buffer_object;
g_Config.backend_info.bSupportsPrimitiveRestart = false; //GLEW_VERSION_3_1;
g_ogl_config.bSupportsGLSLCache = GLEW_ARB_get_program_binary;
g_ogl_config.bSupportsGLPinnedMemory = GLEW_AMD_pinned_memory;
@ -328,9 +329,10 @@ Renderer::Renderer()
g_ogl_config.gl_renderer,
g_ogl_config.gl_version).c_str(), 5000);
OSD::AddMessage(StringFromFormat("Missing Extensions: %s%s%s%s%s%s%s%s",
OSD::AddMessage(StringFromFormat("Missing Extensions: %s%s%s%s%s%s%s%s%s",
g_ActiveConfig.backend_info.bSupportsDualSourceBlend ? "" : "DualSourceBlend ",
g_ActiveConfig.backend_info.bSupportsGLSLUBO ? "" : "UniformBuffer ",
g_ActiveConfig.backend_info.bSupportsPrimitiveRestart ? "" : "PrimitiveRestart ",
g_ogl_config.bSupportsGLPinnedMemory ? "" : "PinnedMemory ",
g_ogl_config.bSupportsGLSLCache ? "" : "ShaderCache ",
g_ogl_config.bSupportsGLBaseVertex ? "" : "BaseVertex ",
@ -1391,6 +1393,12 @@ void Renderer::ResetAPIState()
glDisable(GL_BLEND);
glDepthMask(GL_FALSE);
glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
if(g_ActiveConfig.backend_info.bSupportsPrimitiveRestart)
{
glDisable(GL_PRIMITIVE_RESTART);
}
}
void Renderer::RestoreAPIState()
@ -1411,6 +1419,12 @@ void Renderer::RestoreAPIState()
vm->m_last_vao = 0;
TextureCache::SetStage();
if(g_ActiveConfig.backend_info.bSupportsPrimitiveRestart)
{
glEnable(GL_PRIMITIVE_RESTART);
glPrimitiveRestartIndex(65535);
}
}
void Renderer::SetGenerationMode()

View File

@ -127,7 +127,7 @@ void VertexManager::Draw(u32 stride)
if(g_ogl_config.bSupportsGLBaseVertex) {
if (triangle_index_size > 0)
{
glDrawElementsBaseVertex(GL_TRIANGLES, triangle_index_size, GL_UNSIGNED_SHORT, (u8*)NULL+s_offset[0], s_baseVertex);
glDrawElementsBaseVertex(g_ActiveConfig.backend_info.bSupportsPrimitiveRestart?GL_TRIANGLE_STRIP:GL_TRIANGLES, triangle_index_size, GL_UNSIGNED_SHORT, (u8*)NULL+s_offset[0], s_baseVertex);
INCSTAT(stats.thisFrame.numIndexedDrawCalls);
}
if (line_index_size > 0)
@ -143,7 +143,7 @@ void VertexManager::Draw(u32 stride)
} else {
if (triangle_index_size > 0)
{
glDrawElements(GL_TRIANGLES, triangle_index_size, GL_UNSIGNED_SHORT, (u8*)NULL+s_offset[0]);
glDrawElements(g_ActiveConfig.backend_info.bSupportsPrimitiveRestart?GL_TRIANGLE_STRIP:GL_TRIANGLES, triangle_index_size, GL_UNSIGNED_SHORT, (u8*)NULL+s_offset[0]);
INCSTAT(stats.thisFrame.numIndexedDrawCalls);
}
if (line_index_size > 0)