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.

4 files changed, 87 insertions, 34 deletions

diff --git a/Source/Core/VideoCommon/Src/IndexGenerator.cpp b/Source/Core/VideoCommon/Src/IndexGenerator.cpp
index aa1a971687..05097b7c9a 100644
--- a/Source/Core/VideoCommon/Src/IndexGenerator.cpp
+++ b/Source/Core/VideoCommon/Src/IndexGenerator.cpp
@@ -18,6 +18,7 @@
 #include <cstddef>
 
 #include "Common.h"
+#include "VideoConfig.h"
 #include "IndexGenerator.h"
 
 /*
@@ -90,7 +91,9 @@ __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);
-
-		wind ^= true;
+	if(g_Config.backend_info.bSupportsPrimitiveRestart) {
+		for (u32 i = 0; i < numVerts; ++i)
+		{
+			*Tptr++ = index + i;
+		}
+		*Tptr++ = 65535;
+		numT += numVerts - 2;
+		
+	} 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);
+		}
 	}
 }
 
diff --git a/Source/Core/VideoCommon/Src/VertexLoader.cpp b/Source/Core/VideoCommon/Src/VertexLoader.cpp
index 5df495a44c..74c1bd9705 100644
--- a/Source/Core/VideoCommon/Src/VertexLoader.cpp
+++ b/Source/Core/VideoCommon/Src/VertexLoader.cpp
@@ -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
diff --git a/Source/Core/VideoCommon/Src/VertexManagerBase.cpp b/Source/Core/VideoCommon/Src/VertexManagerBase.cpp
index 87856b5aeb..076c45e4eb 100644
--- a/Source/Core/VideoCommon/Src/VertexManagerBase.cpp
+++ b/Source/Core/VideoCommon/Src/VertexManagerBase.cpp
@@ -74,28 +74,54 @@ 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;
-
-	case GX_DRAW_POINTS:
-		return (MAXIBUFFERSIZE - IndexGenerator::GetPointindexLen());
-
-	default:
-		return 0;
-	}
+	
+	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_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;
+		}
+	} 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;
+		}
+	} 
 }
 
 void VertexManager::AddVertices(int primitive, u32 numVertices)
diff --git a/Source/Core/VideoCommon/Src/VideoConfig.h b/Source/Core/VideoCommon/Src/VideoConfig.h
index 442e95f9bf..88326346f2 100644
--- a/Source/Core/VideoCommon/Src/VideoConfig.h
+++ b/Source/Core/VideoCommon/Src/VideoConfig.h
@@ -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;