From 4f2fb10347a52be612fb79a32f954168451e99da Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Sun, 12 Feb 2017 16:14:14 +0100
Subject: [PATCH] Further D3D11 fixes

---
 GPU/Common/ShaderUniforms.h               |  2 +-
 GPU/Common/SoftwareTransformCommon.cpp    |  2 +-
 GPU/D3D11/DrawEngineD3D11.cpp             | 23 +++++++++++------------
 GPU/Directx9/PixelShaderGeneratorDX9.cpp  |  2 +-
 GPU/Directx9/VertexShaderGeneratorDX9.cpp |  9 +++++++--
 5 files changed, 21 insertions(+), 17 deletions(-)

diff --git a/GPU/Common/ShaderUniforms.h b/GPU/Common/ShaderUniforms.h
index 17ec2955bb..df3d0e495e 100644
--- a/GPU/Common/ShaderUniforms.h
+++ b/GPU/Common/ShaderUniforms.h
@@ -69,7 +69,7 @@ R"(  float4x4 u_proj;
   float4x4 u_tex;
   float4 u_uvscaleoffset;
   float4 u_depthRange;
-  float3 u_fogcoef;
+  float3 u_fogcoef_stencilreplace;
   float4 u_matambientalpha;
   float3 u_fogcolor;
   float3 u_texenv;
diff --git a/GPU/Common/SoftwareTransformCommon.cpp b/GPU/Common/SoftwareTransformCommon.cpp
index 5d9c2cf4c7..bab715e13f 100644
--- a/GPU/Common/SoftwareTransformCommon.cpp
+++ b/GPU/Common/SoftwareTransformCommon.cpp
@@ -502,7 +502,7 @@ void SoftwareTransform(
 		const u16 *indsIn = (const u16 *)inds;
 		u16 *newInds = inds + vertexCount;
 		u16 *indsOut = newInds;
-		maxIndex = 4 * vertexCount;
+		maxIndex = 4 * (vertexCount / 2);
 		for (int i = 0; i < vertexCount; i += 2) {
 			const TransformedVertex &transVtxTL = transformed[indsIn[i + 0]];
 			const TransformedVertex &transVtxBR = transformed[indsIn[i + 1]];
diff --git a/GPU/D3D11/DrawEngineD3D11.cpp b/GPU/D3D11/DrawEngineD3D11.cpp
index a85241956b..4b994a6c61 100644
--- a/GPU/D3D11/DrawEngineD3D11.cpp
+++ b/GPU/D3D11/DrawEngineD3D11.cpp
@@ -40,13 +40,13 @@
 #include "GPU/D3D11/ShaderManagerD3D11.h"
 #include "GPU/D3D11/GPU_D3D11.h"
 
-const D3D11_PRIMITIVE_TOPOLOGY glprim[8] = {
+const D3D11_PRIMITIVE_TOPOLOGY d3d11prim[8] = {
 	D3D11_PRIMITIVE_TOPOLOGY_POINTLIST,
 	D3D11_PRIMITIVE_TOPOLOGY_LINELIST,
 	D3D11_PRIMITIVE_TOPOLOGY_LINESTRIP,
 	D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST,
 	D3D11_PRIMITIVE_TOPOLOGY_TRIANGLESTRIP,
-	D3D11_PRIMITIVE_TOPOLOGY_UNDEFINED,  // Fans not supported
+	D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST,  // Fans not supported
 	D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST,  // Need expansion - though we could do it with geom shaders in most cases
 };
 
@@ -104,7 +104,9 @@ DrawEngineD3D11::DrawEngineD3D11(Draw::DrawContext *draw, ID3D11Device *device,
 
 	tessDataTransfer = new TessellationDataTransferD3D11();
 
-	// Vertex pushing buffers.
+	// Vertex pushing buffers. For uniforms we use short DISCARD buffers, but we could use
+	// this kind of buffer there as well with D3D11.1. We might be able to use the same buffer
+	// for both vertices and indices, and possibly all three data types.
 	pushVerts_ = new PushBufferD3D11(device, VERTEX_PUSH_SIZE, D3D11_BIND_VERTEX_BUFFER);
 	pushInds_ = new PushBufferD3D11(device, INDEX_PUSH_SIZE, D3D11_BIND_INDEX_BUFFER);
 }
@@ -113,10 +115,8 @@ DrawEngineD3D11::~DrawEngineD3D11() {
 	delete pushVerts_;
 	delete pushInds_;
 
-	for (auto decl = inputLayoutMap_.begin(); decl != inputLayoutMap_.end(); ++decl) {
-		if (decl->second) {
-			decl->second->Release();
-		}
+	for (auto &decl : inputLayoutMap_) {
+		decl.second->Release();
 	}
 	for (auto &depth : depthStencilCache_) {
 		depth.second->Release();
@@ -559,12 +559,10 @@ void DrawEngineD3D11::BeginFrame() {
 }
 
 VertexArrayInfoD3D11::~VertexArrayInfoD3D11() {
-	if (vbo) {
+	if (vbo)
 		vbo->Release();
-	}
-	if (ebo) {
+	if (ebo)
 		ebo->Release();
-	}
 }
 
 static uint32_t SwapRB(uint32_t c) {
@@ -775,7 +773,7 @@ rotateVBO:
 
 		context_->IASetInputLayout(inputLayout);
 		UINT stride = dec_->GetDecVtxFmt().stride;
-		context_->IASetPrimitiveTopology(glprim[prim]);
+		context_->IASetPrimitiveTopology(d3d11prim[prim]);
 		if (!vb_) {
 			// Push!
 			UINT vOffset;
@@ -869,6 +867,7 @@ rotateVBO:
 				layout = iter->second;
 			}
 			context_->IASetInputLayout(layout);
+			context_->IASetPrimitiveTopology(d3d11prim[prim]);
 
 			UINT stride = sizeof(TransformedVertex);
 			UINT vOffset = 0;
diff --git a/GPU/Directx9/PixelShaderGeneratorDX9.cpp b/GPU/Directx9/PixelShaderGeneratorDX9.cpp
index 382bf440c1..b099d7a0e7 100644
--- a/GPU/Directx9/PixelShaderGeneratorDX9.cpp
+++ b/GPU/Directx9/PixelShaderGeneratorDX9.cpp
@@ -341,7 +341,7 @@ bool GenerateFragmentShaderHLSL(const ShaderID &id, char *buffer, ShaderLanguage
 	if (stencilToAlpha != REPLACE_ALPHA_NO) {
 		switch (replaceAlphaWithStencilType) {
 		case STENCIL_VALUE_UNIFORM:
-			replacedAlpha = "u_stencilReplaceValue";
+			replacedAlpha = lang == HLSL_D3D11 ? "u_fogcoef_stencilreplace.z" : "u_stencilReplaceValue";
 			break;
 
 		case STENCIL_VALUE_ZERO:
diff --git a/GPU/Directx9/VertexShaderGeneratorDX9.cpp b/GPU/Directx9/VertexShaderGeneratorDX9.cpp
index b9ff0f0dd1..37b2c47645 100644
--- a/GPU/Directx9/VertexShaderGeneratorDX9.cpp
+++ b/GPU/Directx9/VertexShaderGeneratorDX9.cpp
@@ -604,8 +604,13 @@ void GenerateVertexShaderHLSL(const ShaderID &id, char *buffer, ShaderLanguage l
 		}
 
 		// Compute fogdepth
-		if (enableFog)
-			WRITE(p, "  Out.v_fogdepth.x = (viewPos.z + u_fogcoef.x) * u_fogcoef.y;\n");
+		if (enableFog) {
+			if (lang == HLSL_D3D11) {
+				WRITE(p, "  Out.v_fogdepth.x = (viewPos.z + u_fogcoef_stencilreplace.x) * u_fogcoef_stencilreplace.y;\n");
+			} else {
+				WRITE(p, "  Out.v_fogdepth.x = (viewPos.z + u_fogcoef.x) * u_fogcoef.y;\n");
+			}
+		}
 	}
 
 	WRITE(p, "  return Out;\n");