Merge pull request #1748 from Armada651/stereo-format

FramebufferManager: Support stereoscopic EFB format changes.

7 files changed, 66 insertions, 37 deletions

diff --git a/Source/Core/VideoBackends/D3D/D3DUtil.cpp b/Source/Core/VideoBackends/D3D/D3DUtil.cpp
index ad9acc809f..020045d687 100644
--- a/Source/Core/VideoBackends/D3D/D3DUtil.cpp
+++ b/Source/Core/VideoBackends/D3D/D3DUtil.cpp
@@ -697,7 +697,7 @@ void drawClearQuad(u32 Color, float z)
 	}
 
 	stateman->SetVertexShader(VertexShaderCache::GetClearVertexShader());
-	stateman->SetGeometryShader(g_ActiveConfig.iStereoMode > 0 ? GeometryShaderCache::GetClearGeometryShader() : nullptr);
+	stateman->SetGeometryShader(GeometryShaderCache::GetClearGeometryShader());
 	stateman->SetPixelShader(PixelShaderCache::GetClearProgram());
 	stateman->SetInputLayout(VertexShaderCache::GetClearInputLayout());
 
diff --git a/Source/Core/VideoBackends/D3D/GeometryShaderCache.cpp b/Source/Core/VideoBackends/D3D/GeometryShaderCache.cpp
index 59b0e3ff65..a5cfcd235f 100644
--- a/Source/Core/VideoBackends/D3D/GeometryShaderCache.cpp
+++ b/Source/Core/VideoBackends/D3D/GeometryShaderCache.cpp
@@ -12,6 +12,7 @@
 
 #include "VideoBackends/D3D/D3DBase.h"
 #include "VideoBackends/D3D/D3DShader.h"
+#include "VideoBackends/D3D/FramebufferManager.h"
 #include "VideoBackends/D3D/GeometryShaderCache.h"
 #include "VideoBackends/D3D/Globals.h"
 
@@ -35,8 +36,8 @@ ID3D11GeometryShader* CopyGeometryShader = nullptr;
 
 LinearDiskCache<GeometryShaderUid, u8> g_gs_disk_cache;
 
-ID3D11GeometryShader* GeometryShaderCache::GetClearGeometryShader() { return ClearGeometryShader; }
-ID3D11GeometryShader* GeometryShaderCache::GetCopyGeometryShader() { return CopyGeometryShader; }
+ID3D11GeometryShader* GeometryShaderCache::GetClearGeometryShader() { return (FramebufferManager::GetEFBLayers() > 1) ? ClearGeometryShader: nullptr; }
+ID3D11GeometryShader* GeometryShaderCache::GetCopyGeometryShader() { return (FramebufferManager::GetEFBLayers() > 1) ? CopyGeometryShader : nullptr; }
 
 ID3D11Buffer* gscbuf = nullptr;
 
diff --git a/Source/Core/VideoBackends/D3D/PixelShaderCache.cpp b/Source/Core/VideoBackends/D3D/PixelShaderCache.cpp
index ae1919c91e..a94d113c4f 100644
--- a/Source/Core/VideoBackends/D3D/PixelShaderCache.cpp
+++ b/Source/Core/VideoBackends/D3D/PixelShaderCache.cpp
@@ -201,11 +201,11 @@ const char depth_matrix_program_msaa[] = {
 
 const char reint_rgba6_to_rgb8[] = {
 	"sampler samp0 : register(s0);\n"
-	"Texture2D Tex0 : register(t0);\n"
+	"Texture2DArray Tex0 : register(t0);\n"
 	"void main(\n"
 	"	out float4 ocol0 : SV_Target,\n"
 	"	in float4 pos : SV_Position,\n"
-	"	in float2 uv0 : TEXCOORD0)\n"
+	"	in float3 uv0 : TEXCOORD0)\n"
 	"{\n"
 	"	int4 src6 = round(Tex0.Sample(samp0,uv0) * 63.f);\n"
 	"	int4 dst8;\n"
@@ -219,17 +219,17 @@ const char reint_rgba6_to_rgb8[] = {
 
 const char reint_rgba6_to_rgb8_msaa[] = {
 	"sampler samp0 : register(s0);\n"
-	"Texture2DMS<float4, %d> Tex0 : register(t0);\n"
+	"Texture2DMSArray<float4, %d> Tex0 : register(t0);\n"
 	"void main(\n"
 	"	out float4 ocol0 : SV_Target,\n"
 	"	in float4 pos : SV_Position,\n"
-	"	in float2 uv0 : TEXCOORD0)\n"
+	"	in float3 uv0 : TEXCOORD0)\n"
 	"{\n"
-	"	int width, height, samples;\n"
-	"	Tex0.GetDimensions(width, height, samples);\n"
+	"	int width, height, slices, samples;\n"
+	"	Tex0.GetDimensions(width, height, slices, samples);\n"
 	"	float4 texcol = 0;\n"
 	"	for (int i = 0; i < samples; ++i)\n"
-	"		texcol += Tex0.Load(int2(uv0.x*(width), uv0.y*(height)), i);\n"
+	"		texcol += Tex0.Load(int3(uv0.x*(width), uv0.y*(height), uv0.z), i);\n"
 	"	texcol /= samples;\n"
 	"	int4 src6 = round(texcol * 63.f);\n"
 	"	int4 dst8;\n"
@@ -243,11 +243,11 @@ const char reint_rgba6_to_rgb8_msaa[] = {
 
 const char reint_rgb8_to_rgba6[] = {
 	"sampler samp0 : register(s0);\n"
-	"Texture2D Tex0 : register(t0);\n"
+	"Texture2DArray Tex0 : register(t0);\n"
 	"void main(\n"
 	"	out float4 ocol0 : SV_Target,\n"
 	"	in float4 pos : SV_Position,\n"
-	"	in float2 uv0 : TEXCOORD0)\n"
+	"	in float3 uv0 : TEXCOORD0)\n"
 	"{\n"
 	"	int4 src8 = round(Tex0.Sample(samp0,uv0) * 255.f);\n"
 	"	int4 dst6;\n"
@@ -261,17 +261,17 @@ const char reint_rgb8_to_rgba6[] = {
 
 const char reint_rgb8_to_rgba6_msaa[] = {
 	"sampler samp0 : register(s0);\n"
-	"Texture2DMS<float4, %d> Tex0 : register(t0);\n"
+	"Texture2DMSArray<float4, %d> Tex0 : register(t0);\n"
 	"void main(\n"
 	"	out float4 ocol0 : SV_Target,\n"
 	"	in float4 pos : SV_Position,\n"
-	"	in float2 uv0 : TEXCOORD0)\n"
+	"	in float3 uv0 : TEXCOORD0)\n"
 	"{\n"
-	"	int width, height, samples;\n"
-	"	Tex0.GetDimensions(width, height, samples);\n"
+	"	int width, height, slices, samples;\n"
+	"	Tex0.GetDimensions(width, height, slices, samples);\n"
 	"	float4 texcol = 0;\n"
 	"	for (int i = 0; i < samples; ++i)\n"
-	"		texcol += Tex0.Load(int2(uv0.x*(width), uv0.y*(height)), i);\n"
+	"		texcol += Tex0.Load(int3(uv0.x*(width), uv0.y*(height), uv0.z), i);\n"
 	"	texcol /= samples;\n"
 	"	int4 src8 = round(texcol * 255.f);\n"
 	"	int4 dst6;\n"
diff --git a/Source/Core/VideoBackends/D3D/Render.cpp b/Source/Core/VideoBackends/D3D/Render.cpp
index 84b83ef1f0..c309882e75 100644
--- a/Source/Core/VideoBackends/D3D/Render.cpp
+++ b/Source/Core/VideoBackends/D3D/Render.cpp
@@ -577,7 +577,8 @@ void Renderer::ReinterpretPixelData(unsigned int convtype)
 
 	D3D::context->OMSetRenderTargets(1, &FramebufferManager::GetEFBColorTempTexture()->GetRTV(), nullptr);
 	D3D::SetPointCopySampler();
-	D3D::drawShadedTexQuad(FramebufferManager::GetEFBColorTexture()->GetSRV(), &source, g_renderer->GetTargetWidth(), g_renderer->GetTargetHeight(), pixel_shader, VertexShaderCache::GetSimpleVertexShader(), VertexShaderCache::GetSimpleInputLayout());
+	D3D::drawShadedTexQuad(FramebufferManager::GetEFBColorTexture()->GetSRV(), &source, g_renderer->GetTargetWidth(), g_renderer->GetTargetHeight(),
+		pixel_shader, VertexShaderCache::GetSimpleVertexShader(), VertexShaderCache::GetSimpleInputLayout(), GeometryShaderCache::GetCopyGeometryShader());
 
 	g_renderer->RestoreAPIState();
 
diff --git a/Source/Core/VideoBackends/D3D/TextureCache.cpp b/Source/Core/VideoBackends/D3D/TextureCache.cpp
index b251a30682..af1267c1de 100644
--- a/Source/Core/VideoBackends/D3D/TextureCache.cpp
+++ b/Source/Core/VideoBackends/D3D/TextureCache.cpp
@@ -168,8 +168,7 @@ void TextureCache::TCacheEntry::FromRenderTarget(u32 dstAddr, unsigned int dstFo
 			(srcFormat == PEControl::Z24) ? FramebufferManager::GetEFBDepthTexture()->GetSRV() : FramebufferManager::GetEFBColorTexture()->GetSRV(),
 			&sourcerect, Renderer::GetTargetWidth(), Renderer::GetTargetHeight(),
 			(srcFormat == PEControl::Z24) ? PixelShaderCache::GetDepthMatrixProgram(true) : PixelShaderCache::GetColorMatrixProgram(true),
-			VertexShaderCache::GetSimpleVertexShader(), VertexShaderCache::GetSimpleInputLayout(),
-			(g_Config.iStereoMode > 0) ? GeometryShaderCache::GetCopyGeometryShader() : nullptr);
+			VertexShaderCache::GetSimpleVertexShader(), VertexShaderCache::GetSimpleInputLayout(), GeometryShaderCache::GetCopyGeometryShader());
 
 		D3D::context->OMSetRenderTargets(1, &FramebufferManager::GetEFBColorTexture()->GetRTV(), FramebufferManager::GetEFBDepthTexture()->GetDSV());
 
diff --git a/Source/Core/VideoBackends/OGL/FramebufferManager.cpp b/Source/Core/VideoBackends/OGL/FramebufferManager.cpp
index e9d2ba7294..731cf389d2 100644
--- a/Source/Core/VideoBackends/OGL/FramebufferManager.cpp
+++ b/Source/Core/VideoBackends/OGL/FramebufferManager.cpp
@@ -192,10 +192,16 @@ FramebufferManager::FramebufferManager(int targetWidth, int targetHeight, int ms
 	glClear(GL_COLOR_BUFFER_BIT|GL_DEPTH_BUFFER_BIT);
 
 	// reinterpret pixel format
-	char vs[] =
+	const char* vs = m_EFBLayers > 1 ?
 		"void main(void) {\n"
 		"	vec2 rawpos = vec2(gl_VertexID&1, gl_VertexID&2);\n"
 		"	gl_Position = vec4(rawpos*2.0-1.0, 0.0, 1.0);\n"
+		"}\n" :
+		"flat out int layer;\n"
+		"void main(void) {\n"
+		"	layer = 0;\n"
+		"	vec2 rawpos = vec2(gl_VertexID&1, gl_VertexID&2);\n"
+		"	gl_Position = vec4(rawpos*2.0-1.0, 0.0, 1.0);\n"
 		"}\n";
 
 	// The way to sample the EFB is based on the on the current configuration.
@@ -207,8 +213,8 @@ FramebufferManager::FramebufferManager(int targetWidth, int targetHeight, int ms
 		// non-msaa, so just fetch the pixel
 		sampler =
 			"SAMPLER_BINDING(9) uniform sampler2DArray samp9;\n"
-			"vec4 sampleEFB(ivec2 pos) {\n"
-			"	return texelFetch(samp9, ivec3(pos, 0), 0);\n"
+			"vec4 sampleEFB(ivec3 pos) {\n"
+			"	return texelFetch(samp9, pos, 0);\n"
 			"}\n";
 	}
 	else if (g_ogl_config.bSupportSampleShading)
@@ -220,16 +226,16 @@ FramebufferManager::FramebufferManager(int targetWidth, int targetHeight, int ms
 		{
 			sampler =
 				"SAMPLER_BINDING(9) uniform sampler2DMSArray samp9;\n"
-				"vec4 sampleEFB(ivec2 pos) {\n"
-				"	return texelFetch(samp9, ivec3(pos, 0), gl_SampleID);\n"
+				"vec4 sampleEFB(ivec3 pos) {\n"
+				"	return texelFetch(samp9, pos, gl_SampleID);\n"
 				"}\n";
 		}
 		else
 		{
 			sampler =
 				"SAMPLER_BINDING(9) uniform sampler2DMS samp9;\n"
-				"vec4 sampleEFB(ivec2 pos) {\n"
-				"	return texelFetch(samp9, pos, gl_SampleID);\n"
+				"vec4 sampleEFB(ivec3 pos) {\n"
+				"	return texelFetch(samp9, pos.xy, gl_SampleID);\n"
 				"}\n";
 		}
 	}
@@ -242,10 +248,10 @@ FramebufferManager::FramebufferManager(int targetWidth, int targetHeight, int ms
 		{
 			sampler =
 				"SAMPLER_BINDING(9) uniform sampler2DMSArray samp9;\n"
-				"vec4 sampleEFB(ivec2 pos) {\n"
+				"vec4 sampleEFB(ivec3 pos) {\n"
 				"	vec4 color = vec4(0.0, 0.0, 0.0, 0.0);\n"
 				"	for(int i=0; i<" + samples.str() + "; i++)\n"
-				"		color += texelFetch(samp9, ivec3(pos, 0), i);\n"
+				"		color += texelFetch(samp9, pos, 0), i);\n"
 				"	return color / " + samples.str() + ";\n"
 				"}\n";
 		}
@@ -253,20 +259,21 @@ FramebufferManager::FramebufferManager(int targetWidth, int targetHeight, int ms
 		{
 			sampler =
 				"SAMPLER_BINDING(9) uniform sampler2DMS samp9;\n"
-				"vec4 sampleEFB(ivec2 pos) {\n"
+				"vec4 sampleEFB(ivec3 pos) {\n"
 				"	vec4 color = vec4(0.0, 0.0, 0.0, 0.0);\n"
 				"	for(int i=0; i<" + samples.str() + "; i++)\n"
-				"		color += texelFetch(samp9, pos, i);\n"
+				"		color += texelFetch(samp9, pos.xy, i);\n"
 				"	return color / " + samples.str() + ";\n"
 				"}\n";
 		}
 	}
 
 	std::string ps_rgba6_to_rgb8 = sampler +
+		"flat in int layer;\n"
 		"out vec4 ocol0;\n"
 		"void main()\n"
 		"{\n"
-		"	ivec4 src6 = ivec4(round(sampleEFB(ivec2(gl_FragCoord.xy)) * 63.f));\n"
+		"	ivec4 src6 = ivec4(round(sampleEFB(ivec3(gl_FragCoord.xy, layer)) * 63.f));\n"
 		"	ivec4 dst8;\n"
 		"	dst8.r = (src6.r << 2) | (src6.g >> 4);\n"
 		"	dst8.g = ((src6.g & 0xF) << 4) | (src6.b >> 2);\n"
@@ -276,10 +283,11 @@ FramebufferManager::FramebufferManager(int targetWidth, int targetHeight, int ms
 		"}";
 
 	std::string ps_rgb8_to_rgba6 = sampler +
+		"flat in int layer;\n"
 		"out vec4 ocol0;\n"
 		"void main()\n"
 		"{\n"
-		"	ivec4 src8 = ivec4(round(sampleEFB(ivec2(gl_FragCoord.xy)) * 255.f));\n"
+		"	ivec4 src8 = ivec4(round(sampleEFB(ivec3(gl_FragCoord.xy, layer)) * 255.f));\n"
 		"	ivec4 dst6;\n"
 		"	dst6.r = src8.r >> 2;\n"
 		"	dst6.g = ((src8.r & 0x3) << 4) | (src8.g >> 4);\n"
@@ -288,8 +296,28 @@ FramebufferManager::FramebufferManager(int targetWidth, int targetHeight, int ms
 		"	ocol0 = float4(dst6) / 63.f;\n"
 		"}";
 
-	ProgramShaderCache::CompileShader(m_pixel_format_shaders[0], vs, ps_rgb8_to_rgba6.c_str());
-	ProgramShaderCache::CompileShader(m_pixel_format_shaders[1], vs, ps_rgba6_to_rgb8.c_str());
+	std::stringstream vertices, layers;
+	vertices << m_EFBLayers * 3;
+	layers << m_EFBLayers;
+	std::string gs = sampler +
+		"layout(triangles) in;\n"
+		"layout(triangle_strip, max_vertices = " + vertices.str() + ") out;\n"
+		"flat out int layer;\n"
+		"void main()\n"
+		"{\n"
+		"	for (int j = 0; j < " + layers.str() + "; ++j) {\n"
+		"		for (int i = 0; i < 3; ++i) {\n"
+		"			layer = j;\n"
+		"			gl_Layer = j;\n"
+		"			gl_Position = gl_in[i].gl_Position;\n"
+		"			EmitVertex();\n"
+		"		}\n"
+		"		EndPrimitive();\n"
+		"	}\n"
+		"}\n";
+
+	ProgramShaderCache::CompileShader(m_pixel_format_shaders[0], vs, ps_rgb8_to_rgba6.c_str(), (m_EFBLayers > 1) ? gs.c_str() : nullptr);
+	ProgramShaderCache::CompileShader(m_pixel_format_shaders[1], vs, ps_rgba6_to_rgb8.c_str(), (m_EFBLayers > 1) ? gs.c_str() : nullptr);
 }
 
 FramebufferManager::~FramebufferManager()
diff --git a/Source/Core/VideoBackends/OGL/TextureCache.cpp b/Source/Core/VideoBackends/OGL/TextureCache.cpp
index f697d51c73..97bc487675 100644
--- a/Source/Core/VideoBackends/OGL/TextureCache.cpp
+++ b/Source/Core/VideoBackends/OGL/TextureCache.cpp
@@ -400,7 +400,7 @@ void TextureCache::CompileShaders()
 		"	gl_Position = vec4(rawpos*2.0-1.0, 0.0, 1.0);\n"
 		"}\n";
 
-	const char *GProgram = (g_ActiveConfig.iStereoMode > 0) ?
+	const char *GProgram = FramebufferManager::GetEFBLayers() > 1 ?
 		"layout(triangles) in;\n"
 		"layout(triangle_strip, max_vertices = 6) out;\n"
 		"in vec3 v_uv0[3];\n"
@@ -421,7 +421,7 @@ void TextureCache::CompileShaders()
 		"}\n" : nullptr;
 
 	const char* prefix = (GProgram == nullptr) ? "f" : "v";
-	const char* depth_layer = (g_ActiveConfig.bStereoMonoEFBDepth) ? "0" : "f_uv0.z";
+	const char* depth_layer = (g_ActiveConfig.bStereoMonoEFBDepth) ? "0.0" : "f_uv0.z";
 
 	ProgramShaderCache::CompileShader(s_ColorMatrixProgram, StringFromFormat(VProgram, prefix, prefix).c_str(), pColorMatrixProg, GProgram);
 	ProgramShaderCache::CompileShader(s_DepthMatrixProgram, StringFromFormat(VProgram, prefix, prefix).c_str(), StringFromFormat(pDepthMatrixProg, depth_layer).c_str(), GProgram);