Merge pull request #2448 from Armada651/depth-inversion

D3D: Depth range inversion.

6 files changed, 38 insertions, 18 deletions

diff --git a/Source/Core/VideoBackends/D3D/D3DState.cpp b/Source/Core/VideoBackends/D3D/D3DState.cpp
index 34e0eb7680..bd0a80b355 100644
--- a/Source/Core/VideoBackends/D3D/D3DState.cpp
+++ b/Source/Core/VideoBackends/D3D/D3DState.cpp
@@ -433,7 +433,7 @@ ID3D11DepthStencilState* StateCache::Get(ZMode state)
 
 	depthdc.DepthEnable = TRUE;
 	depthdc.DepthWriteMask = D3D11_DEPTH_WRITE_MASK_ALL;
-	depthdc.DepthFunc = D3D11_COMPARISON_LESS;
+	depthdc.DepthFunc = D3D11_COMPARISON_GREATER;
 	depthdc.StencilEnable = FALSE;
 	depthdc.StencilReadMask = D3D11_DEFAULT_STENCIL_READ_MASK;
 	depthdc.StencilWriteMask = D3D11_DEFAULT_STENCIL_WRITE_MASK;
@@ -441,12 +441,12 @@ ID3D11DepthStencilState* StateCache::Get(ZMode state)
 	const D3D11_COMPARISON_FUNC d3dCmpFuncs[8] =
 	{
 		D3D11_COMPARISON_NEVER,
-		D3D11_COMPARISON_LESS,
-		D3D11_COMPARISON_EQUAL,
-		D3D11_COMPARISON_LESS_EQUAL,
 		D3D11_COMPARISON_GREATER,
-		D3D11_COMPARISON_NOT_EQUAL,
+		D3D11_COMPARISON_EQUAL,
 		D3D11_COMPARISON_GREATER_EQUAL,
+		D3D11_COMPARISON_LESS,
+		D3D11_COMPARISON_NOT_EQUAL,
+		D3D11_COMPARISON_LESS_EQUAL,
 		D3D11_COMPARISON_ALWAYS
 	};
 
diff --git a/Source/Core/VideoBackends/D3D/PixelShaderCache.cpp b/Source/Core/VideoBackends/D3D/PixelShaderCache.cpp
index b1d0cf4cd9..918432bb77 100644
--- a/Source/Core/VideoBackends/D3D/PixelShaderCache.cpp
+++ b/Source/Core/VideoBackends/D3D/PixelShaderCache.cpp
@@ -146,7 +146,7 @@ const char depth_matrix_program[] = {
 	" in float4 pos : SV_Position,\n"
 	" in float3 uv0 : TEXCOORD0){\n"
 	"	float4 texcol = Tex0.Sample(samp0,uv0);\n"
-	"	int depth = clamp(int(texcol.x * 16777216.0), 0, 0xFFFFFF);\n"
+	"	int depth = clamp(int((1.0 - texcol.x) * 16777216.0), 0, 0xFFFFFF);\n"
 
 	// Convert to Z24 format
 	"	int4 workspace;\n"
@@ -180,7 +180,7 @@ const char depth_matrix_program_msaa[] = {
 	"	for(int i = 0; i < SAMPLES; ++i)\n"
 	"		texcol += Tex0.Load(int3(uv0.x*(width), uv0.y*(height), uv0.z), i);\n"
 	"	texcol /= SAMPLES;\n"
-	"	int depth = clamp(int(texcol.x * 16777216.0), 0, 0xFFFFFF);\n"
+	"	int depth = clamp(int((1.0 - texcol.x) * 16777216.0), 0, 0xFFFFFF);\n"
 
 	// Convert to Z24 format
 	"	int4 workspace;\n"
diff --git a/Source/Core/VideoBackends/D3D/Render.cpp b/Source/Core/VideoBackends/D3D/Render.cpp
index 6bda211de6..230cb11ec4 100644
--- a/Source/Core/VideoBackends/D3D/Render.cpp
+++ b/Source/Core/VideoBackends/D3D/Render.cpp
@@ -259,7 +259,7 @@ Renderer::Renderer(void *&window_handle)
 	// Clear EFB textures
 	float ClearColor[4] = { 0.f, 0.f, 0.f, 1.f };
 	D3D::context->ClearRenderTargetView(FramebufferManager::GetEFBColorTexture()->GetRTV(), ClearColor);
-	D3D::context->ClearDepthStencilView(FramebufferManager::GetEFBDepthTexture()->GetDSV(), D3D11_CLEAR_DEPTH, 1.f, 0);
+	D3D::context->ClearDepthStencilView(FramebufferManager::GetEFBDepthTexture()->GetDSV(), D3D11_CLEAR_DEPTH, 0.f, 0);
 
 	D3D11_VIEWPORT vp = CD3D11_VIEWPORT(0.f, 0.f, (float)s_target_width, (float)s_target_height);
 	D3D::context->RSSetViewports(1, &vp);
@@ -415,7 +415,8 @@ u32 Renderer::AccessEFB(EFBAccessType type, u32 x, u32 y, u32 poke_data)
 		// read the data from system memory
 		D3D::context->Map(read_tex, 0, D3D11_MAP_READ, 0, &map);
 
-		float val = *(float*)map.pData;
+		// depth buffer is inverted in the d3d backend
+		float val = 1.0f - *(float*)map.pData;
 		u32 ret = 0;
 		if (bpmem.zcontrol.pixel_format == PEControl::RGB565_Z16)
 		{
@@ -522,8 +523,8 @@ void Renderer::SetViewport()
 	Ht = (Y + Ht <= GetTargetHeight()) ? Ht : (GetTargetHeight() - Y);
 
 	D3D11_VIEWPORT vp = CD3D11_VIEWPORT(X, Y, Wd, Ht,
-		MathUtil::Clamp<float>(xfmem.viewport.farZ - xfmem.viewport.zRange, 0.0f, 16777215.0f) / 16777216.0f,
-		MathUtil::Clamp<float>(xfmem.viewport.farZ, 0.0f, 16777215.0f) / 16777216.0f);
+		1.0f - MathUtil::Clamp<float>(xfmem.viewport.farZ, 0.0f, 16777215.0f) / 16777216.0f,
+		1.0f - MathUtil::Clamp<float>(xfmem.viewport.farZ - xfmem.viewport.zRange, 0.0f, 16777215.0f) / 16777216.0f);
 	D3D::context->RSSetViewports(1, &vp);
 }
 
@@ -548,7 +549,7 @@ void Renderer::ClearScreen(const EFBRectangle& rc, bool colorEnable, bool alphaE
 
 	// Color is passed in bgra mode so we need to convert it to rgba
 	u32 rgbaColor = (color & 0xFF00FF00) | ((color >> 16) & 0xFF) | ((color << 16) & 0xFF0000);
-	D3D::drawClearQuad(rgbaColor, (z & 0xFFFFFF) / 16777216.0f);
+	D3D::drawClearQuad(rgbaColor, 1.0f - (z & 0xFFFFFF) / 16777216.0f);
 
 	D3D::stateman->PopDepthState();
 	D3D::stateman->PopBlendState();
@@ -956,7 +957,7 @@ void Renderer::SwapImpl(u32 xfbAddr, u32 fbWidth, u32 fbStride, u32 fbHeight, co
 		g_framebuffer_manager = new FramebufferManager;
 		float clear_col[4] = { 0.f, 0.f, 0.f, 1.f };
 		D3D::context->ClearRenderTargetView(FramebufferManager::GetEFBColorTexture()->GetRTV(), clear_col);
-		D3D::context->ClearDepthStencilView(FramebufferManager::GetEFBDepthTexture()->GetDSV(), D3D11_CLEAR_DEPTH, 1.f, 0);
+		D3D::context->ClearDepthStencilView(FramebufferManager::GetEFBDepthTexture()->GetDSV(), D3D11_CLEAR_DEPTH, 0.f, 0);
 	}
 
 	// begin next frame
diff --git a/Source/Core/VideoCommon/PixelShaderGen.cpp b/Source/Core/VideoCommon/PixelShaderGen.cpp
index ecb47d5584..d9d8c18599 100644
--- a/Source/Core/VideoCommon/PixelShaderGen.cpp
+++ b/Source/Core/VideoCommon/PixelShaderGen.cpp
@@ -582,7 +582,10 @@ static inline void GeneratePixelShader(T& out, DSTALPHA_MODE dstAlphaMode, API_T
 	}
 	else
 	{
-		out.Write("\tint zCoord = int(rawpos.z * 16777216.0);\n");
+		if (ApiType == API_D3D)
+			out.Write("\tint zCoord = int((1.0 - rawpos.z) * 16777216.0);\n");
+		else
+			out.Write("\tint zCoord = int(rawpos.z * 16777216.0);\n");
 	}
 	out.Write("\tzCoord = clamp(zCoord, " I_ZBIAS"[1].x - " I_ZBIAS"[1].y, " I_ZBIAS"[1].x);\n");
 
@@ -600,7 +603,10 @@ static inline void GeneratePixelShader(T& out, DSTALPHA_MODE dstAlphaMode, API_T
 	// Note: z-textures are not written to depth buffer if early depth test is used
 	if (per_pixel_depth && bpmem.UseEarlyDepthTest())
 	{
-		out.Write("\tdepth = float(zCoord) / 16777216.0;\n");
+		if (ApiType == API_D3D)
+			out.Write("\tdepth = 1.0 - float(zCoord) / 16777216.0;\n");
+		else
+			out.Write("\tdepth = float(zCoord) / 16777216.0;\n");
 	}
 
 	// Note: depth texture output is only written to depth buffer if late depth test is used
@@ -616,7 +622,10 @@ static inline void GeneratePixelShader(T& out, DSTALPHA_MODE dstAlphaMode, API_T
 
 	if (per_pixel_depth && bpmem.UseLateDepthTest())
 	{
-		out.Write("\tdepth = float(zCoord) / 16777216.0;\n");
+		if (ApiType == API_D3D)
+			out.Write("\tdepth = 1.0 - float(zCoord) / 16777216.0;\n");
+		else
+			out.Write("\tdepth = float(zCoord) / 16777216.0;\n");
 	}
 
 	if (dstAlphaMode == DSTALPHA_ALPHA_PASS)
@@ -1129,7 +1138,7 @@ static inline void WriteAlphaTest(T& out, pixel_shader_uid_data* uid_data, API_T
 	if (dstAlphaMode == DSTALPHA_DUAL_SOURCE_BLEND)
 		out.Write("\t\tocol1 = float4(0.0, 0.0, 0.0, 0.0);\n");
 	if (per_pixel_depth)
-		out.Write("\t\tdepth = 1.0;\n");
+		out.Write("\t\tdepth = %s;\n", (ApiType == API_D3D) ? "0.0" : "1.0");
 
 	// ZCOMPLOC HACK:
 	// The only way to emulate alpha test + early-z is to force early-z in the shader.
diff --git a/Source/Core/VideoCommon/TextureConversionShader.cpp b/Source/Core/VideoCommon/TextureConversionShader.cpp
index fac5a0823c..f93d465a16 100644
--- a/Source/Core/VideoCommon/TextureConversionShader.cpp
+++ b/Source/Core/VideoCommon/TextureConversionShader.cpp
@@ -462,15 +462,19 @@ static void WriteZ8Encoder(char*& p, const char* multiplier,API_TYPE ApiType)
 	WRITE(p, " float depth;\n");
 
 	WriteSampleColor(p, "r", "depth", 0, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth = 1.0f - depth;\n");
 	WRITE(p, "ocol0.b = frac(depth * %s);\n", multiplier);
 
 	WriteSampleColor(p, "r", "depth", 1, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth = 1.0f - depth;\n");
 	WRITE(p, "ocol0.g = frac(depth * %s);\n", multiplier);
 
 	WriteSampleColor(p, "r", "depth", 2, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth = 1.0f - depth;\n");
 	WRITE(p, "ocol0.r = frac(depth * %s);\n", multiplier);
 
 	WriteSampleColor(p, "r", "depth", 3, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth = 1.0f - depth;\n");
 	WRITE(p, "ocol0.a = frac(depth * %s);\n", multiplier);
 
 	WriteEncoderEnd(p, ApiType);
@@ -486,6 +490,7 @@ static void WriteZ16Encoder(char*& p,API_TYPE ApiType)
 	// byte order is reversed
 
 	WriteSampleColor(p, "r", "depth", 0, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth = 1.0f - depth;\n");
 
 	WRITE(p, "  depth = clamp(depth * 16777216.0, 0.0, float(0xFFFFFF));\n");
 	WRITE(p, "  expanded.r = floor(depth / (256.0 * 256.0));\n");
@@ -496,6 +501,7 @@ static void WriteZ16Encoder(char*& p,API_TYPE ApiType)
 	WRITE(p, "  ocol0.g = expanded.r / 255.0;\n");
 
 	WriteSampleColor(p, "r", "depth", 1, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth = 1.0f - depth;\n");
 
 	WRITE(p, "  depth = clamp(depth * 16777216.0, 0.0, float(0xFFFFFF));\n");
 	WRITE(p, "  expanded.r = floor(depth / (256.0 * 256.0));\n");
@@ -518,6 +524,7 @@ static void WriteZ16LEncoder(char*& p,API_TYPE ApiType)
 	// byte order is reversed
 
 	WriteSampleColor(p, "r", "depth", 0, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth = 1.0f - depth;\n");
 
 	WRITE(p, "  depth = clamp(depth * 16777216.0, 0.0, float(0xFFFFFF));\n");
 	WRITE(p, "  expanded.r = floor(depth / (256.0 * 256.0));\n");
@@ -530,6 +537,7 @@ static void WriteZ16LEncoder(char*& p,API_TYPE ApiType)
 	WRITE(p, "  ocol0.g = expanded.g / 255.0;\n");
 
 	WriteSampleColor(p, "r", "depth", 1, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth = 1.0f - depth;\n");
 
 	WRITE(p, "  depth = clamp(depth * 16777216.0, 0.0, float(0xFFFFFF));\n");
 	WRITE(p, "  expanded.r = floor(depth / (256.0 * 256.0));\n");
@@ -554,7 +562,9 @@ static void WriteZ24Encoder(char*& p, API_TYPE ApiType)
 	WRITE(p, "  float3 expanded1;\n");
 
 	WriteSampleColor(p, "r", "depth0", 0, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth0 = 1.0f - depth0;\n");
 	WriteSampleColor(p, "r", "depth1", 1, ApiType);
+	if (ApiType == API_D3D) WRITE(p, "depth1 = 1.0f - depth1;\n");
 
 	for (int i = 0; i < 2; i++)
 	{
diff --git a/Source/Core/VideoCommon/VertexShaderGen.cpp b/Source/Core/VideoCommon/VertexShaderGen.cpp
index 28c2e1c827..084ac48aaf 100644
--- a/Source/Core/VideoCommon/VertexShaderGen.cpp
+++ b/Source/Core/VideoCommon/VertexShaderGen.cpp
@@ -385,7 +385,7 @@ static inline void GenerateVertexShader(T& out, u32 components, API_TYPE api_typ
 	}
 	else if (api_type == API_D3D)
 	{
-		out.Write("o.pos.z = o.pos.w + o.pos.z;\n");
+		out.Write("o.pos.z = -o.pos.z;\n");
 	}
 	else // OGL
 	{