Decode EFB copies used as paletted textures.

A number of games make an EFB copy in I4/I8 format, then use it as a
texture in C4/C8 format.  Detect when this happens, and decode the copy on
the GPU using the specified palette.

This has a few advantages: it allows using EFB2Tex for a few more games,
it, it preserves the resolution of scaled EFB copies, and it's probably a
bit faster.

D3D only at the moment, but porting to OpenGL should be straightforward..

Source/Core/VideoBackends/D3D/TextureCache.cpp

@@ -4,6 +4,7 @@
 
 #include "Core/HW/Memmap.h"
 #include "VideoBackends/D3D/D3DBase.h"
+#include "VideoBackends/D3D/D3DShader.h"
 #include "VideoBackends/D3D/D3DState.h"
 #include "VideoBackends/D3D/D3DUtil.h"
 #include "VideoBackends/D3D/FramebufferManager.h"
@@ -14,6 +15,7 @@
 #include "VideoBackends/D3D/TextureEncoder.h"
 #include "VideoBackends/D3D/VertexShaderCache.h"
 #include "VideoCommon/ImageWrite.h"
+#include "VideoCommon/LookUpTables.h"
 #include "VideoCommon/RenderBase.h"
 #include "VideoCommon/VideoConfig.h"
 
@@ -179,17 +181,165 @@ void TextureCache::TCacheEntry::FromRenderTarget(u32 dstAddr, unsigned int dstFo
 
 		size_in_bytes = (u32)encoded_size;
 
-		TextureCache::MakeRangeDynamic(addr, (u32)encoded_size);
+		TextureCache::MakeRangeDynamic(dstAddr, (u32)encoded_size);
 
 		this->hash = hash;
 	}
 }
 
+const char palette_shader[] =
+R"HLSL(
+sampler samp0 : register(s0);
+Texture2DArray Tex0 : register(t0);
+Buffer<uint> Tex1 : register(t1);
+uniform float Multiply;
+
+uint Convert3To8(uint v)
+{
+	// Swizzle bits: 00000123 -> 12312312
+	return (v << 5) | (v << 2) | (v >> 1);
+}
+
+uint Convert4To8(uint v)
+{
+	// Swizzle bits: 00001234 -> 12341234
+	return (v << 4) | v;
+}
+
+uint Convert5To8(uint v)
+{
+	// Swizzle bits: 00012345 -> 12345123
+	return (v << 3) | (v >> 2);
+}
+
+uint Convert6To8(uint v)
+{
+	// Swizzle bits: 00123456 -> 12345612
+	return (v << 2) | (v >> 4);
+}
+
+float4 DecodePixel_RGB5A3(uint val)
+{
+	int r,g,b,a;
+	if ((val&0x8000))
+	{
+		r=Convert5To8((val>>10) & 0x1f);
+		g=Convert5To8((val>>5 ) & 0x1f);
+		b=Convert5To8((val    ) & 0x1f);
+		a=0xFF;
+	}
+	else
+	{
+		a=Convert3To8((val>>12) & 0x7);
+		r=Convert4To8((val>>8 ) & 0xf);
+		g=Convert4To8((val>>4 ) & 0xf);
+		b=Convert4To8((val    ) & 0xf);
+	}
+	return float4(r, g, b, a) / 255;
+}
+
+float4 DecodePixel_RGB565(uint val)
+{
+	int r, g, b, a;
+	r = Convert5To8((val >> 11) & 0x1f);
+	g = Convert6To8((val >> 5) & 0x3f);
+	b = Convert5To8((val)& 0x1f);
+	a = 0xFF;
+	return float4(r, g, b, a) / 255;
+}
+
+float4 DecodePixel_IA8(uint val)
+{
+	int i = val & 0xFF;
+	int a = val >> 8;
+	return float4(i, i, i, a) / 255;
+}
+
+void main(
+	out float4 ocol0 : SV_Target,
+	in float4 pos : SV_Position,
+	in float3 uv0 : TEXCOORD0)
+{
+	uint src = round(Tex0.Sample(samp0,uv0) * Multiply).r;
+	src = Tex1.Load(src);
+	src = ((src << 8) & 0xFF00) | (src >> 8);
+	ocol0 = DECODE(src);
+}
+)HLSL";
+
+void TextureCache::ConvertTexture(TCacheEntryBase* entry, TCacheEntryBase* unconverted, void* palette, TlutFormat format)
+{
+	g_renderer->ResetAPIState();
+
+	// stretch picture with increased internal resolution
+	const D3D11_VIEWPORT vp = CD3D11_VIEWPORT(0.f, 0.f, (float)unconverted->config.width, (float)unconverted->config.height);
+	D3D::context->RSSetViewports(1, &vp);
+
+	D3D11_BOX box{ 0, 0, 0, 512, 1, 1 };
+	D3D::context->UpdateSubresource(palette_buf, 0, &box, palette, 0, 0);
+
+	D3D::stateman->SetTexture(1, palette_buf_srv);
+
+	float params[4] = { unconverted->format == 0 ? 15.f : 255.f };
+	D3D::context->UpdateSubresource(palette_uniform, 0, nullptr, &params, 0, 0);
+	D3D::stateman->SetPixelConstants(palette_uniform);
+
+	const D3D11_RECT sourcerect = CD3D11_RECT(0, 0, unconverted->config.width, unconverted->config.height);
+
+	D3D::SetPointCopySampler();
+
+	// Make sure we don't draw with the texture set as both a source and target.
+	// (This can happen because we don't unbind textures when we free them.)
+	D3D::stateman->UnsetTexture(static_cast<TCacheEntry*>(entry)->texture->GetSRV());
+
+	D3D::context->OMSetRenderTargets(1, &static_cast<TCacheEntry*>(entry)->texture->GetRTV(), nullptr);
+
+	// Create texture copy
+	D3D::drawShadedTexQuad(
+		static_cast<TCacheEntry*>(unconverted)->texture->GetSRV(),
+		&sourcerect, unconverted->config.width, unconverted->config.height,
+		palette_pixel_shader[format],
+		VertexShaderCache::GetSimpleVertexShader(), VertexShaderCache::GetSimpleInputLayout(),
+		GeometryShaderCache::GetCopyGeometryShader());
+
+	D3D::context->OMSetRenderTargets(1, &FramebufferManager::GetEFBColorTexture()->GetRTV(), FramebufferManager::GetEFBDepthTexture()->GetDSV());
+
+	g_renderer->RestoreAPIState();
+}
+
+ID3D11PixelShader *GetConvertShader(const char* Type)
+{
+	std::string shader = "#define DECODE DecodePixel_";
+	shader.append(Type);
+	shader.append("\n");
+	shader.append(palette_shader);
+	return D3D::CompileAndCreatePixelShader(shader);
+}
+
 TextureCache::TextureCache()
 {
 	// FIXME: Is it safe here?
 	g_encoder = new PSTextureEncoder;
 	g_encoder->Init();
+
+	palette_buf = nullptr;
+	palette_buf_srv = nullptr;
+	palette_uniform = nullptr;
+	palette_pixel_shader[GX_TL_IA8] = GetConvertShader("IA8");
+	palette_pixel_shader[GX_TL_RGB565] = GetConvertShader("RGB565");
+	palette_pixel_shader[GX_TL_RGB5A3] = GetConvertShader("RGB5A3");
+	auto lutBd = CD3D11_BUFFER_DESC(sizeof(u16) * 256, D3D11_BIND_SHADER_RESOURCE);
+	HRESULT hr = D3D::device->CreateBuffer(&lutBd, nullptr, &palette_buf);
+	CHECK(SUCCEEDED(hr), "create palette decoder lut buffer");
+	D3D::SetDebugObjectName(palette_buf, "texture decoder lut buffer");
+	auto outlutUavDesc = CD3D11_SHADER_RESOURCE_VIEW_DESC(palette_buf, DXGI_FORMAT_R16_UINT, 0, 256, 0);
+	hr = D3D::device->CreateShaderResourceView(palette_buf, &outlutUavDesc, &palette_buf_srv);
+	CHECK(SUCCEEDED(hr), "create palette decoder lut srv");
+	D3D::SetDebugObjectName(palette_buf_srv, "texture decoder lut srv");
+	const D3D11_BUFFER_DESC cbdesc = CD3D11_BUFFER_DESC(16, D3D11_BIND_CONSTANT_BUFFER, D3D11_USAGE_DEFAULT);
+	hr = D3D::device->CreateBuffer(&cbdesc, nullptr, &palette_uniform);
+	CHECK(SUCCEEDED(hr), "Create palette decoder constant buffer");
+	D3D::SetDebugObjectName((ID3D11DeviceChild*)palette_uniform, "a constant buffer used in TextureCache::CopyRenderTargetToTexture");
 }
 
 TextureCache::~TextureCache()
@@ -200,6 +350,12 @@ TextureCache::~TextureCache()
 	g_encoder->Shutdown();
 	delete g_encoder;
 	g_encoder = nullptr;
+
+	SAFE_RELEASE(palette_buf);
+	SAFE_RELEASE(palette_buf_srv);
+	SAFE_RELEASE(palette_uniform);
+	for (ID3D11PixelShader*& shader : palette_pixel_shader)
+		SAFE_RELEASE(shader);
 }
 
 }

Source/Core/VideoBackends/D3D/TextureCache.h

@@ -42,8 +42,15 @@ class TextureCache : public ::TextureCache
 
 	u64 EncodeToRamFromTexture(u32 address, void* source_texture, u32 SourceW, u32 SourceH, bool bFromZBuffer, bool bIsIntensityFmt, u32 copyfmt, int bScaleByHalf, const EFBRectangle& source) {return 0;};
 
+	virtual void ConvertTexture(TCacheEntryBase* entry, TCacheEntryBase* unconverted, void* palette, TlutFormat format) override;
+
 	void CompileShaders() override { }
 	void DeleteShaders() override { }
+
+	ID3D11Buffer *palette_buf;
+	ID3D11ShaderResourceView *palette_buf_srv;
+	ID3D11Buffer *palette_uniform;
+	ID3D11PixelShader* palette_pixel_shader[3];
 };
 
 }

Source/Core/VideoBackends/OGL/TextureCache.cpp

@@ -204,20 +204,20 @@ void TextureCache::TCacheEntry::FromRenderTarget(u32 dstAddr, unsigned int dstFo
 	if (false == g_ActiveConfig.bCopyEFBToTexture)
 	{
 		int encoded_size = TextureConverter::EncodeToRamFromTexture(
-			addr,
+			dstAddr,
 			read_texture,
 			srcFormat == PEControl::Z24,
 			isIntensity,
 			dstFormat,
 			scaleByHalf,
 			srcRect);
 
-		u8* dst = Memory::GetPointer(addr);
+		u8* dst = Memory::GetPointer(dstAddr);
 		u64 const new_hash = GetHash64(dst,encoded_size,g_ActiveConfig.iSafeTextureCache_ColorSamples);
 
 		size_in_bytes = (u32)encoded_size;
 
-		TextureCache::MakeRangeDynamic(addr,encoded_size);
+		TextureCache::MakeRangeDynamic(dstAddr, encoded_size);
 
 		hash = new_hash;
 	}
@@ -359,4 +359,10 @@ void TextureCache::DeleteShaders()
 	s_DepthMatrixProgram.Destroy();
 }
 
+void TextureCache::ConvertTexture(TextureCache::TCacheEntryBase* entry, TCacheEntryBase* unconverted, void* palette, TlutFormat format)
+{
+	// TODO: Implement.
+	return;
+}
+
 }

Source/Core/VideoBackends/OGL/TextureCache.h

@@ -48,6 +48,7 @@ class TextureCache : public ::TextureCache
 	~TextureCache();
 
 	TCacheEntryBase* CreateTexture(const TCacheEntryConfig& config) override;
+	void ConvertTexture(TCacheEntryBase* entry, TCacheEntryBase* unconverted, void* palette, TlutFormat format) override;
 
 	void CompileShaders() override;
 	void DeleteShaders() override;