From 16f68fec5c2a16a8aefc481d000887731013aca7 Mon Sep 17 00:00:00 2001 From: Ced2911 Date: Sat, 24 Aug 2013 13:24:19 +0200 Subject: [PATCH] sync dx9 gpu with gles --- GPU/Directx9/FragmentShaderGenerator.cpp | 46 ++++++------------------ GPU/Directx9/Framebuffer.cpp | 4 +-- GPU/Directx9/ShaderManager.cpp | 3 +- GPU/Directx9/StateMapping.cpp | 11 +++--- GPU/Directx9/TextureCache.cpp | 34 ++++++++++-------- GPU/Directx9/VertexDecoder.h | 2 +- GPU/Directx9/VertexShaderGenerator.cpp | 17 +++++---- 7 files changed, 48 insertions(+), 69 deletions(-) diff --git a/GPU/Directx9/FragmentShaderGenerator.cpp b/GPU/Directx9/FragmentShaderGenerator.cpp index c466930de8..c3615f12e3 100644 --- a/GPU/Directx9/FragmentShaderGenerator.cpp +++ b/GPU/Directx9/FragmentShaderGenerator.cpp @@ -35,17 +35,16 @@ static bool IsAlphaTestTriviallyTrue() { case GE_COMP_ALWAYS: return true; case GE_COMP_GEQUAL: - if (alphaTestRef == 0) - return true; + return alphaTestRef == 0; // This breaks the trees in MotoGP, for example. // case GE_COMP_GREATER: - //if (alphaTestRef == 0 && (gstate.alphaBlendEnable & 1) && gstate.getBlendFuncA() == GE_SRCBLEND_SRCALPHA && gstate.getBlendFuncB() == GE_SRCBLEND_INVSRCALPHA) + //if (alphaTestRef == 0 && (gstate.isAlphaBlendEnabled() & 1) && gstate.getBlendFuncA() == GE_SRCBLEND_SRCALPHA && gstate.getBlendFuncB() == GE_SRCBLEND_INVSRCALPHA) // return true; case GE_COMP_LEQUAL: - if (alphaTestRef == 255) - return true; + return alphaTestRef == 255; + default: return false; } @@ -104,8 +103,8 @@ void ComputeFragmentShaderID(FragmentShaderID *id) { bool enableColorDoubling = gstate.isColorDoublingEnabled(); // This isn't really correct, but it's a hack to get doubled blend modes to work more correctly. bool enableAlphaDoubling = CanDoubleSrcBlendMode(); - bool doTextureProjection = gstate.getUVGenMode() == 1; - bool doTextureAlpha = (gstate.texfunc & 0x100) != 0; + bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX; + bool doTextureAlpha = gstate.isTextureAlphaUsed(); // All texfuncs except replace are the same for RGB as for RGBA with full alpha. if (gstate_c.textureFullAlpha && gstate.getTextureFunction() != GE_TEXFUNC_REPLACE) @@ -133,34 +132,10 @@ void ComputeFragmentShaderID(FragmentShaderID *id) { // Missing: Z depth range // Also, logic ops etc, of course. Urgh. -#if 0 -void GenerateFragmentShader(char *buffer) { - //-------------------------------------------------------------------------------------- - // Pixel shader - //-------------------------------------------------------------------------------------- - const char * pscode = - " sampler s: register(s0); " - " struct PS_IN " - " { " - " float3 Uv : TEXCOORD0; " - " float4 C1 : COLOR0; " // Vertex color - " float4 C2 : COLOR1; " // Vertex color - " }; " - " " - " float4 main( PS_IN In ) : COLOR " - " { " - //" float4 c = In.C1; " - " float4 c = tex2D(s, In.Uv.xy); " - " return c; " - " } "; - - strcpy(buffer, pscode); -} -#else void GenerateFragmentShader(char *buffer) { char *p = buffer; - int lmode = (gstate.lmode & 1) && gstate.isLightingEnabled(); + int lmode = lmode = gstate.isUsingSecondaryColor() && gstate.isLightingEnabled(); int doTexture = gstate.isTextureMapEnabled() && !gstate.isModeClear(); bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough() && !gstate.isModeClear(); bool enableAlphaTest = gstate.isAlphaTestEnabled() && !IsAlphaTestTriviallyTrue() && !gstate.isModeClear(); @@ -168,8 +143,8 @@ void GenerateFragmentShader(char *buffer) { bool enableColorDoubling = gstate.isColorDoublingEnabled(); // This isn't really correct, but it's a hack to get doubled blend modes to work more correctly. bool enableAlphaDoubling = CanDoubleSrcBlendMode(); - bool doTextureProjection = gstate.getUVGenMode() == 1; - bool doTextureAlpha = (gstate.texfunc & 0x100) != 0; + bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX; + bool doTextureAlpha = gstate.isTextureAlphaUsed(); if (gstate_c.textureFullAlpha && gstate.getTextureFunction() != GE_TEXFUNC_REPLACE) doTextureAlpha = false; @@ -179,7 +154,7 @@ void GenerateFragmentShader(char *buffer) { if (enableAlphaTest || enableColorTest) { WRITE(p, "float4 u_alphacolorref;\n"); - WRITE(p, "float3 u_colormask;\n"); + WRITE(p, "float4 u_colormask;\n"); } if (gstate.isTextureMapEnabled()) WRITE(p, "float3 u_texenv;\n"); @@ -324,4 +299,3 @@ void GenerateFragmentShader(char *buffer) { } WRITE(p, "}\n"); } -#endif diff --git a/GPU/Directx9/Framebuffer.cpp b/GPU/Directx9/Framebuffer.cpp index 049517efc9..981df93d76 100644 --- a/GPU/Directx9/Framebuffer.cpp +++ b/GPU/Directx9/Framebuffer.cpp @@ -37,7 +37,7 @@ enum { }; static bool MaskedEqual(u32 addr1, u32 addr2) { - return (addr1 & 0x3FFFFFF) == (addr2 & 0x3FFFFFF); + return (addr1 & 0x03FFFFFF) == (addr2 & 0x03FFFFFF); } inline u16 RGBA8888toRGB565(u32 px) { @@ -876,7 +876,7 @@ void FramebufferManager::DecimateFBOs() { VirtualFramebuffer *vfb = vfbs_[i]; int age = frameLastFramebufUsed - vfb->last_frame_used; - if(useMem && !age && !vfb->memoryUpdated) { + if(useMem && age == 0 && !vfb->memoryUpdated) { ReadFramebufferToMemory(vfb); } diff --git a/GPU/Directx9/ShaderManager.cpp b/GPU/Directx9/ShaderManager.cpp index 0509684d82..13da6efef0 100644 --- a/GPU/Directx9/ShaderManager.cpp +++ b/GPU/Directx9/ShaderManager.cpp @@ -373,7 +373,8 @@ void LinkedShader::updateUniforms() { int h = gstate.getTextureHeight(0); float widthFactor = (float)w / (float)gstate_c.curTextureWidth; float heightFactor = (float)h / (float)gstate_c.curTextureHeight; - if (gstate.getUVGenMode() == 0) { + // Not sure what GE_TEXMAP_UNKNOWN is, but seen in Riviera. Treating the same as GE_TEXMAP_TEXTURE_COORDS works. + if (gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_COORDS || gstate.getUVGenMode() == GE_TEXMAP_UNKNOWN) { static const float rescale[4] = {1.0f, 2*127.5f/128.f, 2*32767.5f/32768.f, 1.0f}; float factor = rescale[(gstate.vertType & GE_VTYPE_TC_MASK) >> GE_VTYPE_TC_SHIFT]; uvscaleoff[0] = gstate_c.uv.uScale * factor * widthFactor; diff --git a/GPU/Directx9/StateMapping.cpp b/GPU/Directx9/StateMapping.cpp index ee74301fd0..1ce829b8c2 100644 --- a/GPU/Directx9/StateMapping.cpp +++ b/GPU/Directx9/StateMapping.cpp @@ -208,10 +208,9 @@ void TransformDrawEngine::ApplyDrawState(int prim) { dxstate.cullMode.set(false, false); // Depth Test - bool depthMask = (gstate.clearmode >> 10) & 1; dxstate.depthTest.enable(); dxstate.depthFunc.set(D3DCMP_ALWAYS); - dxstate.depthWrite.set(depthMask); + dxstate.depthWrite.set(gstate.isClearModeDepthWriteEnabled()); // Color Test bool colorMask = (gstate.clearmode >> 8) & 1; @@ -288,13 +287,13 @@ void TransformDrawEngine::ApplyDrawState(int prim) { // Scissor int scissorX1 = gstate.getScissorX1(); int scissorY1 = gstate.getScissorY1(); - int scissorX2 = gstate.getScissorX2(); - int scissorY2 = gstate.getScissorY2(); + int scissorX2 = gstate.getScissorX2() + 1; + int scissorY2 = gstate.getScissorY2() + 1; // This is a bit of a hack as the render buffer isn't always that size if (scissorX1 == 0 && scissorY1 == 0 - && scissorX2 >= (int) (gstate_c.curRTWidth - 1) - && scissorY2 >= (int) (gstate_c.curRTHeight - 1)) { + && scissorX2 >= (int) gstate_c.curRTWidth + && scissorY2 >= (int) gstate_c.curRTHeight) { dxstate.scissorTest.disable(); } else { dxstate.scissorTest.enable(); diff --git a/GPU/Directx9/TextureCache.cpp b/GPU/Directx9/TextureCache.cpp index 19a8c90f56..dc0bcbe2ca 100644 --- a/GPU/Directx9/TextureCache.cpp +++ b/GPU/Directx9/TextureCache.cpp @@ -129,7 +129,7 @@ void TextureCache::Decimate() { } void TextureCache::Invalidate(u32 addr, int size, GPUInvalidationType type) { - addr &= 0xFFFFFFF; + addr &= 0x0FFFFFFF; u32 addr_end = addr + size; // They could invalidate inside the texture, let's just give a bit of leeway. @@ -194,11 +194,10 @@ inline void TextureCache::AttachFramebuffer(TexCacheEntry *entry, u32 address, V if (entry->format != framebuffer->format) { WARN_LOG_REPORT_ONCE(diffFormat1, HLE, "Render to texture with different formats %d != %d", entry->format, framebuffer->format); // If it already has one, let's hope that one is correct. - // Try to not bind FB now as it seems to be attached some strange stuff on top of the original FB. - //AttachFramebufferInvalid(entry, framebuffer); + AttachFramebufferInvalid(entry, framebuffer); } else { AttachFramebufferValid(entry, framebuffer); - } + } // TODO: Delete the original non-fbo texture too. } } else if (g_Config.iRenderingMode == FB_NON_BUFFERED_MODE || g_Config.iRenderingMode == FB_BUFFERED_MODE) { @@ -212,11 +211,11 @@ inline void TextureCache::AttachFramebuffer(TexCacheEntry *entry, u32 address, V if (framebuffer->format != entry->format) { WARN_LOG_REPORT_ONCE(diffFormat2, HLE, "Render to texture with different formats %d != %d at %08x", entry->format, framebuffer->format, address); // TODO: Use an FBO to translate the palette? - AttachFramebufferValid(entry, framebuffer); + AttachFramebufferValid(entry, framebuffer); } else if ((entry->addr - address) / entry->bufw < framebuffer->height) { WARN_LOG_REPORT_ONCE(subarea, HLE, "Render to area containing texture at %08x", address); // TODO: Keep track of the y offset. - AttachFramebufferInvalid(entry, framebuffer); + AttachFramebufferValid(entry, framebuffer); } } } @@ -242,22 +241,18 @@ void TextureCache::NotifyFramebuffer(u32 address, VirtualFramebuffer *framebuffe case NOTIFY_FB_CREATED: case NOTIFY_FB_UPDATED: for (auto it = cache.lower_bound(cacheKey), end = cache.upper_bound(cacheKeyEnd); it != end; ++it) { - AttachFramebuffer(&it->second, address, framebuffer, it->first == cacheKey); + AttachFramebuffer(&it->second, address | 0x04000000, framebuffer, it->first == cacheKey); } break; case NOTIFY_FB_DESTROYED: for (auto it = cache.lower_bound(cacheKey), end = cache.upper_bound(cacheKeyEnd); it != end; ++it) { - DetachFramebuffer(&it->second, address, framebuffer); + DetachFramebuffer(&it->second, address | 0x04000000, framebuffer); } break; } } -static u32 GetClutAddr() { - return ((gstate.clutaddr & 0xFFFFFF) | ((gstate.clutaddrupper << 8) & 0x0F000000)); -} - static u32 GetClutIndex(u32 index) { const u32 clutBase = gstate.getClutIndexStartPos(); const u32 clutMask = gstate.getClutIndexMask(); @@ -1297,6 +1292,9 @@ void *TextureCache::DecodeTextureLevel(GETextureFormat format, GEPaletteFormat c case GE_TFMT_CLUT4: { dstFmt = getClutDestFormat(clutformat); + // Don't allow this to be less than 16 bytes (32 * 4 / 8 = 16.) + if (bufw < 32) + bufw = 32; const bool mipmapShareClut = (gstate.texmode & 0x100) == 0; const int clutSharingOffset = mipmapShareClut ? 0 : level * 16; @@ -1355,18 +1353,24 @@ void *TextureCache::DecodeTextureLevel(GETextureFormat format, GEPaletteFormat c break; case GE_TFMT_CLUT8: + if (bufw < 8) + bufw = 8; dstFmt = getClutDestFormat(gstate.getClutPaletteFormat()); texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()]; finalBuf = readIndexedTex(level, texaddr, 1, dstFmt); break; case GE_TFMT_CLUT16: + if (bufw < 8) + bufw = 8; dstFmt = getClutDestFormat(gstate.getClutPaletteFormat()); texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()]; finalBuf = readIndexedTex(level, texaddr, 2, dstFmt); break; case GE_TFMT_CLUT32: + if (bufw < 4) + bufw = 4; dstFmt = getClutDestFormat(gstate.getClutPaletteFormat()); texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()]; finalBuf = readIndexedTex(level, texaddr, 4, dstFmt); @@ -1375,6 +1379,8 @@ void *TextureCache::DecodeTextureLevel(GETextureFormat format, GEPaletteFormat c case GE_TFMT_4444: case GE_TFMT_5551: case GE_TFMT_5650: + if (bufw < 8) + bufw = 8; if (format == GE_TFMT_4444) dstFmt = D3DFMT_A4R4G4B4; else if (format == GE_TFMT_5551) @@ -1398,6 +1404,8 @@ void *TextureCache::DecodeTextureLevel(GETextureFormat format, GEPaletteFormat c break; case GE_TFMT_8888: + if (bufw < 4) + bufw = 4; dstFmt = D3DFMT_A8R8G8B8; if (!(gstate.texmode & 1)) { // Special case: if we don't need to deal with packing, we don't need to copy. @@ -1459,7 +1467,6 @@ void *TextureCache::DecodeTextureLevel(GETextureFormat format, GEPaletteFormat c } w = (w + 3) & ~3; finalBuf = tmpTexBuf32.data(); - //TextureConvertColors(finalBuf, finalBuf, dstFmt, bufw * h); } break; @@ -1481,7 +1488,6 @@ void *TextureCache::DecodeTextureLevel(GETextureFormat format, GEPaletteFormat c } w = (w + 3) & ~3; finalBuf = tmpTexBuf32.data(); - //TextureConvertColors(finalBuf, finalBuf, dstFmt, bufw * h); } break; diff --git a/GPU/Directx9/VertexDecoder.h b/GPU/Directx9/VertexDecoder.h index 809ee6d3d3..acfc8ad75b 100644 --- a/GPU/Directx9/VertexDecoder.h +++ b/GPU/Directx9/VertexDecoder.h @@ -103,7 +103,7 @@ public: u32 InjectUVs(u8 *decoded, const void *verts, float *customuv, int count) const; bool hasColor() const { return col != 0; } - int VertexSize() const { return size; } + int VertexSize() const { return size; } // PSP format size void Step_WeightsU8() const; void Step_WeightsU16() const; diff --git a/GPU/Directx9/VertexShaderGenerator.cpp b/GPU/Directx9/VertexShaderGenerator.cpp index 656ac33c29..33b1b5a5c0 100644 --- a/GPU/Directx9/VertexShaderGenerator.cpp +++ b/GPU/Directx9/VertexShaderGenerator.cpp @@ -50,7 +50,7 @@ int TranslateNumBones(int bones) { void ComputeVertexShaderID(VertexShaderID *id, int prim, bool useHWTransform) { const u32 vertType = gstate.vertType; int doTexture = gstate.isTextureMapEnabled() && !gstate.isModeClear(); - bool doTextureProjection = gstate.getUVGenMode() == 1; + bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX; bool hasColor = (vertType & GE_VTYPE_COL_MASK) != 0; bool hasNormal = (vertType & GE_VTYPE_NRM_MASK) != 0; @@ -77,9 +77,9 @@ void ComputeVertexShaderID(VertexShaderID *id, int prim, bool useHWTransform) { id->d[0] |= gstate.getUVGenMode() << 16; // The next bits are used differently depending on UVgen mode - if (gstate.getUVGenMode() == 1) { + if (gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX) { id->d[0] |= gstate.getUVProjMode() << 18; - } else if (gstate.getUVGenMode() == 2) { + } else if (gstate.getUVGenMode() == GE_TEXMAP_ENVIRONMENT_MAP) { id->d[0] |= gstate.getUVLS0() << 18; id->d[0] |= gstate.getUVLS1() << 20; } @@ -90,7 +90,7 @@ void ComputeVertexShaderID(VertexShaderID *id, int prim, bool useHWTransform) { // Okay, d[1] coming up. ============== - if (gstate.isLightingEnabled() || gstate.getUVGenMode() == 2) { + if (gstate.isLightingEnabled() || gstate.getUVGenMode() == GE_TEXMAP_ENVIRONMENT_MAP) { // Light bits for (int i = 0; i < 4; i++) { id->d[1] |= gstate.getLightComputation(i) << (i * 4); @@ -136,12 +136,12 @@ void GenerateVertexShader(int prim, char *buffer, bool useHWTransform) { bool enableFog = gstate.isFogEnabled() && !gstate.isModeThrough() && !gstate.isModeClear(); bool throughmode = (vertType & GE_VTYPE_THROUGH_MASK) != 0; bool flipV = gstate_c.flipTexture; - bool doTextureProjection = gstate.getUVGenMode() == 1; + bool doTextureProjection = gstate.getUVGenMode() == GE_TEXMAP_TEXTURE_MATRIX; DoLightComputation doLight[4] = {LIGHT_OFF, LIGHT_OFF, LIGHT_OFF, LIGHT_OFF}; if (useHWTransform) { - int shadeLight0 = gstate.getUVGenMode() == 2 ? gstate.getUVLS0() : -1; - int shadeLight1 = gstate.getUVGenMode() == 2 ? gstate.getUVLS1() : -1; + int shadeLight0 = gstate.getUVGenMode() == GE_TEXMAP_ENVIRONMENT_MAP ? gstate.getUVLS0() : -1; + int shadeLight1 = gstate.getUVGenMode() == GE_TEXMAP_ENVIRONMENT_MAP ? gstate.getUVLS1() : -1; for (int i = 0; i < 4; i++) { if (i == shadeLight0 || i == shadeLight1) doLight[i] = LIGHT_SHADE; @@ -194,7 +194,7 @@ void GenerateVertexShader(int prim, char *buffer, bool useHWTransform) { if (type != GE_LIGHTTYPE_DIRECTIONAL) WRITE(p, "float3 u_lightatt%i;\n", i); - if (type == GE_LIGHTTYPE_SPOT) { + if (type == GE_LIGHTTYPE_SPOT || type == GE_LIGHTTYPE_UNKNOWN) { WRITE(p, "float u_lightangle%i;\n", i); WRITE(p, "float u_lightspotCoef%i;\n", i); } @@ -529,7 +529,6 @@ void GenerateVertexShader(int prim, char *buffer, bool useHWTransform) { temp_tc = "float4(0.0, 0.0, 1.0, 1.0)"; break; } - //WRITE(p, " Out.Uv.xyz = (u_texmtx * %s).xyz * float3(u_uvscaleoffset.xy, 1.0);\n", temp_tc.c_str()); WRITE(p, " Out.Uv.xyz = mul(%s,u_texmtx).xyz * float3(u_uvscaleoffset.xy, 1.0);\n", temp_tc.c_str()); } // Transform by texture matrix. XYZ as we are doing projection mapping.