Merge pull request #10898 from hrydgard/restore-hw-skinning-part-1

Restore hw skinning part 1

GPU/Common/ShaderId.cpp

@@ -45,6 +45,7 @@ std::string VertexShaderDesc(const ShaderID &id) {
 		}
 	}
 	if (id.Bits(VS_BIT_MATERIAL_UPDATE, 3)) desc << "MatUp:" << id.Bits(VS_BIT_MATERIAL_UPDATE, 3) << " ";
+	if (id.Bits(VS_BIT_WEIGHT_FMTSCALE, 2)) desc << "WScale " << id.Bits(VS_BIT_WEIGHT_FMTSCALE, 2) << " ";
 	if (id.Bit(VS_BIT_FLATSHADE)) desc << "Flat ";
 
 	if (id.Bit(VS_BIT_BEZIER)) desc << "Bezier ";

GPU/Common/ShaderId.h

@@ -44,7 +44,8 @@ enum {
 	VS_BIT_LIGHT2_ENABLE = 54,
 	VS_BIT_LIGHT3_ENABLE = 55,
 	VS_BIT_LIGHTING_ENABLE = 56,
-	// 57 - 61 are free.
+	VS_BIT_WEIGHT_FMTSCALE = 57,  // only two bits
+	// 59 - 61 are free.
 	VS_BIT_FLATSHADE = 62, // 1 bit
 	VS_BIT_BEZIER = 63, // 1 bit
 	// No more free

GPU/Common/VertexDecoderCommon.cpp

@@ -72,17 +72,21 @@ int DecFmtSize(u8 fmt) {
 }
 
 void DecVtxFormat::ComputeID() {
-	id = uvfmt | (c0fmt << 4) | (c1fmt << 8) | (nrmfmt << 12) | (posfmt << 16);
+	id = w0fmt | (w1fmt << 4) | (uvfmt << 8) | (c0fmt << 12) | (c1fmt << 16) | (nrmfmt << 20) | (posfmt << 24);
 }
 
 void DecVtxFormat::InitializeFromID(uint32_t id) {
 	this->id = id;
-	uvfmt = (id & 0xF);
-	c0fmt = ((id >> 4) & 0xF);
-	c1fmt = ((id >> 8) & 0xF);
-	nrmfmt = ((id >> 12) & 0xF);
-	posfmt = ((id >> 16) & 0xF);
-	uvoff = 0;
+	w0fmt = ((id) & 0xF);
+	w1fmt = ((id >> 4) & 0xF);
+	uvfmt = ((id >> 8) & 0xF);
+	c0fmt = ((id >> 12) & 0xF);
+	c1fmt = ((id >> 16) & 0xF);
+	nrmfmt = ((id >> 20) & 0xF);
+	posfmt = ((id >> 24) & 0xF);
+	w0off = 0;
+	w1off = w0off + DecFmtSize(w0fmt);
+	uvoff = w1off + DecFmtSize(w1fmt);
 	c0off = uvoff + DecFmtSize(uvfmt);
 	c1off = c0off + DecFmtSize(c0fmt);
 	nrmoff = c1off + DecFmtSize(c1fmt);

GPU/Common/VertexDecoderCommon.h

@@ -67,6 +67,8 @@ enum {
 int DecFmtSize(u8 fmt);
 
 struct DecVtxFormat {
+	u8 w0fmt; u8 w0off;  // first 4 weights
+	u8 w1fmt; u8 w1off;  // second 4 weights
 	u8 uvfmt; u8 uvoff;
 	u8 c0fmt; u8 c0off;  // First color
 	u8 c1fmt; u8 c1off;
@@ -345,6 +347,61 @@ public:
 		}
 	}
 
+	void ReadWeights(float weights[8]) const {
+		const float *f = (const float *)(data_ + decFmt_.w0off);
+		const u8 *b = (const u8 *)(data_ + decFmt_.w0off);
+		const u16 *s = (const u16 *)(data_ + decFmt_.w0off);
+		switch (decFmt_.w0fmt) {
+		case DEC_FLOAT_1:
+		case DEC_FLOAT_2:
+		case DEC_FLOAT_3:
+		case DEC_FLOAT_4:
+			for (int i = 0; i <= decFmt_.w0fmt - DEC_FLOAT_1; i++)
+				weights[i] = f[i];
+			break;
+		case DEC_U8_1: weights[0] = b[0] * (1.f / 128.f); break;
+		case DEC_U8_2: for (int i = 0; i < 2; i++) weights[i] = b[i] * (1.f / 128.f); break;
+		case DEC_U8_3: for (int i = 0; i < 3; i++) weights[i] = b[i] * (1.f / 128.f); break;
+		case DEC_U8_4: for (int i = 0; i < 4; i++) weights[i] = b[i] * (1.f / 128.f); break;
+		case DEC_U16_1: weights[0] = s[0] * (1.f / 32768.f); break;
+		case DEC_U16_2: for (int i = 0; i < 2; i++) weights[i] = s[i] * (1.f / 32768.f); break;
+		case DEC_U16_3: for (int i = 0; i < 3; i++) weights[i] = s[i] * (1.f / 32768.f); break;
+		case DEC_U16_4: for (int i = 0; i < 4; i++) weights[i] = s[i] * (1.f / 32768.f); break;
+		default:
+			ERROR_LOG_REPORT_ONCE(fmtw0, G3D, "Reader: Unsupported W0 Format %d", decFmt_.w0fmt);
+			memset(weights, 0, sizeof(float) * 8);
+			break;
+		}
+
+		f = (const float *)(data_ + decFmt_.w1off);
+		b = (const u8 *)(data_ + decFmt_.w1off);
+		s = (const u16 *)(data_ + decFmt_.w1off);
+		switch (decFmt_.w1fmt) {
+		case 0:
+			// It's fine for there to be w0 weights but not w1.
+			break;
+		case DEC_FLOAT_1:
+		case DEC_FLOAT_2:
+		case DEC_FLOAT_3:
+		case DEC_FLOAT_4:
+			for (int i = 0; i <= decFmt_.w1fmt - DEC_FLOAT_1; i++)
+				weights[i+4] = f[i];
+			break;
+		case DEC_U8_1: weights[4] = b[0] * (1.f / 128.f); break;
+		case DEC_U8_2: for (int i = 0; i < 2; i++) weights[i+4] = b[i] * (1.f / 128.f); break;
+		case DEC_U8_3: for (int i = 0; i < 3; i++) weights[i+4] = b[i] * (1.f / 128.f); break;
+		case DEC_U8_4: for (int i = 0; i < 4; i++) weights[i+4] = b[i] * (1.f / 128.f); break;
+		case DEC_U16_1: weights[4] = s[0] * (1.f / 32768.f); break;
+		case DEC_U16_2: for (int i = 0; i < 2; i++) weights[i+4] = s[i] * (1.f / 32768.f); break;
+		case DEC_U16_3: for (int i = 0; i < 3; i++) weights[i+4] = s[i] * (1.f / 32768.f); break;
+		case DEC_U16_4: for (int i = 0; i < 4; i++) weights[i+4] = s[i]  * (1.f / 32768.f); break;
+		default:
+			ERROR_LOG_REPORT_ONCE(fmtw1, G3D, "Reader: Unsupported W1 Format %d", decFmt_.w1fmt);
+			memset(weights + 4, 0, sizeof(float) * 4);
+			break;
+		}
+	}
+
 	bool hasColor0() const { return decFmt_.c0fmt != 0; }
 	bool hasColor1() const { return decFmt_.c1fmt != 0; }
 	bool hasNormal() const { return decFmt_.nrmfmt != 0; }

GPU/D3D11/DrawEngineD3D11.cpp

@@ -204,6 +204,18 @@ ID3D11InputLayout *DrawEngineD3D11::SetupDecFmtForDraw(D3D11VertexShader *vshade
 		D3D11_INPUT_ELEMENT_DESC VertexElements[8];
 		D3D11_INPUT_ELEMENT_DESC *VertexElement = &VertexElements[0];
 
+		// Vertices Elements orders
+		// WEIGHT
+		if (decFmt.w0fmt != 0) {
+			VertexAttribSetup(VertexElement, decFmt.w0fmt, decFmt.w0off, "TEXCOORD", 1);
+			VertexElement++;
+		}
+
+		if (decFmt.w1fmt != 0) {
+			VertexAttribSetup(VertexElement, decFmt.w1fmt, decFmt.w1off, "TEXCOORD", 2);
+			VertexElement++;
+		}
+
 		// TC
 		if (decFmt.uvfmt != 0) {
 			VertexAttribSetup(VertexElement, decFmt.uvfmt, decFmt.uvoff, "TEXCOORD", 0);

GPU/D3D11/ShaderManagerD3D11.h

@@ -101,6 +101,7 @@ public:
 	// Applies dirty changes and copies the buffer.
 	bool IsBaseDirty() { return true; }
 	bool IsLightDirty() { return true; }
+	bool IsBoneDirty() { return true; }
 
 private:
 	void Clear();

GPU/Directx9/DrawEngineDX9.cpp

@@ -175,6 +175,16 @@ IDirect3DVertexDeclaration9 *DrawEngineDX9::SetupDecFmtForDraw(VSShader *vshader
 		D3DVERTEXELEMENT9 *VertexElement = &VertexElements[0];
 
 		// Vertices Elements orders
+		// WEIGHT
+		if (decFmt.w0fmt != 0) {
+			VertexAttribSetup(VertexElement, decFmt.w0fmt, decFmt.w0off, D3DDECLUSAGE_TEXCOORD, 1);
+			VertexElement++;
+		}
+
+		if (decFmt.w1fmt != 0) {
+			VertexAttribSetup(VertexElement, decFmt.w1fmt, decFmt.w1off, D3DDECLUSAGE_TEXCOORD, 2);
+			VertexElement++;
+		}
 
 		// TC
 		if (decFmt.uvfmt != 0) {

GPU/Directx9/VertexShaderGeneratorDX9.cpp

@@ -89,6 +89,9 @@ void GenerateVertexShaderHLSL(const VShaderID &id, char *buffer, ShaderLanguage
 		}
 	}
 
+	int numBoneWeights = 0;
+	int boneWeightScale = id.Bits(VS_BIT_WEIGHT_FMTSCALE, 2);
+
 	if (lang == HLSL_DX9) {
 		WRITE(p, "#pragma warning( disable : 3571 )\n");
 		if (isModeThrough) {

GPU/GLES/DrawEngineGLES.cpp

@@ -211,6 +211,8 @@ GLRInputLayout *DrawEngineGLES::SetupDecFmtForDraw(LinkedShader *program, const
 	}
 
 	std::vector<GLRInputLayout::Entry> entries;
+	VertexAttribSetup(ATTR_W1, decFmt.w0fmt, decFmt.stride, decFmt.w0off, entries);
+	VertexAttribSetup(ATTR_W2, decFmt.w1fmt, decFmt.stride, decFmt.w1off, entries);
 	VertexAttribSetup(ATTR_TEXCOORD, decFmt.uvfmt, decFmt.stride, decFmt.uvoff, entries);
 	VertexAttribSetup(ATTR_COLOR0, decFmt.c0fmt, decFmt.stride, decFmt.c0off, entries);
 	VertexAttribSetup(ATTR_COLOR1, decFmt.c1fmt, decFmt.stride, decFmt.c1off, entries);

GPU/GLES/ShaderManagerGLES.cpp

@@ -79,6 +79,8 @@ LinkedShader::LinkedShader(GLRenderManager *render, VShaderID VSID, Shader *vs,
 	semantics.push_back({ ATTR_POSITION, "position" });
 	semantics.push_back({ ATTR_TEXCOORD, "texcoord" });
 	semantics.push_back({ ATTR_NORMAL, "normal" });
+	semantics.push_back({ ATTR_W1, "w1" });
+	semantics.push_back({ ATTR_W2, "w2" });
 	semantics.push_back({ ATTR_COLOR0, "color0" });
 	semantics.push_back({ ATTR_COLOR1, "color1" });
 
@@ -788,7 +790,7 @@ std::string ShaderManagerGLES::DebugGetShaderString(std::string id, DebugShaderT
 // as sometimes these features might have an effect on the ID bits.
 
 #define CACHE_HEADER_MAGIC 0x83277592
-#define CACHE_VERSION 7
+#define CACHE_VERSION 10
 struct CacheHeader {
 	uint32_t magic;
 	uint32_t version;

GPU/GLES/ShaderManagerGLES.h

@@ -34,8 +34,10 @@ enum {
 	ATTR_POSITION = 0,
 	ATTR_TEXCOORD = 1,
 	ATTR_NORMAL = 2,
-	ATTR_COLOR0 = 3,
-	ATTR_COLOR1 = 4,
+	ATTR_W1 = 3,
+	ATTR_W2 = 4,
+	ATTR_COLOR0 = 5,
+	ATTR_COLOR1 = 6,
 
 	ATTR_COUNT,
 };

GPU/Vulkan/PipelineManagerVulkan.cpp

@@ -76,6 +76,8 @@ static const DeclTypeInfo VComp[] = {
 };
 
 static void VertexAttribSetup(VkVertexInputAttributeDescription *attr, int fmt, int offset, PspAttributeLocation location) {
+	assert(fmt != DEC_NONE);
+	assert(fmt < ARRAY_SIZE(VComp));
 	attr->location = (uint32_t)location;
 	attr->binding = 0;
 	attr->format = VComp[fmt].type;
@@ -87,6 +89,12 @@ static void VertexAttribSetup(VkVertexInputAttributeDescription *attr, int fmt,
 // as we will only call this code when we need to create a new VkPipeline.
 static int SetupVertexAttribs(VkVertexInputAttributeDescription attrs[], const DecVtxFormat &decFmt) {
 	int count = 0;
+	if (decFmt.w0fmt != 0) {
+		VertexAttribSetup(&attrs[count++], decFmt.w0fmt, decFmt.w0off, PspAttributeLocation::W1);
+	}
+	if (decFmt.w1fmt != 0) {
+		VertexAttribSetup(&attrs[count++], decFmt.w1fmt, decFmt.w1off, PspAttributeLocation::W2);
+	}
 	if (decFmt.uvfmt != 0) {
 		VertexAttribSetup(&attrs[count++], decFmt.uvfmt, decFmt.uvoff, PspAttributeLocation::TEXCOORD);
 	}

GPU/Vulkan/ShaderManagerVulkan.cpp

@@ -350,7 +350,7 @@ VulkanFragmentShader *ShaderManagerVulkan::GetFragmentShaderFromModule(VkShaderM
 // instantaneous.
 
 #define CACHE_HEADER_MAGIC 0xff51f420 
-#define CACHE_VERSION 9
+#define CACHE_VERSION 12
 struct VulkanCacheHeader {
 	uint32_t magic;
 	uint32_t version;