Exemple #1
0
bool compileGLSLShader(bx::CommandLine& _cmdLine, uint32_t _gles, const std::string& _code, bx::WriterI* _writer)
{
	char ch = tolower(_cmdLine.findOption('\0', "type")[0]);
	const glslopt_shader_type type = ch == 'f'
		? kGlslOptShaderFragment
		: (ch == 'c' ? kGlslOptShaderCompute : kGlslOptShaderVertex);

    glslopt_target target = kGlslTargetOpenGL;
	switch (_gles)
	{
	case BX_MAKEFOURCC('M', 'T', 'L', 0):
		target = kGlslTargetMetal;
		break;

	case 2:
		target = kGlslTargetOpenGLES20;
		break;

	case 3:
		target = kGlslTargetOpenGLES30;
		break;

	default:
		target = kGlslTargetOpenGL;
		break;
	}

	glslopt_ctx* ctx = glslopt_initialize(target);

	glslopt_shader* shader = glslopt_optimize(ctx, type, _code.c_str(), 0);

	if (!glslopt_get_status(shader) )
	{
		const char* log = glslopt_get_log(shader);
		int32_t source = 0;
		int32_t line = 0;
		int32_t column = 0;
		int32_t start = 0;
		int32_t end = INT32_MAX;

		if (3 == sscanf(log, "%u:%u(%u):", &source, &line, &column)
		&&  0 != line)
		{
			start = bx::uint32_imax(1, line-10);
			end = start + 20;
		}

		printCode(_code.c_str(), line, start, end);
		fprintf(stderr, "Error: %s\n", log);
		glslopt_cleanup(ctx);
		return false;
	}

	const char* optimizedShader = glslopt_get_output(shader);

	// Trim all directives.
	while ('#' == *optimizedShader)
	{
		optimizedShader = bx::strnl(optimizedShader);
	}

	if (0 != _gles)
	{
		char* code = const_cast<char*>(optimizedShader);
		strreplace(code, "gl_FragDepthEXT", "gl_FragDepth");

		strreplace(code, "texture2DLodEXT", "texture2DLod");
		strreplace(code, "texture2DProjLodEXT", "texture2DProjLod");
		strreplace(code, "textureCubeLodEXT", "textureCubeLod");
		strreplace(code, "texture2DGradEXT", "texture2DGrad");
		strreplace(code, "texture2DProjGradEXT", "texture2DProjGrad");
		strreplace(code, "textureCubeGradEXT", "textureCubeGrad");

		strreplace(code, "shadow2DEXT", "shadow2D");
		strreplace(code, "shadow2DProjEXT", "shadow2DProj");
	}

	UniformArray uniforms;

	{
		const char* parse = optimizedShader;

		while (NULL != parse
			&&  *parse != '\0')
		{
			parse = bx::strws(parse);
			const char* eol = strchr(parse, ';');
			if (NULL != eol)
			{
				const char* qualifier = parse;
				parse = bx::strws(bx::strword(parse) );

				if (0 == strncmp(qualifier, "attribute", 9)
				||  0 == strncmp(qualifier, "varying", 7) )
				{
					// skip attributes and varyings.
					parse = eol + 1;
					continue;
				}

				if (0 != strncmp(qualifier, "uniform", 7) )
				{
					// end if there is no uniform keyword.
					parse = NULL;
					continue;
				}

				const char* precision = NULL;
				const char* typen = parse;

				if (0 == strncmp(typen, "lowp", 4)
				||  0 == strncmp(typen, "mediump", 7)
				||  0 == strncmp(typen, "highp", 5) )
				{
					precision = typen;
					typen = parse = bx::strws(bx::strword(parse) );
				}

				BX_UNUSED(precision);

				char uniformType[256];
				parse = bx::strword(parse);

				if (0 == strncmp(typen, "sampler", 7) )
				{
					strcpy(uniformType, "int");
				}
				else
				{
					bx::strlcpy(uniformType, typen, parse-typen+1);
				}

				const char* name = parse = bx::strws(parse);

				char uniformName[256];
				uint8_t num = 1;
				const char* array = bx::strnstr(name, "[", eol-parse);
				if (NULL != array)
				{
					bx::strlcpy(uniformName, name, array-name+1);

					char arraySize[32];
					const char* end = bx::strnstr(array, "]", eol-array);
					bx::strlcpy(arraySize, array+1, end-array);
					num = atoi(arraySize);
				}
				else
				{
					bx::strlcpy(uniformName, name, eol-name+1);
				}

				Uniform un;
				un.type = nameToUniformTypeEnum(uniformType);

				if (UniformType::Count != un.type)
				{
					BX_TRACE("name: %s (type %d, num %d)", uniformName, un.type, num);

					un.name = uniformName;
					un.num = num;
					un.regIndex = 0;
					un.regCount = num;
					uniforms.push_back(un);
				}

				parse = eol + 1;
			}
		}
	}

	uint16_t count = (uint16_t)uniforms.size();
	bx::write(_writer, count);

	for (UniformArray::const_iterator it = uniforms.begin(); it != uniforms.end(); ++it)
	{
		const Uniform& un = *it;
		uint8_t nameSize = (uint8_t)un.name.size();
		bx::write(_writer, nameSize);
		bx::write(_writer, un.name.c_str(), nameSize);
		uint8_t uniformType = un.type;
		bx::write(_writer, uniformType);
		bx::write(_writer, un.num);
		bx::write(_writer, un.regIndex);
		bx::write(_writer, un.regCount);

		BX_TRACE("%s, %s, %d, %d, %d"
			, un.name.c_str()
			, getUniformTypeName(un.type)
			, un.num
			, un.regIndex
			, un.regCount
			);
	}

	uint32_t shaderSize = (uint32_t)strlen(optimizedShader);
	bx::write(_writer, shaderSize);
	bx::write(_writer, optimizedShader, shaderSize);
	uint8_t nul = 0;
	bx::write(_writer, nul);

	glslopt_cleanup(ctx);

	return true;
}
Exemple #2
0
	static bool compile(const Options& _options, uint32_t _version, const std::string& _code, bx::WriterI* _writer)
	{
		char ch = _options.shaderType;
		const glslopt_shader_type type = ch == 'f'
			? kGlslOptShaderFragment
			: (ch == 'c' ? kGlslOptShaderCompute : kGlslOptShaderVertex);

		glslopt_target target = kGlslTargetOpenGL;
		switch (_version)
		{
		case BX_MAKEFOURCC('M', 'T', 'L', 0):
			target = kGlslTargetMetal;
			break;

		case 2:
			target = kGlslTargetOpenGLES20;
			break;

		case 3:
			target = kGlslTargetOpenGLES30;
			break;

		default:
			target = kGlslTargetOpenGL;
			break;
		}

		glslopt_ctx* ctx = glslopt_initialize(target);

		glslopt_shader* shader = glslopt_optimize(ctx, type, _code.c_str(), 0);

		if (!glslopt_get_status(shader) )
		{
			const char* log = glslopt_get_log(shader);
			int32_t source  = 0;
			int32_t line    = 0;
			int32_t column  = 0;
			int32_t start   = 0;
			int32_t end     = INT32_MAX;

			bool found = false
				|| 3 == sscanf(log, "%u:%u(%u):", &source, &line, &column)
				|| 2 == sscanf(log, "(%u,%u):", &line, &column)
				;

			if (found
			&&  0 != line)
			{
				start = bx::uint32_imax(1, line-10);
				end   = start + 20;
			}

			printCode(_code.c_str(), line, start, end, column);
			fprintf(stderr, "Error: %s\n", log);
			glslopt_shader_delete(shader);
			glslopt_cleanup(ctx);
			return false;
		}

		const char* optimizedShader = glslopt_get_output(shader);

		// Trim all directives.
		while ('#' == *optimizedShader)
		{
			optimizedShader = bx::strnl(optimizedShader);
		}

		{
			char* code = const_cast<char*>(optimizedShader);
			strReplace(code, "gl_FragDepthEXT", "gl_FragDepth");

			strReplace(code, "texture2DLodARB", "texture2DLod");
			strReplace(code, "texture2DLodEXT", "texture2DLod");
			strReplace(code, "texture2DGradARB", "texture2DGrad");
			strReplace(code, "texture2DGradEXT", "texture2DGrad");

			strReplace(code, "textureCubeLodARB", "textureCubeLod");
			strReplace(code, "textureCubeLodEXT", "textureCubeLod");
			strReplace(code, "textureCubeGradARB", "textureCubeGrad");
			strReplace(code, "textureCubeGradEXT", "textureCubeGrad");

			strReplace(code, "texture2DProjLodARB", "texture2DProjLod");
			strReplace(code, "texture2DProjLodEXT", "texture2DProjLod");
			strReplace(code, "texture2DProjGradARB", "texture2DProjGrad");
			strReplace(code, "texture2DProjGradEXT", "texture2DProjGrad");

			strReplace(code, "shadow2DARB", "shadow2D");
			strReplace(code, "shadow2DEXT", "shadow2D");
			strReplace(code, "shadow2DProjARB", "shadow2DProj");
			strReplace(code, "shadow2DProjEXT", "shadow2DProj");
		}

		UniformArray uniforms;

		if (target != kGlslTargetMetal)
		{
			const char* parse = optimizedShader;

			while (NULL != parse
				&&  *parse != '\0')
			{
				parse = bx::strws(parse);
				const char* eol = bx::strFind(parse, ';');
				if (NULL != eol)
				{
					const char* qualifier = parse;
					parse = bx::strws(bx::strSkipWord(parse) );

					if (0 == bx::strCmp(qualifier, "attribute", 9)
					||  0 == bx::strCmp(qualifier, "varying",   7)
					||  0 == bx::strCmp(qualifier, "in",        2)
					||  0 == bx::strCmp(qualifier, "out",       3)
					   )
					{
						// skip attributes and varyings.
						parse = eol + 1;
						continue;
					}

					if (0 == bx::strCmp(parse, "tmpvar", 6) )
					{
						// skip temporaries
						parse = eol + 1;
						continue;
					}

					if (0 != bx::strCmp(qualifier, "uniform", 7) )
					{
						// end if there is no uniform keyword.
						parse = NULL;
						continue;
					}

					const char* precision = NULL;
					const char* typen = parse;

					if (0 == bx::strCmp(typen, "lowp", 4)
					||  0 == bx::strCmp(typen, "mediump", 7)
					||  0 == bx::strCmp(typen, "highp", 5) )
					{
						precision = typen;
						typen = parse = bx::strws(bx::strSkipWord(parse) );
					}

					BX_UNUSED(precision);

					char uniformType[256];
					parse = bx::strSkipWord(parse);

					if (0 == bx::strCmp(typen, "sampler", 7) )
					{
						bx::strCopy(uniformType, BX_COUNTOF(uniformType), "int");
					}
					else
					{
						bx::strCopy(uniformType, int32_t(parse-typen+1), typen);
					}

					const char* name = parse = bx::strws(parse);

					char uniformName[256];
					uint8_t num = 1;
					const char* array = bx::strFind(bx::StringView(name, int32_t(eol-parse) ), "[");
					if (NULL != array)
					{
						bx::strCopy(uniformName, int32_t(array-name+1), name);

						char arraySize[32];
						const char* end = bx::strFind(bx::StringView(array, int32_t(eol-array) ), "]");
						bx::strCopy(arraySize, int32_t(end-array), array+1);
						num = uint8_t(atoi(arraySize) );
					}
					else
					{
						bx::strCopy(uniformName, int32_t(eol-name+1), name);
					}

					Uniform un;
					un.type = nameToUniformTypeEnum(uniformType);

					if (UniformType::Count != un.type)
					{
						BX_TRACE("name: %s (type %d, num %d)", uniformName, un.type, num);

						un.name = uniformName;
						un.num = num;
						un.regIndex = 0;
						un.regCount = num;
						uniforms.push_back(un);
					}

					parse = eol + 1;
				}
			}
		}
		else
		{
			const char* parse = bx::strFind(optimizedShader, "struct xlatMtlShaderUniform {");
			const char* end   = parse;
			if (NULL != parse)
			{
				parse += bx::strLen("struct xlatMtlShaderUniform {");
				end   =  bx::strFind(parse, "};");
			}

			while ( parse < end
			&&     *parse != '\0')
			{
				parse = bx::strws(parse);
				const char* eol = bx::strFind(parse, ';');
				if (NULL != eol)
				{
					const char* typen = parse;

					char uniformType[256];
					parse = bx::strSkipWord(parse);
					bx::strCopy(uniformType, int32_t(parse-typen+1), typen);
					const char* name = parse = bx::strws(parse);

					char uniformName[256];
					uint8_t num = 1;
					const char* array = bx::strFind(bx::StringView(name, int32_t(eol-parse) ), "[");
					if (NULL != array)
					{
						bx::strCopy(uniformName, int32_t(array-name+1), name);

						char arraySize[32];
						const char* arrayEnd = bx::strFind(bx::StringView(array, int32_t(eol-array) ), "]");
						bx::strCopy(arraySize, int32_t(arrayEnd-array), array+1);
						num = uint8_t(atoi(arraySize) );
					}
					else
					{
						bx::strCopy(uniformName, int32_t(eol-name+1), name);
					}

					Uniform un;
					un.type = nameToUniformTypeEnum(uniformType);

					if (UniformType::Count != un.type)
					{
						BX_TRACE("name: %s (type %d, num %d)", uniformName, un.type, num);

						un.name = uniformName;
						un.num = num;
						un.regIndex = 0;
						un.regCount = num;
						uniforms.push_back(un);
					}

					parse = eol + 1;
				}
			}
		}

		uint16_t count = (uint16_t)uniforms.size();
		bx::write(_writer, count);

		for (UniformArray::const_iterator it = uniforms.begin(); it != uniforms.end(); ++it)
		{
			const Uniform& un = *it;
			uint8_t nameSize = (uint8_t)un.name.size();
			bx::write(_writer, nameSize);
			bx::write(_writer, un.name.c_str(), nameSize);
			uint8_t uniformType = uint8_t(un.type);
			bx::write(_writer, uniformType);
			bx::write(_writer, un.num);
			bx::write(_writer, un.regIndex);
			bx::write(_writer, un.regCount);

			BX_TRACE("%s, %s, %d, %d, %d"
				, un.name.c_str()
				, getUniformTypeName(un.type)
				, un.num
				, un.regIndex
				, un.regCount
				);
		}

		uint32_t shaderSize = (uint32_t)bx::strLen(optimizedShader);
		bx::write(_writer, shaderSize);
		bx::write(_writer, optimizedShader, shaderSize);
		uint8_t nul = 0;
		bx::write(_writer, nul);

		if (_options.disasm )
		{
			std::string disasmfp = _options.outputFilePath + ".disasm";
			writeFile(disasmfp.c_str(), optimizedShader, shaderSize);
		}

		glslopt_shader_delete(shader);
		glslopt_cleanup(ctx);

		return true;
	}
Exemple #3
0
bool compileHLSLShaderDx11(bx::CommandLine& _cmdLine, const std::string& _code, bx::WriterI* _writer)
{
	BX_TRACE("DX11");

	const char* profile = _cmdLine.findOption('p', "profile");
	if (NULL == profile)
	{
		fprintf(stderr, "Shader profile must be specified.\n");
		return false;
	}

	bool debug = _cmdLine.hasArg('\0', "debug");

	uint32_t flags = D3DCOMPILE_ENABLE_BACKWARDS_COMPATIBILITY;
	flags |= debug ? D3DCOMPILE_DEBUG : 0;
	flags |= _cmdLine.hasArg('\0', "avoid-flow-control") ? D3DCOMPILE_AVOID_FLOW_CONTROL : 0;
	flags |= _cmdLine.hasArg('\0', "no-preshader") ? D3DCOMPILE_NO_PRESHADER : 0;
	flags |= _cmdLine.hasArg('\0', "partial-precision") ? D3DCOMPILE_PARTIAL_PRECISION : 0;
	flags |= _cmdLine.hasArg('\0', "prefer-flow-control") ? D3DCOMPILE_PREFER_FLOW_CONTROL : 0;
	flags |= _cmdLine.hasArg('\0', "backwards-compatibility") ? D3DCOMPILE_ENABLE_BACKWARDS_COMPATIBILITY : 0;

	bool werror = _cmdLine.hasArg('\0', "Werror");

	if (werror)
	{
		flags |= D3DCOMPILE_WARNINGS_ARE_ERRORS;
	}

	uint32_t optimization = 3;
	if (_cmdLine.hasArg(optimization, 'O') )
	{
		optimization = bx::uint32_min(optimization, BX_COUNTOF(s_optimizationLevelDx11)-1);
		flags |= s_optimizationLevelDx11[optimization];
	}
	else
	{
		flags |= D3DCOMPILE_SKIP_OPTIMIZATION;
	}

	BX_TRACE("Profile: %s", profile);
	BX_TRACE("Flags: 0x%08x", flags);

	ID3DBlob* code;
	ID3DBlob* errorMsg;

	// Output preprocessed shader so that HLSL can be debugged via GPA
	// or PIX. Compiling through memory won't embed preprocessed shader
	// file path.
	std::string hlslfp;

	if (debug)
	{
		hlslfp = _cmdLine.findOption('o');
		hlslfp += ".hlsl";
		writeFile(hlslfp.c_str(), _code.c_str(), (int32_t)_code.size() );
	}

	HRESULT hr = D3DCompile(_code.c_str()
		, _code.size()
		, hlslfp.c_str()
		, NULL
		, NULL
		, "main"
		, profile
		, flags
		, 0
		, &code
		, &errorMsg
		);
	if (FAILED(hr)
	|| (werror && NULL != errorMsg) )
	{
		const char* log = (char*)errorMsg->GetBufferPointer();

		int32_t line = 0;
		int32_t column = 0;
		int32_t start = 0;
		int32_t end = INT32_MAX;

		if (2 == sscanf(log, "(%u,%u):", &line, &column)
		&&  0 != line)
		{
			start = bx::uint32_imax(1, line-10);
			end = start + 20;
		}

		printCode(_code.c_str(), line, start, end);
		fprintf(stderr, "Error: 0x%08x %s\n", (uint32_t)hr, log);
		errorMsg->Release();
		return false;
	}

	UniformArray uniforms;

	ID3D11ShaderReflection* reflect = NULL;
	hr = D3DReflect(code->GetBufferPointer()
		, code->GetBufferSize()
		, IID_ID3D11ShaderReflection
		, (void**)&reflect
		);
	if (FAILED(hr) )
	{
		fprintf(stderr, "Error: 0x%08x\n", (uint32_t)hr);
		return false;
	}

	D3D11_SHADER_DESC desc;
	hr = reflect->GetDesc(&desc);
	if (FAILED(hr) )
	{
		fprintf(stderr, BX_FILE_LINE_LITERAL "Error: 0x%08x\n", (uint32_t)hr);
		return false;
	}

	BX_TRACE("Creator: %s 0x%08x", desc.Creator, desc.Version);
	BX_TRACE("Num constant buffers: %d", desc.ConstantBuffers);

	BX_TRACE("Input:");
	uint8_t numAttrs = 0;
	uint16_t attrs[bgfx::Attrib::Count];

	if (profile[0] == 'v') // Only care about input semantic on vertex shaders
	{
		for (uint32_t ii = 0; ii < desc.InputParameters; ++ii)
		{
			D3D11_SIGNATURE_PARAMETER_DESC spd;
			reflect->GetInputParameterDesc(ii, &spd);
			BX_TRACE("\t%2d: %s%d, vt %d, ct %d, mask %x, reg %d"
				, ii
				, spd.SemanticName
				, spd.SemanticIndex
				, spd.SystemValueType
				, spd.ComponentType
				, spd.Mask
				, spd.Register
				);

			const RemapInputSemantic& ris = findInputSemantic(spd.SemanticName, spd.SemanticIndex);
			if (ris.m_attr != bgfx::Attrib::Count)
			{
				attrs[numAttrs] = bgfx::attribToId(ris.m_attr);
				++numAttrs;
			}
		}
	}

	BX_TRACE("Output:");
	for (uint32_t ii = 0; ii < desc.OutputParameters; ++ii)
	{
		D3D11_SIGNATURE_PARAMETER_DESC spd;
		reflect->GetOutputParameterDesc(ii, &spd);
		BX_TRACE("\t%2d: %s%d, %d, %d", ii, spd.SemanticName, spd.SemanticIndex, spd.SystemValueType, spd.ComponentType);
	}

	uint16_t size = 0;

	for (uint32_t ii = 0; ii < bx::uint32_min(1, desc.ConstantBuffers); ++ii)
	{
		ID3D11ShaderReflectionConstantBuffer* cbuffer = reflect->GetConstantBufferByIndex(ii);
		D3D11_SHADER_BUFFER_DESC bufferDesc;
		hr = cbuffer->GetDesc(&bufferDesc);

		size = (uint16_t)bufferDesc.Size;

		if (SUCCEEDED(hr) )
		{
			BX_TRACE("%s, %d, vars %d, size %d"
				, bufferDesc.Name
				, bufferDesc.Type
				, bufferDesc.Variables
				, bufferDesc.Size
				);

			for (uint32_t jj = 0; jj < bufferDesc.Variables; ++jj)
			{
				ID3D11ShaderReflectionVariable* var = cbuffer->GetVariableByIndex(jj);
				ID3D11ShaderReflectionType* type = var->GetType();
				D3D11_SHADER_VARIABLE_DESC varDesc;
				hr = var->GetDesc(&varDesc);
				if (SUCCEEDED(hr) )
				{
					D3D11_SHADER_TYPE_DESC constDesc;
					hr = type->GetDesc(&constDesc);
					if (SUCCEEDED(hr) )
					{
						UniformType::Enum uniformType = findUniformTypeDx11(constDesc);

						if (UniformType::Count != uniformType
						&&  0 != (varDesc.uFlags & D3D_SVF_USED) )
						{
							Uniform un;
							un.name = varDesc.Name;
							un.type = uniformType;
							un.num = constDesc.Elements;
							un.regIndex = varDesc.StartOffset;
							un.regCount = BX_ALIGN_16(varDesc.Size)/16;
							uniforms.push_back(un);

							BX_TRACE("\t%s, %d, size %d, flags 0x%08x, %d"
								, varDesc.Name
								, varDesc.StartOffset
								, varDesc.Size
								, varDesc.uFlags
								, uniformType
								);
						}
						else
						{
							BX_TRACE("\t%s, unknown type", varDesc.Name);
						}
					}
				}
			}
		}
	}

	BX_TRACE("Bound:");
	for (uint32_t ii = 0; ii < desc.BoundResources; ++ii)
	{
		D3D11_SHADER_INPUT_BIND_DESC bindDesc;

		hr = reflect->GetResourceBindingDesc(ii, &bindDesc);
		if (SUCCEEDED(hr) )
		{
			//			if (bindDesc.Type == D3D_SIT_SAMPLER)
			{
				BX_TRACE("\t%s, %d, %d, %d"
					, bindDesc.Name
					, bindDesc.Type
					, bindDesc.BindPoint
					, bindDesc.BindCount
					);
			}
		}
	}

	uint16_t count = (uint16_t)uniforms.size();
	bx::write(_writer, count);

	uint32_t fragmentBit = profile[0] == 'p' ? BGFX_UNIFORM_FRAGMENTBIT : 0;
	for (UniformArray::const_iterator it = uniforms.begin(); it != uniforms.end(); ++it)
	{
		const Uniform& un = *it;
		uint8_t nameSize = (uint8_t)un.name.size();
		bx::write(_writer, nameSize);
		bx::write(_writer, un.name.c_str(), nameSize);
		uint8_t type = un.type|fragmentBit;
		bx::write(_writer, type);
		bx::write(_writer, un.num);
		bx::write(_writer, un.regIndex);
		bx::write(_writer, un.regCount);

		BX_TRACE("%s, %s, %d, %d, %d"
			, un.name.c_str()
			, getUniformTypeName(un.type)
			, un.num
			, un.regIndex
			, un.regCount
			);
	}

	{
		ID3DBlob* stripped;
		hr = D3DStripShader(code->GetBufferPointer()
			, code->GetBufferSize()
			, D3DCOMPILER_STRIP_REFLECTION_DATA
			| D3DCOMPILER_STRIP_TEST_BLOBS
			, &stripped
			);

		if (SUCCEEDED(hr) )
		{
			code->Release();
			code = stripped;
		}
	}

	uint16_t shaderSize = (uint16_t)code->GetBufferSize();
	bx::write(_writer, shaderSize);
	bx::write(_writer, code->GetBufferPointer(), shaderSize);
	uint8_t nul = 0;
	bx::write(_writer, nul);

	bx::write(_writer, numAttrs);
	bx::write(_writer, attrs, numAttrs*sizeof(uint16_t) );

	bx::write(_writer, size);

	if (_cmdLine.hasArg('\0', "disasm") )
	{
		ID3DBlob* disasm;
		D3DDisassemble(code->GetBufferPointer()
			, code->GetBufferSize()
			, 0
			, NULL
			, &disasm
			);

		if (NULL != disasm)
		{
			std::string disasmfp = _cmdLine.findOption('o');
			disasmfp += ".disasm";

			writeFile(disasmfp.c_str(), disasm->GetBufferPointer(), (uint32_t)disasm->GetBufferSize() );
			disasm->Release();
		}
	}

	if (NULL != reflect)
	{
		reflect->Release();
	}

	if (NULL != errorMsg)
	{
		errorMsg->Release();
	}

	code->Release();

	return true;
}
bool getReflectionDataD3D11(ID3DBlob* _code, bool _vshader, UniformArray& _uniforms, uint8_t& _numAttrs, uint16_t* _attrs, uint16_t& _size, UniformNameList& unusedUniforms)
{
    ID3D11ShaderReflection* reflect = NULL;
    HRESULT hr = D3DReflect(_code->GetBufferPointer()
                            , _code->GetBufferSize()
                            , s_compiler->IID_ID3D11ShaderReflection
                            , (void**)&reflect
                           );
    if (FAILED(hr) )
    {
        fprintf(stderr, "Error: D3DReflect failed 0x%08x\n", (uint32_t)hr);
        return false;
    }

    D3D11_SHADER_DESC desc;
    hr = reflect->GetDesc(&desc);
    if (FAILED(hr) )
    {
        fprintf(stderr, "Error: ID3D11ShaderReflection::GetDesc failed 0x%08x\n", (uint32_t)hr);
        return false;
    }

    BX_TRACE("Creator: %s 0x%08x", desc.Creator, desc.Version);
    BX_TRACE("Num constant buffers: %d", desc.ConstantBuffers);

    BX_TRACE("Input:");

    if (_vshader) // Only care about input semantic on vertex shaders
    {
        for (uint32_t ii = 0; ii < desc.InputParameters; ++ii)
        {
            D3D11_SIGNATURE_PARAMETER_DESC spd;
            reflect->GetInputParameterDesc(ii, &spd);
            BX_TRACE("\t%2d: %s%d, vt %d, ct %d, mask %x, reg %d"
                     , ii
                     , spd.SemanticName
                     , spd.SemanticIndex
                     , spd.SystemValueType
                     , spd.ComponentType
                     , spd.Mask
                     , spd.Register
                    );

            const RemapInputSemantic& ris = findInputSemantic(spd.SemanticName, spd.SemanticIndex);
            if (ris.m_attr != bgfx::Attrib::Count)
            {
                _attrs[_numAttrs] = bgfx::attribToId(ris.m_attr);
                ++_numAttrs;
            }
        }
    }

    BX_TRACE("Output:");
    for (uint32_t ii = 0; ii < desc.OutputParameters; ++ii)
    {
        D3D11_SIGNATURE_PARAMETER_DESC spd;
        reflect->GetOutputParameterDesc(ii, &spd);
        BX_TRACE("\t%2d: %s%d, %d, %d", ii, spd.SemanticName, spd.SemanticIndex, spd.SystemValueType, spd.ComponentType);
    }

    for (uint32_t ii = 0, num = bx::uint32_min(1, desc.ConstantBuffers); ii < num; ++ii)
    {
        ID3D11ShaderReflectionConstantBuffer* cbuffer = reflect->GetConstantBufferByIndex(ii);
        D3D11_SHADER_BUFFER_DESC bufferDesc;
        hr = cbuffer->GetDesc(&bufferDesc);

        _size = (uint16_t)bufferDesc.Size;

        if (SUCCEEDED(hr) )
        {
            BX_TRACE("%s, %d, vars %d, size %d"
                     , bufferDesc.Name
                     , bufferDesc.Type
                     , bufferDesc.Variables
                     , bufferDesc.Size
                    );

            for (uint32_t jj = 0; jj < bufferDesc.Variables; ++jj)
            {
                ID3D11ShaderReflectionVariable* var = cbuffer->GetVariableByIndex(jj);
                ID3D11ShaderReflectionType* type = var->GetType();
                D3D11_SHADER_VARIABLE_DESC varDesc;
                hr = var->GetDesc(&varDesc);
                if (SUCCEEDED(hr) )
                {
                    D3D11_SHADER_TYPE_DESC constDesc;
                    hr = type->GetDesc(&constDesc);
                    if (SUCCEEDED(hr) )
                    {
                        UniformType::Enum uniformType = findUniformType(constDesc);

                        if (UniformType::Count != uniformType
                                &&  0 != (varDesc.uFlags & D3D_SVF_USED) )
                        {
                            Uniform un;
                            un.name = varDesc.Name;
                            un.type = uniformType;
                            un.num = constDesc.Elements;
                            un.regIndex = varDesc.StartOffset;
                            un.regCount = BX_ALIGN_16(varDesc.Size) / 16;
                            _uniforms.push_back(un);

                            BX_TRACE("\t%s, %d, size %d, flags 0x%08x, %d (used)"
                                     , varDesc.Name
                                     , varDesc.StartOffset
                                     , varDesc.Size
                                     , varDesc.uFlags
                                     , uniformType
                                    );
                        }
                        else
                        {
                            if (0 == (varDesc.uFlags & D3D_SVF_USED) )
                            {
                                unusedUniforms.push_back(varDesc.Name);
                            }

                            BX_TRACE("\t%s, unknown type", varDesc.Name);
                        }
                    }
                }
            }
        }
    }

    BX_TRACE("Bound:");
    for (uint32_t ii = 0; ii < desc.BoundResources; ++ii)
    {
        D3D11_SHADER_INPUT_BIND_DESC bindDesc;

        hr = reflect->GetResourceBindingDesc(ii, &bindDesc);
        if (SUCCEEDED(hr) )
        {
            if (D3D_SIT_SAMPLER == bindDesc.Type)
            {
                BX_TRACE("\t%s, %d, %d, %d"
                         , bindDesc.Name
                         , bindDesc.Type
                         , bindDesc.BindPoint
                         , bindDesc.BindCount
                        );

                const char * end = strstr(bindDesc.Name, "Sampler");
                if (NULL != end)
                {
                    Uniform un;
                    un.name.assign(bindDesc.Name, (end - bindDesc.Name) );
                    un.type = UniformType::Enum(BGFX_UNIFORM_SAMPLERBIT | UniformType::Int1);
                    un.num = 1;
                    un.regIndex = bindDesc.BindPoint;
                    un.regCount = bindDesc.BindCount;
                    _uniforms.push_back(un);
                }
            }
        }
    }

    if (NULL != reflect)
    {
        reflect->Release();
    }

    return true;
}
bool getReflectionDataD3D9(ID3DBlob* _code, UniformArray& _uniforms)
{
    // see reference for magic values: https://msdn.microsoft.com/en-us/library/ff552891(VS.85).aspx
    const uint32_t D3DSIO_COMMENT = 0x0000FFFE;
    const uint32_t D3DSIO_END = 0x0000FFFF;
    const uint32_t D3DSI_OPCODE_MASK = 0x0000FFFF;
    const uint32_t D3DSI_COMMENTSIZE_MASK = 0x7FFF0000;
    const uint32_t CTAB_CONSTANT = MAKEFOURCC('C', 'T', 'A', 'B');

    // parse the shader blob for the constant table
    const size_t codeSize = _code->GetBufferSize();
    const uint32_t* ptr = (const uint32_t*)_code->GetBufferPointer();
    const uint32_t* end = (const uint32_t*)( (const uint8_t*)ptr + codeSize);
    const CTHeader* header = NULL;

    ptr++;	// first byte is shader type / version; skip it since we already know

    while (ptr < end && *ptr != D3DSIO_END)
    {
        uint32_t cur = *ptr++;
        if ( (cur & D3DSI_OPCODE_MASK) != D3DSIO_COMMENT)
        {
            continue;
        }

        // try to find CTAB comment block
        uint32_t commentSize = (cur & D3DSI_COMMENTSIZE_MASK) >> 16;
        uint32_t fourcc = *ptr;
        if (fourcc == CTAB_CONSTANT)
        {
            // found the constant table data
            header = (const CTHeader*)(ptr + 1);
            uint32_t tableSize = (commentSize - 1) * 4;
            if (tableSize < sizeof(CTHeader) || header->Size != sizeof(CTHeader) )
            {
                fprintf(stderr, "Error: Invalid constant table data\n");
                return false;
            }
            break;
        }

        // this is a different kind of comment section, so skip over it
        ptr += commentSize - 1;
    }

    if (!header)
    {
        fprintf(stderr, "Error: Could not find constant table data\n");
        return false;
    }

    const uint8_t* headerBytePtr = (const uint8_t*)header;
    const char* creator = (const char*)(headerBytePtr + header->Creator);

    BX_TRACE("Creator: %s 0x%08x", creator, header->Version);
    BX_TRACE("Num constants: %d", header->Constants);
    BX_TRACE("#   cl ty RxC   S  By Name");

    const CTInfo* ctInfoArray = (const CTInfo*)(headerBytePtr + header->ConstantInfo);
    for (uint32_t ii = 0; ii < header->Constants; ++ii)
    {
        const CTInfo& ctInfo = ctInfoArray[ii];
        const CTType& ctType = *(const CTType*)(headerBytePtr + ctInfo.TypeInfo);
        const char* name = (const char*)(headerBytePtr + ctInfo.Name);

        BX_TRACE("%3d %2d %2d [%dx%d] %d %s[%d] c%d (%d)"
                 , ii
                 , ctType.Class
                 , ctType.Type
                 , ctType.Rows
                 , ctType.Columns
                 , ctType.StructMembers
                 , name
                 , ctType.Elements
                 , ctInfo.RegisterIndex
                 , ctInfo.RegisterCount
                );

        D3D11_SHADER_TYPE_DESC desc;
        desc.Class = (D3D_SHADER_VARIABLE_CLASS)ctType.Class;
        desc.Type = (D3D_SHADER_VARIABLE_TYPE)ctType.Type;
        desc.Rows = ctType.Rows;
        desc.Columns = ctType.Columns;

        UniformType::Enum type = findUniformType(desc);
        if (UniformType::Count != type)
        {
            Uniform un;
            un.name = '$' == name[0] ? name + 1 : name;
            un.type = isSampler(desc.Type)
                      ? UniformType::Enum(BGFX_UNIFORM_SAMPLERBIT | type)
                      : type
                      ;
            un.num = (uint8_t)ctType.Elements;
            un.regIndex = ctInfo.RegisterIndex;
            un.regCount = ctInfo.RegisterCount;

            _uniforms.push_back(un);
        }
    }

    return true;
}
bool compileHLSLShaderDx9(bx::CommandLine& _cmdLine, const std::string& _code, bx::WriterI* _writer)
{
	BX_TRACE("DX9");

	const char* profile = _cmdLine.findOption('p', "profile");
	if (NULL == profile)
	{
		fprintf(stderr, "Shader profile must be specified.\n");
		return false;
	}

	bool debug = _cmdLine.hasArg('\0', "debug");

	uint32_t flags = 0;
	flags |= debug ? D3DXSHADER_DEBUG : 0;
	flags |= _cmdLine.hasArg('\0', "avoid-flow-control") ? D3DXSHADER_AVOID_FLOW_CONTROL : 0;
	flags |= _cmdLine.hasArg('\0', "no-preshader") ? D3DXSHADER_NO_PRESHADER : 0;
	flags |= _cmdLine.hasArg('\0', "partial-precision") ? D3DXSHADER_PARTIALPRECISION : 0;
	flags |= _cmdLine.hasArg('\0', "prefer-flow-control") ? D3DXSHADER_PREFER_FLOW_CONTROL : 0;
	flags |= _cmdLine.hasArg('\0', "backwards-compatibility") ? D3DXSHADER_ENABLE_BACKWARDS_COMPATIBILITY : 0;

	bool werror = _cmdLine.hasArg('\0', "Werror");

	uint32_t optimization = 3;
	if (_cmdLine.hasArg(optimization, 'O') )
	{
		optimization = bx::uint32_min(optimization, BX_COUNTOF(s_optimizationLevelDx9)-1);
		flags |= s_optimizationLevelDx9[optimization];
	}
	else
	{
		flags |= D3DXSHADER_SKIPOPTIMIZATION;
	}

	BX_TRACE("Profile: %s", profile);
	BX_TRACE("Flags: 0x%08x", flags);

	LPD3DXBUFFER code;
	LPD3DXBUFFER errorMsg;
	LPD3DXCONSTANTTABLE constantTable;

	HRESULT hr;

	// Output preprocessed shader so that HLSL can be debugged via GPA
	// or PIX. Compiling through memory won't embed preprocessed shader
	// file path.
	if (debug)
	{
		std::string hlslfp = _cmdLine.findOption('o');
		hlslfp += ".hlsl";
		writeFile(hlslfp.c_str(), _code.c_str(), (int32_t)_code.size() );

		hr = D3DXCompileShaderFromFileA(hlslfp.c_str()
			, NULL
			, NULL
			, "main"
			, profile
			, flags
			, &code
			, &errorMsg
			, &constantTable
			);
	}
	else
	{
		hr = D3DXCompileShader(_code.c_str()
			, (uint32_t)_code.size()
			, NULL
			, NULL
			, "main"
			, profile
			, flags
			, &code
			, &errorMsg
			, &constantTable
			);
	}

	if (FAILED(hr)
	|| (werror && NULL != errorMsg) )
	{
		const char* log = (const char*)errorMsg->GetBufferPointer();

		char source[1024];
		int32_t line = 0;
		int32_t column = 0;
		int32_t start = 0;
		int32_t end = INT32_MAX;

		if (3 == sscanf(log, "%[^(](%u,%u):", source, &line, &column)
		&&  0 != line)
		{
			start = bx::uint32_imax(1, line-10);
			end = start + 20;
		}

		printCode(_code.c_str(), line, start, end);
		fprintf(stderr, "Error: 0x%08x %s\n", (uint32_t)hr, log);
		errorMsg->Release();
		return false;
	}

	UniformArray uniforms;

	if (NULL != constantTable)
	{
		D3DXCONSTANTTABLE_DESC desc;
		hr = constantTable->GetDesc(&desc);
		if (FAILED(hr) )
		{
			fprintf(stderr, "Error 0x%08x\n", (uint32_t)hr);
			return false;
		}

		BX_TRACE("Creator: %s 0x%08x", desc.Creator, (uint32_t /*mingw warning*/)desc.Version);
		BX_TRACE("Num constants: %d", desc.Constants);
		BX_TRACE("#   cl ty RxC   S  By Name");

		for (uint32_t ii = 0; ii < desc.Constants; ++ii)
		{
			D3DXHANDLE handle = constantTable->GetConstant(NULL, ii);
			D3DXCONSTANT_DESC constDesc;
			uint32_t count;
			constantTable->GetConstantDesc(handle, &constDesc, &count);
			BX_TRACE("%3d %2d %2d [%dx%d] %d %3d %s[%d] c%d (%d)"
				, ii
				, constDesc.Class
				, constDesc.Type
				, constDesc.Rows
				, constDesc.Columns
				, constDesc.StructMembers
				, constDesc.Bytes
				, constDesc.Name
				, constDesc.Elements
				, constDesc.RegisterIndex
				, constDesc.RegisterCount
				);

			UniformType::Enum type = findUniformTypeDx9(constDesc);
			if (UniformType::Count != type)
			{
				Uniform un;
				un.name = '$' == constDesc.Name[0] ? constDesc.Name+1 : constDesc.Name;
				un.type = type;
				un.num = constDesc.Elements;
				un.regIndex = constDesc.RegisterIndex;
				un.regCount = constDesc.RegisterCount;
				uniforms.push_back(un);
			}
		}
	}

	uint16_t count = (uint16_t)uniforms.size();
	bx::write(_writer, count);

	uint32_t fragmentBit = profile[0] == 'p' ? BGFX_UNIFORM_FRAGMENTBIT : 0;
	for (UniformArray::const_iterator it = uniforms.begin(); it != uniforms.end(); ++it)
	{
		const Uniform& un = *it;
		uint8_t nameSize = (uint8_t)un.name.size();
		bx::write(_writer, nameSize);
		bx::write(_writer, un.name.c_str(), nameSize);
		uint8_t type = un.type|fragmentBit;
		bx::write(_writer, type);
		bx::write(_writer, un.num);
		bx::write(_writer, un.regIndex);
		bx::write(_writer, un.regCount);

		BX_TRACE("%s, %s, %d, %d, %d"
			, un.name.c_str()
			, getUniformTypeName(un.type)
			, un.num
			, un.regIndex
			, un.regCount
			);
	}

	uint16_t shaderSize = (uint16_t)code->GetBufferSize();
	bx::write(_writer, shaderSize);
	bx::write(_writer, code->GetBufferPointer(), shaderSize);
	uint8_t nul = 0;
	bx::write(_writer, nul);

	if (_cmdLine.hasArg('\0', "disasm") )
	{
		LPD3DXBUFFER disasm;
		D3DXDisassembleShader( (const DWORD*)code->GetBufferPointer()
			, false
			, NULL
			, &disasm
			);

		if (NULL != disasm)
		{
			std::string disasmfp = _cmdLine.findOption('o');
			disasmfp += ".disasm";

			writeFile(disasmfp.c_str(), disasm->GetBufferPointer(), disasm->GetBufferSize() );
			disasm->Release();
		}
	}

	if (NULL != code)
	{
		code->Release();
	}

	if (NULL != errorMsg)
	{
		errorMsg->Release();
	}

	if (NULL != constantTable)
	{
		constantTable->Release();
	}

	return true;
}