// Copyright 2013 Dolphin Emulator Project // Licensed under GPLv2 // Refer to the license.txt file included. #include "ProgramShaderCache.h" #include "DriverDetails.h" #include "MathUtil.h" #include "StreamBuffer.h" #include "Debugger.h" #include "Statistics.h" #include "ImageWrite.h" #include "Render.h" #include "PixelShaderManager.h" #include "VertexShaderManager.h" namespace OGL { static const u32 UBO_LENGTH = 32*1024*1024; u32 ProgramShaderCache::s_ubo_buffer_size; s32 ProgramShaderCache::s_ubo_align; static StreamBuffer *s_buffer; static int num_failures = 0; LinearDiskCache g_program_disk_cache; static GLuint CurrentProgram = 0; ProgramShaderCache::PCache ProgramShaderCache::pshaders; ProgramShaderCache::PCacheEntry* ProgramShaderCache::last_entry; SHADERUID ProgramShaderCache::last_uid; UidChecker ProgramShaderCache::pixel_uid_checker; UidChecker ProgramShaderCache::vertex_uid_checker; static char s_glsl_header[1024] = ""; // Annoying sure, can be removed once we drop our UBO workaround const char *UniformNames[NUM_UNIFORMS] = { // PIXEL SHADER UNIFORMS I_COLORS, I_KCOLORS, I_ALPHA, I_TEXDIMS, I_ZBIAS , I_INDTEXSCALE , I_INDTEXMTX, I_FOG, I_PLIGHTS, I_PMATERIALS, // VERTEX SHADER UNIFORMS I_POSNORMALMATRIX, I_PROJECTION , I_MATERIALS, I_LIGHTS, I_TEXMATRICES, I_TRANSFORMMATRICES , I_NORMALMATRICES , I_POSTTRANSFORMMATRICES, I_DEPTHPARAMS, }; const static int PSVar_Loc[] = { offsetof(PixelShaderConstants, colors)/16, offsetof(PixelShaderConstants, kcolors)/16, offsetof(PixelShaderConstants, alpha)/16, offsetof(PixelShaderConstants, texdims)/16, offsetof(PixelShaderConstants, zbias)/16, offsetof(PixelShaderConstants, indtexscale)/16, offsetof(PixelShaderConstants, indtexmtx)/16, offsetof(PixelShaderConstants, fog)/16, offsetof(PixelShaderConstants, plights)/16, offsetof(PixelShaderConstants, pmaterials)/16, }; const static int VSVar_Loc[] = { offsetof(VertexShaderConstants, posnormalmatrix)/16, offsetof(VertexShaderConstants, projection)/16, offsetof(VertexShaderConstants, materials)/16, offsetof(VertexShaderConstants, lights)/16, offsetof(VertexShaderConstants, texmatrices)/16, offsetof(VertexShaderConstants, transformmatrices)/16, offsetof(VertexShaderConstants, normalmatrices)/16, offsetof(VertexShaderConstants, posttransformmatrices)/16, offsetof(VertexShaderConstants, depthparams)/16, }; // End of UBO workaround void SHADER::SetProgramVariables() { // glsl shader must be bind to set samplers Bind(); // Bind UBO if (g_ActiveConfig.backend_info.bSupportsGLSLUBO) { GLint PSBlock_id = glGetUniformBlockIndex(glprogid, "PSBlock"); GLint VSBlock_id = glGetUniformBlockIndex(glprogid, "VSBlock"); if(PSBlock_id != -1) glUniformBlockBinding(glprogid, PSBlock_id, 1); if(VSBlock_id != -1) glUniformBlockBinding(glprogid, VSBlock_id, 2); } // UBO workaround for (int a = 0; a < NUM_UNIFORMS; ++a) { UniformLocations[a] = glGetUniformLocation(glprogid, UniformNames[a]); UniformSize[a] = 0; if(g_ActiveConfig.backend_info.bSupportsGLSLUBO) break; } if(!g_ActiveConfig.backend_info.bSupportsGLSLUBO) { int max_uniforms = 0; char name[50]; int size; glGetProgramiv(glprogid, GL_ACTIVE_UNIFORMS, &max_uniforms); for(int i=0; iAlloc(s_ubo_buffer_size); size_t offset = s_buffer->Upload((u8*)&PixelShaderManager::constants, ROUND_UP(sizeof(PixelShaderConstants), s_ubo_align)); glBindBufferRange(GL_UNIFORM_BUFFER, 1, s_buffer->getBuffer(), offset, sizeof(PixelShaderConstants)); offset = s_buffer->Upload((u8*)&VertexShaderManager::constants, ROUND_UP(sizeof(VertexShaderConstants), s_ubo_align)); glBindBufferRange(GL_UNIFORM_BUFFER, 2, s_buffer->getBuffer(), offset, sizeof(VertexShaderConstants)); PixelShaderManager::dirty = false; VertexShaderManager::dirty = false; ADDSTAT(stats.thisFrame.bytesUniformStreamed, s_ubo_buffer_size); } } else { // UBO workaround // this must be updated per shader switch, so also update it when it's not dirty for (unsigned int a = 0; a < 10; ++a) { if(last_entry->shader.UniformSize[a] > 0) glUniform4fv(last_entry->shader.UniformLocations[a], last_entry->shader.UniformSize[a], (float*) &PixelShaderManager::constants + 4*PSVar_Loc[a]); } for (unsigned int a = 0; a < 9; ++a) { if(last_entry->shader.UniformSize[a+10] > 0) glUniform4fv(last_entry->shader.UniformLocations[a+10], last_entry->shader.UniformSize[a+10], (float*) &VertexShaderManager::constants + 4*VSVar_Loc[a]); } ADDSTAT(stats.thisFrame.bytesUniformStreamed, s_ubo_buffer_size); } } GLuint ProgramShaderCache::GetCurrentProgram(void) { return CurrentProgram; } SHADER* ProgramShaderCache::SetShader ( DSTALPHA_MODE dstAlphaMode, u32 components ) { SHADERUID uid; GetShaderId(&uid, dstAlphaMode, components); // Check if the shader is already set if (last_entry) { if (uid == last_uid) { GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true); last_entry->shader.Bind(); return &last_entry->shader; } } last_uid = uid; // Check if shader is already in cache PCache::iterator iter = pshaders.find(uid); if (iter != pshaders.end()) { PCacheEntry *entry = &iter->second; last_entry = entry; GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true); last_entry->shader.Bind(); return &last_entry->shader; } // Make an entry in the table PCacheEntry& newentry = pshaders[uid]; last_entry = &newentry; newentry.in_cache = 0; VertexShaderCode vcode; PixelShaderCode pcode; GenerateVertexShaderCode(vcode, components, API_OPENGL); GeneratePixelShaderCode(pcode, dstAlphaMode, API_OPENGL, components); if (g_ActiveConfig.bEnableShaderDebugging) { newentry.shader.strvprog = vcode.GetBuffer(); newentry.shader.strpprog = pcode.GetBuffer(); } #if defined(_DEBUG) || defined(DEBUGFAST) if (g_ActiveConfig.iLog & CONF_SAVESHADERS) { static int counter = 0; char szTemp[MAX_PATH]; sprintf(szTemp, "%svs_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), counter++); SaveData(szTemp, vcode.GetBuffer()); sprintf(szTemp, "%sps_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), counter++); SaveData(szTemp, pcode.GetBuffer()); } #endif if (!CompileShader(newentry.shader, vcode.GetBuffer(), pcode.GetBuffer())) { GFX_DEBUGGER_PAUSE_AT(NEXT_ERROR, true); return NULL; } INCSTAT(stats.numPixelShadersCreated); SETSTAT(stats.numPixelShadersAlive, pshaders.size()); GFX_DEBUGGER_PAUSE_AT(NEXT_PIXEL_SHADER_CHANGE, true); last_entry->shader.Bind(); return &last_entry->shader; } bool ProgramShaderCache::CompileShader ( SHADER& shader, const char* vcode, const char* pcode ) { GLuint vsid = CompileSingleShader(GL_VERTEX_SHADER, vcode); GLuint psid = CompileSingleShader(GL_FRAGMENT_SHADER, pcode); if(!vsid || !psid) { glDeleteShader(vsid); glDeleteShader(psid); return false; } GLuint pid = shader.glprogid = glCreateProgram();; glAttachShader(pid, vsid); glAttachShader(pid, psid); if (g_ogl_config.bSupportsGLSLCache) glProgramParameteri(pid, GL_PROGRAM_BINARY_RETRIEVABLE_HINT, GL_TRUE); shader.SetProgramBindings(); glLinkProgram(pid); // original shaders aren't needed any more glDeleteShader(vsid); glDeleteShader(psid); GLint linkStatus; glGetProgramiv(pid, GL_LINK_STATUS, &linkStatus); GLsizei length = 0; glGetProgramiv(pid, GL_INFO_LOG_LENGTH, &length); if (linkStatus != GL_TRUE || (length > 1 && DEBUG_GLSL)) { GLsizei charsWritten; GLchar* infoLog = new GLchar[length]; glGetProgramInfoLog(pid, length, &charsWritten, infoLog); ERROR_LOG(VIDEO, "Program info log:\n%s", infoLog); char szTemp[MAX_PATH]; sprintf(szTemp, "%sbad_p_%d.txt", File::GetUserPath(D_DUMP_IDX).c_str(), num_failures++); std::ofstream file; OpenFStream(file, szTemp, std::ios_base::out); file << s_glsl_header << vcode << s_glsl_header << pcode << infoLog; file.close(); if(linkStatus != GL_TRUE) PanicAlert("Failed to link shaders!\nThis usually happens when trying to use Dolphin with an outdated GPU or integrated GPU like the Intel GMA series.\n\nIf you're sure this is Dolphin's error anyway, post the contents of %s along with this error message at the forums.\n\nDebug info (%s, %s, %s):\n%s", szTemp, g_ogl_config.gl_vendor, g_ogl_config.gl_renderer, g_ogl_config.gl_version, infoLog); delete [] infoLog; } if (linkStatus != GL_TRUE) { // Compile failed ERROR_LOG(VIDEO, "Program linking failed; see info log"); // Don't try to use this shader glDeleteProgram(pid); return false; } shader.SetProgramVariables(); return true; } GLuint ProgramShaderCache::CompileSingleShader (GLuint type, const char* code ) { GLuint result = glCreateShader(type); const char *src[] = {s_glsl_header, code}; glShaderSource(result, 2, src, NULL); glCompileShader(result); GLint compileStatus; glGetShaderiv(result, GL_COMPILE_STATUS, &compileStatus); GLsizei length = 0; glGetShaderiv(result, GL_INFO_LOG_LENGTH, &length); if (DriverDetails::HasBug(DriverDetails::BUG_BROKENINFOLOG)) length = 1024; if (compileStatus != GL_TRUE || (length > 1 && DEBUG_GLSL)) { GLsizei charsWritten; GLchar* infoLog = new GLchar[length]; glGetShaderInfoLog(result, length, &charsWritten, infoLog); ERROR_LOG(VIDEO, "PS Shader info log:\n%s", infoLog); char szTemp[MAX_PATH]; sprintf(szTemp, "%sbad_%s_%04i.txt", File::GetUserPath(D_DUMP_IDX).c_str(), type==GL_VERTEX_SHADER ? "vs" : "ps", num_failures++); std::ofstream file; OpenFStream(file, szTemp, std::ios_base::out); file << s_glsl_header << code << infoLog; file.close(); if(compileStatus != GL_TRUE) PanicAlert("Failed to compile %s shader!\nThis usually happens when trying to use Dolphin with an outdated GPU or integrated GPU like the Intel GMA series.\n\nIf you're sure this is Dolphin's error anyway, post the contents of %s along with this error message at the forums.\n\nDebug info (%s, %s, %s):\n%s", type==GL_VERTEX_SHADER ? "vertex" : "pixel", szTemp, g_ogl_config.gl_vendor, g_ogl_config.gl_renderer, g_ogl_config.gl_version, infoLog); delete[] infoLog; } if (compileStatus != GL_TRUE) { // Compile failed ERROR_LOG(VIDEO, "Shader compilation failed; see info log"); // Don't try to use this shader glDeleteShader(result); return 0; } (void)GL_REPORT_ERROR(); return result; } void ProgramShaderCache::GetShaderId(SHADERUID* uid, DSTALPHA_MODE dstAlphaMode, u32 components) { GetPixelShaderUid(uid->puid, dstAlphaMode, API_OPENGL, components); GetVertexShaderUid(uid->vuid, components, API_OPENGL); if (g_ActiveConfig.bEnableShaderDebugging) { PixelShaderCode pcode; GeneratePixelShaderCode(pcode, dstAlphaMode, API_OPENGL, components); pixel_uid_checker.AddToIndexAndCheck(pcode, uid->puid, "Pixel", "p"); VertexShaderCode vcode; GenerateVertexShaderCode(vcode, components, API_OPENGL); vertex_uid_checker.AddToIndexAndCheck(vcode, uid->vuid, "Vertex", "v"); } } ProgramShaderCache::PCacheEntry ProgramShaderCache::GetShaderProgram(void) { return *last_entry; } void ProgramShaderCache::Init(void) { // We have to get the UBO alignment here because // if we generate a buffer that isn't aligned // then the UBO will fail. if (g_ActiveConfig.backend_info.bSupportsGLSLUBO) { glGetIntegerv(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, &s_ubo_align); s_ubo_buffer_size = ROUND_UP(sizeof(PixelShaderConstants), s_ubo_align) + ROUND_UP(sizeof(VertexShaderConstants), s_ubo_align); // We multiply by *4*4 because we need to get down to basic machine units. // So multiply by four to get how many floats we have from vec4s // Then once more to get bytes s_buffer = new StreamBuffer(GL_UNIFORM_BUFFER, UBO_LENGTH); } // Read our shader cache, only if supported if (g_ogl_config.bSupportsGLSLCache) { GLint Supported; glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &Supported); if(!Supported) { ERROR_LOG(VIDEO, "GL_ARB_get_program_binary is supported, but no binary format is known. So disable shader cache."); g_ogl_config.bSupportsGLSLCache = false; } else { if (!File::Exists(File::GetUserPath(D_SHADERCACHE_IDX))) File::CreateDir(File::GetUserPath(D_SHADERCACHE_IDX).c_str()); char cache_filename[MAX_PATH]; sprintf(cache_filename, "%sogl-%s-shaders.cache", File::GetUserPath(D_SHADERCACHE_IDX).c_str(), SConfig::GetInstance().m_LocalCoreStartupParameter.m_strUniqueID.c_str()); ProgramShaderCacheInserter inserter; g_program_disk_cache.OpenAndRead(cache_filename, inserter); } SETSTAT(stats.numPixelShadersAlive, pshaders.size()); } CreateHeader(); CurrentProgram = 0; last_entry = NULL; } void ProgramShaderCache::Shutdown(void) { // store all shaders in cache on disk if (g_ogl_config.bSupportsGLSLCache) { PCache::iterator iter = pshaders.begin(); for (; iter != pshaders.end(); ++iter) { if(iter->second.in_cache) continue; GLint binary_size; glGetProgramiv(iter->second.shader.glprogid, GL_PROGRAM_BINARY_LENGTH, &binary_size); if(!binary_size) continue; u8 *data = new u8[binary_size+sizeof(GLenum)]; u8 *binary = data + sizeof(GLenum); GLenum *prog_format = (GLenum*)data; glGetProgramBinary(iter->second.shader.glprogid, binary_size, NULL, prog_format, binary); g_program_disk_cache.Append(iter->first, data, binary_size+sizeof(GLenum)); delete [] data; } g_program_disk_cache.Sync(); g_program_disk_cache.Close(); } glUseProgram(0); PCache::iterator iter = pshaders.begin(); for (; iter != pshaders.end(); ++iter) iter->second.Destroy(); pshaders.clear(); pixel_uid_checker.Invalidate(); vertex_uid_checker.Invalidate(); if (g_ActiveConfig.backend_info.bSupportsGLSLUBO) { delete s_buffer; s_buffer = 0; } } void ProgramShaderCache::CreateHeader ( void ) { GLSL_VERSION v = g_ogl_config.eSupportedGLSLVersion; snprintf(s_glsl_header, sizeof(s_glsl_header), "%s\n" "%s\n" // ubo "%s\n" // early-z // Precision defines for GLSLES2/3 "%s\n" "\n"// A few required defines and ones that will make our lives a lot easier "#define ATTRIN %s\n" "#define ATTROUT %s\n" "#define VARYIN %s\n" "#define VARYOUT %s\n" // Silly differences "#define float2 vec2\n" "#define float3 vec3\n" "#define float4 vec4\n" // hlsl to glsl function translation "#define frac fract\n" "#define lerp mix\n" // texture2d hack "%s\n" "%s\n" "%s\n" // GLSLES2 hacks "%s\n" "%s\n" "%s\n" "%s\n" "%s\n" "%s\n" "%s\n" "#define COLOROUT(name) %s\n" , v==GLSLES2 ? "" : v==GLSLES3 ? "#version 300 es" : v==GLSL_130 ? "#version 130" : v==GLSL_140 ? "#version 140" : "#version 150" , g_ActiveConfig.backend_info.bSupportsGLSLUBO && v