summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorReinUsesLisp <reinuseslisp@airmail.cc>2019-01-14 04:58:15 +0100
committerReinUsesLisp <reinuseslisp@airmail.cc>2019-02-07 02:20:57 +0100
commita3703f5767332dfc5f7e8d37a1f715d8ccb76fcf (patch)
tree84ffe77dabc36e77dc0fb543d96d8774483c67a0 /src
parentgl_shader_disk_cache: Add transferable cache invalidation (diff)
downloadyuzu-a3703f5767332dfc5f7e8d37a1f715d8ccb76fcf.tar
yuzu-a3703f5767332dfc5f7e8d37a1f715d8ccb76fcf.tar.gz
yuzu-a3703f5767332dfc5f7e8d37a1f715d8ccb76fcf.tar.bz2
yuzu-a3703f5767332dfc5f7e8d37a1f715d8ccb76fcf.tar.lz
yuzu-a3703f5767332dfc5f7e8d37a1f715d8ccb76fcf.tar.xz
yuzu-a3703f5767332dfc5f7e8d37a1f715d8ccb76fcf.tar.zst
yuzu-a3703f5767332dfc5f7e8d37a1f715d8ccb76fcf.zip
Diffstat (limited to 'src')
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.cpp450
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.h59
2 files changed, 388 insertions, 121 deletions
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 363b941f3..a70ff79d0 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -11,6 +11,7 @@
#include "video_core/renderer_opengl/gl_rasterizer.h"
#include "video_core/renderer_opengl/gl_shader_cache.h"
#include "video_core/renderer_opengl/gl_shader_decompiler.h"
+#include "video_core/renderer_opengl/gl_shader_disk_cache.h"
#include "video_core/renderer_opengl/gl_shader_manager.h"
#include "video_core/renderer_opengl/utils.h"
#include "video_core/shader/shader_ir.h"
@@ -19,8 +20,19 @@ namespace OpenGL {
using VideoCommon::Shader::ProgramCode;
+// One UBO is always reserved for emulation values
+constexpr u32 RESERVED_UBOS = 1;
+
+struct UnspecializedShader {
+ std::string code;
+ GLShader::ShaderEntries entries;
+ Maxwell::ShaderProgram program_type;
+};
+
+namespace {
+
/// Gets the address for the specified shader stage program
-static VAddr GetShaderAddress(Maxwell::ShaderProgram program) {
+VAddr GetShaderAddress(Maxwell::ShaderProgram program) {
const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
const auto& shader_config = gpu.regs.shader_config[static_cast<std::size_t>(program)];
const auto address = gpu.memory_manager.GpuToCpuAddress(gpu.regs.code_address.CodeAddress() +
@@ -30,7 +42,7 @@ static VAddr GetShaderAddress(Maxwell::ShaderProgram program) {
}
/// Gets the shader program code from memory for the specified address
-static ProgramCode GetShaderCode(VAddr addr) {
+ProgramCode GetShaderCode(VAddr addr) {
ProgramCode program_code(VideoCommon::Shader::MAX_PROGRAM_LENGTH);
Memory::ReadBlock(addr, program_code.data(), program_code.size() * sizeof(u64));
return program_code;
@@ -51,38 +63,193 @@ constexpr GLenum GetShaderType(Maxwell::ShaderProgram program_type) {
}
}
-CachedShader::CachedShader(VAddr addr, Maxwell::ShaderProgram program_type)
- : addr{addr}, program_type{program_type}, setup{GetShaderCode(addr)} {
+/// Gets if the current instruction offset is a scheduler instruction
+constexpr bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
+ // Sched instructions appear once every 4 instructions.
+ constexpr std::size_t SchedPeriod = 4;
+ const std::size_t absolute_offset = offset - main_offset;
+ return (absolute_offset % SchedPeriod) == 0;
+}
- GLShader::ProgramResult program_result;
+/// Describes primitive behavior on geometry shaders
+constexpr std::tuple<const char*, const char*, u32> GetPrimitiveDescription(GLenum primitive_mode) {
+ switch (primitive_mode) {
+ case GL_POINTS:
+ return {"points", "Points", 1};
+ case GL_LINES:
+ case GL_LINE_STRIP:
+ return {"lines", "Lines", 2};
+ case GL_LINES_ADJACENCY:
+ case GL_LINE_STRIP_ADJACENCY:
+ return {"lines_adjacency", "LinesAdj", 4};
+ case GL_TRIANGLES:
+ case GL_TRIANGLE_STRIP:
+ case GL_TRIANGLE_FAN:
+ return {"triangles", "Triangles", 3};
+ case GL_TRIANGLES_ADJACENCY:
+ case GL_TRIANGLE_STRIP_ADJACENCY:
+ return {"triangles_adjacency", "TrianglesAdj", 6};
+ default:
+ return {"points", "Invalid", 1};
+ }
+}
- switch (program_type) {
- case Maxwell::ShaderProgram::VertexA:
+/// Calculates the size of a program stream
+std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
+ constexpr std::size_t start_offset = 10;
+ std::size_t offset = start_offset;
+ std::size_t size = start_offset * sizeof(u64);
+ while (offset < program.size()) {
+ const u64 instruction = program[offset];
+ if (!IsSchedInstruction(offset, start_offset)) {
+ if (instruction == 0 || (instruction >> 52) == 0x50b) {
+ // End on Maxwell's "nop" instruction
+ break;
+ }
+ }
+ size += sizeof(u64);
+ offset++;
+ }
+ // The last instruction is included in the program size
+ return std::min(size + sizeof(u64), program.size() * sizeof(u64));
+}
+
+/// Hashes one (or two) program streams
+u64 GetUniqueIdentifier(Maxwell::ShaderProgram program_type, const ProgramCode& code,
+ const ProgramCode& code_b) {
+ u64 unique_identifier =
+ Common::CityHash64(reinterpret_cast<const char*>(code.data()), CalculateProgramSize(code));
+ if (program_type != Maxwell::ShaderProgram::VertexA) {
+ return unique_identifier;
+ }
+ // VertexA programs include two programs
+
+ std::size_t seed = 0;
+ boost::hash_combine(seed, unique_identifier);
+
+ const u64 identifier_b = Common::CityHash64(reinterpret_cast<const char*>(code_b.data()),
+ CalculateProgramSize(code_b));
+ boost::hash_combine(seed, identifier_b);
+ return static_cast<u64>(seed);
+}
+
+/// Creates an unspecialized program from code streams
+GLShader::ProgramResult CreateProgram(Maxwell::ShaderProgram program_type, ProgramCode program_code,
+ ProgramCode program_code_b) {
+ GLShader::ShaderSetup setup(std::move(program_code));
+ if (program_type == Maxwell::ShaderProgram::VertexA) {
// VertexB is always enabled, so when VertexA is enabled, we have two vertex shaders.
// Conventional HW does not support this, so we combine VertexA and VertexB into one
// stage here.
- setup.SetProgramB(GetShaderCode(GetShaderAddress(Maxwell::ShaderProgram::VertexB)));
+ setup.SetProgramB(std::move(program_code_b));
+ }
+
+ switch (program_type) {
+ case Maxwell::ShaderProgram::VertexA:
case Maxwell::ShaderProgram::VertexB:
- CalculateProperties();
- program_result = GLShader::GenerateVertexShader(setup);
- break;
+ return GLShader::GenerateVertexShader(setup);
case Maxwell::ShaderProgram::Geometry:
- CalculateProperties();
- program_result = GLShader::GenerateGeometryShader(setup);
- break;
+ return GLShader::GenerateGeometryShader(setup);
case Maxwell::ShaderProgram::Fragment:
- CalculateProperties();
- program_result = GLShader::GenerateFragmentShader(setup);
- break;
+ return GLShader::GenerateFragmentShader(setup);
default:
LOG_CRITICAL(HW_GPU, "Unimplemented program_type={}", static_cast<u32>(program_type));
UNREACHABLE();
+ return {};
+ }
+}
+
+CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEntries& entries,
+ Maxwell::ShaderProgram program_type, BaseBindings base_bindings,
+ GLenum primitive_mode, bool hint_retrievable = false) {
+ std::string source = "#version 430 core\n";
+ source += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
+
+ for (const auto& cbuf : entries.const_buffers) {
+ source +=
+ fmt::format("#define CBUF_BINDING_{} {}\n", cbuf.GetIndex(), base_bindings.cbuf++);
+ }
+ for (const auto& gmem : entries.global_memory_entries) {
+ source += fmt::format("#define GMEM_BINDING_{}_{} {}\n", gmem.GetCbufIndex(),
+ gmem.GetCbufOffset(), base_bindings.gmem++);
+ }
+ for (const auto& sampler : entries.samplers) {
+ source += fmt::format("#define SAMPLER_BINDING_{} {}\n", sampler.GetIndex(),
+ base_bindings.sampler++);
+ }
+
+ if (program_type == Maxwell::ShaderProgram::Geometry) {
+ const auto [glsl_topology, _, max_vertices] = GetPrimitiveDescription(primitive_mode);
+
+ source += "layout (" + std::string(glsl_topology) + ") in;\n";
+ source += "#define MAX_VERTEX_INPUT " + std::to_string(max_vertices) + '\n';
+ }
+
+ source += code;
+
+ OGLShader shader;
+ shader.Create(source.c_str(), GetShaderType(program_type));
+
+ auto program = std::make_shared<OGLProgram>();
+ program->Create(true, hint_retrievable, shader.handle);
+ return program;
+}
+
+std::set<GLenum> GetSupportedFormats() {
+ std::set<GLenum> supported_formats;
+
+ GLint num_formats{};
+ glGetIntegerv(GL_NUM_PROGRAM_BINARY_FORMATS, &num_formats);
+
+ std::vector<GLint> formats(num_formats);
+ glGetIntegerv(GL_PROGRAM_BINARY_FORMATS, formats.data());
+
+ for (const GLint format : formats)
+ supported_formats.insert(static_cast<GLenum>(format));
+ return supported_formats;
+}
+
+} // namespace
+
+CachedShader::CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
+ ShaderDiskCacheOpenGL& disk_cache,
+ const PrecompiledPrograms& precompiled_programs,
+ ProgramCode&& program_code, ProgramCode&& program_code_b)
+ : addr{addr}, unique_identifier{unique_identifier}, program_type{program_type},
+ disk_cache{disk_cache}, precompiled_programs{precompiled_programs} {
+
+ const std::size_t code_size = CalculateProgramSize(program_code);
+ const std::size_t code_size_b =
+ program_code_b.empty() ? 0 : CalculateProgramSize(program_code_b);
+
+ GLShader::ProgramResult program_result =
+ CreateProgram(program_type, program_code, program_code_b);
+ if (program_result.first.empty()) {
+ // TODO(Rodrigo): Unimplemented shader stages hit here, avoid using these for now
return;
}
code = program_result.first;
entries = program_result.second;
shader_length = entries.shader_length;
+
+ const ShaderDiskCacheRaw raw(unique_identifier, program_type,
+ static_cast<u32>(code_size / sizeof(u64)),
+ static_cast<u32>(code_size_b / sizeof(u64)),
+ std::move(program_code), std::move(program_code_b));
+ disk_cache.SaveRaw(raw);
+}
+
+CachedShader::CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
+ ShaderDiskCacheOpenGL& disk_cache,
+ const PrecompiledPrograms& precompiled_programs,
+ GLShader::ProgramResult result)
+ : addr{addr}, unique_identifier{unique_identifier}, program_type{program_type},
+ disk_cache{disk_cache}, precompiled_programs{precompiled_programs} {
+
+ code = std::move(result.first);
+ entries = result.second;
+ shader_length = entries.shader_length;
}
std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(GLenum primitive_mode,
@@ -94,138 +261,195 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(GLenum primitive
const auto [entry, is_cache_miss] = programs.try_emplace(base_bindings);
auto& program = entry->second;
if (is_cache_miss) {
- std::string source = AllocateBindings(base_bindings);
- source += code;
+ program = TryLoadProgram(primitive_mode, base_bindings);
+ if (!program) {
+ program =
+ SpecializeShader(code, entries, program_type, base_bindings, primitive_mode);
+ disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings));
+ }
- OGLShader shader;
- shader.Create(source.c_str(), GetShaderType(program_type));
- program.Create(true, shader.handle);
- LabelGLObject(GL_PROGRAM, program.handle, addr);
+ LabelGLObject(GL_PROGRAM, program->handle, addr);
}
- handle = program.handle;
+ handle = program->handle;
}
- // Add const buffer and samplers offset reserved by this shader. One UBO binding is reserved for
- // emulation values
- base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size()) + 1;
+ base_bindings.cbuf += static_cast<u32>(entries.const_buffers.size()) + RESERVED_UBOS;
base_bindings.gmem += static_cast<u32>(entries.global_memory_entries.size());
base_bindings.sampler += static_cast<u32>(entries.samplers.size());
return {handle, base_bindings};
}
-std::string CachedShader::AllocateBindings(BaseBindings base_bindings) {
- std::string code = "#version 430 core\n";
- code += fmt::format("#define EMULATION_UBO_BINDING {}\n", base_bindings.cbuf++);
-
- for (const auto& cbuf : entries.const_buffers) {
- code += fmt::format("#define CBUF_BINDING_{} {}\n", cbuf.GetIndex(), base_bindings.cbuf++);
- }
-
- for (const auto& gmem : entries.global_memory_entries) {
- code += fmt::format("#define GMEM_BINDING_{}_{} {}\n", gmem.GetCbufIndex(),
- gmem.GetCbufOffset(), base_bindings.gmem++);
- }
-
- for (const auto& sampler : entries.samplers) {
- code += fmt::format("#define SAMPLER_BINDING_{} {}\n", sampler.GetIndex(),
- base_bindings.sampler++);
- }
-
- return code;
-}
-
GLuint CachedShader::GetGeometryShader(GLenum primitive_mode, BaseBindings base_bindings) {
const auto [entry, is_cache_miss] = geometry_programs.try_emplace(base_bindings);
auto& programs = entry->second;
switch (primitive_mode) {
case GL_POINTS:
- return LazyGeometryProgram(programs.points, base_bindings, "points", 1, "ShaderPoints");
+ return LazyGeometryProgram(programs.points, base_bindings, primitive_mode);
case GL_LINES:
case GL_LINE_STRIP:
- return LazyGeometryProgram(programs.lines, base_bindings, "lines", 2, "ShaderLines");
+ return LazyGeometryProgram(programs.lines, base_bindings, primitive_mode);
case GL_LINES_ADJACENCY:
case GL_LINE_STRIP_ADJACENCY:
- return LazyGeometryProgram(programs.lines_adjacency, base_bindings, "lines_adjacency", 4,
- "ShaderLinesAdjacency");
+ return LazyGeometryProgram(programs.lines_adjacency, base_bindings, primitive_mode);
case GL_TRIANGLES:
case GL_TRIANGLE_STRIP:
case GL_TRIANGLE_FAN:
- return LazyGeometryProgram(programs.triangles, base_bindings, "triangles", 3,
- "ShaderTriangles");
+ return LazyGeometryProgram(programs.triangles, base_bindings, primitive_mode);
case GL_TRIANGLES_ADJACENCY:
case GL_TRIANGLE_STRIP_ADJACENCY:
- return LazyGeometryProgram(programs.triangles_adjacency, base_bindings,
- "triangles_adjacency", 6, "ShaderTrianglesAdjacency");
+ return LazyGeometryProgram(programs.triangles_adjacency, base_bindings, primitive_mode);
default:
UNREACHABLE_MSG("Unknown primitive mode.");
- return LazyGeometryProgram(programs.points, base_bindings, "points", 1, "ShaderPoints");
+ return LazyGeometryProgram(programs.points, base_bindings, primitive_mode);
}
}
-GLuint CachedShader::LazyGeometryProgram(OGLProgram& target_program, BaseBindings base_bindings,
- const std::string& glsl_topology, u32 max_vertices,
- const std::string& debug_name) {
- if (target_program.handle != 0) {
- return target_program.handle;
+GLuint CachedShader::LazyGeometryProgram(CachedProgram& target_program, BaseBindings base_bindings,
+ GLenum primitive_mode) {
+ if (target_program) {
+ return target_program->handle;
+ }
+ const auto [_, debug_name, __] = GetPrimitiveDescription(primitive_mode);
+ target_program = TryLoadProgram(primitive_mode, base_bindings);
+ if (!target_program) {
+ target_program =
+ SpecializeShader(code, entries, program_type, base_bindings, primitive_mode);
+ disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings));
}
- std::string source = AllocateBindings(base_bindings);
- source += "layout (" + glsl_topology + ") in;\n";
- source += "#define MAX_VERTEX_INPUT " + std::to_string(max_vertices) + '\n';
- source += code;
- OGLShader shader;
- shader.Create(source.c_str(), GL_GEOMETRY_SHADER);
- target_program.Create(true, shader.handle);
- LabelGLObject(GL_PROGRAM, target_program.handle, addr, debug_name);
- return target_program.handle;
+ LabelGLObject(GL_PROGRAM, target_program->handle, addr, debug_name);
+
+ return target_program->handle;
};
-static bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
- // sched instructions appear once every 4 instructions.
- static constexpr std::size_t SchedPeriod = 4;
- const std::size_t absolute_offset = offset - main_offset;
- return (absolute_offset % SchedPeriod) == 0;
+CachedProgram CachedShader::TryLoadProgram(GLenum primitive_mode,
+ BaseBindings base_bindings) const {
+ const auto found = precompiled_programs.find(GetUsage(primitive_mode, base_bindings));
+ if (found == precompiled_programs.end()) {
+ return {};
+ }
+ return found->second;
}
-static std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) {
- constexpr std::size_t start_offset = 10;
- std::size_t offset = start_offset;
- std::size_t size = start_offset * sizeof(u64);
- while (offset < program.size()) {
- const u64 inst = program[offset];
- if (!IsSchedInstruction(offset, start_offset)) {
- if (inst == 0 || (inst >> 52) == 0x50b) {
- break;
- }
+ShaderDiskCacheUsage CachedShader::GetUsage(GLenum primitive_mode,
+ BaseBindings base_bindings) const {
+ return {unique_identifier, base_bindings, primitive_mode};
+}
+
+ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer) : RasterizerCache{rasterizer} {}
+
+void ShaderCacheOpenGL::LoadDiskCache() {
+ std::vector<ShaderDiskCacheRaw> raws;
+ std::vector<ShaderDiskCacheUsage> usages;
+ if (!disk_cache.LoadTransferable(raws, usages)) {
+ return;
+ }
+
+ std::vector<ShaderDiskCachePrecompiledEntry> precompiled = disk_cache.LoadPrecompiled();
+ const auto SearchPrecompiled = [&precompiled](const ShaderDiskCacheUsage& usage) {
+ return std::find_if(
+ precompiled.begin(), precompiled.end(),
+ [&usage](const auto& precompiled_entry) { return precompiled_entry.usage == usage; });
+ };
+
+ const std::set<GLenum> supported_formats{GetSupportedFormats()};
+ const auto unspecialized{GenerateUnspecializedShaders(raws)};
+
+ // Build shaders
+ for (std::size_t i = 0; i < usages.size(); ++i) {
+ const auto& usage{usages[i]};
+ LOG_INFO(Render_OpenGL, "Building shader {:016x} ({} of {})", usage.unique_identifier,
+ i + 1, usages.size());
+
+ const auto& unspec{unspecialized.at(usage.unique_identifier)};
+
+ const auto precompiled_it = SearchPrecompiled(usage);
+ const bool is_precompiled = precompiled_it != precompiled.end();
+
+ CachedProgram shader;
+ if (is_precompiled) {
+ shader = GeneratePrecompiledProgram(precompiled, *precompiled_it, supported_formats);
+ }
+ if (!shader) {
+ shader = SpecializeShader(unspec.code, unspec.entries, unspec.program_type,
+ usage.bindings, usage.primitive, true);
+ }
+ precompiled_programs.insert({usage, std::move(shader)});
+ }
+
+ // TODO(Rodrigo): Do state tracking for transferable shaders and do a dummy draw before
+ // precompiling them
+
+ for (std::size_t i = 0; i < usages.size(); ++i) {
+ const auto& usage{usages[i]};
+ if (SearchPrecompiled(usage) == precompiled.end()) {
+ const auto& program = precompiled_programs.at(usage);
+ disk_cache.SavePrecompiled(usage, program->handle);
}
- size += sizeof(inst);
- offset++;
}
- return size;
}
-void CachedShader::CalculateProperties() {
- setup.program.real_size = CalculateProgramSize(setup.program.code);
- setup.program.real_size_b = 0;
- setup.program.unique_identifier = Common::CityHash64(
- reinterpret_cast<const char*>(setup.program.code.data()), setup.program.real_size);
- if (program_type == Maxwell::ShaderProgram::VertexA) {
- std::size_t seed = 0;
- boost::hash_combine(seed, setup.program.unique_identifier);
- setup.program.real_size_b = CalculateProgramSize(setup.program.code_b);
- const u64 identifier_b = Common::CityHash64(
- reinterpret_cast<const char*>(setup.program.code_b.data()), setup.program.real_size_b);
- boost::hash_combine(seed, identifier_b);
- setup.program.unique_identifier = static_cast<u64>(seed);
+CachedProgram ShaderCacheOpenGL::GeneratePrecompiledProgram(
+ std::vector<ShaderDiskCachePrecompiledEntry>& precompiled,
+ const ShaderDiskCachePrecompiledEntry& precompiled_entry,
+ const std::set<GLenum>& supported_formats) {
+
+ if (supported_formats.find(precompiled_entry.binary_format) == supported_formats.end()) {
+ LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format - removing");
+ disk_cache.InvalidatePrecompiled();
+ precompiled.clear();
+ return {};
}
+
+ CachedProgram shader = std::make_shared<OGLProgram>();
+ shader->handle = glCreateProgram();
+ glProgramBinary(shader->handle, precompiled_entry.binary_format,
+ precompiled_entry.binary.data(),
+ static_cast<GLsizei>(precompiled_entry.binary.size()));
+
+ GLint link_status{};
+ glGetProgramiv(shader->handle, GL_LINK_STATUS, &link_status);
+ if (link_status == GL_FALSE) {
+ LOG_INFO(Render_OpenGL, "Precompiled cache rejected by the driver - removing");
+ disk_cache.InvalidatePrecompiled();
+ precompiled.clear();
+
+ shader.reset();
+ }
+
+ return shader;
}
-ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer) : RasterizerCache{rasterizer} {}
+std::map<u64, UnspecializedShader> ShaderCacheOpenGL::GenerateUnspecializedShaders(
+ const std::vector<ShaderDiskCacheRaw>& raws) {
+
+ std::map<u64, UnspecializedShader> unspecialized;
+ for (const auto& raw : raws) {
+ const u64 calculated_hash =
+ GetUniqueIdentifier(raw.GetProgramType(), raw.GetProgramCode(), raw.GetProgramCodeB());
+ if (raw.GetUniqueIdentifier() != calculated_hash) {
+ LOG_ERROR(
+ Render_OpenGL,
+ "Invalid hash in entry={:016x} (obtained hash={:016x}) - removing shader cache",
+ raw.GetUniqueIdentifier(), calculated_hash);
+ disk_cache.InvalidateTransferable();
+ return {};
+ }
+
+ auto result =
+ CreateProgram(raw.GetProgramType(), raw.GetProgramCode(), raw.GetProgramCodeB());
+
+ precompiled_shaders.insert({raw.GetUniqueIdentifier(), result});
-void ShaderCacheOpenGL::LoadDiskCache() {}
+ unspecialized.insert(
+ {raw.GetUniqueIdentifier(),
+ {std::move(result.first), std::move(result.second), raw.GetProgramType()}});
+ }
+ return unspecialized;
+}
Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
if (!Core::System::GetInstance().GPU().Maxwell3D().dirty_flags.shaders) {
@@ -239,7 +463,23 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
if (!shader) {
// No shader found - create a new one
- shader = std::make_shared<CachedShader>(program_addr, program);
+ ProgramCode program_code = GetShaderCode(program_addr);
+ ProgramCode program_code_b;
+ if (program == Maxwell::ShaderProgram::VertexA) {
+ program_code_b = GetShaderCode(GetShaderAddress(Maxwell::ShaderProgram::VertexB));
+ }
+ const u64 unique_identifier = GetUniqueIdentifier(program, program_code, program_code_b);
+
+ const auto found = precompiled_shaders.find(unique_identifier);
+ if (found != precompiled_shaders.end()) {
+ shader =
+ std::make_shared<CachedShader>(program_addr, unique_identifier, program, disk_cache,
+ precompiled_programs, found->second);
+ } else {
+ shader = std::make_shared<CachedShader>(
+ program_addr, unique_identifier, program, disk_cache, precompiled_programs,
+ std::move(program_code), std::move(program_code_b));
+ }
Register(shader);
}
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h
index 18fb80bcc..763a47bce 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_cache.h
@@ -7,6 +7,7 @@
#include <array>
#include <map>
#include <memory>
+#include <set>
#include <tuple>
#include <glad/glad.h>
@@ -23,13 +24,25 @@ namespace OpenGL {
class CachedShader;
class RasterizerOpenGL;
+struct UnspecializedShader;
using Shader = std::shared_ptr<CachedShader>;
+using CachedProgram = std::shared_ptr<OGLProgram>;
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+using PrecompiledPrograms = std::map<ShaderDiskCacheUsage, CachedProgram>;
+using PrecompiledShaders = std::map<u64, GLShader::ProgramResult>;
class CachedShader final : public RasterizerCacheObject {
public:
- CachedShader(VAddr addr, Maxwell::ShaderProgram program_type);
+ explicit CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
+ ShaderDiskCacheOpenGL& disk_cache,
+ const PrecompiledPrograms& precompiled_programs,
+ ProgramCode&& program_code, ProgramCode&& program_code_b);
+
+ explicit CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
+ ShaderDiskCacheOpenGL& disk_cache,
+ const PrecompiledPrograms& precompiled_programs,
+ GLShader::ProgramResult result);
VAddr GetAddr() const override {
return addr;
@@ -56,33 +69,35 @@ private:
// declared by the hardware. Workaround this issue by generating a different shader per input
// topology class.
struct GeometryPrograms {
- OGLProgram points;
- OGLProgram lines;
- OGLProgram lines_adjacency;
- OGLProgram triangles;
- OGLProgram triangles_adjacency;
+ CachedProgram points;
+ CachedProgram lines;
+ CachedProgram lines_adjacency;
+ CachedProgram triangles;
+ CachedProgram triangles_adjacency;
};
- std::string AllocateBindings(BaseBindings base_bindings);
-
GLuint GetGeometryShader(GLenum primitive_mode, BaseBindings base_bindings);
/// Generates a geometry shader or returns one that already exists.
- GLuint LazyGeometryProgram(OGLProgram& target_program, BaseBindings base_bindings,
- const std::string& glsl_topology, u32 max_vertices,
- const std::string& debug_name);
+ GLuint LazyGeometryProgram(CachedProgram& target_program, BaseBindings base_bindings,
+ GLenum primitive_mode);
+
+ CachedProgram TryLoadProgram(GLenum primitive_mode, BaseBindings base_bindings) const;
+
+ ShaderDiskCacheUsage GetUsage(GLenum primitive_mode, BaseBindings base_bindings) const;
- void CalculateProperties();
+ const VAddr addr;
+ const u64 unique_identifier;
+ const Maxwell::ShaderProgram program_type;
+ ShaderDiskCacheOpenGL& disk_cache;
+ const PrecompiledPrograms& precompiled_programs;
- VAddr addr{};
std::size_t shader_length{};
- Maxwell::ShaderProgram program_type{};
- GLShader::ShaderSetup setup;
GLShader::ShaderEntries entries;
std::string code;
- std::map<BaseBindings, OGLProgram> programs;
+ std::map<BaseBindings, CachedProgram> programs;
std::map<BaseBindings, GeometryPrograms> geometry_programs;
std::map<u32, GLuint> cbuf_resource_cache;
@@ -101,7 +116,19 @@ public:
Shader GetStageProgram(Maxwell::ShaderProgram program);
private:
+ std::map<u64, UnspecializedShader> GenerateUnspecializedShaders(
+ const std::vector<ShaderDiskCacheRaw>& raws);
+
+ CachedProgram GeneratePrecompiledProgram(
+ std::vector<ShaderDiskCachePrecompiledEntry>& precompiled,
+ const ShaderDiskCachePrecompiledEntry& precompiled_entry,
+ const std::set<GLenum>& supported_formats);
+
std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
+
+ ShaderDiskCacheOpenGL disk_cache;
+ PrecompiledShaders precompiled_shaders;
+ PrecompiledPrograms precompiled_programs;
};
} // namespace OpenGL