diff options
Diffstat (limited to 'src/video_core/shader_cache.cpp')
| -rw-r--r-- | src/video_core/shader_cache.cpp | 233 |
1 files changed, 233 insertions, 0 deletions
diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp new file mode 100644 index 000000000..b8b8eace5 --- /dev/null +++ b/src/video_core/shader_cache.cpp | |||
| @@ -0,0 +1,233 @@ | |||
| 1 | // Copyright 2021 yuzu Emulator Project | ||
| 2 | // Licensed under GPLv2 or any later version | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <array> | ||
| 7 | #include <vector> | ||
| 8 | |||
| 9 | #include "common/assert.h" | ||
| 10 | #include "shader_recompiler/frontend/maxwell/control_flow.h" | ||
| 11 | #include "shader_recompiler/object_pool.h" | ||
| 12 | #include "video_core/dirty_flags.h" | ||
| 13 | #include "video_core/engines/kepler_compute.h" | ||
| 14 | #include "video_core/engines/maxwell_3d.h" | ||
| 15 | #include "video_core/memory_manager.h" | ||
| 16 | #include "video_core/shader_cache.h" | ||
| 17 | #include "video_core/shader_environment.h" | ||
| 18 | |||
| 19 | namespace VideoCommon { | ||
| 20 | |||
| 21 | void ShaderCache::InvalidateRegion(VAddr addr, size_t size) { | ||
| 22 | std::scoped_lock lock{invalidation_mutex}; | ||
| 23 | InvalidatePagesInRegion(addr, size); | ||
| 24 | RemovePendingShaders(); | ||
| 25 | } | ||
| 26 | |||
| 27 | void ShaderCache::OnCPUWrite(VAddr addr, size_t size) { | ||
| 28 | std::lock_guard lock{invalidation_mutex}; | ||
| 29 | InvalidatePagesInRegion(addr, size); | ||
| 30 | } | ||
| 31 | |||
| 32 | void ShaderCache::SyncGuestHost() { | ||
| 33 | std::scoped_lock lock{invalidation_mutex}; | ||
| 34 | RemovePendingShaders(); | ||
| 35 | } | ||
| 36 | |||
| 37 | ShaderCache::ShaderCache(VideoCore::RasterizerInterface& rasterizer_, | ||
| 38 | Tegra::MemoryManager& gpu_memory_, Tegra::Engines::Maxwell3D& maxwell3d_, | ||
| 39 | Tegra::Engines::KeplerCompute& kepler_compute_) | ||
| 40 | : gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_}, | ||
| 41 | rasterizer{rasterizer_} {} | ||
| 42 | |||
| 43 | bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) { | ||
| 44 | auto& dirty{maxwell3d.dirty.flags}; | ||
| 45 | if (!dirty[VideoCommon::Dirty::Shaders]) { | ||
| 46 | return last_shaders_valid; | ||
| 47 | } | ||
| 48 | dirty[VideoCommon::Dirty::Shaders] = false; | ||
| 49 | |||
| 50 | const GPUVAddr base_addr{maxwell3d.regs.code_address.CodeAddress()}; | ||
| 51 | for (size_t index = 0; index < Tegra::Engines::Maxwell3D::Regs::MaxShaderProgram; ++index) { | ||
| 52 | if (!maxwell3d.regs.IsShaderConfigEnabled(index)) { | ||
| 53 | unique_hashes[index] = 0; | ||
| 54 | continue; | ||
| 55 | } | ||
| 56 | const auto& shader_config{maxwell3d.regs.shader_config[index]}; | ||
| 57 | const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderProgram>(index)}; | ||
| 58 | const GPUVAddr shader_addr{base_addr + shader_config.offset}; | ||
| 59 | const std::optional<VAddr> cpu_shader_addr{gpu_memory.GpuToCpuAddress(shader_addr)}; | ||
| 60 | if (!cpu_shader_addr) { | ||
| 61 | LOG_ERROR(HW_GPU, "Invalid GPU address for shader 0x{:016x}", shader_addr); | ||
| 62 | last_shaders_valid = false; | ||
| 63 | return false; | ||
| 64 | } | ||
| 65 | const ShaderInfo* shader_info{TryGet(*cpu_shader_addr)}; | ||
| 66 | if (!shader_info) { | ||
| 67 | const u32 start_address{shader_config.offset}; | ||
| 68 | GraphicsEnvironment env{maxwell3d, gpu_memory, program, base_addr, start_address}; | ||
| 69 | shader_info = MakeShaderInfo(env, *cpu_shader_addr); | ||
| 70 | } | ||
| 71 | shader_infos[index] = shader_info; | ||
| 72 | unique_hashes[index] = shader_info->unique_hash; | ||
| 73 | } | ||
| 74 | last_shaders_valid = true; | ||
| 75 | return true; | ||
| 76 | } | ||
| 77 | |||
| 78 | const ShaderInfo* ShaderCache::ComputeShader() { | ||
| 79 | const GPUVAddr program_base{kepler_compute.regs.code_loc.Address()}; | ||
| 80 | const auto& qmd{kepler_compute.launch_description}; | ||
| 81 | const GPUVAddr shader_addr{program_base + qmd.program_start}; | ||
| 82 | const std::optional<VAddr> cpu_shader_addr{gpu_memory.GpuToCpuAddress(shader_addr)}; | ||
| 83 | if (!cpu_shader_addr) { | ||
| 84 | LOG_ERROR(HW_GPU, "Invalid GPU address for shader 0x{:016x}", shader_addr); | ||
| 85 | return nullptr; | ||
| 86 | } | ||
| 87 | if (const ShaderInfo* const shader = TryGet(*cpu_shader_addr)) { | ||
| 88 | return shader; | ||
| 89 | } | ||
| 90 | ComputeEnvironment env{kepler_compute, gpu_memory, program_base, qmd.program_start}; | ||
| 91 | return MakeShaderInfo(env, *cpu_shader_addr); | ||
| 92 | } | ||
| 93 | |||
| 94 | ShaderInfo* ShaderCache::TryGet(VAddr addr) const { | ||
| 95 | std::scoped_lock lock{lookup_mutex}; | ||
| 96 | |||
| 97 | const auto it = lookup_cache.find(addr); | ||
| 98 | if (it == lookup_cache.end()) { | ||
| 99 | return nullptr; | ||
| 100 | } | ||
| 101 | return it->second->data; | ||
| 102 | } | ||
| 103 | |||
| 104 | void ShaderCache::Register(std::unique_ptr<ShaderInfo> data, VAddr addr, size_t size) { | ||
| 105 | std::scoped_lock lock{invalidation_mutex, lookup_mutex}; | ||
| 106 | |||
| 107 | const VAddr addr_end = addr + size; | ||
| 108 | Entry* const entry = NewEntry(addr, addr_end, data.get()); | ||
| 109 | |||
| 110 | const u64 page_end = (addr_end + PAGE_SIZE - 1) >> PAGE_BITS; | ||
| 111 | for (u64 page = addr >> PAGE_BITS; page < page_end; ++page) { | ||
| 112 | invalidation_cache[page].push_back(entry); | ||
| 113 | } | ||
| 114 | |||
| 115 | storage.push_back(std::move(data)); | ||
| 116 | |||
| 117 | rasterizer.UpdatePagesCachedCount(addr, size, 1); | ||
| 118 | } | ||
| 119 | |||
| 120 | void ShaderCache::InvalidatePagesInRegion(VAddr addr, size_t size) { | ||
| 121 | const VAddr addr_end = addr + size; | ||
| 122 | const u64 page_end = (addr_end + PAGE_SIZE - 1) >> PAGE_BITS; | ||
| 123 | for (u64 page = addr >> PAGE_BITS; page < page_end; ++page) { | ||
| 124 | auto it = invalidation_cache.find(page); | ||
| 125 | if (it == invalidation_cache.end()) { | ||
| 126 | continue; | ||
| 127 | } | ||
| 128 | InvalidatePageEntries(it->second, addr, addr_end); | ||
| 129 | } | ||
| 130 | } | ||
| 131 | |||
| 132 | void ShaderCache::RemovePendingShaders() { | ||
| 133 | if (marked_for_removal.empty()) { | ||
| 134 | return; | ||
| 135 | } | ||
| 136 | // Remove duplicates | ||
| 137 | std::ranges::sort(marked_for_removal); | ||
| 138 | marked_for_removal.erase(std::unique(marked_for_removal.begin(), marked_for_removal.end()), | ||
| 139 | marked_for_removal.end()); | ||
| 140 | |||
| 141 | std::vector<ShaderInfo*> removed_shaders; | ||
| 142 | removed_shaders.reserve(marked_for_removal.size()); | ||
| 143 | |||
| 144 | std::scoped_lock lock{lookup_mutex}; | ||
| 145 | |||
| 146 | for (Entry* const entry : marked_for_removal) { | ||
| 147 | removed_shaders.push_back(entry->data); | ||
| 148 | |||
| 149 | const auto it = lookup_cache.find(entry->addr_start); | ||
| 150 | ASSERT(it != lookup_cache.end()); | ||
| 151 | lookup_cache.erase(it); | ||
| 152 | } | ||
| 153 | marked_for_removal.clear(); | ||
| 154 | |||
| 155 | if (!removed_shaders.empty()) { | ||
| 156 | RemoveShadersFromStorage(std::move(removed_shaders)); | ||
| 157 | } | ||
| 158 | } | ||
| 159 | |||
| 160 | void ShaderCache::InvalidatePageEntries(std::vector<Entry*>& entries, VAddr addr, VAddr addr_end) { | ||
| 161 | size_t index = 0; | ||
| 162 | while (index < entries.size()) { | ||
| 163 | Entry* const entry = entries[index]; | ||
| 164 | if (!entry->Overlaps(addr, addr_end)) { | ||
| 165 | ++index; | ||
| 166 | continue; | ||
| 167 | } | ||
| 168 | |||
| 169 | UnmarkMemory(entry); | ||
| 170 | RemoveEntryFromInvalidationCache(entry); | ||
| 171 | marked_for_removal.push_back(entry); | ||
| 172 | } | ||
| 173 | } | ||
| 174 | |||
| 175 | void ShaderCache::RemoveEntryFromInvalidationCache(const Entry* entry) { | ||
| 176 | const u64 page_end = (entry->addr_end + PAGE_SIZE - 1) >> PAGE_BITS; | ||
| 177 | for (u64 page = entry->addr_start >> PAGE_BITS; page < page_end; ++page) { | ||
| 178 | const auto entries_it = invalidation_cache.find(page); | ||
| 179 | ASSERT(entries_it != invalidation_cache.end()); | ||
| 180 | std::vector<Entry*>& entries = entries_it->second; | ||
| 181 | |||
| 182 | const auto entry_it = std::ranges::find(entries, entry); | ||
| 183 | ASSERT(entry_it != entries.end()); | ||
| 184 | entries.erase(entry_it); | ||
| 185 | } | ||
| 186 | } | ||
| 187 | |||
| 188 | void ShaderCache::UnmarkMemory(Entry* entry) { | ||
| 189 | if (!entry->is_memory_marked) { | ||
| 190 | return; | ||
| 191 | } | ||
| 192 | entry->is_memory_marked = false; | ||
| 193 | |||
| 194 | const VAddr addr = entry->addr_start; | ||
| 195 | const size_t size = entry->addr_end - addr; | ||
| 196 | rasterizer.UpdatePagesCachedCount(addr, size, -1); | ||
| 197 | } | ||
| 198 | |||
| 199 | void ShaderCache::RemoveShadersFromStorage(std::vector<ShaderInfo*> removed_shaders) { | ||
| 200 | // Remove them from the cache | ||
| 201 | std::erase_if(storage, [&removed_shaders](const std::unique_ptr<ShaderInfo>& shader) { | ||
| 202 | return std::ranges::find(removed_shaders, shader.get()) != removed_shaders.end(); | ||
| 203 | }); | ||
| 204 | } | ||
| 205 | |||
| 206 | ShaderCache::Entry* ShaderCache::NewEntry(VAddr addr, VAddr addr_end, ShaderInfo* data) { | ||
| 207 | auto entry = std::make_unique<Entry>(Entry{addr, addr_end, data}); | ||
| 208 | Entry* const entry_pointer = entry.get(); | ||
| 209 | |||
| 210 | lookup_cache.emplace(addr, std::move(entry)); | ||
| 211 | return entry_pointer; | ||
| 212 | } | ||
| 213 | |||
| 214 | const ShaderInfo* ShaderCache::MakeShaderInfo(GenericEnvironment& env, VAddr cpu_addr) { | ||
| 215 | auto info = std::make_unique<ShaderInfo>(); | ||
| 216 | if (const std::optional<u64> cached_hash{env.Analyze()}) { | ||
| 217 | info->unique_hash = *cached_hash; | ||
| 218 | info->size_bytes = env.CachedSize(); | ||
| 219 | } else { | ||
| 220 | // Slow path, not really hit on commercial games | ||
| 221 | // Build a control flow graph to get the real shader size | ||
| 222 | Shader::ObjectPool<Shader::Maxwell::Flow::Block> flow_block; | ||
| 223 | Shader::Maxwell::Flow::CFG cfg{env, flow_block, env.StartAddress()}; | ||
| 224 | info->unique_hash = env.CalculateHash(); | ||
| 225 | info->size_bytes = env.ReadSize(); | ||
| 226 | } | ||
| 227 | const size_t size_bytes{info->size_bytes}; | ||
| 228 | const ShaderInfo* const result{info.get()}; | ||
| 229 | Register(std::move(info), cpu_addr, size_bytes); | ||
| 230 | return result; | ||
| 231 | } | ||
| 232 | |||
| 233 | } // namespace VideoCommon | ||