This commit is contained in:
Fire Cube 2025-05-03 15:29:57 +02:00
parent 3eb8000b75
commit a500b98e9b
6 changed files with 582 additions and 90 deletions

View File

@ -15,33 +15,4 @@ void writeBin(std::ostream& os, const T& v) {
template <typename T>
void readBin(std::istream& is, T& v) {
is.read(reinterpret_cast<char*>(&v), sizeof(T));
}
// Spezialfall für Arrays/Blöcke
template <typename T>
void writeBlock(std::ostream& os, const T* data, size_t count) {
os.write(reinterpret_cast<const char*>(data), sizeof(T) * count);
}
template <typename T>
void readBlock(std::istream& is, T* data, size_t count) {
is.read(reinterpret_cast<char*>(data), sizeof(T) * count);
}
// Spezialfall für Container
template <typename T>
void writeContainer(std::ostream& os, const std::vector<T>& v) {
u32 n = static_cast<u32>(v.size());
writeBin(os, n);
if (n)
writeBlock(os, v.data(), n);
}
template <typename T>
void readContainer(std::istream& is, std::vector<T>& v) {
u32 n;
readBin(is, n);
v.resize(n);
if (n)
readBlock(is, v.data(), n);
}
}

View File

@ -5,10 +5,7 @@
#include "common/types.h"
[[nodiscard]] inline u64 HashCombine(const u64 seed, const u64 hash) {
return seed ^ (hash + 0x9e3779b9 + (seed << 12) + (seed >> 4));
}
[[nodiscard]] inline u32 HashCombine(const u32 seed, const u32 hash) {
return seed ^ (hash + 0x9e3779b9 + (seed << 6) + (seed >> 2));
template <typename T, typename U>
T HashCombine(const T& seed, const U& value) {
return seed ^ (static_cast<T>(value) + 0x9e3779b9 + (seed << 6) + (seed >> 2));
}

View File

@ -1,12 +1,14 @@
#include <iostream>
#include <vector>
#include <functional>
#include <string>
#include "common/hash.h"
#include "common/path_util.h"
#include "common/io_file.h"
#include "common/binary_helper.h"
#include "common/logging/log.h"
#include "shader_recompiler/ir/type.h"
#include "shader_recompiler/info.h"
#include "shader_recompiler/specialization.h"
using u64 = uint64_t;
using u32 = uint32_t;
@ -15,13 +17,242 @@ namespace ShaderCache {
const auto shader_cache_dir = Common::FS::GetUserPath(Common::FS::PathType::ShaderDir) / "cache";
std::string CreateShaderID(u64 pgm_hash, size_t perm_idx, std::ostream& info_serialized) {
std::ostringstream data_stream;
data_stream << pgm_hash << perm_idx;
data_stream << info_serialized.rdbuf();
std::hash<std::string> hasher;
size_t shader_id = hasher(data_stream.str());
return std::to_string(shader_id);
u64 CalculateSpecializationHash(const Shader::StageSpecialization& spec) {
u64 hash = 0;
// Start mit dem Hash der RuntimeInfo
// Die RuntimeInfo enthält verschiedene Unions, daher müssen wir basierend auf dem Stage-Typ
// hashen
const auto& runtime_info = spec.runtime_info;
hash = HashCombine(hash, static_cast<u32>(runtime_info.stage));
hash = HashCombine(hash, runtime_info.num_user_data);
hash = HashCombine(hash, runtime_info.num_input_vgprs);
hash = HashCombine(hash, runtime_info.num_allocated_vgprs);
hash = HashCombine(hash, static_cast<u32>(runtime_info.fp_denorm_mode32));
hash = HashCombine(hash, static_cast<u32>(runtime_info.fp_round_mode32));
// Abhängig vom Stage-Typ die spezifischen RuntimeInfo-Felder hashen
switch (runtime_info.stage) {
case Shader::Stage::Local:
hash = HashCombine(hash, runtime_info.ls_info.ls_stride);
hash = HashCombine(hash, runtime_info.ls_info.links_with_tcs);
break;
case Shader::Stage::Export:
hash = HashCombine(hash, runtime_info.es_info.vertex_data_size);
break;
case Shader::Stage::Vertex:
hash = HashCombine(hash, runtime_info.vs_info.num_outputs);
// Hash der Output-Maps
for (size_t i = 0;
i < runtime_info.vs_info.num_outputs && i < runtime_info.vs_info.outputs.size(); ++i) {
const auto& output_map = runtime_info.vs_info.outputs[i];
for (const auto& output : output_map) {
hash = HashCombine(hash, static_cast<u32>(output));
}
}
hash = HashCombine(hash, runtime_info.vs_info.emulate_depth_negative_one_to_one);
hash = HashCombine(hash, runtime_info.vs_info.clip_disable);
hash = HashCombine(hash, static_cast<u32>(runtime_info.vs_info.tess_type));
hash = HashCombine(hash, static_cast<u32>(runtime_info.vs_info.tess_topology));
hash = HashCombine(hash, static_cast<u32>(runtime_info.vs_info.tess_partitioning));
hash = HashCombine(hash, runtime_info.vs_info.hs_output_cp_stride);
break;
case Shader::Stage::Hull:
hash = HashCombine(hash, runtime_info.hs_info.num_input_control_points);
hash = HashCombine(hash, runtime_info.hs_info.num_threads);
hash = HashCombine(hash, static_cast<u32>(runtime_info.hs_info.tess_type));
hash = HashCombine(hash, runtime_info.hs_info.ls_stride);
hash = HashCombine(hash, runtime_info.hs_info.hs_output_cp_stride);
hash = HashCombine(hash, runtime_info.hs_info.hs_output_base);
break;
case Shader::Stage::Geometry:
hash = HashCombine(hash, runtime_info.gs_info.num_invocations);
hash = HashCombine(hash, runtime_info.gs_info.output_vertices);
hash = HashCombine(hash, runtime_info.gs_info.in_vertex_data_size);
hash = HashCombine(hash, runtime_info.gs_info.out_vertex_data_size);
hash = HashCombine(hash, static_cast<u32>(runtime_info.gs_info.in_primitive));
// Hash der Output-Primitive-Types für alle Streams
for (const auto& out_prim : runtime_info.gs_info.out_primitive) {
hash = HashCombine(hash, static_cast<u32>(out_prim));
}
hash = HashCombine(hash, runtime_info.gs_info.vs_copy_hash);
break;
case Shader::Stage::Fragment:
hash = HashCombine(hash, runtime_info.fs_info.en_flags.raw);
hash = HashCombine(hash, runtime_info.fs_info.addr_flags.raw);
hash = HashCombine(hash, runtime_info.fs_info.num_inputs);
// Hash der PS-Inputs
for (u32 i = 0;
i < runtime_info.fs_info.num_inputs && i < runtime_info.fs_info.inputs.size(); ++i) {
const auto& input = runtime_info.fs_info.inputs[i];
hash = HashCombine(hash, input.param_index);
hash = HashCombine(hash, input.is_default);
hash = HashCombine(hash, input.is_flat);
hash = HashCombine(hash, input.default_value);
}
// Hash der ColorBuffers
for (const auto& color_buffer : runtime_info.fs_info.color_buffers) {
hash = HashCombine(hash, static_cast<u32>(color_buffer.num_format));
hash = HashCombine(hash, static_cast<u32>(color_buffer.num_conversion));
hash = HashCombine(hash, static_cast<u32>(color_buffer.export_format));
hash = HashCombine(hash, color_buffer.needs_unorm_fixup);
hash = HashCombine(hash, color_buffer.swizzle.r);
hash = HashCombine(hash, color_buffer.swizzle.g);
hash = HashCombine(hash, color_buffer.swizzle.b);
hash = HashCombine(hash, color_buffer.swizzle.a);
}
break;
case Shader::Stage::Compute:
hash = HashCombine(hash, runtime_info.cs_info.shared_memory_size);
for (u32 i = 0; i < 3; ++i) {
hash = HashCombine(hash, runtime_info.cs_info.workgroup_size[i]);
hash = HashCombine(hash, runtime_info.cs_info.tgid_enable[i]);
}
break;
}
// Hash des FetchShader-Daten
if (spec.fetch_shader_data) {
const auto& fetch_shader = *spec.fetch_shader_data;
hash = HashCombine(hash, fetch_shader.size);
hash = HashCombine(hash, static_cast<u64>(fetch_shader.vertex_offset_sgpr));
hash = HashCombine(hash, static_cast<u64>(fetch_shader.instance_offset_sgpr));
// Hash der Attribute
for (const auto& attr : fetch_shader.attributes) {
hash = HashCombine(hash, static_cast<u64>(attr.semantic));
hash = HashCombine(hash, static_cast<u64>(attr.dest_vgpr));
hash = HashCombine(hash, static_cast<u64>(attr.num_elements));
hash = HashCombine(hash, static_cast<u64>(attr.sgpr_base));
hash = HashCombine(hash, static_cast<u64>(attr.dword_offset));
hash = HashCombine(hash, static_cast<u64>(attr.instance_data));
}
}
// Hash der VS-Attribut-Spezialisierungen
for (const auto& vs_attrib : spec.vs_attribs) {
hash = HashCombine(hash, vs_attrib.num_components);
hash = HashCombine(hash, static_cast<u32>(vs_attrib.num_class));
hash = HashCombine(hash, vs_attrib.dst_select.r);
hash = HashCombine(hash, vs_attrib.dst_select.g);
hash = HashCombine(hash, vs_attrib.dst_select.b);
hash = HashCombine(hash, vs_attrib.dst_select.a);
}
// Hash des Bitsets
const std::string bitset_str = spec.bitset.to_string();
for (size_t i = 0; i < bitset_str.size(); i += 8) {
size_t end = std::min(i + 8, bitset_str.size());
std::string chunk = bitset_str.substr(i, end - i);
u8 value = 0;
for (size_t j = 0; j < chunk.size(); ++j) {
if (chunk[j] == '1') {
value |= (1 << j);
}
}
hash = HashCombine(hash, value);
}
// Hash der Buffer-Spezialisierungen
for (const auto& buffer : spec.buffers) {
hash = HashCombine(hash, buffer.stride);
hash = HashCombine(hash, buffer.is_storage);
hash = HashCombine(hash, buffer.is_formatted);
hash = HashCombine(hash, buffer.swizzle_enable);
if (buffer.is_formatted) {
hash = HashCombine(hash, buffer.data_format);
hash = HashCombine(hash, buffer.num_format);
hash = HashCombine(hash, buffer.dst_select.r);
hash = HashCombine(hash, buffer.dst_select.g);
hash = HashCombine(hash, buffer.dst_select.b);
hash = HashCombine(hash, buffer.dst_select.a);
hash = HashCombine(hash, static_cast<u32>(buffer.num_conversion));
}
if (buffer.swizzle_enable) {
hash = HashCombine(hash, buffer.index_stride);
hash = HashCombine(hash, buffer.element_size);
}
}
// Hash der Bild-Spezialisierungen
for (const auto& image : spec.images) {
hash = HashCombine(hash, static_cast<u32>(image.type));
hash = HashCombine(hash, image.is_integer);
hash = HashCombine(hash, image.is_storage);
hash = HashCombine(hash, image.is_cube);
if (image.is_storage) {
hash = HashCombine(hash, image.dst_select.r);
hash = HashCombine(hash, image.dst_select.g);
hash = HashCombine(hash, image.dst_select.b);
hash = HashCombine(hash, image.dst_select.a);
}
hash = HashCombine(hash, static_cast<u32>(image.num_conversion));
}
// Hash der FMask-Spezialisierungen
for (const auto& fmask : spec.fmasks) {
hash = HashCombine(hash, fmask.width);
hash = HashCombine(hash, fmask.height);
}
// Hash der Sampler-Spezialisierungen
for (const auto& sampler : spec.samplers) {
hash = HashCombine(hash, sampler.force_unnormalized);
}
// Hash der Start-Bindings
hash = HashCombine(hash, spec.start.buffer);
hash = HashCombine(hash, spec.start.unified);
hash = HashCombine(hash, spec.start.user_data);
// Hash vom info pointer
if (spec.info) {
hash = HashCombine(hash, spec.info->pgm_hash);
hash = HashCombine(hash, static_cast<u32>(spec.info->stage));
hash = HashCombine(hash, static_cast<u32>(spec.info->l_stage));
hash = HashCombine(hash, spec.info->has_storage_images);
hash = HashCombine(hash, spec.info->has_discard);
hash = HashCombine(hash, spec.info->has_image_gather);
hash = HashCombine(hash, spec.info->has_image_query);
hash = HashCombine(hash, spec.info->uses_lane_id);
hash = HashCombine(hash, spec.info->uses_group_quad);
hash = HashCombine(hash, spec.info->uses_group_ballot);
hash = HashCombine(hash, spec.info->uses_shared);
hash = HashCombine(hash, spec.info->uses_fp16);
hash = HashCombine(hash, spec.info->uses_fp64);
hash = HashCombine(hash, spec.info->uses_pack_10_11_11);
hash = HashCombine(hash, spec.info->uses_unpack_10_11_11);
hash = HashCombine(hash, spec.info->stores_tess_level_outer);
hash = HashCombine(hash, spec.info->stores_tess_level_inner);
hash = HashCombine(hash, spec.info->translation_failed);
hash = HashCombine(hash, spec.info->has_readconst);
hash = HashCombine(hash, spec.info->mrt_mask);
hash = HashCombine(hash, spec.info->has_fetch_shader);
hash = HashCombine(hash, spec.info->fetch_shader_sgpr_base);
// Hash der Flags für loads und stores
for (size_t i = 0; i < spec.info->loads.flags.size(); ++i) {
hash = HashCombine(hash, spec.info->loads.flags[i]);
}
for (size_t i = 0; i < spec.info->stores.flags.size(); ++i) {
hash = HashCombine(hash, spec.info->stores.flags[i]);
}
// Hash der UserDataMask
hash = HashCombine(hash, spec.info->ud_mask.mask);
// Hash der uses_patches
hash = HashCombine(hash, spec.info->uses_patches);
}
return hash;
}
void SerializeInfo(std::ostream& info_serialized, Shader::Info info) {
@ -29,9 +260,8 @@ void SerializeInfo(std::ostream& info_serialized, Shader::Info info) {
writeBin(info_serialized, info.ud_mask.mask);
// Buffer-Resources
u32 count = static_cast<u32>(info.buffers.size());
writeBin(info_serialized, count); // Buffer Amount
LOG_INFO(Render_Recompiler, "ShaderCache: Buffer count: {}", info.buffers.size());
u32 bufferCount = static_cast<u32>(info.buffers.size());
writeBin(info_serialized, bufferCount); // Buffer Amount
for (const auto& buffer : info.buffers) {
writeBin(info_serialized, buffer.sharp_idx);
@ -43,9 +273,9 @@ void SerializeInfo(std::ostream& info_serialized, Shader::Info info) {
}
// Image-Resources
count = static_cast<u32>(info.images.size());
writeBin(info_serialized, count); // Image Amount
LOG_INFO(Render_Recompiler, "ShaderCache: Image count: {}", info.images.size());
u32 imageCount = static_cast<u32>(info.images.size());
writeBin(info_serialized, imageCount); // Image Amount
for (const auto& image : info.images) {
writeBin(info_serialized, image.sharp_idx);
writeBin(info_serialized, static_cast<u8>(image.is_depth ? 1 : 0));
@ -55,35 +285,36 @@ void SerializeInfo(std::ostream& info_serialized, Shader::Info info) {
}
// Sampler-Resources
count = static_cast<u32>(info.samplers.size());
writeBin(info_serialized, count); // Sampler Amount
LOG_INFO(Render_Recompiler, "ShaderCache: Sampler count: {}", info.samplers.size());
u32 samplerCount = static_cast<u32>(info.samplers.size());
writeBin(info_serialized, samplerCount); // Sampler Amount
for (const auto& sampler : info.samplers) {
writeBin(info_serialized, sampler.sharp_idx);
}
// FMask-Resources
count = static_cast<u32>(info.fmasks.size());
writeBin(info_serialized, count); // FMask Amount
LOG_INFO(Render_Recompiler, "ShaderCache: FMask count: {}", info.fmasks.size());
u32 fmaskCount = static_cast<u32>(info.fmasks.size());
writeBin(info_serialized, fmaskCount); // FMask Amount
for (const auto& fmask : info.fmasks) {
writeBin(info_serialized, fmask.sharp_idx);
}
// GS Copy Data
writeBin(info_serialized, info.gs_copy_data.num_attrs);
u32 mapCount = static_cast<u32>(info.gs_copy_data.attr_map.size());
writeBin(info_serialized, mapCount);
for (auto const& [loc, idx] : info.gs_copy_data.attr_map) {
for (auto const& [loc, attr_pair] : info.gs_copy_data.attr_map) {
writeBin(info_serialized, loc);
writeBin(info_serialized, idx);
// Das erste Element des Paars ist ein Shader::IR::Attribute, ein Enum
writeBin(info_serialized, static_cast<u32>(attr_pair.first));
// Das zweite Element ist ein u32
writeBin(info_serialized, attr_pair.second);
}
// SRT Info
u32 srtCount = static_cast<u32>(info.srt_info.srt_reservations.size());
writeBin(info_serialized, count);
writeBin(info_serialized, srtCount);
for (const auto& res : info.srt_info.srt_reservations) {
writeBin(info_serialized, res.sgpr_base);
@ -91,24 +322,313 @@ void SerializeInfo(std::ostream& info_serialized, Shader::Info info) {
writeBin(info_serialized, res.num_dwords);
}
// Flat UD
u32 flatCount = static_cast<u32>(info.flattened_ud_buf.size());
writeBin(info_serialized, flatCount);
for (const auto& flat : info.flattened_ud_buf) {
writeBin(info_serialized, flat);
}
// Tessellation Data
writeBin(info_serialized, info.tess_consts_ptr_base);
writeBin(info_serialized, info.tess_consts_dword_offset);
// Flags
writeBin(info_serialized, static_cast<u8>(info.has_storage_images ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.has_discard ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.has_image_gather ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.has_image_query ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.uses_lane_id ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.uses_group_quad ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.uses_group_ballot ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.uses_shared ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.uses_fp16 ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.uses_fp64 ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.uses_pack_10_11_11 ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.uses_unpack_10_11_11 ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.stores_tess_level_outer ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.stores_tess_level_inner ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.translation_failed ? 1 : 0));
writeBin(info_serialized, static_cast<u8>(info.has_readconst ? 1 : 0));
// MRT Mask
writeBin(info_serialized, info.mrt_mask);
}
void DeserializeInfo(std::istream& info_serialized, Shader::Info& info) {
readBin(info_serialized, info.mrt_mask);
// Fetch
writeBin(info_serialized, static_cast<u8>(info.has_fetch_shader ? 1 : 0));
writeBin(info_serialized, info.fetch_shader_sgpr_base);
// Stage
writeBin(info_serialized, info.stage);
writeBin(info_serialized, info.l_stage);
writeBin(info_serialized, info.pgm_hash);
// AttributeFlags für loads
u32 loads_size = static_cast<u32>(info.loads.flags.size());
writeBin(info_serialized, loads_size);
for (size_t i = 0; i < info.loads.flags.size(); ++i) {
writeBin(info_serialized, info.loads.flags[i]);
}
// AttributeFlags für stores
u32 stores_size = static_cast<u32>(info.stores.flags.size());
writeBin(info_serialized, stores_size);
for (size_t i = 0; i < info.stores.flags.size(); ++i) {
writeBin(info_serialized, info.stores.flags[i]);
}
}
bool CheckShaderCache(std::string shader_id) {
return 0;
// Überprüfen, ob das Verzeichnis existiert
if (!std::filesystem::exists(shader_cache_dir)) {
LOG_INFO(Render_Vulkan, "Shader-Cache-Verzeichnis existiert nicht");
return false;
}
// Überprüfen, ob sowohl die SPIR-V-Datei als auch die Ressourcendatei existieren
std::filesystem::path spirv_cache_file_path = shader_cache_dir / (shader_id + ".spv");
std::filesystem::path resources_file_path = shader_cache_dir / (shader_id + ".resources");
if (!std::filesystem::exists(spirv_cache_file_path)) {
LOG_DEBUG(Render_Vulkan, "SPIR-V-Datei nicht gefunden: {}", spirv_cache_file_path.string());
return false;
}
if (!std::filesystem::exists(resources_file_path)) {
LOG_DEBUG(Render_Vulkan, "Ressourcendatei nicht gefunden: {}",
resources_file_path.string());
return false;
}
// Überprüfen, ob die Dateien lesbar und nicht leer sind
Common::FS::IOFile spirv_file(spirv_cache_file_path, Common::FS::FileAccessMode::Read);
Common::FS::IOFile resources_file(resources_file_path, Common::FS::FileAccessMode::Read);
const bool spirv_valid = spirv_file.IsOpen() && spirv_file.GetSize() > 0;
const bool resources_valid = resources_file.IsOpen() && resources_file.GetSize() > 0;
spirv_file.Close();
resources_file.Close();
if (!spirv_valid || !resources_valid) {
LOG_WARNING(Render_Vulkan, "Ungueltige Dateien im Shader-Cache für ID: {}", shader_id);
// Fehlerhafte Dateien entfernen, um zukünftige Probleme zu vermeiden
if (std::filesystem::exists(spirv_cache_file_path)) {
std::filesystem::remove(spirv_cache_file_path);
}
if (std::filesystem::exists(resources_file_path)) {
std::filesystem::remove(resources_file_path);
}
return false;
}
LOG_INFO(Render_Vulkan, "Shader mit ID {} im Cache gefunden", shader_id);
return true;
}
void GetShader(std::string shader_id, Shader::Info& info) {
std::string spirv_cache_filename = shader_id + ".spv ";
void DeserializeInfo(std::istream& info_serialized, Shader::Info& info) {
// UD Mask
readBin(info_serialized, info.ud_mask.mask);
// Buffer-Resources
u32 bufferCount;
readBin(info_serialized, bufferCount);
info.buffers.clear();
info.buffers.reserve(bufferCount);
for (u32 i = 0; i < bufferCount; ++i) {
Shader::BufferResource buffer;
readBin(info_serialized, buffer.sharp_idx);
u32 used_types;
readBin(info_serialized, used_types);
buffer.used_types = static_cast<Shader::IR::Type>(used_types);
u32 buffer_type;
readBin(info_serialized, buffer_type);
buffer.buffer_type = static_cast<Shader::BufferType>(buffer_type);
readBin(info_serialized, buffer.instance_attrib);
u8 is_written;
readBin(info_serialized, is_written);
buffer.is_written = (is_written == 1);
u8 is_formatted;
readBin(info_serialized, is_formatted);
buffer.is_formatted = (is_formatted == 1);
info.buffers.push_back(std::move(buffer));
}
// Image-Resources
u32 imageCount;
readBin(info_serialized, imageCount);
info.images.clear();
info.images.reserve(imageCount);
for (u32 i = 0; i < imageCount; ++i) {
Shader::ImageResource image;
readBin(info_serialized, image.sharp_idx);
u8 is_depth;
readBin(info_serialized, is_depth);
image.is_depth = (is_depth == 1);
u8 is_atomic;
readBin(info_serialized, is_atomic);
image.is_atomic = (is_atomic == 1);
u8 is_array;
readBin(info_serialized, is_array);
image.is_array = (is_array == 1);
u8 is_written;
readBin(info_serialized, is_written);
image.is_written = (is_written == 1);
info.images.push_back(std::move(image));
}
// Sampler-Resources
u32 samplerCount;
readBin(info_serialized, samplerCount);
info.samplers.clear();
info.samplers.reserve(samplerCount);
for (u32 i = 0; i < samplerCount; ++i) {
Shader::SamplerResource sampler;
readBin(info_serialized, sampler.sharp_idx);
info.samplers.push_back(std::move(sampler));
}
// FMask-Resources
u32 fmaskCount;
readBin(info_serialized, fmaskCount);
info.fmasks.clear();
info.fmasks.reserve(fmaskCount);
for (u32 i = 0; i < fmaskCount; ++i) {
Shader::FMaskResource fmask;
readBin(info_serialized, fmask.sharp_idx);
info.fmasks.push_back(std::move(fmask));
}
// GS Copy Data
u32 mapCount;
readBin(info_serialized, mapCount);
info.gs_copy_data.attr_map.clear();
for (u32 i = 0; i < mapCount; ++i) {
u32 loc;
u32 attribute_value;
u32 idx;
readBin(info_serialized, loc);
readBin(info_serialized, attribute_value);
readBin(info_serialized, idx);
// Umwandeln des numerischen Werts zurück in das Shader::IR::Attribute-Enum
Shader::IR::Attribute attribute = static_cast<Shader::IR::Attribute>(attribute_value);
// Einfügen in die Map mit dem richtigen Paar-Typ
info.gs_copy_data.attr_map.emplace(loc, std::make_pair(attribute, idx));
}
// SRT Info
u32 srtCount;
readBin(info_serialized, srtCount);
info.srt_info.srt_reservations.clear();
info.srt_info.srt_reservations.resize(srtCount);
for (u32 i = 0; i < srtCount; ++i) {
auto& res = info.srt_info.srt_reservations[i];
readBin(info_serialized, res.sgpr_base);
readBin(info_serialized, res.dword_offset);
readBin(info_serialized, res.num_dwords);
}
// Flat UD
u32 flatCount;
readBin(info_serialized, flatCount);
info.flattened_ud_buf.clear();
info.flattened_ud_buf.resize(flatCount);
for (u32 i = 0; i < flatCount; ++i) {
readBin(info_serialized, info.flattened_ud_buf[i]);
}
// Tessellation Data
readBin(info_serialized, info.tess_consts_ptr_base);
readBin(info_serialized, info.tess_consts_dword_offset);
// Flags
u8 flag_value;
readBin(info_serialized, flag_value);
info.has_storage_images = (flag_value == 1);
readBin(info_serialized, flag_value);
info.has_discard = (flag_value == 1);
readBin(info_serialized, flag_value);
info.has_image_gather = (flag_value == 1);
readBin(info_serialized, flag_value);
info.has_image_query = (flag_value == 1);
readBin(info_serialized, flag_value);
info.uses_lane_id = (flag_value == 1);
readBin(info_serialized, flag_value);
info.uses_group_quad = (flag_value == 1);
readBin(info_serialized, flag_value);
info.uses_group_ballot = (flag_value == 1);
readBin(info_serialized, flag_value);
info.uses_shared = (flag_value == 1);
readBin(info_serialized, flag_value);
info.uses_fp16 = (flag_value == 1);
readBin(info_serialized, flag_value);
info.uses_fp64 = (flag_value == 1);
readBin(info_serialized, flag_value);
info.uses_pack_10_11_11 = (flag_value == 1);
readBin(info_serialized, flag_value);
info.uses_unpack_10_11_11 = (flag_value == 1);
readBin(info_serialized, flag_value);
info.stores_tess_level_outer = (flag_value == 1);
readBin(info_serialized, flag_value);
info.stores_tess_level_inner = (flag_value == 1);
readBin(info_serialized, flag_value);
info.translation_failed = (flag_value == 1);
readBin(info_serialized, flag_value);
info.has_readconst = (flag_value == 1);
// MRT Mask
readBin(info_serialized, info.mrt_mask);
// Fetch Shader
u8 has_fetch_shader;
readBin(info_serialized, has_fetch_shader);
info.has_fetch_shader = (has_fetch_shader == 1);
readBin(info_serialized, info.fetch_shader_sgpr_base);
// Stage
readBin(info_serialized, info.stage);
readBin(info_serialized, info.l_stage);
readBin(info_serialized, info.pgm_hash);
// AttributeFlags für loads
u32 loads_size;
readBin(info_serialized, loads_size);
for (size_t i = 0; i < loads_size && i < info.loads.flags.size(); ++i) {
readBin(info_serialized, info.loads.flags[i]);
}
// AttributeFlags für stores
u32 stores_size;
readBin(info_serialized, stores_size);
for (size_t i = 0; i < stores_size && i < info.stores.flags.size(); ++i) {
readBin(info_serialized, info.stores.flags[i]);
}
// Check if there are any remaining bytes in the stream
if (info_serialized.peek() != EOF) {
LOG_WARNING(Render_Vulkan, "Es sind noch {} Bytes im Stream übrig",
info_serialized.gcount());
}
}
void GetShader(std::string shader_id, Shader::Info& info, std::vector<u32>& spv) {
std::string spirv_cache_filename = shader_id + ".spv";
std::filesystem::path spirv_cache_file_path = shader_cache_dir / spirv_cache_filename;
Common::FS::IOFile spirv_cache_file(spirv_cache_file_path,
Common::FS::FileAccessMode::Read);
std::vector<u32> spv;
spv.resize(spirv_cache_file.GetSize() / sizeof(u32));
spirv_cache_file.Read(spv);
spirv_cache_file.Close();
@ -126,7 +646,6 @@ void GetShader(std::string shader_id, Shader::Info& info) {
// Verarbeite die gespeicherten Daten
std::istringstream combined_stream(std::string(resources_data.begin(), resources_data.end()));
// Deserialisiere info und profile
std::istringstream info_stream;
info_stream.str(std::string(resources_data.begin(), resources_data.end()));
DeserializeInfo(info_stream, info);
@ -146,7 +665,7 @@ void AddShader(std::string shader_id, std::vector<u32> spv, std::ostream& info_s
Common::FS::IOFile resources_dump_file(resources_dump_file_path,
Common::FS::FileAccessMode::Write);
// Die Streams müssen zurückgesetzt werden, bevor wir sie lesen können
// Die Streams müssen zurückgesetzt werden, bevor wir sie lesen können
if (std::ostringstream* info_oss = dynamic_cast<std::ostringstream*>(&info_serialized)) {
std::string info_data = info_oss->str();
resources_dump_file.WriteSpan(std::span<const char>(info_data.data(), info_data.size()));

View File

@ -5,15 +5,17 @@
#include <string>
#include <vector>
#include "shader_recompiler/info.h"
#include <shader_recompiler/specialization.h>
namespace ShaderCache {
std::string CreateShaderID(u64 pgm_hash, size_t perm_idx, std::ostream& info_dump);
void SerializeInfo(std::ostream& info_serialized, Shader::Info info);
u64 CalculateSpecializationHash(const Shader::StageSpecialization& spec);
void SerializeInfo(
std::ostream& info_serialized, Shader::Info info);
void DeserializeInfo(std::istream& info_serialized, Shader::Info& info);
bool CheckShaderCache(std::string shader_id);
void GetShader(std::string shader_id, Shader::Info& info);
void GetShader(std::string shader_id, Shader::Info& info, std::vector<u32>& spv);
void AddShader(std::string shader_id, std::vector<u32> spv, std::ostream& info_serialized);
} // namespace ShaderCache

View File

@ -488,34 +488,36 @@ bool PipelineCache::RefreshComputeKey() {
vk::ShaderModule PipelineCache::CompileModule(Shader::Info& info, Shader::RuntimeInfo& runtime_info,
std::span<const u32> code, size_t perm_idx,
Shader::Backend::Bindings& binding) {
Shader::Backend::Bindings& binding, Shader::StageSpecialization spec) {
LOG_INFO(Render_Vulkan, "Compiling {} shader {:#x} {}", info.stage, info.pgm_hash,
perm_idx != 0 ? "(permutation)" : "");
DumpShader(code, info.pgm_hash, info.stage, perm_idx, "bin");
const auto ir_program = Shader::TranslateProgram(code, pools, info, runtime_info, profile);
std::string shader_name = GetShaderName(info.stage, info.pgm_hash, perm_idx);
std::vector<u32> spv;
if (false){ //(::ShaderCache::CheckShaderCache(shader_id)) {
LOG_INFO(Render_Vulkan, "Loaded SPIR-V from cache");
std::string shader_id = std::to_string(::ShaderCache::CalculateSpecializationHash(spec));
if (::ShaderCache::CheckShaderCache(shader_id)) {
LOG_INFO(Render_Vulkan, "Loaded shader {} {:#x} {} from cache", info.stage, info.pgm_hash,
perm_idx != 0 ? "(permutation)" : "");
::ShaderCache::GetShader(shader_id, info, spv);
info.RefreshFlatBuf();
} else {
LOG_INFO(Render_Vulkan, "Shader {} {:#x} {} not in cache", info.stage,
info.pgm_hash, perm_idx != 0 ? "(permutation)" : "");
const auto ir_program = Shader::TranslateProgram(code, pools, info, runtime_info, profile);
spv = Shader::Backend::SPIRV::EmitSPIRV(profile, runtime_info, ir_program, binding);
std::ostringstream info_serialized;
::ShaderCache::SerializeInfo(info_serialized, info);
std::string shader_id =
::ShaderCache::CreateShaderID(info.pgm_hash, perm_idx, info_serialized);
::ShaderCache::AddShader(shader_id, spv, info_serialized);
LOG_INFO(Render_Vulkan, "Shader ID: {}", shader_id);
DumpShader(spv, info.pgm_hash, info.stage, perm_idx, "spv");
LOG_INFO(Render_Vulkan, "Compiled SPIR-V and stored in cache");
LOG_INFO(Render_Vulkan, "Compiled shader {} {:#x} {} and saved it to cache", info.stage, info.pgm_hash,
perm_idx != 0 ? "(permutation)" : "");
}
vk::ShaderModule module;
@ -543,13 +545,14 @@ PipelineCache::Result PipelineCache::GetProgram(Stage stage, LogicalStage l_stag
Shader::ShaderParams params,
Shader::Backend::Bindings& binding) {
auto runtime_info = BuildRuntimeInfo(stage, l_stage);
auto [it_pgm, new_program] = program_cache.try_emplace(params.hash);
auto [it_pgm, new_program] = program_cache.try_emplace(params.hash); // code in vs
if (new_program) {
it_pgm.value() = std::make_unique<Program>(stage, l_stage, params);
auto& program = it_pgm.value();
auto start = binding;
const auto module = CompileModule(program->info, runtime_info, params.code, 0, binding);
const auto spec = Shader::StageSpecialization(program->info, runtime_info, profile, start);
Shader::StageSpecialization spec =
Shader::StageSpecialization(program->info, runtime_info, profile, start);
const auto module = CompileModule(program->info, runtime_info, params.code, 0, binding, spec);
program->AddPermut(module, std::move(spec));
return std::make_tuple(&program->info, module, spec.fetch_shader_data,
HashCombine(params.hash, 0));
@ -566,7 +569,7 @@ PipelineCache::Result PipelineCache::GetProgram(Stage stage, LogicalStage l_stag
const auto it = std::ranges::find(program->modules, spec, &Program::Module::spec);
if (it == program->modules.end()) {
auto new_info = Shader::Info(stage, l_stage, params);
module = CompileModule(new_info, runtime_info, params.code, perm_idx, binding);
module = CompileModule(new_info, runtime_info, params.code, perm_idx, binding, spec);
program->AddPermut(module, std::move(spec));
} else {
info.AddBindings(binding);

View File

@ -82,7 +82,7 @@ private:
std::string_view ext);
vk::ShaderModule CompileModule(Shader::Info& info, Shader::RuntimeInfo& runtime_info,
std::span<const u32> code, size_t perm_idx,
Shader::Backend::Bindings& binding);
Shader::Backend::Bindings& binding, Shader::StageSpecialization spec);
const Shader::RuntimeInfo& BuildRuntimeInfo(Shader::Stage stage, Shader::LogicalStage l_stage);
private: