Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

gl_shader_cache: Rework shader cache and remove post-specializations

Browse source 
Instead of pre-specializing shaders and then post-specializing them,
drop the later and only "specialize" the shader while decoding it.
This commit is contained in:
ReinUsesLisp 2020-02-26 16:13:47 -03:00
parent 22e825a3bc
commit bd8b9bbcee
19 changed files with 548 additions and 1100 deletions
Download patch file Download diff file Expand all files Collapse all files

7
src/video_core/shader/const_buffer_locker.cpp
View file

@ -14,8 +14,9 @@ namespace VideoCommon::Shader {
using Tegra::Engines::SamplerDescriptor;
ConstBufferLocker::ConstBufferLocker(Tegra::Engines::ShaderType shader_stage)
: stage{shader_stage} {}
ConstBufferLocker::ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
VideoCore::GuestDriverProfile stored_guest_driver_profile)
: stage{shader_stage}, stored_guest_driver_profile{stored_guest_driver_profile} {}
ConstBufferLocker::ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
Tegra::Engines::ConstBufferEngineInterface& engine)
@ -97,7 +98,7 @@ void ConstBufferLocker::SetBoundBuffer(u32 buffer) {
bool ConstBufferLocker::IsConsistent() const {
if (!engine) {
return false;
return true;
}
return std::all_of(keys.begin(), keys.end(),
[this](const auto& pair) {

11
src/video_core/shader/const_buffer_locker.h
View file

@ -26,7 +26,8 @@ using BindlessSamplerMap =
*/
class ConstBufferLocker {
public:
explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage);
explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
VideoCore::GuestDriverProfile stored_guest_driver_profile);
explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
Tegra::Engines::ConstBufferEngineInterface& engine);
@ -83,15 +84,13 @@ public:
}
/// Obtains access to the guest driver's profile.
VideoCore::GuestDriverProfile* AccessGuestDriverProfile() const {
if (engine) {
return &engine->AccessGuestDriverProfile();
}
return nullptr;
VideoCore::GuestDriverProfile& AccessGuestDriverProfile() {
return engine ? engine->AccessGuestDriverProfile() : stored_guest_driver_profile;
}
private:
const Tegra::Engines::ShaderType stage;
VideoCore::GuestDriverProfile stored_guest_driver_profile;
Tegra::Engines::ConstBufferEngineInterface* engine = nullptr;
KeyMap keys;
BoundSamplerMap bound_samplers;

18
src/video_core/shader/decode.cpp
View file

@ -34,13 +34,9 @@ constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
return (absolute_offset % SchedPeriod) == 0;
}
void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile* gpu_driver,
void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile& gpu_driver,
const std::list<Sampler>& used_samplers) {
if (gpu_driver == nullptr) {
LOG_CRITICAL(HW_GPU, "GPU driver profile has not been created yet");
return;
}
if (gpu_driver->TextureHandlerSizeKnown() || used_samplers.size() <= 1) {
if (gpu_driver.IsTextureHandlerSizeKnown() || used_samplers.size() <= 1) {
return;
}
u32 count{};
@ -53,17 +49,13 @@ void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile* gpu_driver,
bound_offsets.emplace_back(sampler.GetOffset());
}
if (count > 1) {
gpu_driver->DeduceTextureHandlerSize(std::move(bound_offsets));
gpu_driver.DeduceTextureHandlerSize(std::move(bound_offsets));
}
}
std::optional<u32> TryDeduceSamplerSize(const Sampler& sampler_to_deduce,
VideoCore::GuestDriverProfile* gpu_driver,
VideoCore::GuestDriverProfile& gpu_driver,
const std::list<Sampler>& used_samplers) {
if (gpu_driver == nullptr) {
LOG_CRITICAL(HW_GPU, "GPU Driver profile has not been created yet");
return std::nullopt;
}
const u32 base_offset = sampler_to_deduce.GetOffset();
u32 max_offset{std::numeric_limits<u32>::max()};
for (const auto& sampler : used_samplers) {
@ -77,7 +69,7 @@ std::optional<u32> TryDeduceSamplerSize(const Sampler& sampler_to_deduce,
if (max_offset == std::numeric_limits<u32>::max()) {
return std::nullopt;
}
return ((max_offset - base_offset) * 4) / gpu_driver->GetTextureHandlerSize();
return ((max_offset - base_offset) * 4) / gpu_driver.GetTextureHandlerSize();
}
} // Anonymous namespace

9
src/video_core/shader/track.cpp
View file

@ -94,13 +94,10 @@ std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, cons
}
auto [gpr, base_offset] = *pair;
const auto offset_inm = std::get_if<ImmediateNode>(&*base_offset);
auto gpu_driver = locker.AccessGuestDriverProfile();
if (gpu_driver == nullptr) {
return {};
}
const auto& gpu_driver = locker.AccessGuestDriverProfile();
const u32 bindless_cv = NewCustomVariable();
const Node op = Operation(OperationCode::UDiv, NO_PRECISE, gpr,
Immediate(gpu_driver->GetTextureHandlerSize()));
const Node op =
Operation(OperationCode::UDiv, gpr, Immediate(gpu_driver.GetTextureHandlerSize()));
const Node cv_node = GetCustomVariable(bindless_cv);
Node amend_op = Operation(OperationCode::Assign, cv_node, std::move(op));