mirror of
https://git.h3cjp.net/H3cJP/citra.git
synced 2024-11-25 21:12:53 +00:00
Shader_IR: Address Feedback
This commit is contained in:
parent
b97608ca64
commit
64496f2456
|
@ -3,7 +3,7 @@
|
|||
// Refer to the license.txt file included.
|
||||
|
||||
#include <algorithm>
|
||||
#include <climits>
|
||||
#include <limits>
|
||||
|
||||
#include "video_core/guest_driver.h"
|
||||
|
||||
|
@ -17,10 +17,9 @@ void GuestDriverProfile::DeduceTextureHandlerSize(std::vector<u32>&& bound_offse
|
|||
if (size < 2) {
|
||||
return;
|
||||
}
|
||||
std::sort(bound_offsets.begin(), bound_offsets.end(),
|
||||
[](const u32& a, const u32& b) { return a < b; });
|
||||
u32 min_val = UINT_MAX;
|
||||
for (std::size_t i = 1; i < size; i++) {
|
||||
std::sort(bound_offsets.begin(), bound_offsets.end(), std::less{});
|
||||
u32 min_val = std::numeric_limits<u32>::max();
|
||||
for (std::size_t i = 1; i < size; ++i) {
|
||||
if (bound_offsets[i] == bound_offsets[i - 1]) {
|
||||
continue;
|
||||
}
|
||||
|
|
|
@ -12,10 +12,13 @@ namespace VideoCore {
|
|||
|
||||
/**
|
||||
* The GuestDriverProfile class is used to learn about the GPU drivers behavior and collect
|
||||
* information necessary for impossible to avoid HLE methods like shader tracks.
|
||||
* information necessary for impossible to avoid HLE methods like shader tracks as they are
|
||||
* Entscheidungsproblems.
|
||||
*/
|
||||
class GuestDriverProfile {
|
||||
public:
|
||||
void DeduceTextureHandlerSize(std::vector<u32>&& bound_offsets);
|
||||
|
||||
u32 GetTextureHandlerSize() const {
|
||||
return texture_handler_size;
|
||||
}
|
||||
|
@ -24,16 +27,14 @@ public:
|
|||
return texture_handler_size_deduced;
|
||||
}
|
||||
|
||||
void DeduceTextureHandlerSize(std::vector<u32>&& bound_offsets);
|
||||
|
||||
private:
|
||||
// Minimum size of texture handler any driver can use.
|
||||
static constexpr u32 min_texture_handler_size = 4;
|
||||
// This goes with Vulkan and OpenGL standards but Nvidia GPUs can easily
|
||||
// use 4 bytes instead. Thus, certain drivers may squish the size.
|
||||
static constexpr u32 default_texture_handler_size = 8;
|
||||
u32 texture_handler_size{default_texture_handler_size};
|
||||
bool texture_handler_size_deduced{};
|
||||
u32 texture_handler_size = default_texture_handler_size;
|
||||
bool texture_handler_size_deduced = false;
|
||||
};
|
||||
|
||||
} // namespace VideoCore
|
||||
|
|
|
@ -80,10 +80,12 @@ public:
|
|||
virtual void LoadDiskResources(const std::atomic_bool& stop_loading = false,
|
||||
const DiskResourceLoadCallback& callback = {}) {}
|
||||
|
||||
/// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver.
|
||||
GuestDriverProfile& AccessGuestDriverProfile() {
|
||||
return guest_driver_profile;
|
||||
}
|
||||
|
||||
/// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver.
|
||||
const GuestDriverProfile& AccessGuestDriverProfile() const {
|
||||
return guest_driver_profile;
|
||||
}
|
||||
|
|
|
@ -83,7 +83,7 @@ public:
|
|||
|
||||
VideoCore::GuestDriverProfile* AccessGuestDriverProfile() const {
|
||||
if (engine) {
|
||||
return &(engine->AccessGuestDriverProfile());
|
||||
return &engine->AccessGuestDriverProfile();
|
||||
}
|
||||
return nullptr;
|
||||
}
|
||||
|
|
|
@ -33,6 +33,29 @@ constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
|
|||
return (absolute_offset % SchedPeriod) == 0;
|
||||
}
|
||||
|
||||
void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile* gpu_driver,
|
||||
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) {
|
||||
return;
|
||||
}
|
||||
u32 count{};
|
||||
std::vector<u32> bound_offsets;
|
||||
for (const auto& sampler : used_samplers) {
|
||||
if (sampler.IsBindless()) {
|
||||
continue;
|
||||
}
|
||||
++count;
|
||||
bound_offsets.emplace_back(sampler.GetOffset());
|
||||
}
|
||||
if (count > 1) {
|
||||
gpu_driver->DeduceTextureHandlerSize(std::move(bound_offsets));
|
||||
}
|
||||
}
|
||||
|
||||
} // Anonymous namespace
|
||||
|
||||
class ASTDecoder {
|
||||
|
@ -315,32 +338,9 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
|
|||
return pc + 1;
|
||||
}
|
||||
|
||||
void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile* gpu_driver,
|
||||
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) {
|
||||
return;
|
||||
}
|
||||
u32 count{};
|
||||
std::vector<u32> bound_offsets;
|
||||
for (const auto& sampler : used_samplers) {
|
||||
if (sampler.IsBindless()) {
|
||||
continue;
|
||||
}
|
||||
count++;
|
||||
bound_offsets.emplace_back(sampler.GetOffset());
|
||||
}
|
||||
if (count > 1) {
|
||||
gpu_driver->DeduceTextureHandlerSize(std::move(bound_offsets));
|
||||
}
|
||||
}
|
||||
|
||||
void ShaderIR::PostDecode() {
|
||||
// Deduce texture handler size if needed
|
||||
auto* gpu_driver = locker.AccessGuestDriverProfile();
|
||||
auto gpu_driver = locker.AccessGuestDriverProfile();
|
||||
DeduceTextureHandlerSize(gpu_driver, used_samplers);
|
||||
}
|
||||
|
||||
|
|
Loading…
Reference in a new issue