mirror of
https://github.com/shadps4-emu/shadPS4.git
synced 2025-05-17 17:05:02 +00:00
shader_recompiler: Implement render target swizzles when no format is available (#739)
* shader_recompiler: Use null image when shader is compiled with unbound sharp * video_core: Refactor and render target swizzles * liverpool_to_vk: Add missing swap format from RDR * video_core: Refactor shader recompiler interface * Makes it much easier to pass runtime information to the recompiler and have it treated as part of the shader key. Also pulls out most runtime state from Info struct * shader_recompiler: Avoid some asserts
This commit is contained in:
TheTurtle
GitHub
parent
3f8a8d3a24
commit
f087f43736
30 changed files with 704 additions and 560 deletions
|
|
@ -3,20 +3,14 @@
|
|||
|
||||
#pragma once
|
||||
|
||||
#include <span>
|
||||
#include <boost/container/small_vector.hpp>
|
||||
#include <algorithm>
|
||||
#include <boost/container/static_vector.hpp>
|
||||
|
||||
#include "common/assert.h"
|
||||
#include "common/types.h"
|
||||
#include "shader_recompiler/ir/attribute.h"
|
||||
#include "shader_recompiler/ir/reg.h"
|
||||
#include "shader_recompiler/ir/type.h"
|
||||
#include "video_core/amdgpu/resource.h"
|
||||
|
||||
namespace Shader {
|
||||
|
||||
static constexpr size_t NumUserDataRegs = 16;
|
||||
|
||||
enum class Stage : u32 {
|
||||
Fragment,
|
||||
Vertex,
|
||||
|
|
@ -29,21 +23,18 @@ enum class Stage : u32 {
|
|||
constexpr u32 MaxStageTypes = 6;
|
||||
|
||||
[[nodiscard]] constexpr Stage StageFromIndex(size_t index) noexcept {
|
||||
return static_cast<Stage>(static_cast<size_t>(Stage::Vertex) + index);
|
||||
return static_cast<Stage>(index);
|
||||
}
|
||||
|
||||
enum class TextureType : u32 {
|
||||
Color1D,
|
||||
ColorArray1D,
|
||||
Color2D,
|
||||
ColorArray2D,
|
||||
Color3D,
|
||||
ColorCube,
|
||||
Buffer,
|
||||
enum class MrtSwizzle : u8 {
|
||||
Identity = 0,
|
||||
Alt = 1,
|
||||
Reverse = 2,
|
||||
ReverseAlt = 3,
|
||||
};
|
||||
constexpr u32 NUM_TEXTURE_TYPES = 7;
|
||||
static constexpr u32 MaxColorBuffers = 8;
|
||||
|
||||
enum class VsOutput : u32 {
|
||||
enum class VsOutput : u8 {
|
||||
None,
|
||||
PointSprite,
|
||||
EdgeFlag,
|
||||
|
|
@ -70,211 +61,69 @@ enum class VsOutput : u32 {
|
|||
};
|
||||
using VsOutputMap = std::array<VsOutput, 4>;
|
||||
|
||||
struct Info;
|
||||
struct VertexRuntimeInfo {
|
||||
boost::container::static_vector<VsOutputMap, 3> outputs;
|
||||
|
||||
struct BufferResource {
|
||||
u32 sgpr_base;
|
||||
u32 dword_offset;
|
||||
IR::Type used_types;
|
||||
AmdGpu::Buffer inline_cbuf;
|
||||
bool is_instance_data{};
|
||||
bool is_written{};
|
||||
|
||||
bool IsStorage(AmdGpu::Buffer buffer) const noexcept {
|
||||
static constexpr size_t MaxUboSize = 65536;
|
||||
return buffer.GetSize() > MaxUboSize || is_written;
|
||||
}
|
||||
|
||||
constexpr AmdGpu::Buffer GetSharp(const Info& info) const noexcept;
|
||||
};
|
||||
using BufferResourceList = boost::container::small_vector<BufferResource, 16>;
|
||||
|
||||
struct TextureBufferResource {
|
||||
u32 sgpr_base;
|
||||
u32 dword_offset;
|
||||
AmdGpu::NumberFormat nfmt;
|
||||
bool is_written{};
|
||||
|
||||
constexpr AmdGpu::Buffer GetSharp(const Info& info) const noexcept;
|
||||
};
|
||||
using TextureBufferResourceList = boost::container::small_vector<TextureBufferResource, 16>;
|
||||
|
||||
struct ImageResource {
|
||||
u32 sgpr_base;
|
||||
u32 dword_offset;
|
||||
AmdGpu::ImageType type;
|
||||
AmdGpu::NumberFormat nfmt;
|
||||
bool is_storage;
|
||||
bool is_depth;
|
||||
bool is_atomic{};
|
||||
|
||||
constexpr AmdGpu::Image GetSharp(const Info& info) const noexcept;
|
||||
};
|
||||
using ImageResourceList = boost::container::small_vector<ImageResource, 16>;
|
||||
|
||||
struct SamplerResource {
|
||||
u32 sgpr_base;
|
||||
u32 dword_offset;
|
||||
AmdGpu::Sampler inline_sampler{};
|
||||
u32 associated_image : 4;
|
||||
u32 disable_aniso : 1;
|
||||
|
||||
constexpr AmdGpu::Sampler GetSharp(const Info& info) const noexcept;
|
||||
};
|
||||
using SamplerResourceList = boost::container::small_vector<SamplerResource, 16>;
|
||||
|
||||
struct PushData {
|
||||
static constexpr size_t BufOffsetIndex = 2;
|
||||
|
||||
u32 step0;
|
||||
u32 step1;
|
||||
std::array<u8, 32> buf_offsets;
|
||||
|
||||
void AddOffset(u32 binding, u32 offset) {
|
||||
ASSERT(offset < 256 && binding < buf_offsets.size());
|
||||
buf_offsets[binding] = offset;
|
||||
bool operator==(const VertexRuntimeInfo& other) const noexcept {
|
||||
return true;
|
||||
}
|
||||
};
|
||||
|
||||
struct Info {
|
||||
struct VsInput {
|
||||
enum InstanceIdType : u8 {
|
||||
None = 0,
|
||||
OverStepRate0 = 1,
|
||||
OverStepRate1 = 2,
|
||||
Plain = 3,
|
||||
};
|
||||
|
||||
AmdGpu::NumberFormat fmt;
|
||||
u16 binding;
|
||||
u16 num_components;
|
||||
u8 sgpr_base;
|
||||
u8 dword_offset;
|
||||
InstanceIdType instance_step_rate;
|
||||
s32 instance_data_buf;
|
||||
};
|
||||
boost::container::static_vector<VsInput, 32> vs_inputs{};
|
||||
|
||||
struct FragmentRuntimeInfo {
|
||||
struct PsInput {
|
||||
u32 param_index;
|
||||
u8 param_index;
|
||||
bool is_default;
|
||||
bool is_flat;
|
||||
u32 default_value;
|
||||
u8 default_value;
|
||||
|
||||
auto operator<=>(const PsInput&) const noexcept = default;
|
||||
};
|
||||
boost::container::static_vector<PsInput, 32> ps_inputs{};
|
||||
boost::container::static_vector<PsInput, 32> inputs;
|
||||
std::array<MrtSwizzle, MaxColorBuffers> mrt_swizzles;
|
||||
|
||||
struct AttributeFlags {
|
||||
bool Get(IR::Attribute attrib, u32 comp = 0) const {
|
||||
return flags[Index(attrib)] & (1 << comp);
|
||||
}
|
||||
bool operator==(const FragmentRuntimeInfo& other) const noexcept {
|
||||
return std::ranges::equal(mrt_swizzles, other.mrt_swizzles) &&
|
||||
std::ranges::equal(inputs, other.inputs);
|
||||
}
|
||||
};
|
||||
|
||||
bool GetAny(IR::Attribute attrib) const {
|
||||
return flags[Index(attrib)];
|
||||
}
|
||||
|
||||
void Set(IR::Attribute attrib, u32 comp = 0) {
|
||||
flags[Index(attrib)] |= (1 << comp);
|
||||
}
|
||||
|
||||
u32 NumComponents(IR::Attribute attrib) const {
|
||||
return 4;
|
||||
}
|
||||
|
||||
static size_t Index(IR::Attribute attrib) {
|
||||
return static_cast<size_t>(attrib);
|
||||
}
|
||||
|
||||
std::array<u8, IR::NumAttributes> flags;
|
||||
};
|
||||
AttributeFlags loads{};
|
||||
AttributeFlags stores{};
|
||||
boost::container::static_vector<VsOutputMap, 3> vs_outputs;
|
||||
|
||||
s8 vertex_offset_sgpr = -1;
|
||||
s8 instance_offset_sgpr = -1;
|
||||
|
||||
BufferResourceList buffers;
|
||||
TextureBufferResourceList texture_buffers;
|
||||
ImageResourceList images;
|
||||
SamplerResourceList samplers;
|
||||
|
||||
std::array<u32, 3> workgroup_size{};
|
||||
struct ComputeRuntimeInfo {
|
||||
u32 shared_memory_size;
|
||||
std::array<u32, 3> workgroup_size;
|
||||
std::array<bool, 3> tgid_enable;
|
||||
|
||||
bool operator==(const ComputeRuntimeInfo& other) const noexcept {
|
||||
return workgroup_size == other.workgroup_size && tgid_enable == other.tgid_enable;
|
||||
}
|
||||
};
|
||||
|
||||
/**
|
||||
* Stores information relevant to shader compilation sourced from liverpool registers.
|
||||
* It may potentially differ with the same shader module so must be checked.
|
||||
* It's also possible to store any other custom information that needs to be part of shader key.
|
||||
*/
|
||||
struct RuntimeInfo {
|
||||
Stage stage;
|
||||
u32 num_user_data;
|
||||
u32 num_input_vgprs;
|
||||
std::span<const u32> user_data;
|
||||
Stage stage;
|
||||
VertexRuntimeInfo vs_info;
|
||||
FragmentRuntimeInfo fs_info;
|
||||
ComputeRuntimeInfo cs_info;
|
||||
|
||||
uintptr_t pgm_base{};
|
||||
u64 pgm_hash{};
|
||||
u32 shared_memory_size{};
|
||||
bool has_storage_images{};
|
||||
bool has_image_buffers{};
|
||||
bool has_texel_buffers{};
|
||||
bool has_discard{};
|
||||
bool has_image_gather{};
|
||||
bool has_image_query{};
|
||||
bool uses_lane_id{};
|
||||
bool uses_group_quad{};
|
||||
bool uses_shared{};
|
||||
bool uses_fp16{};
|
||||
bool uses_step_rates{};
|
||||
bool translation_failed{}; // indicates that shader has unsupported instructions
|
||||
RuntimeInfo(Stage stage_) : stage{stage_} {}
|
||||
|
||||
template <typename T>
|
||||
T ReadUd(u32 ptr_index, u32 dword_offset) const noexcept {
|
||||
T data;
|
||||
const u32* base = user_data.data();
|
||||
if (ptr_index != IR::NumScalarRegs) {
|
||||
std::memcpy(&base, &user_data[ptr_index], sizeof(base));
|
||||
bool operator==(const RuntimeInfo& other) const noexcept {
|
||||
switch (stage) {
|
||||
case Stage::Fragment:
|
||||
return fs_info == other.fs_info;
|
||||
case Stage::Vertex:
|
||||
return vs_info == other.vs_info;
|
||||
case Stage::Compute:
|
||||
return cs_info == other.cs_info;
|
||||
default:
|
||||
return true;
|
||||
}
|
||||
std::memcpy(&data, base + dword_offset, sizeof(T));
|
||||
return data;
|
||||
}
|
||||
|
||||
size_t NumBindings() const noexcept {
|
||||
return buffers.size() + texture_buffers.size() + images.size() + samplers.size();
|
||||
}
|
||||
|
||||
[[nodiscard]] std::pair<u32, u32> GetDrawOffsets() const noexcept {
|
||||
u32 vertex_offset = 0;
|
||||
u32 instance_offset = 0;
|
||||
if (vertex_offset_sgpr != -1) {
|
||||
vertex_offset = user_data[vertex_offset_sgpr];
|
||||
}
|
||||
if (instance_offset_sgpr != -1) {
|
||||
instance_offset = user_data[instance_offset_sgpr];
|
||||
}
|
||||
return {vertex_offset, instance_offset};
|
||||
}
|
||||
};
|
||||
|
||||
constexpr AmdGpu::Buffer BufferResource::GetSharp(const Info& info) const noexcept {
|
||||
return inline_cbuf ? inline_cbuf : info.ReadUd<AmdGpu::Buffer>(sgpr_base, dword_offset);
|
||||
}
|
||||
|
||||
constexpr AmdGpu::Buffer TextureBufferResource::GetSharp(const Info& info) const noexcept {
|
||||
return info.ReadUd<AmdGpu::Buffer>(sgpr_base, dword_offset);
|
||||
}
|
||||
|
||||
constexpr AmdGpu::Image ImageResource::GetSharp(const Info& info) const noexcept {
|
||||
return info.ReadUd<AmdGpu::Image>(sgpr_base, dword_offset);
|
||||
}
|
||||
|
||||
constexpr AmdGpu::Sampler SamplerResource::GetSharp(const Info& info) const noexcept {
|
||||
return inline_sampler ? inline_sampler : info.ReadUd<AmdGpu::Sampler>(sgpr_base, dword_offset);
|
||||
}
|
||||
|
||||
} // namespace Shader
|
||||
|
||||
template <>
|
||||
struct fmt::formatter<Shader::Stage> {
|
||||
constexpr auto parse(format_parse_context& ctx) {
|
||||
return ctx.begin();
|
||||
}
|
||||
auto format(const Shader::Stage stage, format_context& ctx) const {
|
||||
constexpr static std::array names = {"fs", "vs", "gs", "es", "hs", "ls", "cs"};
|
||||
return fmt::format_to(ctx.out(), "{}", names[static_cast<size_t>(stage)]);
|
||||
}
|
||||
};
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue