forked from ShuriZma/suyu
1
0
Fork 0

Shader_IR: allow lookup of texture samplers within the shader_ir for instructions that don't provide it

This commit is contained in:
Fernando Sahmkow 2019-09-25 09:53:18 -04:00 committed by FernandoS27
parent 8909f52166
commit 33fcec3502
9 changed files with 363 additions and 46 deletions

View File

@ -4,7 +4,10 @@
#pragma once
#include "common/bit_field.h"
#include "common/common_types.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/textures/texture.h"
namespace Tegra::Engines {
@ -17,10 +20,100 @@ enum class ShaderType : u32 {
Compute = 5,
};
struct SamplerDescriptor {
union {
BitField<0, 20, Tegra::Shader::TextureType> texture_type;
BitField<20, 1, u32> is_array;
BitField<21, 1, u32> is_buffer;
BitField<22, 1, u32> is_shadow;
u32 raw{};
};
static SamplerDescriptor FromTicTexture(Tegra::Texture::TextureType tic_texture_type) {
SamplerDescriptor result{};
switch (tic_texture_type) {
case Tegra::Texture::TextureType::Texture1D: {
result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::Texture2D: {
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::Texture3D: {
result.texture_type.Assign(Tegra::Shader::TextureType::Texture3D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::TextureCubemap: {
result.texture_type.Assign(Tegra::Shader::TextureType::TextureCube);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::Texture1DArray: {
result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
result.is_array.Assign(1);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::Texture2DArray: {
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.is_array.Assign(1);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::Texture1DBuffer: {
result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
result.is_array.Assign(0);
result.is_buffer.Assign(1);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::Texture2DNoMipmap: {
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::TextureCubeArray: {
result.texture_type.Assign(Tegra::Shader::TextureType::TextureCube);
result.is_array.Assign(1);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
default: {
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.is_array.Assign(0);
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
}
}
}
};
class ConstBufferEngineInterface {
public:
virtual ~ConstBufferEngineInterface() {}
virtual u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const = 0;
virtual SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const = 0;
virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const = 0;
virtual u32 GetBoundBuffer() const = 0;
};
}
} // namespace Tegra::Engines

View File

@ -78,6 +78,24 @@ u32 KeplerCompute::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 o
return result;
}
SamplerDescriptor KeplerCompute::AccessBoundSampler(ShaderType stage, u64 offset) const {
return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle));
}
SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const {
ASSERT(stage == ShaderType::Compute);
const auto& tex_info_buffer = launch_description.const_buffer_config[const_buffer];
const GPUVAddr tex_info_address =
tex_info_buffer.Address() + offset;
const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle, offset);
SamplerDescriptor result = SamplerDescriptor::FromTicTexture(tex_info.tic.texture_type.Value());
result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
return result;
}
void KeplerCompute::ProcessLaunch() {
const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address();
memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description,

View File

@ -10,8 +10,8 @@
#include "common/bit_field.h"
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "video_core/engines/engine_upload.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/engine_upload.h"
#include "video_core/gpu.h"
#include "video_core/textures/texture.h"
@ -38,7 +38,7 @@ namespace Tegra::Engines {
#define KEPLER_COMPUTE_REG_INDEX(field_name) \
(offsetof(Tegra::Engines::KeplerCompute::Regs, field_name) / sizeof(u32))
class KeplerCompute final : public ConstBufferEngineInterface {
class KeplerCompute final : public ConstBufferEngineInterface {
public:
explicit KeplerCompute(Core::System& system, VideoCore::RasterizerInterface& rasterizer,
MemoryManager& memory_manager);
@ -204,6 +204,15 @@ public:
u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const override;
u32 GetBoundBuffer() const override {
return regs.tex_cb_index;
}
private:
Core::System& system;
VideoCore::RasterizerInterface& rasterizer;

View File

@ -856,4 +856,22 @@ u32 Maxwell3D::AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offse
return result;
}
SamplerDescriptor Maxwell3D::AccessBoundSampler(ShaderType stage, u64 offset) const {
return AccessBindlessSampler(stage, regs.tex_cb_index, offset * sizeof(Texture::TextureHandle));
}
SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const {
ASSERT(stage != ShaderType::Compute);
const auto& shader = state.shader_stages[static_cast<std::size_t>(stage)];
const auto& tex_info_buffer = shader.const_buffers[const_buffer];
const GPUVAddr tex_info_address = tex_info_buffer.address + offset;
const Texture::TextureHandle tex_handle{memory_manager.Read<u32>(tex_info_address)};
const Texture::FullTextureInfo tex_info = GetTextureInfo(tex_handle, offset);
SamplerDescriptor result = SamplerDescriptor::FromTicTexture(tex_info.tic.texture_type.Value());
result.is_shadow.Assign(tex_info.tsc.depth_compare_enabled.Value());
return result;
}
} // namespace Tegra::Engines

View File

@ -15,8 +15,8 @@
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/math_util.h"
#include "video_core/engines/const_buffer_info.h"
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/const_buffer_info.h"
#include "video_core/engines/engine_upload.h"
#include "video_core/gpu.h"
#include "video_core/macro_interpreter.h"
@ -1260,6 +1260,15 @@ public:
u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const override;
SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const override;
SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
u64 offset) const override;
u32 GetBoundBuffer() const override {
return regs.tex_cb_index;
}
/// Memory for macro code - it's undetermined how big this is, however 1MB is much larger than
/// we've seen used.
using MacroMemory = std::array<u32, 0x40000>;

View File

@ -27,43 +27,121 @@ void ConstBufferLocker::SetEngine(Tegra::Engines::ConstBufferEngineInterface* en
}
std::optional<u32> ConstBufferLocker::ObtainKey(u32 buffer, u32 offset) {
if (!keys) {
keys = std::make_shared<KeyMap>();
}
auto& key_map = *keys;
const std::pair<u32, u32> key = {buffer, offset};
const auto iter = keys.find(key);
if (iter != keys.end()) {
const auto iter = key_map.find(key);
if (iter != key_map.end()) {
return {iter->second};
}
if (!IsEngineSet()) {
return {};
}
const u32 value = engine->AccessConstBuffer32(shader_stage, buffer, offset);
keys.emplace(key, value);
key_map.emplace(key, value);
return {value};
}
std::optional<Tegra::Engines::SamplerDescriptor> ConstBufferLocker::ObtainBoundSampler(u32 offset) {
if (!bound_samplers) {
bound_samplers = std::make_shared<BoundSamplerMap>();
}
auto& key_map = *bound_samplers;
const u32 key = offset;
const auto iter = key_map.find(key);
if (iter != key_map.end()) {
return {iter->second};
}
if (!IsEngineSet()) {
return {};
}
const Tegra::Engines::SamplerDescriptor value =
engine->AccessBoundSampler(shader_stage, offset);
key_map.emplace(key, value);
return {value};
}
std::optional<Tegra::Engines::SamplerDescriptor> ConstBufferLocker::ObtainBindlessSampler(
u32 buffer, u32 offset) {
if (!bindless_samplers) {
bindless_samplers = std::make_shared<BindlessSamplerMap>();
}
auto& key_map = *bindless_samplers;
const std::pair<u32, u32> key = {buffer, offset};
const auto iter = key_map.find(key);
if (iter != key_map.end()) {
return {iter->second};
}
if (!IsEngineSet()) {
return {};
}
const Tegra::Engines::SamplerDescriptor value =
engine->AccessBindlessSampler(shader_stage, buffer, offset);
key_map.emplace(key, value);
return {value};
}
void ConstBufferLocker::InsertKey(u32 buffer, u32 offset, u32 value) {
if (!keys) {
keys = std::make_shared<KeyMap>();
}
const std::pair<u32, u32> key = {buffer, offset};
keys[key] = value;
(*keys)[key] = value;
}
u32 ConstBufferLocker::NumKeys() const {
return keys.size();
void ConstBufferLocker::InsertBoundSampler(u32 offset, Tegra::Engines::SamplerDescriptor sampler) {
if (!bound_samplers) {
bound_samplers = std::make_shared<BoundSamplerMap>();
}
(*bound_samplers)[offset] = sampler;
}
const std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>&
ConstBufferLocker::AccessKeys() const {
return keys;
void ConstBufferLocker::InsertBindlessSampler(u32 buffer, u32 offset,
Tegra::Engines::SamplerDescriptor sampler) {
if (!bindless_samplers) {
bindless_samplers = std::make_shared<BindlessSamplerMap>();
}
const std::pair<u32, u32> key = {buffer, offset};
(*bindless_samplers)[key] = sampler;
}
bool ConstBufferLocker::AreKeysConsistant() const {
bool ConstBufferLocker::IsConsistant() const {
if (!IsEngineSet()) {
return false;
}
for (const auto& key_val : keys) {
const std::pair<u32, u32> key = key_val.first;
const u32 value = key_val.second;
const u32 other_value = engine->AccessConstBuffer32(shader_stage, key.first, key.second);
if (other_value != value) {
return false;
if (keys) {
for (const auto& key_val : *keys) {
const std::pair<u32, u32> key = key_val.first;
const u32 value = key_val.second;
const u32 other_value =
engine->AccessConstBuffer32(shader_stage, key.first, key.second);
if (other_value != value) {
return false;
}
}
}
if (bound_samplers) {
for (const auto& sampler_val : *bound_samplers) {
const u32 key = sampler_val.first;
const Tegra::Engines::SamplerDescriptor value = sampler_val.second;
const Tegra::Engines::SamplerDescriptor other_value =
engine->AccessBoundSampler(shader_stage, key);
if (other_value.raw != value.raw) {
return false;
}
}
}
if (bindless_samplers) {
for (const auto& sampler_val : *bindless_samplers) {
const std::pair<u32, u32> key = sampler_val.first;
const Tegra::Engines::SamplerDescriptor value = sampler_val.second;
const Tegra::Engines::SamplerDescriptor other_value =
engine->AccessBindlessSampler(shader_stage, key.first, key.second);
if (other_value.raw != value.raw) {
return false;
}
}
}
return true;

View File

@ -11,6 +11,11 @@
namespace VideoCommon::Shader {
using KeyMap = std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>;
using BoundSamplerMap = std::unordered_map<u32, Tegra::Engines::SamplerDescriptor>;
using BindlessSamplerMap =
std::unordered_map<std::pair<u32, u32>, Tegra::Engines::SamplerDescriptor, Common::PairHash>;
class ConstBufferLocker {
public:
explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage);
@ -29,22 +34,67 @@ public:
// registered value, if not it will obtain it from maxwell3d and register it.
std::optional<u32> ObtainKey(u32 buffer, u32 offset);
std::optional<Tegra::Engines::SamplerDescriptor> ObtainBoundSampler(u32 offset);
std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset);
// Manually inserts a key.
void InsertKey(u32 buffer, u32 offset, u32 value);
void InsertBoundSampler(u32 offset, Tegra::Engines::SamplerDescriptor sampler);
void InsertBindlessSampler(u32 buffer, u32 offset, Tegra::Engines::SamplerDescriptor sampler);
// Retrieves the number of keys registered.
u32 NumKeys() const;
std::size_t NumKeys() const {
if (!keys) {
return 0;
}
return keys->size();
}
std::size_t NumBoundSamplers() const {
if (!bound_samplers) {
return 0;
}
return bound_samplers->size();
}
std::size_t NumBindlessSamplers() const {
if (!bindless_samplers) {
return 0;
}
return bindless_samplers->size();
}
// Gives an accessor to the key's database.
const std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash>& AccessKeys() const;
// Pre: NumKeys > 0
const KeyMap& AccessKeys() const {
return *keys;
}
// Checks keys against maxwell3d's current const buffers. Returns true if they
// Gives an accessor to the sampler's database.
// Pre: NumBindlessSamplers > 0
const BoundSamplerMap& AccessBoundSamplers() const {
return *bound_samplers;
}
// Gives an accessor to the sampler's database.
// Pre: NumBindlessSamplers > 0
const BindlessSamplerMap& AccessBindlessSamplers() const {
return *bindless_samplers;
}
// Checks keys & samplers against engine's current const buffers. Returns true if they
// are the same value, false otherwise;
bool AreKeysConsistant() const;
bool IsConsistant() const;
private:
Tegra::Engines::ConstBufferEngineInterface* engine;
Tegra::Engines::ShaderType shader_stage;
std::unordered_map<std::pair<u32, u32>, u32, Common::PairHash> keys{};
// All containers are lazy initialized as most shaders don't use them.
std::shared_ptr<KeyMap> keys{};
std::shared_ptr<BoundSamplerMap> bound_samplers{};
std::shared_ptr<BindlessSamplerMap> bindless_samplers{};
};
} // namespace VideoCommon::Shader

View File

@ -141,7 +141,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
const Node component = Immediate(static_cast<u32>(instr.tld4s.component));
const auto& sampler =
GetSampler(instr.sampler, TextureType::Texture2D, false, depth_compare);
GetSampler(instr.sampler, {{TextureType::Texture2D, false, depth_compare}});
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
@ -165,10 +165,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
// Sadly, not all texture instructions specify the type of texture their sampler
// uses. This must be fixed at a later instance.
const auto& sampler =
is_bindless
? GetBindlessSampler(instr.gpr8, Tegra::Shader::TextureType::Texture2D, false,
false)
: GetSampler(instr.sampler, Tegra::Shader::TextureType::Texture2D, false, false);
is_bindless ? GetBindlessSampler(instr.gpr8, {}) : GetSampler(instr.sampler, {});
u32 indexer = 0;
switch (instr.txq.query_type) {
@ -207,9 +204,9 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
auto texture_type = instr.tmml.texture_type.Value();
const bool is_array = instr.tmml.array != 0;
const auto& sampler = is_bindless
? GetBindlessSampler(instr.gpr20, texture_type, is_array, false)
: GetSampler(instr.sampler, texture_type, is_array, false);
const auto& sampler =
is_bindless ? GetBindlessSampler(instr.gpr20, {{texture_type, is_array, false}})
: GetSampler(instr.sampler, {{texture_type, is_array, false}});
std::vector<Node> coords;
@ -285,10 +282,30 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
return pc;
}
const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, TextureType type,
bool is_array, bool is_shadow) {
const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
std::optional<SamplerInfo> sampler_info) {
const auto offset = static_cast<std::size_t>(sampler.index.Value());
Tegra::Shader::TextureType type;
bool is_array;
bool is_shadow;
if (sampler_info) {
type = sampler_info->type;
is_array = sampler_info->is_array;
is_shadow = sampler_info->is_shadow;
} else {
auto sampler = locker.ObtainBoundSampler(offset);
if (sampler) {
type = sampler->texture_type.Value();
is_array = sampler->is_array.Value() != 0;
is_shadow = sampler->is_shadow.Value() != 0;
} else {
type = Tegra::Shader::TextureType::Texture2D;
is_array = false;
is_shadow = false;
}
}
// If this sampler has already been used, return the existing mapping.
const auto itr =
std::find_if(used_samplers.begin(), used_samplers.end(),
@ -305,13 +322,32 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, Textu
return *used_samplers.emplace(entry).first;
}
const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg, TextureType type,
bool is_array, bool is_shadow) {
const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg,
std::optional<SamplerInfo> sampler_info) {
const Node sampler_register = GetRegister(reg);
const auto [base_sampler, cbuf_index, cbuf_offset] =
TrackCbuf(sampler_register, global_code, static_cast<s64>(global_code.size()));
ASSERT(base_sampler != nullptr);
const auto cbuf_key = (static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset);
Tegra::Shader::TextureType type;
bool is_array;
bool is_shadow;
if (sampler_info) {
type = sampler_info->type;
is_array = sampler_info->is_array;
is_shadow = sampler_info->is_shadow;
} else {
auto sampler = locker.ObtainBindlessSampler(cbuf_index, cbuf_offset);
if (sampler) {
type = sampler->texture_type.Value();
is_array = sampler->is_array.Value() != 0;
is_shadow = sampler->is_shadow.Value() != 0;
} else {
type = Tegra::Shader::TextureType::Texture2D;
is_array = false;
is_shadow = false;
}
}
// If this sampler has already been used, return the existing mapping.
const auto itr =
@ -411,9 +447,9 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
(texture_type == TextureType::TextureCube && is_array && is_shadow),
"This method is not supported.");
const auto& sampler = is_bindless
? GetBindlessSampler(*bindless_reg, texture_type, is_array, is_shadow)
: GetSampler(instr.sampler, texture_type, is_array, is_shadow);
const auto& sampler =
is_bindless ? GetBindlessSampler(*bindless_reg, {{texture_type, is_array, is_shadow}})
: GetSampler(instr.sampler, {{texture_type, is_array, is_shadow}});
const bool lod_needed = process_mode == TextureProcessMode::LZ ||
process_mode == TextureProcessMode::LL ||
@ -577,7 +613,7 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
dc = GetRegister(parameter_register++);
}
const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, depth_compare);
const auto& sampler = GetSampler(instr.sampler, {{texture_type, is_array, depth_compare}});
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
@ -610,7 +646,7 @@ Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
// const Node aoffi_register{is_aoffi ? GetRegister(gpr20_cursor++) : nullptr};
// const Node multisample{is_multisample ? GetRegister(gpr20_cursor++) : nullptr};
const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, false);
const auto& sampler = GetSampler(instr.sampler, {{texture_type, is_array, false}});
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {
@ -646,7 +682,7 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is
// When lod is used always is in gpr20
const Node lod = lod_enabled ? GetRegister(instr.gpr20) : Immediate(0);
const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, false);
const auto& sampler = GetSampler(instr.sampler, {{texture_type, is_array, false}});
Node4 values;
for (u32 element = 0; element < values.size(); ++element) {

View File

@ -173,6 +173,13 @@ public:
private:
friend class ASTDecoder;
struct SamplerInfo {
Tegra::Shader::TextureType type;
bool is_array;
bool is_shadow;
};
void Decode();
NodeBlock DecodeRange(u32 begin, u32 end);
@ -297,12 +304,11 @@ private:
/// Accesses a texture sampler
const Sampler& GetSampler(const Tegra::Shader::Sampler& sampler,
Tegra::Shader::TextureType type, bool is_array, bool is_shadow);
std::optional<SamplerInfo> sampler_info);
// Accesses a texture sampler for a bindless texture.
const Sampler& GetBindlessSampler(const Tegra::Shader::Register& reg,
Tegra::Shader::TextureType type, bool is_array,
bool is_shadow);
std::optional<SamplerInfo> sampler_info);
/// Accesses an image.
Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type);