Merge pull request #3208 from ReinUsesLisp/vk-shader-decompiler
vk_shader_decompiler: Add tessellation and misc changes
This commit is contained in:
commit
6edadef96d
11 changed files with 1737 additions and 750 deletions
2
externals/sirit
vendored
2
externals/sirit
vendored
|
@ -1 +1 @@
|
||||||
Subproject commit f7c4b07a7e14edb1dcd93bc9879c823423705c2e
|
Subproject commit e1a6729df7f11e33f6dc0939b18995a57c8bf3d8
|
|
@ -98,10 +98,11 @@ union Attribute {
|
||||||
BitField<20, 10, u64> immediate;
|
BitField<20, 10, u64> immediate;
|
||||||
BitField<22, 2, u64> element;
|
BitField<22, 2, u64> element;
|
||||||
BitField<24, 6, Index> index;
|
BitField<24, 6, Index> index;
|
||||||
|
BitField<31, 1, u64> patch;
|
||||||
BitField<47, 3, AttributeSize> size;
|
BitField<47, 3, AttributeSize> size;
|
||||||
|
|
||||||
bool IsPhysical() const {
|
bool IsPhysical() const {
|
||||||
return element == 0 && static_cast<u64>(index.Value()) == 0;
|
return patch == 0 && element == 0 && static_cast<u64>(index.Value()) == 0;
|
||||||
}
|
}
|
||||||
} fmt20;
|
} fmt20;
|
||||||
|
|
||||||
|
|
|
@ -1915,6 +1915,10 @@ private:
|
||||||
return {};
|
return {};
|
||||||
}
|
}
|
||||||
|
|
||||||
|
Expression InvocationId(Operation operation) {
|
||||||
|
return {"gl_InvocationID", Type::Int};
|
||||||
|
}
|
||||||
|
|
||||||
Expression YNegate(Operation operation) {
|
Expression YNegate(Operation operation) {
|
||||||
return {"y_direction", Type::Float};
|
return {"y_direction", Type::Float};
|
||||||
}
|
}
|
||||||
|
@ -2153,6 +2157,7 @@ private:
|
||||||
&GLSLDecompiler::EmitVertex,
|
&GLSLDecompiler::EmitVertex,
|
||||||
&GLSLDecompiler::EndPrimitive,
|
&GLSLDecompiler::EndPrimitive,
|
||||||
|
|
||||||
|
&GLSLDecompiler::InvocationId,
|
||||||
&GLSLDecompiler::YNegate,
|
&GLSLDecompiler::YNegate,
|
||||||
&GLSLDecompiler::LocalInvocationId<0>,
|
&GLSLDecompiler::LocalInvocationId<0>,
|
||||||
&GLSLDecompiler::LocalInvocationId<1>,
|
&GLSLDecompiler::LocalInvocationId<1>,
|
||||||
|
|
File diff suppressed because it is too large
Load diff
|
@ -5,29 +5,28 @@
|
||||||
#pragma once
|
#pragma once
|
||||||
|
|
||||||
#include <array>
|
#include <array>
|
||||||
|
#include <bitset>
|
||||||
#include <memory>
|
#include <memory>
|
||||||
#include <set>
|
#include <set>
|
||||||
|
#include <type_traits>
|
||||||
#include <utility>
|
#include <utility>
|
||||||
#include <vector>
|
#include <vector>
|
||||||
|
|
||||||
#include <sirit/sirit.h>
|
|
||||||
|
|
||||||
#include "common/common_types.h"
|
#include "common/common_types.h"
|
||||||
#include "video_core/engines/maxwell_3d.h"
|
#include "video_core/engines/maxwell_3d.h"
|
||||||
|
#include "video_core/engines/shader_type.h"
|
||||||
#include "video_core/shader/shader_ir.h"
|
#include "video_core/shader/shader_ir.h"
|
||||||
|
|
||||||
namespace VideoCommon::Shader {
|
|
||||||
class ShaderIR;
|
|
||||||
}
|
|
||||||
|
|
||||||
namespace Vulkan {
|
namespace Vulkan {
|
||||||
class VKDevice;
|
class VKDevice;
|
||||||
}
|
}
|
||||||
|
|
||||||
namespace Vulkan::VKShader {
|
namespace Vulkan {
|
||||||
|
|
||||||
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
|
using Maxwell = Tegra::Engines::Maxwell3D::Regs;
|
||||||
|
using TexelBufferEntry = VideoCommon::Shader::Sampler;
|
||||||
using SamplerEntry = VideoCommon::Shader::Sampler;
|
using SamplerEntry = VideoCommon::Shader::Sampler;
|
||||||
|
using ImageEntry = VideoCommon::Shader::Image;
|
||||||
|
|
||||||
constexpr u32 DESCRIPTOR_SET = 0;
|
constexpr u32 DESCRIPTOR_SET = 0;
|
||||||
|
|
||||||
|
@ -46,39 +45,74 @@ private:
|
||||||
|
|
||||||
class GlobalBufferEntry {
|
class GlobalBufferEntry {
|
||||||
public:
|
public:
|
||||||
explicit GlobalBufferEntry(u32 cbuf_index, u32 cbuf_offset)
|
constexpr explicit GlobalBufferEntry(u32 cbuf_index, u32 cbuf_offset, bool is_written)
|
||||||
: cbuf_index{cbuf_index}, cbuf_offset{cbuf_offset} {}
|
: cbuf_index{cbuf_index}, cbuf_offset{cbuf_offset}, is_written{is_written} {}
|
||||||
|
|
||||||
u32 GetCbufIndex() const {
|
constexpr u32 GetCbufIndex() const {
|
||||||
return cbuf_index;
|
return cbuf_index;
|
||||||
}
|
}
|
||||||
|
|
||||||
u32 GetCbufOffset() const {
|
constexpr u32 GetCbufOffset() const {
|
||||||
return cbuf_offset;
|
return cbuf_offset;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
constexpr bool IsWritten() const {
|
||||||
|
return is_written;
|
||||||
|
}
|
||||||
|
|
||||||
private:
|
private:
|
||||||
u32 cbuf_index{};
|
u32 cbuf_index{};
|
||||||
u32 cbuf_offset{};
|
u32 cbuf_offset{};
|
||||||
|
bool is_written{};
|
||||||
};
|
};
|
||||||
|
|
||||||
struct ShaderEntries {
|
struct ShaderEntries {
|
||||||
u32 const_buffers_base_binding{};
|
u32 NumBindings() const {
|
||||||
u32 global_buffers_base_binding{};
|
return static_cast<u32>(const_buffers.size() + global_buffers.size() +
|
||||||
u32 samplers_base_binding{};
|
texel_buffers.size() + samplers.size() + images.size());
|
||||||
|
}
|
||||||
|
|
||||||
std::vector<ConstBufferEntry> const_buffers;
|
std::vector<ConstBufferEntry> const_buffers;
|
||||||
std::vector<GlobalBufferEntry> global_buffers;
|
std::vector<GlobalBufferEntry> global_buffers;
|
||||||
|
std::vector<TexelBufferEntry> texel_buffers;
|
||||||
std::vector<SamplerEntry> samplers;
|
std::vector<SamplerEntry> samplers;
|
||||||
|
std::vector<ImageEntry> images;
|
||||||
std::set<u32> attributes;
|
std::set<u32> attributes;
|
||||||
std::array<bool, Maxwell::NumClipDistances> clip_distances{};
|
std::array<bool, Maxwell::NumClipDistances> clip_distances{};
|
||||||
std::size_t shader_length{};
|
std::size_t shader_length{};
|
||||||
Sirit::Id entry_function{};
|
bool uses_warps{};
|
||||||
std::vector<Sirit::Id> interfaces;
|
|
||||||
};
|
};
|
||||||
|
|
||||||
using DecompilerResult = std::pair<std::unique_ptr<Sirit::Module>, ShaderEntries>;
|
struct Specialization final {
|
||||||
|
u32 base_binding{};
|
||||||
|
|
||||||
DecompilerResult Decompile(const VKDevice& device, const VideoCommon::Shader::ShaderIR& ir,
|
// Compute specific
|
||||||
Tegra::Engines::ShaderType stage);
|
std::array<u32, 3> workgroup_size{};
|
||||||
|
u32 shared_memory_size{};
|
||||||
|
|
||||||
} // namespace Vulkan::VKShader
|
// Graphics specific
|
||||||
|
Maxwell::PrimitiveTopology primitive_topology{};
|
||||||
|
std::optional<float> point_size{};
|
||||||
|
std::array<Maxwell::VertexAttribute::Type, Maxwell::NumVertexAttributes> attribute_types{};
|
||||||
|
|
||||||
|
// Tessellation specific
|
||||||
|
struct {
|
||||||
|
Maxwell::TessellationPrimitive primitive{};
|
||||||
|
Maxwell::TessellationSpacing spacing{};
|
||||||
|
bool clockwise{};
|
||||||
|
} tessellation;
|
||||||
|
};
|
||||||
|
// Old gcc versions don't consider this trivially copyable.
|
||||||
|
// static_assert(std::is_trivially_copyable_v<Specialization>);
|
||||||
|
|
||||||
|
struct SPIRVShader {
|
||||||
|
std::vector<u32> code;
|
||||||
|
ShaderEntries entries;
|
||||||
|
};
|
||||||
|
|
||||||
|
ShaderEntries GenerateShaderEntries(const VideoCommon::Shader::ShaderIR& ir);
|
||||||
|
|
||||||
|
std::vector<u32> Decompile(const VKDevice& device, const VideoCommon::Shader::ShaderIR& ir,
|
||||||
|
Tegra::Engines::ShaderType stage, const Specialization& specialization);
|
||||||
|
|
||||||
|
} // namespace Vulkan
|
||||||
|
|
|
@ -21,6 +21,7 @@ using Tegra::Shader::OpCode;
|
||||||
using Tegra::Shader::Register;
|
using Tegra::Shader::Register;
|
||||||
|
|
||||||
namespace {
|
namespace {
|
||||||
|
|
||||||
u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) {
|
u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) {
|
||||||
switch (uniform_type) {
|
switch (uniform_type) {
|
||||||
case Tegra::Shader::UniformType::Single:
|
case Tegra::Shader::UniformType::Single:
|
||||||
|
@ -35,6 +36,7 @@ u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) {
|
||||||
return 1;
|
return 1;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
} // Anonymous namespace
|
} // Anonymous namespace
|
||||||
|
|
||||||
u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
|
u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
|
||||||
|
@ -196,28 +198,28 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
|
||||||
UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0,
|
UNIMPLEMENTED_IF_MSG((instr.attribute.fmt20.immediate.Value() % sizeof(u32)) != 0,
|
||||||
"Unaligned attribute loads are not supported");
|
"Unaligned attribute loads are not supported");
|
||||||
|
|
||||||
u64 next_element = instr.attribute.fmt20.element;
|
u64 element = instr.attribute.fmt20.element;
|
||||||
auto next_index = static_cast<u64>(instr.attribute.fmt20.index.Value());
|
auto index = static_cast<u64>(instr.attribute.fmt20.index.Value());
|
||||||
|
|
||||||
const auto StoreNextElement = [&](u32 reg_offset) {
|
const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1;
|
||||||
const auto dest = GetOutputAttribute(static_cast<Attribute::Index>(next_index),
|
for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) {
|
||||||
next_element, GetRegister(instr.gpr39));
|
Node dest;
|
||||||
|
if (instr.attribute.fmt20.patch) {
|
||||||
|
const u32 offset = static_cast<u32>(index) * 4 + static_cast<u32>(element);
|
||||||
|
dest = MakeNode<PatchNode>(offset);
|
||||||
|
} else {
|
||||||
|
dest = GetOutputAttribute(static_cast<Attribute::Index>(index), element,
|
||||||
|
GetRegister(instr.gpr39));
|
||||||
|
}
|
||||||
const auto src = GetRegister(instr.gpr0.Value() + reg_offset);
|
const auto src = GetRegister(instr.gpr0.Value() + reg_offset);
|
||||||
|
|
||||||
bb.push_back(Operation(OperationCode::Assign, dest, src));
|
bb.push_back(Operation(OperationCode::Assign, dest, src));
|
||||||
|
|
||||||
// Load the next attribute element into the following register. If the element
|
// Load the next attribute element into the following register. If the element to load
|
||||||
// to load goes beyond the vec4 size, load the first element of the next
|
// goes beyond the vec4 size, load the first element of the next attribute.
|
||||||
// attribute.
|
element = (element + 1) % 4;
|
||||||
next_element = (next_element + 1) % 4;
|
index = index + (element == 0 ? 1 : 0);
|
||||||
next_index = next_index + (next_element == 0 ? 1 : 0);
|
|
||||||
};
|
|
||||||
|
|
||||||
const u32 num_words = static_cast<u32>(instr.attribute.fmt20.size.Value()) + 1;
|
|
||||||
for (u32 reg_offset = 0; reg_offset < num_words; ++reg_offset) {
|
|
||||||
StoreNextElement(reg_offset);
|
|
||||||
}
|
}
|
||||||
|
|
||||||
break;
|
break;
|
||||||
}
|
}
|
||||||
case OpCode::Id::ST_L:
|
case OpCode::Id::ST_L:
|
||||||
|
|
|
@ -69,6 +69,8 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
|
||||||
case OpCode::Id::MOV_SYS: {
|
case OpCode::Id::MOV_SYS: {
|
||||||
const Node value = [this, instr] {
|
const Node value = [this, instr] {
|
||||||
switch (instr.sys20) {
|
switch (instr.sys20) {
|
||||||
|
case SystemVariable::InvocationId:
|
||||||
|
return Operation(OperationCode::InvocationId);
|
||||||
case SystemVariable::Ydirection:
|
case SystemVariable::Ydirection:
|
||||||
return Operation(OperationCode::YNegate);
|
return Operation(OperationCode::YNegate);
|
||||||
case SystemVariable::InvocationInfo:
|
case SystemVariable::InvocationInfo:
|
||||||
|
|
|
@ -38,6 +38,9 @@ u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
|
||||||
const Instruction instr = {program_code[pc]};
|
const Instruction instr = {program_code[pc]};
|
||||||
const auto opcode = OpCode::Decode(instr);
|
const auto opcode = OpCode::Decode(instr);
|
||||||
|
|
||||||
|
// Signal the backend that this shader uses warp instructions.
|
||||||
|
uses_warps = true;
|
||||||
|
|
||||||
switch (opcode->get().GetId()) {
|
switch (opcode->get().GetId()) {
|
||||||
case OpCode::Id::VOTE: {
|
case OpCode::Id::VOTE: {
|
||||||
const Node value = GetPredicate(instr.vote.value, instr.vote.negate_value != 0);
|
const Node value = GetPredicate(instr.vote.value, instr.vote.negate_value != 0);
|
||||||
|
|
|
@ -172,6 +172,7 @@ enum class OperationCode {
|
||||||
EmitVertex, /// () -> void
|
EmitVertex, /// () -> void
|
||||||
EndPrimitive, /// () -> void
|
EndPrimitive, /// () -> void
|
||||||
|
|
||||||
|
InvocationId, /// () -> int
|
||||||
YNegate, /// () -> float
|
YNegate, /// () -> float
|
||||||
LocalInvocationIdX, /// () -> uint
|
LocalInvocationIdX, /// () -> uint
|
||||||
LocalInvocationIdY, /// () -> uint
|
LocalInvocationIdY, /// () -> uint
|
||||||
|
@ -213,13 +214,14 @@ class PredicateNode;
|
||||||
class AbufNode;
|
class AbufNode;
|
||||||
class CbufNode;
|
class CbufNode;
|
||||||
class LmemNode;
|
class LmemNode;
|
||||||
|
class PatchNode;
|
||||||
class SmemNode;
|
class SmemNode;
|
||||||
class GmemNode;
|
class GmemNode;
|
||||||
class CommentNode;
|
class CommentNode;
|
||||||
|
|
||||||
using NodeData =
|
using NodeData = std::variant<OperationNode, ConditionalNode, GprNode, ImmediateNode,
|
||||||
std::variant<OperationNode, ConditionalNode, GprNode, ImmediateNode, InternalFlagNode,
|
InternalFlagNode, PredicateNode, AbufNode, PatchNode, CbufNode,
|
||||||
PredicateNode, AbufNode, CbufNode, LmemNode, SmemNode, GmemNode, CommentNode>;
|
LmemNode, SmemNode, GmemNode, CommentNode>;
|
||||||
using Node = std::shared_ptr<NodeData>;
|
using Node = std::shared_ptr<NodeData>;
|
||||||
using Node4 = std::array<Node, 4>;
|
using Node4 = std::array<Node, 4>;
|
||||||
using NodeBlock = std::vector<Node>;
|
using NodeBlock = std::vector<Node>;
|
||||||
|
@ -542,6 +544,19 @@ private:
|
||||||
u32 element{};
|
u32 element{};
|
||||||
};
|
};
|
||||||
|
|
||||||
|
/// Patch memory (used to communicate tessellation stages).
|
||||||
|
class PatchNode final {
|
||||||
|
public:
|
||||||
|
explicit PatchNode(u32 offset) : offset{offset} {}
|
||||||
|
|
||||||
|
u32 GetOffset() const {
|
||||||
|
return offset;
|
||||||
|
}
|
||||||
|
|
||||||
|
private:
|
||||||
|
u32 offset{};
|
||||||
|
};
|
||||||
|
|
||||||
/// Constant buffer node, usually mapped to uniform buffers in GLSL
|
/// Constant buffer node, usually mapped to uniform buffers in GLSL
|
||||||
class CbufNode final {
|
class CbufNode final {
|
||||||
public:
|
public:
|
||||||
|
|
|
@ -137,6 +137,10 @@ public:
|
||||||
return uses_vertex_id;
|
return uses_vertex_id;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
bool UsesWarps() const {
|
||||||
|
return uses_warps;
|
||||||
|
}
|
||||||
|
|
||||||
bool HasPhysicalAttributes() const {
|
bool HasPhysicalAttributes() const {
|
||||||
return uses_physical_attributes;
|
return uses_physical_attributes;
|
||||||
}
|
}
|
||||||
|
@ -415,6 +419,7 @@ private:
|
||||||
bool uses_physical_attributes{}; // Shader uses AL2P or physical attribute read/writes
|
bool uses_physical_attributes{}; // Shader uses AL2P or physical attribute read/writes
|
||||||
bool uses_instance_id{};
|
bool uses_instance_id{};
|
||||||
bool uses_vertex_id{};
|
bool uses_vertex_id{};
|
||||||
|
bool uses_warps{};
|
||||||
|
|
||||||
Tegra::Shader::Header header;
|
Tegra::Shader::Header header;
|
||||||
};
|
};
|
||||||
|
|
|
@ -7,6 +7,7 @@
|
||||||
#include <variant>
|
#include <variant>
|
||||||
|
|
||||||
#include "common/common_types.h"
|
#include "common/common_types.h"
|
||||||
|
#include "video_core/shader/node.h"
|
||||||
#include "video_core/shader/shader_ir.h"
|
#include "video_core/shader/shader_ir.h"
|
||||||
|
|
||||||
namespace VideoCommon::Shader {
|
namespace VideoCommon::Shader {
|
||||||
|
|
Loading…
Reference in a new issue