From e1b3be7ced2f6c036dce456de16886451b205791 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Thu, 6 Jun 2019 19:31:14 -0300 Subject: [PATCH 1/3] shader: Move Node declarations out of the shader IR header Analysis passes do not have a good reason to depend on shader_ir.h to work on top of nodes. This splits node-related declarations to their own file and leaves the IR in shader_ir.h --- src/video_core/CMakeLists.txt | 1 + src/video_core/shader/node.h | 510 ++++++++++++++++++++++++++++ src/video_core/shader/node_helper.h | 7 +- src/video_core/shader/shader_ir.h | 493 +-------------------------- 4 files changed, 518 insertions(+), 493 deletions(-) create mode 100644 src/video_core/shader/node.h diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt index fa80c3814..24ce47682 100644 --- a/src/video_core/CMakeLists.txt +++ b/src/video_core/CMakeLists.txt @@ -104,6 +104,7 @@ add_library(video_core STATIC shader/decode.cpp shader/node_helper.cpp shader/node_helper.h + shader/node.h shader/shader_ir.cpp shader/shader_ir.h shader/track.cpp diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h new file mode 100644 index 000000000..1e7a5f699 --- /dev/null +++ b/src/video_core/shader/node.h @@ -0,0 +1,510 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include +#include +#include +#include +#include +#include +#include +#include + +#include "common/common_types.h" +#include "video_core/engines/shader_bytecode.h" + +namespace VideoCommon::Shader { + +class OperationNode; +class ConditionalNode; +class GprNode; +class ImmediateNode; +class InternalFlagNode; +class PredicateNode; +class AbufNode; ///< Attribute buffer +class CbufNode; ///< Constant buffer +class LmemNode; ///< Local memory +class GmemNode; ///< Global memory +class CommentNode; + +using NodeData = + std::variant; +using Node = std::shared_ptr; +using Node4 = std::array; +using NodeBlock = std::vector; + +enum class OperationCode { + Assign, /// (float& dest, float src) -> void + + Select, /// (MetaArithmetic, bool pred, float a, float b) -> float + + FAdd, /// (MetaArithmetic, float a, float b) -> float + FMul, /// (MetaArithmetic, float a, float b) -> float + FDiv, /// (MetaArithmetic, float a, float b) -> float + FFma, /// (MetaArithmetic, float a, float b, float c) -> float + FNegate, /// (MetaArithmetic, float a) -> float + FAbsolute, /// (MetaArithmetic, float a) -> float + FClamp, /// (MetaArithmetic, float value, float min, float max) -> float + FMin, /// (MetaArithmetic, float a, float b) -> float + FMax, /// (MetaArithmetic, float a, float b) -> float + FCos, /// (MetaArithmetic, float a) -> float + FSin, /// (MetaArithmetic, float a) -> float + FExp2, /// (MetaArithmetic, float a) -> float + FLog2, /// (MetaArithmetic, float a) -> float + FInverseSqrt, /// (MetaArithmetic, float a) -> float + FSqrt, /// (MetaArithmetic, float a) -> float + FRoundEven, /// (MetaArithmetic, float a) -> float + FFloor, /// (MetaArithmetic, float a) -> float + FCeil, /// (MetaArithmetic, float a) -> float + FTrunc, /// (MetaArithmetic, float a) -> float + FCastInteger, /// (MetaArithmetic, int a) -> float + FCastUInteger, /// (MetaArithmetic, uint a) -> float + + IAdd, /// (MetaArithmetic, int a, int b) -> int + IMul, /// (MetaArithmetic, int a, int b) -> int + IDiv, /// (MetaArithmetic, int a, int b) -> int + INegate, /// (MetaArithmetic, int a) -> int + IAbsolute, /// (MetaArithmetic, int a) -> int + IMin, /// (MetaArithmetic, int a, int b) -> int + IMax, /// (MetaArithmetic, int a, int b) -> int + ICastFloat, /// (MetaArithmetic, float a) -> int + ICastUnsigned, /// (MetaArithmetic, uint a) -> int + ILogicalShiftLeft, /// (MetaArithmetic, int a, uint b) -> int + ILogicalShiftRight, /// (MetaArithmetic, int a, uint b) -> int + IArithmeticShiftRight, /// (MetaArithmetic, int a, uint b) -> int + IBitwiseAnd, /// (MetaArithmetic, int a, int b) -> int + IBitwiseOr, /// (MetaArithmetic, int a, int b) -> int + IBitwiseXor, /// (MetaArithmetic, int a, int b) -> int + IBitwiseNot, /// (MetaArithmetic, int a) -> int + IBitfieldInsert, /// (MetaArithmetic, int base, int insert, int offset, int bits) -> int + IBitfieldExtract, /// (MetaArithmetic, int value, int offset, int offset) -> int + IBitCount, /// (MetaArithmetic, int) -> int + + UAdd, /// (MetaArithmetic, uint a, uint b) -> uint + UMul, /// (MetaArithmetic, uint a, uint b) -> uint + UDiv, /// (MetaArithmetic, uint a, uint b) -> uint + UMin, /// (MetaArithmetic, uint a, uint b) -> uint + UMax, /// (MetaArithmetic, uint a, uint b) -> uint + UCastFloat, /// (MetaArithmetic, float a) -> uint + UCastSigned, /// (MetaArithmetic, int a) -> uint + ULogicalShiftLeft, /// (MetaArithmetic, uint a, uint b) -> uint + ULogicalShiftRight, /// (MetaArithmetic, uint a, uint b) -> uint + UArithmeticShiftRight, /// (MetaArithmetic, uint a, uint b) -> uint + UBitwiseAnd, /// (MetaArithmetic, uint a, uint b) -> uint + UBitwiseOr, /// (MetaArithmetic, uint a, uint b) -> uint + UBitwiseXor, /// (MetaArithmetic, uint a, uint b) -> uint + UBitwiseNot, /// (MetaArithmetic, uint a) -> uint + UBitfieldInsert, /// (MetaArithmetic, uint base, uint insert, int offset, int bits) -> uint + UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint + UBitCount, /// (MetaArithmetic, uint) -> uint + + HAdd, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 + HMul, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 + HFma, /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2 + HAbsolute, /// (f16vec2 a) -> f16vec2 + HNegate, /// (f16vec2 a, bool first, bool second) -> f16vec2 + HClamp, /// (f16vec2 src, float min, float max) -> f16vec2 + HUnpack, /// (Tegra::Shader::HalfType, T value) -> f16vec2 + HMergeF32, /// (f16vec2 src) -> float + HMergeH0, /// (f16vec2 dest, f16vec2 src) -> f16vec2 + HMergeH1, /// (f16vec2 dest, f16vec2 src) -> f16vec2 + HPack2, /// (float a, float b) -> f16vec2 + + LogicalAssign, /// (bool& dst, bool src) -> void + LogicalAnd, /// (bool a, bool b) -> bool + LogicalOr, /// (bool a, bool b) -> bool + LogicalXor, /// (bool a, bool b) -> bool + LogicalNegate, /// (bool a) -> bool + LogicalPick2, /// (bool2 pair, uint index) -> bool + LogicalAll2, /// (bool2 a) -> bool + LogicalAny2, /// (bool2 a) -> bool + + LogicalFLessThan, /// (float a, float b) -> bool + LogicalFEqual, /// (float a, float b) -> bool + LogicalFLessEqual, /// (float a, float b) -> bool + LogicalFGreaterThan, /// (float a, float b) -> bool + LogicalFNotEqual, /// (float a, float b) -> bool + LogicalFGreaterEqual, /// (float a, float b) -> bool + LogicalFIsNan, /// (float a) -> bool + + LogicalILessThan, /// (int a, int b) -> bool + LogicalIEqual, /// (int a, int b) -> bool + LogicalILessEqual, /// (int a, int b) -> bool + LogicalIGreaterThan, /// (int a, int b) -> bool + LogicalINotEqual, /// (int a, int b) -> bool + LogicalIGreaterEqual, /// (int a, int b) -> bool + + LogicalULessThan, /// (uint a, uint b) -> bool + LogicalUEqual, /// (uint a, uint b) -> bool + LogicalULessEqual, /// (uint a, uint b) -> bool + LogicalUGreaterThan, /// (uint a, uint b) -> bool + LogicalUNotEqual, /// (uint a, uint b) -> bool + LogicalUGreaterEqual, /// (uint a, uint b) -> bool + + Logical2HLessThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HLessEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HGreaterThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HNotEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HGreaterEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HLessThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HLessEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HGreaterThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HNotEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HGreaterEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + + Texture, /// (MetaTexture, float[N] coords) -> float4 + TextureLod, /// (MetaTexture, float[N] coords) -> float4 + TextureGather, /// (MetaTexture, float[N] coords) -> float4 + TextureQueryDimensions, /// (MetaTexture, float a) -> float4 + TextureQueryLod, /// (MetaTexture, float[N] coords) -> float4 + TexelFetch, /// (MetaTexture, int[N], int) -> float4 + + Branch, /// (uint branch_target) -> void + PushFlowStack, /// (uint branch_target) -> void + PopFlowStack, /// () -> void + Exit, /// () -> void + Discard, /// () -> void + + EmitVertex, /// () -> void + EndPrimitive, /// () -> void + + YNegate, /// () -> float + LocalInvocationIdX, /// () -> uint + LocalInvocationIdY, /// () -> uint + LocalInvocationIdZ, /// () -> uint + WorkGroupIdX, /// () -> uint + WorkGroupIdY, /// () -> uint + WorkGroupIdZ, /// () -> uint + + Amount, +}; + +enum class InternalFlag { + Zero = 0, + Sign = 1, + Carry = 2, + Overflow = 3, + Amount = 4, +}; + +class Sampler { +public: + // Use this constructor for bounded Samplers + explicit Sampler(std::size_t offset, std::size_t index, Tegra::Shader::TextureType type, + bool is_array, bool is_shadow) + : offset{offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow}, + is_bindless{false} {} + + // Use this constructor for bindless Samplers + explicit Sampler(u32 cbuf_index, u32 cbuf_offset, std::size_t index, + Tegra::Shader::TextureType type, bool is_array, bool is_shadow) + : offset{(static_cast(cbuf_index) << 32) | cbuf_offset}, index{index}, type{type}, + is_array{is_array}, is_shadow{is_shadow}, is_bindless{true} {} + + // Use this only for serialization/deserialization + explicit Sampler(std::size_t offset, std::size_t index, Tegra::Shader::TextureType type, + bool is_array, bool is_shadow, bool is_bindless) + : offset{offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow}, + is_bindless{is_bindless} {} + + std::size_t GetOffset() const { + return offset; + } + + std::size_t GetIndex() const { + return index; + } + + Tegra::Shader::TextureType GetType() const { + return type; + } + + bool IsArray() const { + return is_array; + } + + bool IsShadow() const { + return is_shadow; + } + + bool IsBindless() const { + return is_bindless; + } + + std::pair GetBindlessCBuf() const { + return {static_cast(offset >> 32), static_cast(offset)}; + } + + bool operator<(const Sampler& rhs) const { + return std::tie(index, offset, type, is_array, is_shadow, is_bindless) < + std::tie(rhs.index, rhs.offset, rhs.type, rhs.is_array, rhs.is_shadow, + rhs.is_bindless); + } + +private: + /// Offset in TSC memory from which to read the sampler object, as specified by the sampling + /// instruction. + std::size_t offset{}; + std::size_t index{}; ///< Value used to index into the generated GLSL sampler array. + Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc) + bool is_array{}; ///< Whether the texture is being sampled as an array texture or not. + bool is_shadow{}; ///< Whether the texture is being sampled as a depth texture or not. + bool is_bindless{}; ///< Whether this sampler belongs to a bindless texture or not. +}; + +struct GlobalMemoryBase { + u32 cbuf_index{}; + u32 cbuf_offset{}; + + bool operator<(const GlobalMemoryBase& rhs) const { + return std::tie(cbuf_index, cbuf_offset) < std::tie(rhs.cbuf_index, rhs.cbuf_offset); + } +}; + +struct MetaArithmetic { + bool precise{}; +}; + +struct MetaTexture { + const Sampler& sampler; + Node array{}; + Node depth_compare{}; + std::vector aoffi; + Node bias{}; + Node lod{}; + Node component{}; + u32 element{}; +}; + +using Meta = std::variant; + +/// Holds any kind of operation that can be done in the IR +class OperationNode final { +public: + explicit OperationNode(OperationCode code) : OperationNode(code, Meta{}) {} + + explicit OperationNode(OperationCode code, Meta meta) + : OperationNode(code, meta, std::vector{}) {} + + explicit OperationNode(OperationCode code, std::vector operands) + : OperationNode(code, Meta{}, std::move(operands)) {} + + explicit OperationNode(OperationCode code, Meta meta, std::vector operands) + : code{code}, meta{std::move(meta)}, operands{std::move(operands)} {} + + template + explicit OperationNode(OperationCode code, Meta meta, Args&&... operands) + : code{code}, meta{std::move(meta)}, operands{operands...} {} + + OperationCode GetCode() const { + return code; + } + + const Meta& GetMeta() const { + return meta; + } + + std::size_t GetOperandsCount() const { + return operands.size(); + } + + const Node& operator[](std::size_t operand_index) const { + return operands.at(operand_index); + } + +private: + OperationCode code{}; + Meta meta{}; + std::vector operands; +}; + +/// Encloses inside any kind of node that returns a boolean conditionally-executed code +class ConditionalNode final { +public: + explicit ConditionalNode(Node condition, std::vector&& code) + : condition{condition}, code{std::move(code)} {} + + Node GetCondition() const { + return condition; + } + + const std::vector& GetCode() const { + return code; + } + +private: + const Node condition; ///< Condition to be satisfied + std::vector code; ///< Code to execute +}; + +/// A general purpose register +class GprNode final { +public: + explicit constexpr GprNode(Tegra::Shader::Register index) : index{index} {} + + u32 GetIndex() const { + return static_cast(index); + } + +private: + const Tegra::Shader::Register index; +}; + +/// A 32-bits value that represents an immediate value +class ImmediateNode final { +public: + explicit constexpr ImmediateNode(u32 value) : value{value} {} + + u32 GetValue() const { + return value; + } + +private: + const u32 value; +}; + +/// One of Maxwell's internal flags +class InternalFlagNode final { +public: + explicit constexpr InternalFlagNode(InternalFlag flag) : flag{flag} {} + + InternalFlag GetFlag() const { + return flag; + } + +private: + const InternalFlag flag; +}; + +/// A predicate register, it can be negated without additional nodes +class PredicateNode final { +public: + explicit constexpr PredicateNode(Tegra::Shader::Pred index, bool negated) + : index{index}, negated{negated} {} + + Tegra::Shader::Pred GetIndex() const { + return index; + } + + bool IsNegated() const { + return negated; + } + +private: + const Tegra::Shader::Pred index; + const bool negated; +}; + +/// Attribute buffer memory (known as attributes or varyings in GLSL terms) +class AbufNode final { +public: + // Initialize for standard attributes (index is explicit). + explicit AbufNode(Tegra::Shader::Attribute::Index index, u32 element, Node buffer = {}) + : buffer{std::move(buffer)}, index{index}, element{element} {} + + // Initialize for physical attributes (index is a variable value). + explicit AbufNode(Node physical_address, Node buffer = {}) + : physical_address{physical_address}, buffer{std::move(buffer)} {} + + Tegra::Shader::Attribute::Index GetIndex() const { + return index; + } + + u32 GetElement() const { + return element; + } + + Node GetBuffer() const { + return buffer; + } + + bool IsPhysicalBuffer() const { + return static_cast(physical_address); + } + + const Node& GetPhysicalAddress() const { + return physical_address; + } + +private: + Node physical_address; + Node buffer; + Tegra::Shader::Attribute::Index index{}; + u32 element{}; +}; + +/// Constant buffer node, usually mapped to uniform buffers in GLSL +class CbufNode final { +public: + explicit CbufNode(u32 index, Node offset) : index{index}, offset{offset} {} + + u32 GetIndex() const { + return index; + } + + Node GetOffset() const { + return offset; + } + +private: + const u32 index; + const Node offset; +}; + +/// Local memory node +class LmemNode final { +public: + explicit LmemNode(Node address) : address{address} {} + + Node GetAddress() const { + return address; + } + +private: + const Node address; +}; + +/// Global memory node +class GmemNode final { +public: + explicit GmemNode(Node real_address, Node base_address, const GlobalMemoryBase& descriptor) + : real_address{real_address}, base_address{base_address}, descriptor{descriptor} {} + + Node GetRealAddress() const { + return real_address; + } + + Node GetBaseAddress() const { + return base_address; + } + + const GlobalMemoryBase& GetDescriptor() const { + return descriptor; + } + +private: + const Node real_address; + const Node base_address; + const GlobalMemoryBase descriptor; +}; + +/// Commentary, can be dropped +class CommentNode final { +public: + explicit CommentNode(std::string text) : text{std::move(text)} {} + + const std::string& GetText() const { + return text; + } + +private: + std::string text; +}; + +} // namespace VideoCommon::Shader diff --git a/src/video_core/shader/node_helper.h b/src/video_core/shader/node_helper.h index 70547a03d..0c2aa749b 100644 --- a/src/video_core/shader/node_helper.h +++ b/src/video_core/shader/node_helper.h @@ -12,10 +12,15 @@ #include #include "common/common_types.h" -#include "video_core/shader/shader_ir.h" +#include "video_core/shader/node.h" namespace VideoCommon::Shader { +/// This arithmetic operation cannot be constraint +inline constexpr MetaArithmetic PRECISE = {true}; +/// This arithmetic operation can be optimized away +inline constexpr MetaArithmetic NO_PRECISE = {false}; + /// Creates a conditional node Node Conditional(Node condition, std::vector code); diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index 1b2745c23..edcf2288e 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -18,188 +18,14 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/engines/shader_bytecode.h" #include "video_core/engines/shader_header.h" +#include "video_core/shader/node.h" namespace VideoCommon::Shader { -class OperationNode; -class ConditionalNode; -class GprNode; -class ImmediateNode; -class InternalFlagNode; -class PredicateNode; -class AbufNode; ///< Attribute buffer -class CbufNode; ///< Constant buffer -class LmemNode; ///< Local memory -class GmemNode; ///< Global memory -class CommentNode; - using ProgramCode = std::vector; -using NodeData = - std::variant; -using Node = std::shared_ptr; -using Node4 = std::array; -using NodeBlock = std::vector; - constexpr u32 MAX_PROGRAM_LENGTH = 0x1000; -enum class OperationCode { - Assign, /// (float& dest, float src) -> void - - Select, /// (MetaArithmetic, bool pred, float a, float b) -> float - - FAdd, /// (MetaArithmetic, float a, float b) -> float - FMul, /// (MetaArithmetic, float a, float b) -> float - FDiv, /// (MetaArithmetic, float a, float b) -> float - FFma, /// (MetaArithmetic, float a, float b, float c) -> float - FNegate, /// (MetaArithmetic, float a) -> float - FAbsolute, /// (MetaArithmetic, float a) -> float - FClamp, /// (MetaArithmetic, float value, float min, float max) -> float - FMin, /// (MetaArithmetic, float a, float b) -> float - FMax, /// (MetaArithmetic, float a, float b) -> float - FCos, /// (MetaArithmetic, float a) -> float - FSin, /// (MetaArithmetic, float a) -> float - FExp2, /// (MetaArithmetic, float a) -> float - FLog2, /// (MetaArithmetic, float a) -> float - FInverseSqrt, /// (MetaArithmetic, float a) -> float - FSqrt, /// (MetaArithmetic, float a) -> float - FRoundEven, /// (MetaArithmetic, float a) -> float - FFloor, /// (MetaArithmetic, float a) -> float - FCeil, /// (MetaArithmetic, float a) -> float - FTrunc, /// (MetaArithmetic, float a) -> float - FCastInteger, /// (MetaArithmetic, int a) -> float - FCastUInteger, /// (MetaArithmetic, uint a) -> float - - IAdd, /// (MetaArithmetic, int a, int b) -> int - IMul, /// (MetaArithmetic, int a, int b) -> int - IDiv, /// (MetaArithmetic, int a, int b) -> int - INegate, /// (MetaArithmetic, int a) -> int - IAbsolute, /// (MetaArithmetic, int a) -> int - IMin, /// (MetaArithmetic, int a, int b) -> int - IMax, /// (MetaArithmetic, int a, int b) -> int - ICastFloat, /// (MetaArithmetic, float a) -> int - ICastUnsigned, /// (MetaArithmetic, uint a) -> int - ILogicalShiftLeft, /// (MetaArithmetic, int a, uint b) -> int - ILogicalShiftRight, /// (MetaArithmetic, int a, uint b) -> int - IArithmeticShiftRight, /// (MetaArithmetic, int a, uint b) -> int - IBitwiseAnd, /// (MetaArithmetic, int a, int b) -> int - IBitwiseOr, /// (MetaArithmetic, int a, int b) -> int - IBitwiseXor, /// (MetaArithmetic, int a, int b) -> int - IBitwiseNot, /// (MetaArithmetic, int a) -> int - IBitfieldInsert, /// (MetaArithmetic, int base, int insert, int offset, int bits) -> int - IBitfieldExtract, /// (MetaArithmetic, int value, int offset, int offset) -> int - IBitCount, /// (MetaArithmetic, int) -> int - - UAdd, /// (MetaArithmetic, uint a, uint b) -> uint - UMul, /// (MetaArithmetic, uint a, uint b) -> uint - UDiv, /// (MetaArithmetic, uint a, uint b) -> uint - UMin, /// (MetaArithmetic, uint a, uint b) -> uint - UMax, /// (MetaArithmetic, uint a, uint b) -> uint - UCastFloat, /// (MetaArithmetic, float a) -> uint - UCastSigned, /// (MetaArithmetic, int a) -> uint - ULogicalShiftLeft, /// (MetaArithmetic, uint a, uint b) -> uint - ULogicalShiftRight, /// (MetaArithmetic, uint a, uint b) -> uint - UArithmeticShiftRight, /// (MetaArithmetic, uint a, uint b) -> uint - UBitwiseAnd, /// (MetaArithmetic, uint a, uint b) -> uint - UBitwiseOr, /// (MetaArithmetic, uint a, uint b) -> uint - UBitwiseXor, /// (MetaArithmetic, uint a, uint b) -> uint - UBitwiseNot, /// (MetaArithmetic, uint a) -> uint - UBitfieldInsert, /// (MetaArithmetic, uint base, uint insert, int offset, int bits) -> uint - UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint - UBitCount, /// (MetaArithmetic, uint) -> uint - - HAdd, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 - HMul, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 - HFma, /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2 - HAbsolute, /// (f16vec2 a) -> f16vec2 - HNegate, /// (f16vec2 a, bool first, bool second) -> f16vec2 - HClamp, /// (f16vec2 src, float min, float max) -> f16vec2 - HUnpack, /// (Tegra::Shader::HalfType, T value) -> f16vec2 - HMergeF32, /// (f16vec2 src) -> float - HMergeH0, /// (f16vec2 dest, f16vec2 src) -> f16vec2 - HMergeH1, /// (f16vec2 dest, f16vec2 src) -> f16vec2 - HPack2, /// (float a, float b) -> f16vec2 - - LogicalAssign, /// (bool& dst, bool src) -> void - LogicalAnd, /// (bool a, bool b) -> bool - LogicalOr, /// (bool a, bool b) -> bool - LogicalXor, /// (bool a, bool b) -> bool - LogicalNegate, /// (bool a) -> bool - LogicalPick2, /// (bool2 pair, uint index) -> bool - LogicalAll2, /// (bool2 a) -> bool - LogicalAny2, /// (bool2 a) -> bool - - LogicalFLessThan, /// (float a, float b) -> bool - LogicalFEqual, /// (float a, float b) -> bool - LogicalFLessEqual, /// (float a, float b) -> bool - LogicalFGreaterThan, /// (float a, float b) -> bool - LogicalFNotEqual, /// (float a, float b) -> bool - LogicalFGreaterEqual, /// (float a, float b) -> bool - LogicalFIsNan, /// (float a) -> bool - - LogicalILessThan, /// (int a, int b) -> bool - LogicalIEqual, /// (int a, int b) -> bool - LogicalILessEqual, /// (int a, int b) -> bool - LogicalIGreaterThan, /// (int a, int b) -> bool - LogicalINotEqual, /// (int a, int b) -> bool - LogicalIGreaterEqual, /// (int a, int b) -> bool - - LogicalULessThan, /// (uint a, uint b) -> bool - LogicalUEqual, /// (uint a, uint b) -> bool - LogicalULessEqual, /// (uint a, uint b) -> bool - LogicalUGreaterThan, /// (uint a, uint b) -> bool - LogicalUNotEqual, /// (uint a, uint b) -> bool - LogicalUGreaterEqual, /// (uint a, uint b) -> bool - - Logical2HLessThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HLessEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HNotEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HLessThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HLessEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HNotEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - - Texture, /// (MetaTexture, float[N] coords) -> float4 - TextureLod, /// (MetaTexture, float[N] coords) -> float4 - TextureGather, /// (MetaTexture, float[N] coords) -> float4 - TextureQueryDimensions, /// (MetaTexture, float a) -> float4 - TextureQueryLod, /// (MetaTexture, float[N] coords) -> float4 - TexelFetch, /// (MetaTexture, int[N], int) -> float4 - - Branch, /// (uint branch_target) -> void - PushFlowStack, /// (uint branch_target) -> void - PopFlowStack, /// () -> void - Exit, /// () -> void - Discard, /// () -> void - - EmitVertex, /// () -> void - EndPrimitive, /// () -> void - - YNegate, /// () -> float - LocalInvocationIdX, /// () -> uint - LocalInvocationIdY, /// () -> uint - LocalInvocationIdZ, /// () -> uint - WorkGroupIdX, /// () -> uint - WorkGroupIdY, /// () -> uint - WorkGroupIdZ, /// () -> uint - - Amount, -}; - -enum class InternalFlag { - Zero = 0, - Sign = 1, - Carry = 2, - Overflow = 3, - Amount = 4, -}; - /// Describes the behaviour of code path of a given entry point and a return point. enum class ExitMethod { Undetermined, ///< Internal value. Only occur when analyzing JMP loop. @@ -208,71 +34,6 @@ enum class ExitMethod { AlwaysEnd, ///< All code paths reach a END instruction. }; -class Sampler { -public: - // Use this constructor for bounded Samplers - explicit Sampler(std::size_t offset, std::size_t index, Tegra::Shader::TextureType type, - bool is_array, bool is_shadow) - : offset{offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow}, - is_bindless{false} {} - - // Use this constructor for bindless Samplers - explicit Sampler(u32 cbuf_index, u32 cbuf_offset, std::size_t index, - Tegra::Shader::TextureType type, bool is_array, bool is_shadow) - : offset{(static_cast(cbuf_index) << 32) | cbuf_offset}, index{index}, type{type}, - is_array{is_array}, is_shadow{is_shadow}, is_bindless{true} {} - - // Use this only for serialization/deserialization - explicit Sampler(std::size_t offset, std::size_t index, Tegra::Shader::TextureType type, - bool is_array, bool is_shadow, bool is_bindless) - : offset{offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow}, - is_bindless{is_bindless} {} - - std::size_t GetOffset() const { - return offset; - } - - std::size_t GetIndex() const { - return index; - } - - Tegra::Shader::TextureType GetType() const { - return type; - } - - bool IsArray() const { - return is_array; - } - - bool IsShadow() const { - return is_shadow; - } - - bool IsBindless() const { - return is_bindless; - } - - std::pair GetBindlessCBuf() const { - return {static_cast(offset >> 32), static_cast(offset)}; - } - - bool operator<(const Sampler& rhs) const { - return std::tie(index, offset, type, is_array, is_shadow, is_bindless) < - std::tie(rhs.index, rhs.offset, rhs.type, rhs.is_array, rhs.is_shadow, - rhs.is_bindless); - } - -private: - /// Offset in TSC memory from which to read the sampler object, as specified by the sampling - /// instruction. - std::size_t offset{}; - std::size_t index{}; ///< Value used to index into the generated GLSL sampler array. - Tegra::Shader::TextureType type{}; ///< The type used to sample this texture (Texture2D, etc) - bool is_array{}; ///< Whether the texture is being sampled as an array texture or not. - bool is_shadow{}; ///< Whether the texture is being sampled as a depth texture or not. - bool is_bindless{}; ///< Whether this sampler belongs to a bindless texture or not. -}; - class ConstBuffer { public: explicit ConstBuffer(u32 max_offset, bool is_indirect) @@ -305,263 +66,11 @@ private: bool is_indirect{}; }; -struct GlobalMemoryBase { - u32 cbuf_index{}; - u32 cbuf_offset{}; - - bool operator<(const GlobalMemoryBase& rhs) const { - return std::tie(cbuf_index, cbuf_offset) < std::tie(rhs.cbuf_index, rhs.cbuf_offset); - } -}; - struct GlobalMemoryUsage { bool is_read{}; bool is_written{}; }; -struct MetaArithmetic { - bool precise{}; -}; - -struct MetaTexture { - const Sampler& sampler; - Node array{}; - Node depth_compare{}; - std::vector aoffi; - Node bias{}; - Node lod{}; - Node component{}; - u32 element{}; -}; - -constexpr MetaArithmetic PRECISE = {true}; -constexpr MetaArithmetic NO_PRECISE = {false}; - -using Meta = std::variant; - -/// Holds any kind of operation that can be done in the IR -class OperationNode final { -public: - explicit OperationNode(OperationCode code) : OperationNode(code, Meta{}) {} - - explicit OperationNode(OperationCode code, Meta meta) - : OperationNode(code, meta, std::vector{}) {} - - explicit OperationNode(OperationCode code, std::vector operands) - : OperationNode(code, Meta{}, std::move(operands)) {} - - explicit OperationNode(OperationCode code, Meta meta, std::vector operands) - : code{code}, meta{std::move(meta)}, operands{std::move(operands)} {} - - template - explicit OperationNode(OperationCode code, Meta meta, Args&&... operands) - : code{code}, meta{std::move(meta)}, operands{operands...} {} - - OperationCode GetCode() const { - return code; - } - - const Meta& GetMeta() const { - return meta; - } - - std::size_t GetOperandsCount() const { - return operands.size(); - } - - const Node& operator[](std::size_t operand_index) const { - return operands.at(operand_index); - } - -private: - OperationCode code{}; - Meta meta{}; - std::vector operands; -}; - -/// Encloses inside any kind of node that returns a boolean conditionally-executed code -class ConditionalNode final { -public: - explicit ConditionalNode(Node condition, std::vector&& code) - : condition{condition}, code{std::move(code)} {} - - Node GetCondition() const { - return condition; - } - - const std::vector& GetCode() const { - return code; - } - -private: - const Node condition; ///< Condition to be satisfied - std::vector code; ///< Code to execute -}; - -/// A general purpose register -class GprNode final { -public: - explicit constexpr GprNode(Tegra::Shader::Register index) : index{index} {} - - u32 GetIndex() const { - return static_cast(index); - } - -private: - const Tegra::Shader::Register index; -}; - -/// A 32-bits value that represents an immediate value -class ImmediateNode final { -public: - explicit constexpr ImmediateNode(u32 value) : value{value} {} - - u32 GetValue() const { - return value; - } - -private: - const u32 value; -}; - -/// One of Maxwell's internal flags -class InternalFlagNode final { -public: - explicit constexpr InternalFlagNode(InternalFlag flag) : flag{flag} {} - - InternalFlag GetFlag() const { - return flag; - } - -private: - const InternalFlag flag; -}; - -/// A predicate register, it can be negated without additional nodes -class PredicateNode final { -public: - explicit constexpr PredicateNode(Tegra::Shader::Pred index, bool negated) - : index{index}, negated{negated} {} - - Tegra::Shader::Pred GetIndex() const { - return index; - } - - bool IsNegated() const { - return negated; - } - -private: - const Tegra::Shader::Pred index; - const bool negated; -}; - -/// Attribute buffer memory (known as attributes or varyings in GLSL terms) -class AbufNode final { -public: - // Initialize for standard attributes (index is explicit). - explicit AbufNode(Tegra::Shader::Attribute::Index index, u32 element, Node buffer = {}) - : buffer{std::move(buffer)}, index{index}, element{element} {} - - // Initialize for physical attributes (index is a variable value). - explicit AbufNode(Node physical_address, Node buffer = {}) - : physical_address{physical_address}, buffer{std::move(buffer)} {} - - Tegra::Shader::Attribute::Index GetIndex() const { - return index; - } - - u32 GetElement() const { - return element; - } - - Node GetBuffer() const { - return buffer; - } - - bool IsPhysicalBuffer() const { - return static_cast(physical_address); - } - - const Node& GetPhysicalAddress() const { - return physical_address; - } - -private: - Node physical_address; - Node buffer; - Tegra::Shader::Attribute::Index index{}; - u32 element{}; -}; - -/// Constant buffer node, usually mapped to uniform buffers in GLSL -class CbufNode final { -public: - explicit CbufNode(u32 index, Node offset) : index{index}, offset{offset} {} - - u32 GetIndex() const { - return index; - } - - Node GetOffset() const { - return offset; - } - -private: - const u32 index; - const Node offset; -}; - -/// Local memory node -class LmemNode final { -public: - explicit LmemNode(Node address) : address{address} {} - - Node GetAddress() const { - return address; - } - -private: - const Node address; -}; - -/// Global memory node -class GmemNode final { -public: - explicit GmemNode(Node real_address, Node base_address, const GlobalMemoryBase& descriptor) - : real_address{real_address}, base_address{base_address}, descriptor{descriptor} {} - - Node GetRealAddress() const { - return real_address; - } - - Node GetBaseAddress() const { - return base_address; - } - - const GlobalMemoryBase& GetDescriptor() const { - return descriptor; - } - -private: - const Node real_address; - const Node base_address; - const GlobalMemoryBase descriptor; -}; - -/// Commentary, can be dropped -class CommentNode final { -public: - explicit CommentNode(std::string text) : text{std::move(text)} {} - - const std::string& GetText() const { - return text; - } - -private: - std::string text; -}; - class ShaderIR final { public: explicit ShaderIR(const ProgramCode& program_code, u32 main_offset); From 769a50661a90f06aa660419d869e266abcddf9f3 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Thu, 6 Jun 2019 20:01:28 -0300 Subject: [PATCH 2/3] shader/node: Minor changes Reflect std::shared_ptr nature of Node on initializers and remove constant members in nodes. Add some commentaries. --- src/video_core/shader/node.h | 104 ++++++++++++++++++----------------- 1 file changed, 54 insertions(+), 50 deletions(-) diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h index 1e7a5f699..c002f90f9 100644 --- a/src/video_core/shader/node.h +++ b/src/video_core/shader/node.h @@ -18,25 +18,6 @@ namespace VideoCommon::Shader { -class OperationNode; -class ConditionalNode; -class GprNode; -class ImmediateNode; -class InternalFlagNode; -class PredicateNode; -class AbufNode; ///< Attribute buffer -class CbufNode; ///< Constant buffer -class LmemNode; ///< Local memory -class GmemNode; ///< Global memory -class CommentNode; - -using NodeData = - std::variant; -using Node = std::shared_ptr; -using Node4 = std::array; -using NodeBlock = std::vector; - enum class OperationCode { Assign, /// (float& dest, float src) -> void @@ -193,21 +174,40 @@ enum class InternalFlag { Amount = 4, }; +class OperationNode; +class ConditionalNode; +class GprNode; +class ImmediateNode; +class InternalFlagNode; +class PredicateNode; +class AbufNode; +class CbufNode; +class LmemNode; +class GmemNode; +class CommentNode; + +using NodeData = + std::variant; +using Node = std::shared_ptr; +using Node4 = std::array; +using NodeBlock = std::vector; + class Sampler { public: - // Use this constructor for bounded Samplers + /// This constructor is for bound samplers explicit Sampler(std::size_t offset, std::size_t index, Tegra::Shader::TextureType type, bool is_array, bool is_shadow) : offset{offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow}, is_bindless{false} {} - // Use this constructor for bindless Samplers + /// This constructor is for bindless samplers explicit Sampler(u32 cbuf_index, u32 cbuf_offset, std::size_t index, Tegra::Shader::TextureType type, bool is_array, bool is_shadow) : offset{(static_cast(cbuf_index) << 32) | cbuf_offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow}, is_bindless{true} {} - // Use this only for serialization/deserialization + /// This constructor is for serialization/deserialization explicit Sampler(std::size_t offset, std::size_t index, Tegra::Shader::TextureType type, bool is_array, bool is_shadow, bool is_bindless) : offset{offset}, index{index}, type{type}, is_array{is_array}, is_shadow{is_shadow}, @@ -267,21 +267,24 @@ struct GlobalMemoryBase { } }; +/// Parameters describing an arithmetic operation struct MetaArithmetic { - bool precise{}; + bool precise{}; ///< Whether the operation can be constraint or not }; +/// Parameters describing a texture sampler struct MetaTexture { const Sampler& sampler; - Node array{}; - Node depth_compare{}; + Node array; + Node depth_compare; std::vector aoffi; - Node bias{}; - Node lod{}; + Node bias; + Node lod; Node component{}; u32 element{}; }; +/// Parameters that modify an operation but are not part of any particular operand using Meta = std::variant; /// Holds any kind of operation that can be done in the IR @@ -328,9 +331,9 @@ private: class ConditionalNode final { public: explicit ConditionalNode(Node condition, std::vector&& code) - : condition{condition}, code{std::move(code)} {} + : condition{std::move(condition)}, code{std::move(code)} {} - Node GetCondition() const { + const Node& GetCondition() const { return condition; } @@ -339,7 +342,7 @@ public: } private: - const Node condition; ///< Condition to be satisfied + Node condition; ///< Condition to be satisfied std::vector code; ///< Code to execute }; @@ -353,7 +356,7 @@ public: } private: - const Tegra::Shader::Register index; + Tegra::Shader::Register index{}; }; /// A 32-bits value that represents an immediate value @@ -366,7 +369,7 @@ public: } private: - const u32 value; + u32 value{}; }; /// One of Maxwell's internal flags @@ -379,7 +382,7 @@ public: } private: - const InternalFlag flag; + InternalFlag flag{}; }; /// A predicate register, it can be negated without additional nodes @@ -397,8 +400,8 @@ public: } private: - const Tegra::Shader::Pred index; - const bool negated; + Tegra::Shader::Pred index{}; + bool negated{}; }; /// Attribute buffer memory (known as attributes or varyings in GLSL terms) @@ -410,7 +413,7 @@ public: // Initialize for physical attributes (index is a variable value). explicit AbufNode(Node physical_address, Node buffer = {}) - : physical_address{physical_address}, buffer{std::move(buffer)} {} + : physical_address{std::move(physical_address)}, buffer{std::move(buffer)} {} Tegra::Shader::Attribute::Index GetIndex() const { return index; @@ -420,7 +423,7 @@ public: return element; } - Node GetBuffer() const { + const Node& GetBuffer() const { return buffer; } @@ -442,45 +445,46 @@ private: /// Constant buffer node, usually mapped to uniform buffers in GLSL class CbufNode final { public: - explicit CbufNode(u32 index, Node offset) : index{index}, offset{offset} {} + explicit CbufNode(u32 index, Node offset) : index{index}, offset{std::move(offset)} {} u32 GetIndex() const { return index; } - Node GetOffset() const { + const Node& GetOffset() const { return offset; } private: - const u32 index; - const Node offset; + u32 index{}; + Node offset; }; /// Local memory node class LmemNode final { public: - explicit LmemNode(Node address) : address{address} {} + explicit LmemNode(Node address) : address{std::move(address)} {} - Node GetAddress() const { + const Node& GetAddress() const { return address; } private: - const Node address; + Node address; }; /// Global memory node class GmemNode final { public: explicit GmemNode(Node real_address, Node base_address, const GlobalMemoryBase& descriptor) - : real_address{real_address}, base_address{base_address}, descriptor{descriptor} {} + : real_address{std::move(real_address)}, base_address{std::move(base_address)}, + descriptor{descriptor} {} - Node GetRealAddress() const { + const Node& GetRealAddress() const { return real_address; } - Node GetBaseAddress() const { + const Node& GetBaseAddress() const { return base_address; } @@ -489,9 +493,9 @@ public: } private: - const Node real_address; - const Node base_address; - const GlobalMemoryBase descriptor; + Node real_address; + Node base_address; + GlobalMemoryBase descriptor; }; /// Commentary, can be dropped From dec1cbaf7f80847dc9426cf87105c3457aade04f Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Thu, 6 Jun 2019 20:11:48 -0300 Subject: [PATCH 3/3] cmake: Add missing shader hash file entries --- CMakeModules/GenerateSCMRev.cmake | 3 +++ src/common/CMakeLists.txt | 3 +++ 2 files changed, 6 insertions(+) diff --git a/CMakeModules/GenerateSCMRev.cmake b/CMakeModules/GenerateSCMRev.cmake index 5e00d839f..31edeb63d 100644 --- a/CMakeModules/GenerateSCMRev.cmake +++ b/CMakeModules/GenerateSCMRev.cmake @@ -82,6 +82,9 @@ set(HASH_FILES "${VIDEO_CORE}/shader/decode/video.cpp" "${VIDEO_CORE}/shader/decode/xmad.cpp" "${VIDEO_CORE}/shader/decode.cpp" + "${VIDEO_CORE}/shader/node.h" + "${VIDEO_CORE}/shader/node_helper.cpp" + "${VIDEO_CORE}/shader/node_helper.h" "${VIDEO_CORE}/shader/shader_ir.cpp" "${VIDEO_CORE}/shader/shader_ir.h" "${VIDEO_CORE}/shader/track.cpp" diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt index cb514a0d2..198b3fe07 100644 --- a/src/common/CMakeLists.txt +++ b/src/common/CMakeLists.txt @@ -56,6 +56,9 @@ add_custom_command(OUTPUT scm_rev.cpp "${VIDEO_CORE}/shader/decode/video.cpp" "${VIDEO_CORE}/shader/decode/xmad.cpp" "${VIDEO_CORE}/shader/decode.cpp" + "${VIDEO_CORE}/shader/node.h" + "${VIDEO_CORE}/shader/node_helper.cpp" + "${VIDEO_CORE}/shader/node_helper.h" "${VIDEO_CORE}/shader/shader_ir.cpp" "${VIDEO_CORE}/shader/shader_ir.h" "${VIDEO_CORE}/shader/track.cpp"