yuzu/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp
2021-07-22 21:51:23 -04:00

382 lines
13 KiB
C++

// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <algorithm>
#include <tuple>
#include <type_traits>
#include "common/bit_cast.h"
#include "common/bit_util.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/ir/ir_emitter.h"
#include "shader_recompiler/frontend/ir/microinstruction.h"
#include "shader_recompiler/ir_opt/passes.h"
namespace Shader::Optimization {
namespace {
// Metaprogramming stuff to get arguments information out of a lambda
template <typename Func>
struct LambdaTraits : LambdaTraits<decltype(&std::remove_reference_t<Func>::operator())> {};
template <typename ReturnType, typename LambdaType, typename... Args>
struct LambdaTraits<ReturnType (LambdaType::*)(Args...) const> {
template <size_t I>
using ArgType = std::tuple_element_t<I, std::tuple<Args...>>;
static constexpr size_t NUM_ARGS{sizeof...(Args)};
};
template <typename T>
[[nodiscard]] T Arg(const IR::Value& value) {
if constexpr (std::is_same_v<T, bool>) {
return value.U1();
} else if constexpr (std::is_same_v<T, u32>) {
return value.U32();
} else if constexpr (std::is_same_v<T, s32>) {
return static_cast<s32>(value.U32());
} else if constexpr (std::is_same_v<T, f32>) {
return value.F32();
} else if constexpr (std::is_same_v<T, u64>) {
return value.U64();
}
}
template <typename T, typename ImmFn>
bool FoldCommutative(IR::Inst& inst, ImmFn&& imm_fn) {
const IR::Value lhs{inst.Arg(0)};
const IR::Value rhs{inst.Arg(1)};
const bool is_lhs_immediate{lhs.IsImmediate()};
const bool is_rhs_immediate{rhs.IsImmediate()};
if (is_lhs_immediate && is_rhs_immediate) {
const auto result{imm_fn(Arg<T>(lhs), Arg<T>(rhs))};
inst.ReplaceUsesWith(IR::Value{result});
return false;
}
if (is_lhs_immediate && !is_rhs_immediate) {
IR::Inst* const rhs_inst{rhs.InstRecursive()};
if (rhs_inst->Opcode() == inst.Opcode() && rhs_inst->Arg(1).IsImmediate()) {
const auto combined{imm_fn(Arg<T>(lhs), Arg<T>(rhs_inst->Arg(1)))};
inst.SetArg(0, rhs_inst->Arg(0));
inst.SetArg(1, IR::Value{combined});
} else {
// Normalize
inst.SetArg(0, rhs);
inst.SetArg(1, lhs);
}
}
if (!is_lhs_immediate && is_rhs_immediate) {
const IR::Inst* const lhs_inst{lhs.InstRecursive()};
if (lhs_inst->Opcode() == inst.Opcode() && lhs_inst->Arg(1).IsImmediate()) {
const auto combined{imm_fn(Arg<T>(rhs), Arg<T>(lhs_inst->Arg(1)))};
inst.SetArg(0, lhs_inst->Arg(0));
inst.SetArg(1, IR::Value{combined});
}
}
return true;
}
template <typename Func>
bool FoldWhenAllImmediates(IR::Inst& inst, Func&& func) {
if (!inst.AreAllArgsImmediates() || inst.HasAssociatedPseudoOperation()) {
return false;
}
using Indices = std::make_index_sequence<LambdaTraits<decltype(func)>::NUM_ARGS>;
inst.ReplaceUsesWith(EvalImmediates(inst, func, Indices{}));
return true;
}
void FoldGetRegister(IR::Inst& inst) {
if (inst.Arg(0).Reg() == IR::Reg::RZ) {
inst.ReplaceUsesWith(IR::Value{u32{0}});
}
}
void FoldGetPred(IR::Inst& inst) {
if (inst.Arg(0).Pred() == IR::Pred::PT) {
inst.ReplaceUsesWith(IR::Value{true});
}
}
/// Replaces the pattern generated by two XMAD multiplications
bool FoldXmadMultiply(IR::Block& block, IR::Inst& inst) {
/*
* We are looking for this pattern:
* %rhs_bfe = BitFieldUExtract %factor_a, #0, #16
* %rhs_mul = IMul32 %rhs_bfe, %factor_b
* %lhs_bfe = BitFieldUExtract %factor_a, #16, #16
* %rhs_mul = IMul32 %lhs_bfe, %factor_b
* %lhs_shl = ShiftLeftLogical32 %rhs_mul, #16
* %result = IAdd32 %lhs_shl, %rhs_mul
*
* And replacing it with
* %result = IMul32 %factor_a, %factor_b
*
* This optimization has been proven safe by LLVM and MSVC.
*/
const IR::Value lhs_arg{inst.Arg(0)};
const IR::Value rhs_arg{inst.Arg(1)};
if (lhs_arg.IsImmediate() || rhs_arg.IsImmediate()) {
return false;
}
IR::Inst* const lhs_shl{lhs_arg.InstRecursive()};
if (lhs_shl->Opcode() != IR::Opcode::ShiftLeftLogical32 || lhs_shl->Arg(1) != IR::Value{16U}) {
return false;
}
if (lhs_shl->Arg(0).IsImmediate()) {
return false;
}
IR::Inst* const lhs_mul{lhs_shl->Arg(0).InstRecursive()};
IR::Inst* const rhs_mul{rhs_arg.InstRecursive()};
if (lhs_mul->Opcode() != IR::Opcode::IMul32 || rhs_mul->Opcode() != IR::Opcode::IMul32) {
return false;
}
if (lhs_mul->Arg(1).Resolve() != rhs_mul->Arg(1).Resolve()) {
return false;
}
const IR::U32 factor_b{lhs_mul->Arg(1)};
if (lhs_mul->Arg(0).IsImmediate() || rhs_mul->Arg(0).IsImmediate()) {
return false;
}
IR::Inst* const lhs_bfe{lhs_mul->Arg(0).InstRecursive()};
IR::Inst* const rhs_bfe{rhs_mul->Arg(0).InstRecursive()};
if (lhs_bfe->Opcode() != IR::Opcode::BitFieldUExtract) {
return false;
}
if (rhs_bfe->Opcode() != IR::Opcode::BitFieldUExtract) {
return false;
}
if (lhs_bfe->Arg(1) != IR::Value{16U} || lhs_bfe->Arg(2) != IR::Value{16U}) {
return false;
}
if (rhs_bfe->Arg(1) != IR::Value{0U} || rhs_bfe->Arg(2) != IR::Value{16U}) {
return false;
}
if (lhs_bfe->Arg(0).Resolve() != rhs_bfe->Arg(0).Resolve()) {
return false;
}
const IR::U32 factor_a{lhs_bfe->Arg(0)};
IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
inst.ReplaceUsesWith(ir.IMul(factor_a, factor_b));
return true;
}
template <typename T>
void FoldAdd(IR::Block& block, IR::Inst& inst) {
if (inst.HasAssociatedPseudoOperation()) {
return;
}
if (!FoldCommutative<T>(inst, [](T a, T b) { return a + b; })) {
return;
}
const IR::Value rhs{inst.Arg(1)};
if (rhs.IsImmediate() && Arg<T>(rhs) == 0) {
inst.ReplaceUsesWith(inst.Arg(0));
return;
}
if constexpr (std::is_same_v<T, u32>) {
if (FoldXmadMultiply(block, inst)) {
return;
}
}
}
void FoldISub32(IR::Inst& inst) {
if (FoldWhenAllImmediates(inst, [](u32 a, u32 b) { return a - b; })) {
return;
}
if (inst.Arg(0).IsImmediate() || inst.Arg(1).IsImmediate()) {
return;
}
// ISub32 is generally used to subtract two constant buffers, compare and replace this with
// zero if they equal.
const auto equal_cbuf{[](IR::Inst* a, IR::Inst* b) {
return a->Opcode() == IR::Opcode::GetCbufU32 && b->Opcode() == IR::Opcode::GetCbufU32 &&
a->Arg(0) == b->Arg(0) && a->Arg(1) == b->Arg(1);
}};
IR::Inst* op_a{inst.Arg(0).InstRecursive()};
IR::Inst* op_b{inst.Arg(1).InstRecursive()};
if (equal_cbuf(op_a, op_b)) {
inst.ReplaceUsesWith(IR::Value{u32{0}});
return;
}
// It's also possible a value is being added to a cbuf and then subtracted
if (op_b->Opcode() == IR::Opcode::IAdd32) {
// Canonicalize local variables to simplify the following logic
std::swap(op_a, op_b);
}
if (op_b->Opcode() != IR::Opcode::GetCbufU32) {
return;
}
IR::Inst* const inst_cbuf{op_b};
if (op_a->Opcode() != IR::Opcode::IAdd32) {
return;
}
IR::Value add_op_a{op_a->Arg(0)};
IR::Value add_op_b{op_a->Arg(1)};
if (add_op_b.IsImmediate()) {
// Canonicalize
std::swap(add_op_a, add_op_b);
}
if (add_op_b.IsImmediate()) {
return;
}
IR::Inst* const add_cbuf{add_op_b.InstRecursive()};
if (equal_cbuf(add_cbuf, inst_cbuf)) {
inst.ReplaceUsesWith(add_op_a);
}
}
template <typename T>
void FoldSelect(IR::Inst& inst) {
const IR::Value cond{inst.Arg(0)};
if (cond.IsImmediate()) {
inst.ReplaceUsesWith(cond.U1() ? inst.Arg(1) : inst.Arg(2));
}
}
void FoldLogicalAnd(IR::Inst& inst) {
if (!FoldCommutative<bool>(inst, [](bool a, bool b) { return a && b; })) {
return;
}
const IR::Value rhs{inst.Arg(1)};
if (rhs.IsImmediate()) {
if (rhs.U1()) {
inst.ReplaceUsesWith(inst.Arg(0));
} else {
inst.ReplaceUsesWith(IR::Value{false});
}
}
}
void FoldLogicalOr(IR::Inst& inst) {
if (!FoldCommutative<bool>(inst, [](bool a, bool b) { return a || b; })) {
return;
}
const IR::Value rhs{inst.Arg(1)};
if (rhs.IsImmediate()) {
if (rhs.U1()) {
inst.ReplaceUsesWith(IR::Value{true});
} else {
inst.ReplaceUsesWith(inst.Arg(0));
}
}
}
void FoldLogicalNot(IR::Inst& inst) {
const IR::U1 value{inst.Arg(0)};
if (value.IsImmediate()) {
inst.ReplaceUsesWith(IR::Value{!value.U1()});
return;
}
IR::Inst* const arg{value.InstRecursive()};
if (arg->Opcode() == IR::Opcode::LogicalNot) {
inst.ReplaceUsesWith(arg->Arg(0));
}
}
template <IR::Opcode op, typename Dest, typename Source>
void FoldBitCast(IR::Inst& inst, IR::Opcode reverse) {
const IR::Value value{inst.Arg(0)};
if (value.IsImmediate()) {
inst.ReplaceUsesWith(IR::Value{Common::BitCast<Dest>(Arg<Source>(value))});
return;
}
IR::Inst* const arg_inst{value.InstRecursive()};
if (arg_inst->Opcode() == reverse) {
inst.ReplaceUsesWith(arg_inst->Arg(0));
return;
}
if constexpr (op == IR::Opcode::BitCastF32U32) {
if (arg_inst->Opcode() == IR::Opcode::GetCbufU32) {
// Replace the bitcast with a typed constant buffer read
inst.ReplaceOpcode(IR::Opcode::GetCbufF32);
inst.SetArg(0, arg_inst->Arg(0));
inst.SetArg(1, arg_inst->Arg(1));
return;
}
}
}
template <typename Func, size_t... I>
IR::Value EvalImmediates(const IR::Inst& inst, Func&& func, std::index_sequence<I...>) {
using Traits = LambdaTraits<decltype(func)>;
return IR::Value{func(Arg<Traits::ArgType<I>>(inst.Arg(I))...)};
}
void FoldBranchConditional(IR::Inst& inst) {
const IR::U1 cond{inst.Arg(0)};
if (cond.IsImmediate()) {
// TODO: Convert to Branch
return;
}
const IR::Inst* cond_inst{cond.InstRecursive()};
if (cond_inst->Opcode() == IR::Opcode::LogicalNot) {
const IR::Value true_label{inst.Arg(1)};
const IR::Value false_label{inst.Arg(2)};
// Remove negation on the conditional (take the parameter out of LogicalNot) and swap
// the branches
inst.SetArg(0, cond_inst->Arg(0));
inst.SetArg(1, false_label);
inst.SetArg(2, true_label);
}
}
void ConstantPropagation(IR::Block& block, IR::Inst& inst) {
switch (inst.Opcode()) {
case IR::Opcode::GetRegister:
return FoldGetRegister(inst);
case IR::Opcode::GetPred:
return FoldGetPred(inst);
case IR::Opcode::IAdd32:
return FoldAdd<u32>(block, inst);
case IR::Opcode::ISub32:
return FoldISub32(inst);
case IR::Opcode::BitCastF32U32:
return FoldBitCast<IR::Opcode::BitCastF32U32, f32, u32>(inst, IR::Opcode::BitCastU32F32);
case IR::Opcode::BitCastU32F32:
return FoldBitCast<IR::Opcode::BitCastU32F32, u32, f32>(inst, IR::Opcode::BitCastF32U32);
case IR::Opcode::IAdd64:
return FoldAdd<u64>(block, inst);
case IR::Opcode::SelectU32:
return FoldSelect<u32>(inst);
case IR::Opcode::LogicalAnd:
return FoldLogicalAnd(inst);
case IR::Opcode::LogicalOr:
return FoldLogicalOr(inst);
case IR::Opcode::LogicalNot:
return FoldLogicalNot(inst);
case IR::Opcode::SLessThan:
FoldWhenAllImmediates(inst, [](s32 a, s32 b) { return a < b; });
return;
case IR::Opcode::ULessThan:
FoldWhenAllImmediates(inst, [](u32 a, u32 b) { return a < b; });
return;
case IR::Opcode::BitFieldUExtract:
FoldWhenAllImmediates(inst, [](u32 base, u32 shift, u32 count) {
if (static_cast<size_t>(shift) + static_cast<size_t>(count) > Common::BitSize<u32>()) {
throw LogicError("Undefined result in {}({}, {}, {})", IR::Opcode::BitFieldUExtract,
base, shift, count);
}
return (base >> shift) & ((1U << count) - 1);
});
return;
case IR::Opcode::BranchConditional:
return FoldBranchConditional(inst);
default:
break;
}
}
} // Anonymous namespace
void ConstantPropagationPass(IR::Program& program) {
for (IR::Block* const block : program.post_order_blocks) {
for (IR::Inst& inst : block->Instructions()) {
ConstantPropagation(*block, inst);
}
}
}
} // namespace Shader::Optimization