shader: Add int64 to int32 lowering pass

5 years ago · e3df2285ca
3 changed files with 218 additions and 0 deletions
--- a/src/shader_recompiler/CMakeLists.txt
+++ b/src/shader_recompiler/CMakeLists.txt
@ -219,6 +219,7 @@ add_library(shader_recompiler STATIC
    ir_opt/global_memory_to_storage_buffer_pass.cpp
    ir_opt/identity_removal_pass.cpp
    ir_opt/lower_fp16_to_fp32.cpp
+    ir_opt/lower_int64_to_int32.cpp
    ir_opt/passes.h
    ir_opt/ssa_rewrite_pass.cpp
    ir_opt/texture_pass.cpp
--- a/src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp
+++ b/src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp
@ -0,0 +1,216 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <ranges>
+#include <utility>
+
+#include "shader_recompiler/exception.h"
+#include "shader_recompiler/frontend/ir/basic_block.h"
+#include "shader_recompiler/frontend/ir/ir_emitter.h"
+#include "shader_recompiler/frontend/ir/program.h"
+#include "shader_recompiler/frontend/ir/value.h"
+
+namespace Shader::Optimization {
+namespace {
+std::pair<IR::U32, IR::U32> Unpack(IR::IREmitter& ir, const IR::Value& packed) {
+    if (packed.IsImmediate()) {
+        const u64 value{packed.U64()};
+        return {
+            ir.Imm32(static_cast<u32>(value)),
+            ir.Imm32(static_cast<u32>(value >> 32)),
+        };
+    } else {
+        return std::pair<IR::U32, IR::U32>{
+            ir.CompositeExtract(packed, 0u),
+            ir.CompositeExtract(packed, 1u),
+        };
+    }
+}
+
+void IAdd64To32(IR::Block& block, IR::Inst& inst) {
+    if (inst.HasAssociatedPseudoOperation()) {
+        throw NotImplementedException("IAdd64 emulation with pseudo instructions");
+    }
+    IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
+    const auto [a_lo, a_hi]{Unpack(ir, inst.Arg(0))};
+    const auto [b_lo, b_hi]{Unpack(ir, inst.Arg(1))};
+
+    const IR::U32 ret_lo{ir.IAdd(a_lo, b_lo)};
+    const IR::U32 carry{ir.Select(ir.GetCarryFromOp(ret_lo), ir.Imm32(1u), ir.Imm32(0u))};
+
+    const IR::U32 ret_hi{ir.IAdd(ir.IAdd(a_hi, b_hi), carry)};
+    inst.ReplaceUsesWith(ir.CompositeConstruct(ret_lo, ret_hi));
+}
+
+void ISub64To32(IR::Block& block, IR::Inst& inst) {
+    if (inst.HasAssociatedPseudoOperation()) {
+        throw NotImplementedException("ISub64 emulation with pseudo instructions");
+    }
+    IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
+    const auto [a_lo, a_hi]{Unpack(ir, inst.Arg(0))};
+    const auto [b_lo, b_hi]{Unpack(ir, inst.Arg(1))};
+
+    const IR::U32 ret_lo{ir.ISub(a_lo, b_lo)};
+    const IR::U1 underflow{ir.IGreaterThan(ret_lo, a_lo, false)};
+    const IR::U32 underflow_bit{ir.Select(underflow, ir.Imm32(1u), ir.Imm32(0u))};
+
+    const IR::U32 ret_hi{ir.ISub(ir.ISub(a_hi, b_hi), underflow_bit)};
+    inst.ReplaceUsesWith(ir.CompositeConstruct(ret_lo, ret_hi));
+}
+
+void INeg64To32(IR::Block& block, IR::Inst& inst) {
+    if (inst.HasAssociatedPseudoOperation()) {
+        throw NotImplementedException("INeg64 emulation with pseudo instructions");
+    }
+    IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
+    auto [lo, hi]{Unpack(ir, inst.Arg(0))};
+    lo = ir.BitwiseNot(lo);
+    hi = ir.BitwiseNot(hi);
+
+    lo = ir.IAdd(lo, ir.Imm32(1));
+
+    const IR::U32 carry{ir.Select(ir.GetCarryFromOp(lo), ir.Imm32(1u), ir.Imm32(0u))};
+    hi = ir.IAdd(hi, carry);
+
+    inst.ReplaceUsesWith(ir.CompositeConstruct(lo, hi));
+}
+
+void ShiftLeftLogical64To32(IR::Block& block, IR::Inst& inst) {
+    if (inst.HasAssociatedPseudoOperation()) {
+        throw NotImplementedException("ShiftLeftLogical64 emulation with pseudo instructions");
+    }
+    IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
+    const auto [lo, hi]{Unpack(ir, inst.Arg(0))};
+    const IR::U32 shift{inst.Arg(1)};
+
+    const IR::U32 shifted_lo{ir.ShiftLeftLogical(lo, shift)};
+    const IR::U32 shifted_hi{ir.ShiftLeftLogical(hi, shift)};
+
+    const IR::U32 inv_shift{ir.ISub(shift, ir.Imm32(32))};
+    const IR::U1 is_long{ir.IGreaterThanEqual(inv_shift, ir.Imm32(0), true)};
+    const IR::U1 is_zero{ir.IEqual(shift, ir.Imm32(0))};
+
+    const IR::U32 long_ret_lo{ir.Imm32(0)};
+    const IR::U32 long_ret_hi{ir.ShiftLeftLogical(lo, inv_shift)};
+
+    const IR::U32 shift_complement{ir.ISub(ir.Imm32(32), shift)};
+    const IR::U32 lo_extract{ir.BitFieldExtract(lo, shift_complement, shift, false)};
+    const IR::U32 short_ret_lo{shifted_lo};
+    const IR::U32 short_ret_hi{ir.BitwiseOr(shifted_hi, lo_extract)};
+
+    const IR::U32 zero_ret_lo{lo};
+    const IR::U32 zero_ret_hi{hi};
+
+    const IR::U32 non_zero_lo{ir.Select(is_long, long_ret_lo, short_ret_lo)};
+    const IR::U32 non_zero_hi{ir.Select(is_long, long_ret_hi, short_ret_hi)};
+
+    const IR::U32 ret_lo{ir.Select(is_zero, zero_ret_lo, non_zero_lo)};
+    const IR::U32 ret_hi{ir.Select(is_zero, zero_ret_hi, non_zero_hi)};
+    inst.ReplaceUsesWith(ir.CompositeConstruct(ret_lo, ret_hi));
+}
+
+void ShiftRightLogical64To32(IR::Block& block, IR::Inst& inst) {
+    if (inst.HasAssociatedPseudoOperation()) {
+        throw NotImplementedException("ShiftRightLogical64 emulation with pseudo instructions");
+    }
+    IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
+    const auto [lo, hi]{Unpack(ir, inst.Arg(0))};
+    const IR::U32 shift{inst.Arg(1)};
+
+    const IR::U32 shifted_lo{ir.ShiftRightLogical(lo, shift)};
+    const IR::U32 shifted_hi{ir.ShiftRightLogical(hi, shift)};
+
+    const IR::U32 inv_shift{ir.ISub(shift, ir.Imm32(32))};
+    const IR::U1 is_long{ir.IGreaterThanEqual(inv_shift, ir.Imm32(0), true)};
+    const IR::U1 is_zero{ir.IEqual(shift, ir.Imm32(0))};
+
+    const IR::U32 long_ret_hi{ir.Imm32(0)};
+    const IR::U32 long_ret_lo{ir.ShiftRightLogical(hi, inv_shift)};
+
+    const IR::U32 shift_complement{ir.ISub(ir.Imm32(32), shift)};
+    const IR::U32 short_hi_extract{ir.BitFieldExtract(hi, ir.Imm32(0), shift)};
+    const IR::U32 short_ret_hi{shifted_hi};
+    const IR::U32 short_ret_lo{
+        ir.BitFieldInsert(shifted_lo, short_hi_extract, shift_complement, shift)};
+
+    const IR::U32 zero_ret_lo{lo};
+    const IR::U32 zero_ret_hi{hi};
+
+    const IR::U32 non_zero_lo{ir.Select(is_long, long_ret_lo, short_ret_lo)};
+    const IR::U32 non_zero_hi{ir.Select(is_long, long_ret_hi, short_ret_hi)};
+
+    const IR::U32 ret_lo{ir.Select(is_zero, zero_ret_lo, non_zero_lo)};
+    const IR::U32 ret_hi{ir.Select(is_zero, zero_ret_hi, non_zero_hi)};
+    inst.ReplaceUsesWith(ir.CompositeConstruct(ret_lo, ret_hi));
+}
+
+void ShiftRightArithmetic64To32(IR::Block& block, IR::Inst& inst) {
+    if (inst.HasAssociatedPseudoOperation()) {
+        throw NotImplementedException("ShiftRightArithmetic64 emulation with pseudo instructions");
+    }
+    IR::IREmitter ir(block, IR::Block::InstructionList::s_iterator_to(inst));
+    const auto [lo, hi]{Unpack(ir, inst.Arg(0))};
+    const IR::U32 shift{inst.Arg(1)};
+
+    const IR::U32 shifted_lo{ir.ShiftRightLogical(lo, shift)};
+    const IR::U32 shifted_hi{ir.ShiftRightArithmetic(hi, shift)};
+
+    const IR::U32 sign_extension{ir.ShiftRightArithmetic(hi, ir.Imm32(31))};
+
+    const IR::U32 inv_shift{ir.ISub(shift, ir.Imm32(32))};
+    const IR::U1 is_long{ir.IGreaterThanEqual(inv_shift, ir.Imm32(0), true)};
+    const IR::U1 is_zero{ir.IEqual(shift, ir.Imm32(0))};
+
+    const IR::U32 long_ret_hi{sign_extension};
+    const IR::U32 long_ret_lo{ir.ShiftRightArithmetic(hi, inv_shift)};
+
+    const IR::U32 shift_complement{ir.ISub(ir.Imm32(32), shift)};
+    const IR::U32 short_hi_extract(ir.BitFieldExtract(hi, ir.Imm32(0), shift));
+    const IR::U32 short_ret_hi{shifted_hi};
+    const IR::U32 short_ret_lo{
+        ir.BitFieldInsert(shifted_lo, short_hi_extract, shift_complement, shift)};
+
+    const IR::U32 zero_ret_lo{lo};
+    const IR::U32 zero_ret_hi{hi};
+
+    const IR::U32 non_zero_lo{ir.Select(is_long, long_ret_lo, short_ret_lo)};
+    const IR::U32 non_zero_hi{ir.Select(is_long, long_ret_hi, short_ret_hi)};
+
+    const IR::U32 ret_lo{ir.Select(is_zero, zero_ret_lo, non_zero_lo)};
+    const IR::U32 ret_hi{ir.Select(is_zero, zero_ret_hi, non_zero_hi)};
+    inst.ReplaceUsesWith(ir.CompositeConstruct(ret_lo, ret_hi));
+}
+
+void Lower(IR::Block& block, IR::Inst& inst) {
+    switch (inst.GetOpcode()) {
+    case IR::Opcode::PackUint2x32:
+    case IR::Opcode::UnpackUint2x32:
+        return inst.ReplaceOpcode(IR::Opcode::Identity);
+    case IR::Opcode::IAdd64:
+        return IAdd64To32(block, inst);
+    case IR::Opcode::ISub64:
+        return ISub64To32(block, inst);
+    case IR::Opcode::INeg64:
+        return INeg64To32(block, inst);
+    case IR::Opcode::ShiftLeftLogical64:
+        return ShiftLeftLogical64To32(block, inst);
+    case IR::Opcode::ShiftRightLogical64:
+        return ShiftRightLogical64To32(block, inst);
+    case IR::Opcode::ShiftRightArithmetic64:
+        return ShiftRightArithmetic64To32(block, inst);
+    default:
+        break;
+    }
+}
+} // Anonymous namespace
+
+void LowerInt64ToInt32(IR::Program& program) {
+    for (IR::Block* const block : program.post_order_blocks | std::views::reverse) {
+        for (IR::Inst& inst : block->Instructions()) {
+            Lower(*block, inst);
+        }
+    }
+}
+
+} // namespace Shader::Optimization
--- a/src/shader_recompiler/ir_opt/passes.h
+++ b/src/shader_recompiler/ir_opt/passes.h
@ -18,6 +18,7 @@ void DeadCodeEliminationPass(IR::Program& program);
 void GlobalMemoryToStorageBufferPass(IR::Program& program);
 void IdentityRemovalPass(IR::Program& program);
 void LowerFp16ToFp32(IR::Program& program);
+void LowerInt64ToInt32(IR::Program& program);
 void SsaRewritePass(IR::Program& program);
 void TexturePass(Environment& env, IR::Program& program);
 void VerificationPass(const IR::Program& program);