Revert "[AArch64][Machine-Combiner] Split gather patterns into neon regs to multiple vectors (#142941)" #150505
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…egs to multiple vectors (#142941)" Reverting due to reported miscompiles, will reland with a fix.
Member
|
@llvm/pr-subscribers-backend-aarch64 Author: Jonathan Cohen (jcohen-apple) ChangesReverting due to reported miscompiles, will reland once it is fixed. Patch is 67.56 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/150505.diff 10 Files Affected:
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
index 802e4a973cdc3..8685d7a04ac9c 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
@@ -20,7 +20,6 @@
#include "Utils/AArch64BaseInfo.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/CodeGen/CFIInstBuilder.h"
#include "llvm/CodeGen/LivePhysRegs.h"
@@ -36,7 +35,6 @@
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/RegisterScavenging.h"
#include "llvm/CodeGen/StackMaps.h"
-#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/IR/DebugInfoMetadata.h"
@@ -7354,9 +7352,6 @@ bool AArch64InstrInfo::isThroughputPattern(unsigned Pattern) const {
case AArch64MachineCombinerPattern::MULSUBv2i32_indexed_OP2:
case AArch64MachineCombinerPattern::MULSUBv4i32_indexed_OP1:
case AArch64MachineCombinerPattern::MULSUBv4i32_indexed_OP2:
- case AArch64MachineCombinerPattern::GATHER_LANE_i32:
- case AArch64MachineCombinerPattern::GATHER_LANE_i16:
- case AArch64MachineCombinerPattern::GATHER_LANE_i8:
return true;
} // end switch (Pattern)
return false;
@@ -7397,252 +7392,11 @@ static bool getMiscPatterns(MachineInstr &Root,
return false;
}
-static bool getGatherPattern(MachineInstr &Root,
- SmallVectorImpl<unsigned> &Patterns,
- unsigned LoadLaneOpCode, unsigned NumLanes) {
- const MachineFunction *MF = Root.getMF();
-
- // Early exit if optimizing for size.
- if (MF->getFunction().hasMinSize())
- return false;
-
- const MachineRegisterInfo &MRI = MF->getRegInfo();
- const TargetRegisterInfo *TRI = MF->getSubtarget().getRegisterInfo();
-
- // The root of the pattern must load into the last lane of the vector.
- if (Root.getOperand(2).getImm() != NumLanes - 1)
- return false;
-
- // Check that we have load into all lanes except lane 0.
- // For each load we also want to check that:
- // 1. It has a single non-debug use (since we will be replacing the virtual
- // register)
- // 2. That the addressing mode only uses a single offset register.
- auto *CurrInstr = MRI.getUniqueVRegDef(Root.getOperand(1).getReg());
- auto Range = llvm::seq<unsigned>(1, NumLanes - 1);
- SmallSet<unsigned, 4> RemainingLanes(Range.begin(), Range.end());
- while (!RemainingLanes.empty() && CurrInstr &&
- CurrInstr->getOpcode() == LoadLaneOpCode &&
- MRI.hasOneNonDBGUse(CurrInstr->getOperand(0).getReg()) &&
- CurrInstr->getNumOperands() == 4) {
- RemainingLanes.erase(CurrInstr->getOperand(2).getImm());
- CurrInstr = MRI.getUniqueVRegDef(CurrInstr->getOperand(1).getReg());
- }
-
- if (!RemainingLanes.empty())
- return false;
-
- // Match the SUBREG_TO_REG sequence.
- if (CurrInstr->getOpcode() != TargetOpcode::SUBREG_TO_REG)
- return false;
-
- // Verify that the subreg to reg loads an integer into the first lane.
- auto Lane0LoadReg = CurrInstr->getOperand(2).getReg();
- unsigned SingleLaneSizeInBits = 128 / NumLanes;
- if (TRI->getRegSizeInBits(Lane0LoadReg, MRI) != SingleLaneSizeInBits)
- return false;
-
- // Verify that it also has a single non debug use.
- if (!MRI.hasOneNonDBGUse(Lane0LoadReg))
- return false;
-
- switch (NumLanes) {
- case 4:
- Patterns.push_back(AArch64MachineCombinerPattern::GATHER_LANE_i32);
- break;
- case 8:
- Patterns.push_back(AArch64MachineCombinerPattern::GATHER_LANE_i16);
- break;
- case 16:
- Patterns.push_back(AArch64MachineCombinerPattern::GATHER_LANE_i8);
- break;
- default:
- llvm_unreachable("Got bad number of lanes for gather pattern.");
- }
-
- return true;
-}
-
-/// Search for patterns where we use LD1 instructions to load into
-/// separate lanes of an 128 bit Neon register. We can increase Memory Level
-/// Parallelism by loading into 2 Neon registers instead.
-static bool getLoadPatterns(MachineInstr &Root,
- SmallVectorImpl<unsigned> &Patterns) {
-
- // The pattern searches for loads into single lanes.
- switch (Root.getOpcode()) {
- case AArch64::LD1i32:
- return getGatherPattern(Root, Patterns, Root.getOpcode(), 4);
- case AArch64::LD1i16:
- return getGatherPattern(Root, Patterns, Root.getOpcode(), 8);
- case AArch64::LD1i8:
- return getGatherPattern(Root, Patterns, Root.getOpcode(), 16);
- default:
- return false;
- }
-}
-
-static void
-generateGatherPattern(MachineInstr &Root,
- SmallVectorImpl<MachineInstr *> &InsInstrs,
- SmallVectorImpl<MachineInstr *> &DelInstrs,
- DenseMap<Register, unsigned> &InstrIdxForVirtReg,
- unsigned Pattern, unsigned NumLanes) {
-
- MachineFunction &MF = *Root.getParent()->getParent();
- MachineRegisterInfo &MRI = MF.getRegInfo();
- const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
-
- // Gather the initial load instructions to build the pattern
- SmallVector<MachineInstr *, 16> LoadToLaneInstrs;
- MachineInstr *CurrInstr = &Root;
- for (unsigned i = 0; i < NumLanes - 1; ++i) {
- LoadToLaneInstrs.push_back(CurrInstr);
- CurrInstr = MRI.getUniqueVRegDef(CurrInstr->getOperand(1).getReg());
- }
-
- // Sort the load instructions according to the lane.
- llvm::sort(LoadToLaneInstrs,
- [](const MachineInstr *A, const MachineInstr *B) {
- return A->getOperand(2).getImm() > B->getOperand(2).getImm();
- });
-
- MachineInstr *SubregToReg = CurrInstr;
- LoadToLaneInstrs.push_back(
- MRI.getUniqueVRegDef(SubregToReg->getOperand(2).getReg()));
- auto LoadToLaneInstrsAscending = llvm::reverse(LoadToLaneInstrs);
-
- const TargetRegisterClass *FPR128RegClass =
- MRI.getRegClass(Root.getOperand(0).getReg());
-
- auto LoadLaneToRegister = [&](MachineInstr *OriginalInstr,
- Register SrcRegister, unsigned Lane,
- Register OffsetRegister) {
- auto NewRegister = MRI.createVirtualRegister(FPR128RegClass);
- MachineInstrBuilder LoadIndexIntoRegister =
- BuildMI(MF, MIMetadata(*OriginalInstr), TII->get(Root.getOpcode()),
- NewRegister)
- .addReg(SrcRegister)
- .addImm(Lane)
- .addReg(OffsetRegister, getKillRegState(true));
- InstrIdxForVirtReg.insert(std::make_pair(NewRegister, InsInstrs.size()));
- InsInstrs.push_back(LoadIndexIntoRegister);
- return NewRegister;
- };
-
- // Helper to create load instruction based on opcode
- auto CreateLoadInstruction = [&](unsigned NumLanes, Register DestReg,
- Register OffsetReg) -> MachineInstrBuilder {
- unsigned Opcode;
- switch (NumLanes) {
- case 4:
- Opcode = AArch64::LDRSui;
- break;
- case 8:
- Opcode = AArch64::LDRHui;
- break;
- case 16:
- Opcode = AArch64::LDRBui;
- break;
- default:
- llvm_unreachable(
- "Got unsupported number of lanes in machine-combiner gather pattern");
- }
- // Immediate offset load
- return BuildMI(MF, MIMetadata(Root), TII->get(Opcode), DestReg)
- .addReg(OffsetReg)
- .addImm(0); // immediate offset
- };
-
- // Load the remaining lanes into register 0.
- auto LanesToLoadToReg0 =
- llvm::make_range(LoadToLaneInstrsAscending.begin() + 1,
- LoadToLaneInstrsAscending.begin() + NumLanes / 2);
- auto PrevReg = SubregToReg->getOperand(0).getReg();
- for (auto [Index, LoadInstr] : llvm::enumerate(LanesToLoadToReg0)) {
- PrevReg = LoadLaneToRegister(LoadInstr, PrevReg, Index + 1,
- LoadInstr->getOperand(3).getReg());
- DelInstrs.push_back(LoadInstr);
- }
- auto LastLoadReg0 = PrevReg;
-
- // First load into register 1. Perform a LDRSui to zero out the upper lanes in
- // a single instruction.
- auto Lane0Load = *LoadToLaneInstrsAscending.begin();
- auto OriginalSplitLoad =
- *std::next(LoadToLaneInstrsAscending.begin(), NumLanes / 2);
- auto DestRegForMiddleIndex = MRI.createVirtualRegister(
- MRI.getRegClass(Lane0Load->getOperand(0).getReg()));
-
- MachineInstrBuilder MiddleIndexLoadInstr =
- CreateLoadInstruction(NumLanes, DestRegForMiddleIndex,
- OriginalSplitLoad->getOperand(3).getReg());
-
- InstrIdxForVirtReg.insert(
- std::make_pair(DestRegForMiddleIndex, InsInstrs.size()));
- InsInstrs.push_back(MiddleIndexLoadInstr);
- DelInstrs.push_back(OriginalSplitLoad);
-
- // Subreg To Reg instruction for register 1.
- auto DestRegForSubregToReg = MRI.createVirtualRegister(FPR128RegClass);
- unsigned SubregType;
- switch (NumLanes) {
- case 4:
- SubregType = AArch64::ssub;
- break;
- case 8:
- SubregType = AArch64::hsub;
- break;
- case 16:
- SubregType = AArch64::bsub;
- break;
- default:
- llvm_unreachable(
- "Got invalid NumLanes for machine-combiner gather pattern");
- }
-
- auto SubRegToRegInstr =
- BuildMI(MF, MIMetadata(Root), TII->get(SubregToReg->getOpcode()),
- DestRegForSubregToReg)
- .addImm(0)
- .addReg(DestRegForMiddleIndex, getKillRegState(true))
- .addImm(SubregType);
- InstrIdxForVirtReg.insert(
- std::make_pair(DestRegForSubregToReg, InsInstrs.size()));
- InsInstrs.push_back(SubRegToRegInstr);
-
- // Load remaining lanes into register 1.
- auto LanesToLoadToReg1 =
- llvm::make_range(LoadToLaneInstrsAscending.begin() + NumLanes / 2 + 1,
- LoadToLaneInstrsAscending.end());
- PrevReg = SubRegToRegInstr->getOperand(0).getReg();
- for (auto [Index, LoadInstr] : llvm::enumerate(LanesToLoadToReg1)) {
- PrevReg = LoadLaneToRegister(LoadInstr, PrevReg, Index + 1,
- LoadInstr->getOperand(3).getReg());
- if (Index == NumLanes / 2 - 2) {
- break;
- }
- DelInstrs.push_back(LoadInstr);
- }
- auto LastLoadReg1 = PrevReg;
-
- // Create the final zip instruction to combine the results.
- MachineInstrBuilder ZipInstr =
- BuildMI(MF, MIMetadata(Root), TII->get(AArch64::ZIP1v2i64),
- Root.getOperand(0).getReg())
- .addReg(LastLoadReg0)
- .addReg(LastLoadReg1);
- InsInstrs.push_back(ZipInstr);
-}
-
CombinerObjective
AArch64InstrInfo::getCombinerObjective(unsigned Pattern) const {
switch (Pattern) {
case AArch64MachineCombinerPattern::SUBADD_OP1:
case AArch64MachineCombinerPattern::SUBADD_OP2:
- case AArch64MachineCombinerPattern::GATHER_LANE_i32:
- case AArch64MachineCombinerPattern::GATHER_LANE_i16:
- case AArch64MachineCombinerPattern::GATHER_LANE_i8:
return CombinerObjective::MustReduceDepth;
default:
return TargetInstrInfo::getCombinerObjective(Pattern);
@@ -7672,10 +7426,6 @@ bool AArch64InstrInfo::getMachineCombinerPatterns(
if (getMiscPatterns(Root, Patterns))
return true;
- // Load patterns
- if (getLoadPatterns(Root, Patterns))
- return true;
-
return TargetInstrInfo::getMachineCombinerPatterns(Root, Patterns,
DoRegPressureReduce);
}
@@ -8931,21 +8681,6 @@ void AArch64InstrInfo::genAlternativeCodeSequence(
MUL = genFNegatedMAD(MF, MRI, TII, Root, InsInstrs);
break;
}
- case AArch64MachineCombinerPattern::GATHER_LANE_i32: {
- generateGatherPattern(Root, InsInstrs, DelInstrs, InstrIdxForVirtReg,
- Pattern, 4);
- break;
- }
- case AArch64MachineCombinerPattern::GATHER_LANE_i16: {
- generateGatherPattern(Root, InsInstrs, DelInstrs, InstrIdxForVirtReg,
- Pattern, 8);
- break;
- }
- case AArch64MachineCombinerPattern::GATHER_LANE_i8: {
- generateGatherPattern(Root, InsInstrs, DelInstrs, InstrIdxForVirtReg,
- Pattern, 16);
- break;
- }
} // end switch (Pattern)
// Record MUL and ADD/SUB for deletion
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.h b/llvm/lib/Target/AArch64/AArch64InstrInfo.h
index 02734866e7122..7c255da333e4b 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.h
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.h
@@ -172,10 +172,6 @@ enum AArch64MachineCombinerPattern : unsigned {
FMULv8i16_indexed_OP2,
FNMADD,
-
- GATHER_LANE_i32,
- GATHER_LANE_i16,
- GATHER_LANE_i8
};
class AArch64InstrInfo final : public AArch64GenInstrInfo {
const AArch64RegisterInfo RI;
diff --git a/llvm/test/CodeGen/AArch64/aarch64-combine-gather-lanes.mir b/llvm/test/CodeGen/AArch64/aarch64-combine-gather-lanes.mir
deleted file mode 100644
index 09eb18b0e3574..0000000000000
--- a/llvm/test/CodeGen/AArch64/aarch64-combine-gather-lanes.mir
+++ /dev/null
@@ -1,364 +0,0 @@
-# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
-# RUN: llc -run-pass=machine-combiner -mcpu=neoverse-n2 -mtriple=aarch64-none-linux-gnu -verify-machineinstrs %s -o - | FileCheck %s
-
----
-name: split_loads_to_fpr128
-body: |
- bb.0.entry:
- liveins: $x0, $x1, $x2, $x3, $x4
-
- ; CHECK-LABEL: name: split_loads_to_fpr128
- ; CHECK: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
- ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64common = COPY $x1
- ; CHECK-NEXT: [[COPY2:%[0-9]+]]:gpr64common = COPY $x2
- ; CHECK-NEXT: [[COPY3:%[0-9]+]]:gpr64common = COPY $x3
- ; CHECK-NEXT: [[COPY4:%[0-9]+]]:gpr64common = COPY $x4
- ; CHECK-NEXT: [[LD_i32:%[0-9]+]]:fpr32 = LDRSroX [[COPY]], killed [[COPY1]], 0, 1
- ; CHECK-NEXT: [[FIRST_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, killed [[LD_i32]], %subreg.ssub
- ; CHECK-NEXT: [[LD0_1:%[0-9]+]]:fpr128 = LD1i32 [[FIRST_REG]], 1, killed [[COPY2]]
- ; CHECK-NEXT: [[LD1_0:%[0-9]+]]:fpr32 = LDRSui [[COPY3]], 0
- ; CHECK-NEXT: [[SECOND_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, killed [[LD1_0]], %subreg.ssub
- ; CHECK-NEXT: [[LD1_1:%[0-9]+]]:fpr128 = LD1i32 [[SECOND_REG]], 1, killed [[COPY4]]
- ; CHECK-NEXT: [[ZIP:%[0-9]+]]:fpr128 = ZIP1v2i64 [[LD0_1]], [[LD1_1]]
- ; CHECK-NEXT: $q0 = COPY [[ZIP]]
- ; CHECK-NEXT: RET_ReallyLR implicit $q0
- %0:gpr64common = COPY $x0
- %1:gpr64common = COPY $x1
- %2:gpr64common = COPY $x2
- %3:gpr64common = COPY $x3
- %4:gpr64common = COPY $x4
- %5:fpr32 = LDRSroX %0, killed %1, 0, 1
- %6:fpr128 = SUBREG_TO_REG 0, killed %5, %subreg.ssub
- %7:fpr128 = LD1i32 %6, 1, killed %2
- %8:fpr128 = LD1i32 %7, 2, killed %3
- %9:fpr128 = LD1i32 %8, 3, killed %4
- $q0 = COPY %9
- RET_ReallyLR implicit $q0
-
----
-name: split_loads_to_fpr128_ui
-body: |
- bb.0.entry:
- liveins: $x0, $x1, $x2, $x3, $x4
-
- ; CHECK-LABEL: name: split_loads_to_fpr128_ui
- ; CHECK: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
- ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64common = COPY $x1
- ; CHECK-NEXT: [[COPY2:%[0-9]+]]:gpr64common = COPY $x2
- ; CHECK-NEXT: [[COPY3:%[0-9]+]]:gpr64common = COPY $x3
- ; CHECK-NEXT: [[COPY4:%[0-9]+]]:gpr64common = COPY $x4
- ; CHECK-NEXT: [[LD_i32:%[0-9]+]]:fpr32 = LDRSui [[COPY]], 0
- ; CHECK-NEXT: [[FIRST_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, killed [[LD_i32]], %subreg.ssub
- ; CHECK-NEXT: [[LD0_1:%[0-9]+]]:fpr128 = LD1i32 [[FIRST_REG]], 1, killed [[COPY1]]
- ; CHECK-NEXT: [[LD1_0:%[0-9]+]]:fpr32 = LDRSui [[COPY2]], 0
- ; CHECK-NEXT: [[SECOND_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, killed [[LD1_0]], %subreg.ssub
- ; CHECK-NEXT: [[LD1_1:%[0-9]+]]:fpr128 = LD1i32 [[SECOND_REG]], 1, killed [[COPY3]]
- ; CHECK-NEXT: [[ZIP:%[0-9]+]]:fpr128 = ZIP1v2i64 [[LD0_1]], [[LD1_1]]
- ; CHECK-NEXT: $q0 = COPY [[ZIP]]
- ; CHECK-NEXT: RET_ReallyLR implicit $q0
- %0:gpr64common = COPY $x0
- %1:gpr64common = COPY $x1
- %2:gpr64common = COPY $x2
- %3:gpr64common = COPY $x3
- %4:gpr64common = COPY $x4
- %5:fpr32 = LDRSui %0, 0
- %6:fpr128 = SUBREG_TO_REG 0, killed %5, %subreg.ssub
- %7:fpr128 = LD1i32 %6, 1, killed %1
- %8:fpr128 = LD1i32 %7, 2, killed %2
- %9:fpr128 = LD1i32 %8, 3, killed %3
- $q0 = COPY %9
- RET_ReallyLR implicit $q0
-
----
-name: split_loads_to_fpr128_i16
-body: |
- bb.0.entry:
- liveins: $x0, $x1, $x2, $x3, $x4, $x5, $x6, $x7, $x8
-
- ; CHECK-LABEL: name: split_loads_to_fpr128_i16
- ; CHECK: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
- ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64common = COPY $x1
- ; CHECK-NEXT: [[COPY2:%[0-9]+]]:gpr64common = COPY $x2
- ; CHECK-NEXT: [[COPY3:%[0-9]+]]:gpr64common = COPY $x3
- ; CHECK-NEXT: [[COPY4:%[0-9]+]]:gpr64common = COPY $x4
- ; CHECK-NEXT: [[COPY5:%[0-9]+]]:gpr64common = COPY $x5
- ; CHECK-NEXT: [[COPY6:%[0-9]+]]:gpr64common = COPY $x6
- ; CHECK-NEXT: [[COPY7:%[0-9]+]]:gpr64common = COPY $x7
- ; CHECK-NEXT: [[COPY8:%[0-9]+]]:gpr64common = COPY $x8
- ; CHECK-NEXT: [[LD_i16:%[0-9]+]]:fpr16 = LDRHroX [[COPY]], killed [[COPY1]], 0, 1
- ; CHECK-NEXT: [[FIRST_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, killed [[LD_i16]], %subreg.hsub
- ; CHECK-NEXT: [[LD0_1:%[0-9]+]]:fpr128 = LD1i16 [[FIRST_REG]], 1, killed [[COPY2]]
- ; CHECK-NEXT: [[LD0_2:%[0-9]+]]:fpr128 = LD1i16 [[LD0_1]], 2, killed [[COPY3]]
- ; CHECK-NEXT: [[LD0_3:%[0-9]+]]:fpr128 = LD1i16 [[LD0_2]], 3, killed [[COPY4]]
- ; CHECK-NEXT: [[LD1_0:%[0-9]+]]:fpr16 = LDRHui [[COPY5]], 0
- ; CHECK-NEXT: [[SECOND_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, killed [[LD1_0]], %subreg.hsub
- ; CHECK-NEXT: [[LD1_1:%[0-9]+]]:fpr128 = LD1i16 [[SECOND_REG]], 1, killed [[COPY6]]
- ; CHECK-NEXT: [[LD1_2:%[0-9]+]]:fpr128 = LD1i16 [[LD1_1]], 2, killed [[COPY7]]
- ; CHECK-NEXT: [[LD1_3:%[0-9]+]]:fpr128 = LD1i16 [[LD1_2]], 3, killed [[COPY8]]
- ; CHECK-NEXT: [[ZIP:%[0-9]+]]:fpr128 = ZIP1v2i64 [[LD0_3]], [[LD1_3]]
- ; CHECK-NEXT: $q0 = COPY [[ZIP]]
- ; CHECK-NEXT: RET_ReallyLR implicit $q0
- %0:gpr64common = COPY $x0
- %1:gpr64common = COPY $x1
- %2:gpr64common = COPY $x2
- %3:gpr64common = COPY $x3
- %4:gpr64common = COPY $x4
- %5:gpr64common = COPY $x5
- %6:gpr64common = COPY $x6
- %7:gpr64common = COPY $x7
- %8:gpr64common = COPY $x8
- %9:fpr16 = LDRHroX %0, killed %1, 0, 1
- %10:fpr128 = SUBREG_TO_REG 0, killed %9, %subreg.hsub
- %11:fpr128 = LD1i16 %10, 1, killed %2
- %12:fpr128 = LD1i16 %11, 2, killed %3
- %13:fpr128 = LD1i16 %12, 3, killed %4
- %14:fpr128 = LD1i16 %13, 4, killed %5
- %15:fpr128 = LD1i16 %14, 5, killed %6
- %16:fpr128 = LD1i16 %15, 6, killed %7
- %17:fpr128 = LD1i16 %16, 7, killed %8
- $q0 = COPY %17
- RET_ReallyLR implicit $q0
-
----
-name: split_loads_to_fpr128_i16_ui
-body: |
- bb.0.entry:
- liveins: $x0, $x1, $x2, $x3, $x4, $x5, $x6, $x7, $x8
-
- ; CHECK-LABEL: name: split_loads_to_fpr128_i16_ui
- ; CHECK: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
- ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64common = COPY $x1
- ; CHECK-NEXT: [[COPY2:%[0-9]+]]:gpr64common = COPY $x2
- ; CHECK-NEXT: [[COPY3:%[0-9]+]]:gpr64common = COPY $x3
- ; CHECK-NEXT: [[COPY4:%[0-9]+]]:gpr64common = COPY $x4
- ; CHECK-NEXT: [[COPY5:%[0-9]+]]:gpr64common = COPY $x5
- ; CHECK-NEXT: [[COPY6:%[0-9]+]]:gpr64common = COPY $x6
- ; CHECK-NEXT: [[COPY7:%[0-9]+]]:gpr64common = COPY $x7
- ; CHECK-NEXT: [[COPY8:%[0-9]+]]:gpr64common = COPY $x8
- ; CHECK-NEXT: [[LD_i16:%[0-9]+]]:fpr16 = LDRHui [[COPY]], 0
- ; CHECK-NEXT: [[FIRST_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, killed [[LD_i16]], %subreg.hsub
- ; CHECK-NEXT: [[LD0_1:%[0-9]+]]:fpr128 = LD1i16 [[FIRST_REG]], 1, killed [[COPY1]]
- ; CHECK-NEXT: [[LD0_2:%[0-9]+]]:fpr128 = LD1i16 [[LD0_1]], 2, killed [[COPY2]]
- ; CHECK-NEXT: [[LD0_3:%[0-9]+]]:fpr128 = LD1i16 [[LD0_2]], 3, killed [[COPY3]]
- ; CHECK-NEXT: [[LD1_0:%[0-9]+]]:fpr16 = LDRHui [[COPY4]], 0
- ; CHECK-NEXT: [[SECOND_REG:%[0-9]+]]:fpr128 = SUBREG_TO_REG 0, killed [[LD1_0]], %subreg.hsub
- ; CHECK-NEXT: [[LD1_1:%[0-9]+]]:fpr128 = LD1i16 [[SECOND_REG]], 1, killed [[COPY5]]
- ; CHECK-NEXT: [[LD1_2:%[0-9]+]]:fpr128 = LD1i16 [[LD1_1]], 2, killed [[COPY6]]
- ; CHECK-NEXT: [[LD1_3:%[0-9]+]]:fpr128 = LD1i16 [[LD1_2]], 3, killed [[COPY7]]
- ; CHECK-NEXT: [[ZIP:%[0-9]+]]:fpr128 = ZIP1v2i64 [[LD0_3]], [[LD1_3]]
- ; CH...
[truncated]
|
davemgreen
approved these changes
Jul 24, 2025
Collaborator
davemgreen
left a comment
davemgreen
left a comment
There was a problem hiding this comment.
LGTM, thanks for doing this. (And reverts don't usually need review unless they might be contentious).
mahesh-attarde
pushed a commit
to mahesh-attarde/llvm-project
that referenced
this pull request
Jul 28, 2025
…egs to multiple vectors (llvm#142941)" (llvm#150505) Reverting due to reported miscompiles, will reland once it is fixed.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Reverting due to reported miscompiles, will reland once it is fixed.