[llvm] [GlobalISel] Make sure to check for load barriers when merging G_EXTRACT_VECTOR_ELT into G_LOAD. (PR #82306)
Owen Anderson via llvm-commits
llvm-commits at lists.llvm.org
Tue Feb 20 19:34:44 PST 2024
https://github.com/resistor updated https://github.com/llvm/llvm-project/pull/82306
>From 0409fd1ae27e65ea4a06c3d19883c9b1b7ba35d9 Mon Sep 17 00:00:00 2001
From: Owen Anderson <resistor at mac.com>
Date: Tue, 20 Feb 2024 00:03:19 -0500
Subject: [PATCH] [GlobalISel] Make sure to check for load barriers when
merging G_EXTRACT_VECTOR_ELT into G_LOAD.
Fixes https://github.com/llvm/llvm-project/issues/78477
---
.../lib/CodeGen/GlobalISel/CombinerHelper.cpp | 12 +++++
.../CodeGen/AArch64/extractvector-of-load.mir | 46 +++++++++++++++++++
2 files changed, 58 insertions(+)
create mode 100644 llvm/test/CodeGen/AArch64/extractvector-of-load.mir
diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
index b400eb34e2901b..45427fc2466bd1 100644
--- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
@@ -1198,6 +1198,18 @@ bool CombinerHelper::matchCombineExtractedVectorLoad(MachineInstr &MI,
if (!VecEltTy.isByteSized())
return false;
+ // Check for load fold barriers between the extraction and the load.
+ if (MI.getParent() != LoadMI->getParent())
+ return false;
+ const unsigned MaxIter = 20;
+ unsigned Iter = 0;
+ for (auto II = LoadMI->getIterator(), IE = MI.getIterator(); II != IE; ++II) {
+ if (II->isLoadFoldBarrier())
+ return false;
+ if (Iter++ == MaxIter)
+ return false;
+ }
+
// Check if the new load that we are going to create is legal
// if we are in the post-legalization phase.
MachineMemOperand MMO = LoadMI->getMMO();
diff --git a/llvm/test/CodeGen/AArch64/extractvector-of-load.mir b/llvm/test/CodeGen/AArch64/extractvector-of-load.mir
new file mode 100644
index 00000000000000..43051232b436df
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/extractvector-of-load.mir
@@ -0,0 +1,46 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 4
+# RUN: llc -mtriple=aarch64-linux-gnu -run-pass=aarch64-prelegalizer-combiner -verify-machineinstrs %s -o - | FileCheck %s
+
+---
+name: f
+alignment: 4
+tracksRegLiveness: true
+registers:
+ - { id: 0, class: _ }
+ - { id: 1, class: _ }
+ - { id: 2, class: _ }
+ - { id: 3, class: _ }
+ - { id: 4, class: _ }
+ - { id: 5, class: _ }
+liveins:
+ - { reg: '$x0' }
+frameInfo:
+ maxAlignment: 1
+machineFunctionInfo: {}
+body: |
+ bb.0:
+ liveins: $x0
+
+ ; CHECK-LABEL: name: f
+ ; CHECK: liveins: $x0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[C]](s32), [[C]](s32), [[C]](s32), [[C]](s32)
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
+ ; CHECK-NEXT: [[LOAD:%[0-9]+]]:_(<2 x s32>) = G_LOAD [[COPY]](p0) :: (load (<2 x s32>))
+ ; CHECK-NEXT: G_STORE [[BUILD_VECTOR]](<4 x s32>), [[COPY]](p0) :: (store (<4 x s32>))
+ ; CHECK-NEXT: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[LOAD]](<2 x s32>), [[C1]](s64)
+ ; CHECK-NEXT: $w0 = COPY [[EVEC]](s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0
+ %0:_(p0) = COPY $x0
+ %3:_(s32) = G_CONSTANT i32 0
+ %2:_(<4 x s32>) = G_BUILD_VECTOR %3(s32), %3(s32), %3(s32), %3(s32)
+ %5:_(s64) = G_CONSTANT i64 0
+ %1:_(<2 x s32>) = G_LOAD %0(p0) :: (load (<2 x s32>))
+ G_STORE %2(<4 x s32>), %0(p0) :: (store (<4 x s32>))
+ %4:_(s32) = G_EXTRACT_VECTOR_ELT %1(<2 x s32>), %5(s64)
+ $w0 = COPY %4(s32)
+ RET_ReallyLR implicit $w0
+
+...
More information about the llvm-commits
mailing list