https://github.com/c-rhodes updated
https://github.com/llvm/llvm-project/pull/187651
>From 5f04300518b574249f116c5434d76590a1dcdf2a Mon Sep 17 00:00:00 2001
From: David Green
Date: Fri, 20 Mar 2026 07:28:16 +
Subject: [PATCH] [AArch64] Use an unknown size for memcpy ops with
non-constant sizes. (#187445)
The previous value of 0 was allowing loads to move past the mops
operations where it is not valid. Use a LocationSize::afterPointer()
size instead.
The GISel lowering currently loses the MMO, which is fine as it should
be conservatively treated as a load/store to any location.
(cherry picked from commit 441790b31f06ed2300e0e8d74f3bdf446bc105aa)
---
.../AArch64/AArch64SelectionDAGInfo.cpp | 8 +++---
llvm/test/CodeGen/AArch64/mops-mmo-size.ll| 28 +++
2 files changed, 32 insertions(+), 4 deletions(-)
create mode 100644 llvm/test/CodeGen/AArch64/mops-mmo-size.ll
diff --git a/llvm/lib/Target/AArch64/AArch64SelectionDAGInfo.cpp
b/llvm/lib/Target/AArch64/AArch64SelectionDAGInfo.cpp
index df8ba2c8e922f..f236118fd6dce 100644
--- a/llvm/lib/Target/AArch64/AArch64SelectionDAGInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64SelectionDAGInfo.cpp
@@ -123,9 +123,9 @@ SDValue AArch64SelectionDAGInfo::EmitMOPS(unsigned Opcode,
SelectionDAG &DAG,
MachinePointerInfo SrcPtrInfo) const
{
// Get the constant size of the copy/set.
- uint64_t ConstSize = 0;
+ LocationSize MemSize = LocationSize::afterPointer();
if (auto *C = dyn_cast(Size))
-ConstSize = C->getZExtValue();
+MemSize = LocationSize::precise(C->getZExtValue());
const bool IsSet = Opcode == AArch64::MOPSMemorySetPseudo ||
Opcode == AArch64::MOPSMemorySetTaggingPseudo;
@@ -136,7 +136,7 @@ SDValue AArch64SelectionDAGInfo::EmitMOPS(unsigned Opcode,
SelectionDAG &DAG,
isVolatile ? MachineMemOperand::MOVolatile : MachineMemOperand::MONone;
auto DstFlags = MachineMemOperand::MOStore | Vol;
auto *DstOp =
- MF.getMachineMemOperand(DstPtrInfo, DstFlags, ConstSize, Alignment);
+ MF.getMachineMemOperand(DstPtrInfo, DstFlags, MemSize, Alignment);
if (IsSet) {
// Extend value to i64, if required.
@@ -154,7 +154,7 @@ SDValue AArch64SelectionDAGInfo::EmitMOPS(unsigned Opcode,
SelectionDAG &DAG,
auto SrcFlags = MachineMemOperand::MOLoad | Vol;
auto *SrcOp =
-MF.getMachineMemOperand(SrcPtrInfo, SrcFlags, ConstSize, Alignment);
+MF.getMachineMemOperand(SrcPtrInfo, SrcFlags, MemSize, Alignment);
DAG.setNodeMemRefs(Node, {DstOp, SrcOp});
return SDValue(Node, 3);
}
diff --git a/llvm/test/CodeGen/AArch64/mops-mmo-size.ll
b/llvm/test/CodeGen/AArch64/mops-mmo-size.ll
new file mode 100644
index 0..1d203c7cdb955
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/mops-mmo-size.ll
@@ -0,0 +1,28 @@
+; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
UTC_ARGS: --version 6
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mattr=+mops
-stop-after=aarch64-isel -o - %s | FileCheck %s --check-prefix=CHECK-SD
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mattr=+mops -global-isel
-stop-after=finalize-isel -o - %s | FileCheck %s --check-prefix=CHECK-GI
+
+define void @test(ptr %out, ptr %in, i64 %a) {
+ ; CHECK-SD-LABEL: name: test
+ ; CHECK-SD: bb.0.entry:
+ ; CHECK-SD-NEXT: liveins: $x0, $x1, $x2
+ ; CHECK-SD-NEXT: {{ $}}
+ ; CHECK-SD-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $x2
+ ; CHECK-SD-NEXT: [[COPY1:%[0-9]+]]:gpr64common = COPY $x1
+ ; CHECK-SD-NEXT: [[COPY2:%[0-9]+]]:gpr64common = COPY $x0
+ ; CHECK-SD-NEXT: [[MOPSMemoryMovePseudo:%[0-9]+]]:gpr64common,
[[MOPSMemoryMovePseudo1:%[0-9]+]]:gpr64common,
[[MOPSMemoryMovePseudo2:%[0-9]+]]:gpr64 = MOPSMemoryMovePseudo [[COPY2]],
[[COPY1]], [[COPY]], implicit-def dead $nzcv :: (store unknown-size into
%ir.out, align 1), (load unknown-size from %ir.in, align 1)
+ ; CHECK-SD-NEXT: RET_ReallyLR
+ ;
+ ; CHECK-GI-LABEL: name: test
+ ; CHECK-GI: bb.1.entry:
+ ; CHECK-GI-NEXT: liveins: $x0, $x1, $x2
+ ; CHECK-GI-NEXT: {{ $}}
+ ; CHECK-GI-NEXT: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
+ ; CHECK-GI-NEXT: [[COPY1:%[0-9]+]]:gpr64common = COPY $x1
+ ; CHECK-GI-NEXT: [[COPY2:%[0-9]+]]:gpr64 = COPY $x2
+ ; CHECK-GI-NEXT: [[MOPSMemoryMovePseudo:%[0-9]+]]:gpr64common,
[[MOPSMemoryMovePseudo1:%[0-9]+]]:gpr64common,
[[MOPSMemoryMovePseudo2:%[0-9]+]]:gpr64 = MOPSMemoryMovePseudo [[COPY]],
[[COPY1]], [[COPY2]], implicit-def dead $nzcv
+ ; CHECK-GI-NEXT: RET_ReallyLR
+entry:
+ call void @llvm.memmove.p0.p0.i64(ptr %out, ptr %in, i64 %a, i1 false)
+ ret void
+}
___
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
