[llvm] [LoopUtils] Minimize expanded RT-checks returned (PR #157518)
Ramkumar Ramachandra via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 8 10:35:24 PDT 2025
https://github.com/artagnon created https://github.com/llvm/llvm-project/pull/157518
Follow up on 528b13d ([SCEVExp] Add helper to clean up dead instructions after expansion.) to hoist the SCEVExapnder::eraseDeadInstructions call from LoopVectorize into the LoopUtils APIs add[Diff]RuntimeChecks, so that other callers (LoopDistribute and LoopVersioning) can benefit from the patch.
>From 4011c4dc3bdbd62438af08dfe7d9984a50225302 Mon Sep 17 00:00:00 2001
From: Ramkumar Ramachandra <ramkumar.ramachandra at codasip.com>
Date: Mon, 8 Sep 2025 18:26:01 +0100
Subject: [PATCH] [LoopUtils] Minimize expanded RT-checks returned
Follow up on 528b13d ([SCEVExp] Add helper to clean up dead instructions
after expansion.) to hoist the SCEVExapnder::eraseDeadInstructions call
from LoopVectorize into the LoopUtils APIs add[Diff]RuntimeChecks, so
that other callers (LoopDistribute and LoopVersioning) can benefit from
the patch.
---
llvm/lib/Transforms/Utils/LoopUtils.cpp | 2 ++
llvm/lib/Transforms/Utils/LoopVersioning.cpp | 2 ++
.../lib/Transforms/Vectorize/LoopVectorize.cpp | 1 -
.../scev-inserted-runtime-check.ll | 2 --
.../Transforms/LoopVersioning/incorrect-phi.ll | 18 +++++-------------
.../wrapping-pointer-versioning.ll | 3 ---
6 files changed, 9 insertions(+), 19 deletions(-)
diff --git a/llvm/lib/Transforms/Utils/LoopUtils.cpp b/llvm/lib/Transforms/Utils/LoopUtils.cpp
index 843364eb34f83..b172ef6ba0803 100644
--- a/llvm/lib/Transforms/Utils/LoopUtils.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUtils.cpp
@@ -2032,6 +2032,7 @@ Value *llvm::addRuntimeChecks(
MemoryRuntimeCheck = IsConflict;
}
+ Exp.eraseDeadInstructions(MemoryRuntimeCheck);
return MemoryRuntimeCheck;
}
@@ -2077,6 +2078,7 @@ Value *llvm::addDiffRuntimeChecks(
MemoryRuntimeCheck = IsConflict;
}
+ Expander.eraseDeadInstructions(MemoryRuntimeCheck);
return MemoryRuntimeCheck;
}
diff --git a/llvm/lib/Transforms/Utils/LoopVersioning.cpp b/llvm/lib/Transforms/Utils/LoopVersioning.cpp
index 1711163fb9f59..ec2e6c1ab796b 100644
--- a/llvm/lib/Transforms/Utils/LoopVersioning.cpp
+++ b/llvm/lib/Transforms/Utils/LoopVersioning.cpp
@@ -81,6 +81,8 @@ void LoopVersioning::versionLoop(
} else
RuntimeCheck = MemRuntimeCheck ? MemRuntimeCheck : SCEVRuntimeCheck;
+ Exp.eraseDeadInstructions(SCEVRuntimeCheck);
+
assert(RuntimeCheck && "called even though we don't need "
"any runtime checks");
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index dd4b3f8e3077b..6c96214cea13c 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1849,7 +1849,6 @@ class GeneratedRTChecks {
}
SCEVExp.eraseDeadInstructions(SCEVCheckCond);
- MemCheckExp.eraseDeadInstructions(MemRuntimeCheckCond);
if (!MemCheckBlock && !SCEVCheckBlock)
return;
diff --git a/llvm/test/Transforms/LoopDistribute/scev-inserted-runtime-check.ll b/llvm/test/Transforms/LoopDistribute/scev-inserted-runtime-check.ll
index bfd0ec0340fbb..697f4e117acf4 100644
--- a/llvm/test/Transforms/LoopDistribute/scev-inserted-runtime-check.ll
+++ b/llvm/test/Transforms/LoopDistribute/scev-inserted-runtime-check.ll
@@ -12,7 +12,6 @@ define void @f(ptr noalias %a, ptr noalias %b, ptr noalias %c, ptr noalias %d, p
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], -1
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP0]] to i32
; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[TMP2]])
-; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
; CHECK-NEXT: [[TMP3:%.*]] = or i1 [[MUL_OVERFLOW]], [[TMP1]]
@@ -142,7 +141,6 @@ define void @f_with_offset(ptr noalias %b, ptr noalias %c, ptr noalias %d, ptr n
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], -1
; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[TMP0]] to i32
; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[TMP2]])
-; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
; CHECK-NEXT: [[TMP3:%.*]] = or i1 [[MUL_OVERFLOW]], [[TMP1]]
diff --git a/llvm/test/Transforms/LoopVersioning/incorrect-phi.ll b/llvm/test/Transforms/LoopVersioning/incorrect-phi.ll
index b6896c467812b..de7b3c1dc581d 100644
--- a/llvm/test/Transforms/LoopVersioning/incorrect-phi.ll
+++ b/llvm/test/Transforms/LoopVersioning/incorrect-phi.ll
@@ -1,4 +1,4 @@
-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 --check-globals none
; RUN: opt -passes=loop-versioning -S < %s | FileCheck %s
; Make sure all PHIs are properly updated in the exit block. Based on
@@ -9,10 +9,6 @@
define void @phi_with_poison() {
; CHECK-LABEL: define void @phi_with_poison() {
; CHECK-NEXT: [[BB6_LVER_CHECK:.*:]]
-; CHECK-NEXT: [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 10, i64 0)
-; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL]], 0
-; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; CHECK-NEXT: [[TMP0:%.*]] = sub i64 0, [[MUL_RESULT]]
; CHECK-NEXT: br i1 poison, label %[[BB6_PH_LVER_ORIG:.*]], label %[[BB6_PH:.*]]
; CHECK: [[BB6_PH_LVER_ORIG]]:
; CHECK-NEXT: br label %[[BB6_LVER_ORIG:.*]]
@@ -32,8 +28,8 @@ define void @phi_with_poison() {
; CHECK-NEXT: [[_TMP123:%.*]] = getelementptr [2 x [3 x [5 x i16]]], ptr @x, i16 0, i64 poison
; CHECK-NEXT: [[_TMP126:%.*]] = getelementptr [3 x [5 x i16]], ptr [[_TMP123]], i16 0, i64 [[_TMP1423]]
; CHECK-NEXT: [[_TMP129:%.*]] = getelementptr [5 x i16], ptr [[_TMP126]], i16 0, i64 poison
-; CHECK-NEXT: [[_TMP130:%.*]] = load i16, ptr [[_TMP129]], align 2
-; CHECK-NEXT: store i16 poison, ptr @x, align 2
+; CHECK-NEXT: [[_TMP130:%.*]] = load i16, ptr [[_TMP129]], align 2, !alias.scope [[META0:![0-9]+]]
+; CHECK-NEXT: store i16 poison, ptr @x, align 2, !alias.scope [[META3:![0-9]+]]
; CHECK-NEXT: [[_TMP142]] = add i64 [[_TMP1423]], 1
; CHECK-NEXT: br i1 false, label %[[BB6]], label %[[LOOP_EXIT_LOOPEXIT1:.*]]
; CHECK: [[LOOP_EXIT_LOOPEXIT]]:
@@ -77,10 +73,6 @@ define void @phi_with_non_loop_defined_value() {
; CHECK-LABEL: define void @phi_with_non_loop_defined_value() {
; CHECK-NEXT: [[BB6_LVER_CHECK:.*:]]
; CHECK-NEXT: [[T:%.*]] = add i16 1, 1
-; CHECK-NEXT: [[MUL:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 10, i64 0)
-; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL]], 0
-; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL]], 1
-; CHECK-NEXT: [[TMP0:%.*]] = sub i64 0, [[MUL_RESULT]]
; CHECK-NEXT: br i1 poison, label %[[BB6_PH_LVER_ORIG:.*]], label %[[BB6_PH:.*]]
; CHECK: [[BB6_PH_LVER_ORIG]]:
; CHECK-NEXT: br label %[[BB6_LVER_ORIG:.*]]
@@ -100,8 +92,8 @@ define void @phi_with_non_loop_defined_value() {
; CHECK-NEXT: [[_TMP123:%.*]] = getelementptr [2 x [3 x [5 x i16]]], ptr @x, i16 0, i64 poison
; CHECK-NEXT: [[_TMP126:%.*]] = getelementptr [3 x [5 x i16]], ptr [[_TMP123]], i16 0, i64 [[_TMP1423]]
; CHECK-NEXT: [[_TMP129:%.*]] = getelementptr [5 x i16], ptr [[_TMP126]], i16 0, i64 poison
-; CHECK-NEXT: [[_TMP130:%.*]] = load i16, ptr [[_TMP129]], align 2
-; CHECK-NEXT: store i16 poison, ptr @x, align 2
+; CHECK-NEXT: [[_TMP130:%.*]] = load i16, ptr [[_TMP129]], align 2, !alias.scope [[META5:![0-9]+]]
+; CHECK-NEXT: store i16 poison, ptr @x, align 2, !alias.scope [[META8:![0-9]+]]
; CHECK-NEXT: [[_TMP142]] = add i64 [[_TMP1423]], 1
; CHECK-NEXT: br i1 false, label %[[BB6]], label %[[LOOP_EXIT_LOOPEXIT1:.*]]
; CHECK: [[LOOP_EXIT_LOOPEXIT]]:
diff --git a/llvm/test/Transforms/LoopVersioning/wrapping-pointer-versioning.ll b/llvm/test/Transforms/LoopVersioning/wrapping-pointer-versioning.ll
index 3e25695304854..d851a936969d1 100644
--- a/llvm/test/Transforms/LoopVersioning/wrapping-pointer-versioning.ll
+++ b/llvm/test/Transforms/LoopVersioning/wrapping-pointer-versioning.ll
@@ -31,14 +31,12 @@ define void @f1(ptr noalias %a,
; LV-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], -1
; LV-NEXT: [[TMP5:%.*]] = trunc i64 [[TMP0]] to i32
; LV-NEXT: [[MUL2:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[TMP5]])
-; LV-NEXT: [[MUL_RESULT1:%.*]] = extractvalue { i32, i1 } [[MUL2]], 0
; LV-NEXT: [[MUL_OVERFLOW1:%.*]] = extractvalue { i32, i1 } [[MUL2]], 1
; LV-NEXT: [[TMP1:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
; LV-NEXT: [[TMP8:%.*]] = or i1 [[MUL_OVERFLOW1]], [[TMP1]]
; LV-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[TMP0]])
; LV-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0
; LV-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1
-; LV-NEXT: [[TMP2:%.*]] = sub i64 0, [[MUL_RESULT]]
; LV-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[MUL_RESULT]]
; LV-NEXT: [[TMP4:%.*]] = icmp ult ptr [[TMP3]], [[A]]
; LV-NEXT: [[TMP6:%.*]] = or i1 [[TMP4]], [[MUL_OVERFLOW]]
@@ -264,7 +262,6 @@ define void @f3(ptr noalias %a,
; LV-NEXT: [[MUL2:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[TMP0]])
; LV-NEXT: [[MUL_RESULT3:%.*]] = extractvalue { i64, i1 } [[MUL2]], 0
; LV-NEXT: [[MUL_OVERFLOW4:%.*]] = extractvalue { i64, i1 } [[MUL2]], 1
-; LV-NEXT: [[TMP6:%.*]] = sub i64 0, [[MUL_RESULT3]]
; LV-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[MUL_RESULT3]]
; LV-NEXT: [[TMP8:%.*]] = icmp ult ptr [[TMP7]], [[A]]
; LV-NEXT: [[TMP9:%.*]] = or i1 [[TMP8]], [[MUL_OVERFLOW4]]
More information about the llvm-commits
mailing list