[LSR] Do not consider uses in lifetime intrinsics #149492
Merged
+64
−0
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
We should ignore uses of pointers in lifetime intrinsics, as these are not actually materialized in the final code, so don't affect register pressure or anything else LSR needs to model. Handling these only results in peculiar rewrites where additional intermediate GEPs are introduced.
|
@llvm/pr-subscribers-llvm-transforms Author: Nikita Popov (nikic) ChangesWe should ignore uses of pointers in lifetime intrinsics, as these are not actually materialized in the final code, so don't affect register pressure or anything else LSR needs to model. Handling these only results in peculiar rewrites where additional intermediate GEPs are introduced. Full diff: https://github.com/llvm/llvm-project/pull/149492.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index dc8fa4379752f..4eab10c0bca57 100644
--- a/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -3790,6 +3790,11 @@ LSRInstance::CollectLoopInvariantFixupsAndFormulae() {
continue;
}
+ // Do not consider uses inside lifetime intrinsics. These are not
+ // actually materialized.
+ if (UserInst->isLifetimeStartOrEnd())
+ continue;
+
std::pair<size_t, Immediate> P =
getUse(S, LSRUse::Basic, MemAccessTy());
size_t LUIdx = P.first;
diff --git a/llvm/test/Transforms/LoopStrengthReduce/X86/lifetime-use.ll b/llvm/test/Transforms/LoopStrengthReduce/X86/lifetime-use.ll
new file mode 100644
index 0000000000000..c7a0de22b200b
--- /dev/null
+++ b/llvm/test/Transforms/LoopStrengthReduce/X86/lifetime-use.ll
@@ -0,0 +1,59 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -passes=loop-reduce -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
+
+define void @test(ptr %p, i64 %idx) {
+; CHECK-LABEL: define void @test(
+; CHECK-SAME: ptr [[P:%.*]], i64 [[IDX:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [4 x [4 x i32]], align 16
+; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 64, ptr [[ALLOCA]])
+; CHECK-NEXT: [[TMP0:%.*]] = shl i64 [[IDX]], 6
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[TMP0]], 48
+; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP1]]
+; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr nuw i8, ptr [[ALLOCA]], i64 48
+; CHECK-NEXT: br label %[[LOOP:.*]]
+; CHECK: [[LOOP]]:
+; CHECK-NEXT: [[LSR_IV:%.*]] = phi i64 [ [[LSR_IV_NEXT:%.*]], %[[LOOP]] ], [ -8, %[[ENTRY]] ]
+; CHECK-NEXT: [[TMP2:%.*]] = shl nsw i64 [[LSR_IV]], 2
+; CHECK-NEXT: [[SCEVGEP8:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP2]]
+; CHECK-NEXT: [[SCEVGEP9:%.*]] = getelementptr i8, ptr [[SCEVGEP8]], i64 32
+; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[SCEVGEP9]], align 4
+; CHECK-NEXT: [[SCEVGEP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[LSR_IV]]
+; CHECK-NEXT: [[SCEVGEP7:%.*]] = getelementptr i8, ptr [[SCEVGEP6]], i64 8
+; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[SCEVGEP7]], align 4
+; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i8, ptr [[SCEVGEP3]], i64 [[LSR_IV]]
+; CHECK-NEXT: [[SCEVGEP5:%.*]] = getelementptr i8, ptr [[SCEVGEP4]], i64 8
+; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[SCEVGEP5]], align 4
+; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[LSR_IV]]
+; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[SCEVGEP1]], i64 8
+; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr [[SCEVGEP2]], align 4
+; CHECK-NEXT: [[LSR_IV_NEXT]] = add nsw i64 [[LSR_IV]], 4
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[LSR_IV_NEXT]], 0
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[LOOP]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 64, ptr [[ALLOCA]])
+; CHECK-NEXT: ret void
+;
+entry:
+ %alloca = alloca [4 x [4 x i32]], align 16
+ call void @llvm.lifetime.start.p0(i64 64, ptr %alloca)
+ br label %loop
+
+loop:
+ %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %loop ]
+ %gep1 = getelementptr [4 x [12 x [4 x [4 x i32]]]], ptr %p, i64 0, i64 0, i64 0, i64 %indvars.iv, i64 0
+ %0 = load i32, ptr %gep1, align 4
+ %gep2 = getelementptr [6 x [4 x [4 x i32]]], ptr %p, i64 0, i64 0, i64 0, i64 %indvars.iv
+ %1 = load i32, ptr %gep2, align 4
+ %gep3 = getelementptr [4 x [4 x i32]], ptr %alloca, i64 0, i64 3, i64 %indvars.iv
+ %2 = load i32, ptr %gep3, align 4
+ %gep4 = getelementptr [6 x [4 x [4 x i32]]], ptr %p, i64 0, i64 %idx, i64 3, i64 %indvars.iv
+ %3 = load i32, ptr %gep4, align 4
+ %indvars.iv.next = add i64 %indvars.iv, 1
+ %exitcond.not = icmp eq i64 %indvars.iv, 1
+ br i1 %exitcond.not, label %exit, label %loop
+
+exit:
+ call void @llvm.lifetime.end.p0(i64 64, ptr %alloca)
+ ret void
+}
|
|
We have other intrinsics like this that convey information, but get dropped late - assumes spring to mind. Do we have some utility for an ignorable use somewhere? |
We have isAssumeLikeIntrinsic(). |
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.
We should ignore uses of pointers in lifetime intrinsics, as these are not actually materialized in the final code, so don't affect register pressure or anything else LSR needs to model.
Handling these only results in peculiar rewrites where additional intermediate GEPs are introduced.