[llvm] [GlobalOpt] Preven widenDestArray from shrinking an alloca. (PR #144641)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Craig Topper (topperc)

<details>
<summary>Changes</summary>

If the destination alloca for one of the memcpy calls we are
modifying is already larger than our desired size, we shouldn't
replace it with a smaller alloca.

---
Full diff: https://github.com/llvm/llvm-project/pull/144641.diff


2 Files Affected:

- (modified) llvm/lib/Transforms/IPO/GlobalOpt.cpp (+4) 
- (added) llvm/test/Transforms/GlobalOpt/ARM/arm-widen-large-alloca.ll (+27) 


``````````diff
diff --git a/llvm/lib/Transforms/IPO/GlobalOpt.cpp b/llvm/lib/Transforms/IPO/GlobalOpt.cpp
index 7db0586386506..f0cd00df23959 100644
--- a/llvm/lib/Transforms/IPO/GlobalOpt.cpp
+++ b/llvm/lib/Transforms/IPO/GlobalOpt.cpp
@@ -2103,6 +2103,10 @@ static void widenDestArray(CallInst *CI, const unsigned NumBytesToPad,
     unsigned ElementByteWidth = SourceDataArray->getElementByteSize();
     unsigned int TotalBytes = NumBytesToCopy + NumBytesToPad;
     unsigned NumElementsToCopy = divideCeil(TotalBytes, ElementByteWidth);
+    // Don't change size if already wide enough.
+    if (Alloca->getAllocatedType()->getArrayNumElements() >= NumElementsToCopy)
+      return;
+
     // Update destination array to be word aligned (memcpy(X,...,...))
     IRBuilder<> BuildAlloca(Alloca);
     AllocaInst *NewAlloca = BuildAlloca.CreateAlloca(ArrayType::get(
diff --git a/llvm/test/Transforms/GlobalOpt/ARM/arm-widen-large-alloca.ll b/llvm/test/Transforms/GlobalOpt/ARM/arm-widen-large-alloca.ll
new file mode 100644
index 0000000000000..4fca1ffefdcaf
--- /dev/null
+++ b/llvm/test/Transforms/GlobalOpt/ARM/arm-widen-large-alloca.ll
@@ -0,0 +1,27 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -mtriple=arm-none-eabi -passes=globalopt -S | FileCheck %s
+
+ at .i8 = private unnamed_addr constant [3 x i8] [i8 1, i8 2, i8 3] , align 1
+
+define void @memcpy()  {
+; CHECK-LABEL: define void @memcpy() local_unnamed_addr {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[ALLOCA1:%.*]] = alloca [4 x i8], align 1
+; CHECK-NEXT:    [[ALLOCA2:%.*]] = alloca [5 x i8], align 1
+; CHECK-NEXT:    [[CALL1:%.*]] = call i32 @bar(ptr nonnull [[ALLOCA1]])
+; CHECK-NEXT:    [[CALL2:%.*]] = call i32 @bar(ptr nonnull [[ALLOCA2]])
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(3) [[ALLOCA1]], ptr noundef nonnull align 1 dereferenceable(3) @.i8, i32 4, i1 false)
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(5) [[ALLOCA2]], ptr noundef nonnull align 1 dereferenceable(3) @.i8, i32 4, i1 false)
+; CHECK-NEXT:    ret void
+;
+entry:
+  %alloca1 = alloca [3 x i8], align 1
+  %alloca2 = alloca [5 x i8], align 1
+  %call1 = call i32 @bar(ptr nonnull %alloca1)
+  %call2 = call i32 @bar(ptr nonnull %alloca2)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(3) %alloca1, ptr noundef nonnull align 1 dereferenceable(3) @.i8, i32 3, i1 false)
+  call void @llvm.memcpy.p0.p0.i32(ptr noundef nonnull align 1 dereferenceable(5) %alloca2, ptr noundef nonnull align 1 dereferenceable(3) @.i8, i32 3, i1 false)
+  ret void
+}
+
+declare i32 @bar(...)

``````````

</details>


https://github.com/llvm/llvm-project/pull/144641