[llvm] 8256ddf - Resolve a long-standing FIXME in memcpyopt.

[Owen Anderson via llvm-commits]

Author: Owen Anderson
Date: 2022-12-23T16:15:24-07:00
New Revision: 8256ddf78cc558fa24070feb966075aa0c26589e

URL: https://github.com/llvm/llvm-project/commit/8256ddf78cc558fa24070feb966075aa0c26589e
DIFF: https://github.com/llvm/llvm-project/commit/8256ddf78cc558fa24070feb966075aa0c26589e.diff

LOG: Resolve a long-standing FIXME in memcpyopt.

Inspecting the downstream use of the cpyAlign, it is clear that
`performCallSlotOptzn` is expecting it to represent the alignment
of the copy destination, not the minimum of the src and dest
alignments. This patch renames the parameter to make this more
obvious.

I believe this change is NFC, because the downstream code has
alignment checks such that it all works out in the end. I have not
been able to construct a test case that actually triggers a change
in output.

Reviewed By: nikic

Differential Revision: https://reviews.llvm.org/D140603

Added: 
    

Modified: 
    llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
    llvm/test/Transforms/MemCpyOpt/callslot.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 34b46ad22869..288ee5ceac5d 100644
--- a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -873,7 +873,7 @@ bool MemCpyOptPass::processMemSet(MemSetInst *MSI, BasicBlock::iterator &BBI) {
 bool MemCpyOptPass::performCallSlotOptzn(Instruction *cpyLoad,
                                          Instruction *cpyStore, Value *cpyDest,
                                          Value *cpySrc, TypeSize cpySize,
-                                         Align cpyAlign, BatchAAResults &BAA,
+                                         Align cpyDestAlign, BatchAAResults &BAA,
                                          std::function<CallInst *()> GetC) {
   // The general transformation to keep in mind is
   //
@@ -978,7 +978,7 @@ bool MemCpyOptPass::performCallSlotOptzn(Instruction *cpyLoad,
 
   // Check that dest points to memory that is at least as aligned as src.
   Align srcAlign = srcAlloca->getAlign();
-  bool isDestSufficientlyAligned = srcAlign <= cpyAlign;
+  bool isDestSufficientlyAligned = srcAlign <= cpyDestAlign;
   // If dest is not aligned enough and we can't increase its alignment then
   // bail out.
   if (!isDestSufficientlyAligned && !isa<AllocaInst>(cpyDest)) {
@@ -1501,13 +1501,9 @@ bool MemCpyOptPass::processMemCpy(MemCpyInst *M, BasicBlock::iterator &BBI) {
     if (Instruction *MI = MD->getMemoryInst()) {
       if (auto *CopySize = dyn_cast<ConstantInt>(M->getLength())) {
         if (auto *C = dyn_cast<CallInst>(MI)) {
-          // FIXME: Can we pass in either of dest/src alignment here instead
-          // of conservatively taking the minimum?
-          Align Alignment = std::min(M->getDestAlign().valueOrOne(),
-                                     M->getSourceAlign().valueOrOne());
           if (performCallSlotOptzn(M, M, M->getDest(), M->getSource(),
                                    TypeSize::getFixed(CopySize->getZExtValue()),
-                                   Alignment, BAA,
+                                   M->getDestAlign().valueOrOne(), BAA,
                                    [C]() -> CallInst * { return C; })) {
             LLVM_DEBUG(dbgs() << "Performed call slot optimization:\n"
                               << "    call: " << *C << "\n"

diff  --git a/llvm/test/Transforms/MemCpyOpt/callslot.ll b/llvm/test/Transforms/MemCpyOpt/callslot.ll
index 30df4c4ac245..3a6b3c3804b5 100644
--- a/llvm/test/Transforms/MemCpyOpt/callslot.ll
+++ b/llvm/test/Transforms/MemCpyOpt/callslot.ll
@@ -211,6 +211,18 @@ nocaptures:
   ret void
 }
 
+define void @source_alignment(ptr noalias dereferenceable(128) %dst) {
+; CHECK-LABEL: @source_alignment(
+; CHECK-NEXT:    [[SRC:%.*]] = alloca [128 x i8], align 4
+; CHECK-NEXT:    call void @accept_ptr(ptr nocapture [[DST:%.*]]) #[[ATTR3]]
+; CHECK-NEXT:    ret void
+;
+  %src = alloca [128 x i8], align 4
+  call void @accept_ptr(ptr nocapture %src) nounwind
+  call void @llvm.memcpy.p0.p0.i64(ptr align 4 %dst, ptr %src, i64 128, i1 false)
+  ret void
+}
+
 declare void @may_throw()
 declare void @accept_ptr(ptr)
 declare void @llvm.memcpy.p0.p0.i64(ptr, ptr, i64, i1)