[ARM] Fix lowering of misaligned memcpy/memset

Currently getOptimalMemOpType returns i32 for large enough sizes without
checking for alignment, leading to poor code generation when misaligned accesses
aren't permitted as we generate a word store then later split it up into byte
stores. This means we inadvertantly go over the MaxStoresPerMemcpy limit and for
memset we splat the memset value into a word then immediately split it up
again.

Fix this by leaving it up to FindOptimalMemOpLowering to figure out which type
to use, but also fix a bug there where it wasn't correctly checking if
misaligned memory accesses are allowed.

Differential Revision: https://reviews.llvm.org/D33442

llvm-svn: 303990

1 files changed, 0 insertions, 6 deletions

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 4bfb9b01656..62e774d869d 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -12147,12 +12147,6 @@ EVT ARMTargetLowering::getOptimalMemOpType(uint64_t Size,
     }
   }
 
-  // Lowering to i32/i16 if the size permits.
-  if (Size >= 4)
-    return MVT::i32;
-  else if (Size >= 2)
-    return MVT::i16;
-
   // Let the target-independent logic figure it out.
   return MVT::Other;
 }