Revise alignment checking/calculation on 256-bit unaligned memory access
authorMichael Liao <michael.liao@intel.com>
Mon, 25 Mar 2013 23:50:10 +0000 (23:50 +0000)
committerMichael Liao <michael.liao@intel.com>
Mon, 25 Mar 2013 23:50:10 +0000 (23:50 +0000)
commitd4584c9e5658887ec50c43760c988d04eaa13e34
tree1f64d66547a6aed3e1b1ee10531534229ad01f18
parentb4f98ea1213c866f39aa5b341ec0116f9c2335d7
Revise alignment checking/calculation on 256-bit unaligned memory access

- It's still considered aligned when the specified alignment is larger
  than the natural alignment;
- The new alignment for the high 128-bit vector should be min(16,
  alignment) as the pointer is advanced by 16, a power-of-2 offset.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177947 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/X86/X86ISelLowering.cpp
test/CodeGen/X86/avx-load-store.ll