[PATCH] D14322: AVX512 : VMOVSHDUP/VMOVSLDUP implementation.

[Simon Pilgrim via llvm-commits]

RKSimon added inline comments.

================
Comment at: lib/Target/X86/InstPrinter/X86InstComments.cpp:348
@@ -323,2 +347,3 @@
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVSLDUPMask(MVT::v8f32, ShuffleMask);
+    if (MI->getNumOperands() < 6)
+      Src1Name = getRegName(MI->getOperand(1).getReg());
----------------
delena wrote:
> missing comments
Wouldn't it be better to match the 'fall through' pattern that we have for most rm / rr shuffle pairs instead of using operand count to re-determine if its rm? It would mean a splitting of the CASE_MOVDUP into 2 versions though.


Repository:
  rL LLVM

http://reviews.llvm.org/D14322