[PATCH] D17899: [x86, AVX] optimize masked loads with constant masks

[Sanjay Patel via llvm-commits]

spatel added inline comments.

================
Comment at: lib/Target/X86/X86ISelLowering.cpp:27256
@@ +27255,3 @@
+/// Convert a masked load with a constant mask and a real (not undefined)
+/// pass-through operand into a masked load and a shuffle.
+static SDValue
----------------
RKSimon wrote:
> Why is it important that we don't blend with an undefined pass-through?
Ah, I'll make that explicit in the comment: this transform is *creating* a masked load with an undef pass-through. If we don't guard against that case, we'll infinite loop.


http://reviews.llvm.org/D17899