[PATCH] D30189: [AVX512] Fix EXTRACT_VECTOR_ELT for v2i1/v4i1/v32i1/v64i1 with variable index.

Igor Breger via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue Feb 21 04:25:43 PST 2017


igorb added inline comments.


================
Comment at: lib/Target/X86/X86ISelLowering.cpp:13741
+    unsigned NumElts = VecVT.getVectorNumElements();
+    unsigned VecSize = (NumElts <= 4 ? 128 : 512);
+    MVT ExtVT = MVT::getVectorVT(MVT::getIntegerVT(VecSize/NumElts), NumElts);
----------------
delena wrote:
> You can extend v8i1 and v16i1 to 128 or 256 or 512.
> Is one of the variants is preferable in terms of performance?
> If all the same, please leave a comment inside. 
For KNL extend  v8i1 and v16i1 to 512bit vector generate better code .


Repository:
  rL LLVM

https://reviews.llvm.org/D30189





More information about the llvm-commits mailing list