[PATCH] D14942: Fixed a bug in VPERMT2 intrinsic

[Igor Breger via llvm-commits]

igorb added inline comments.

================
Comment at: lib/Target/X86/X86InstrAVX512.td:1204
@@ +1203,3 @@
+// VPERMT
+multiclass avx512_perm_t<bits<8> opc, string OpcodeStr,
+                            SDNode OpNode, X86VectorVTInfo _,
----------------
Is it possible to use this class  for VPERM2I  definition ?

   avx512_perm_i<opc, OpcodeStr, OpNode, VTInfo.info512>
  
   avx512_perm_t<opc, OpcodeStr, OpNode, VTInfo.info512, VTInfo.info512>

================
Comment at: lib/Target/X86/X86InstrAVX512.td:1218
@@ +1217,3 @@
+            OpcodeStr, "$src3, $src2", "$src2, $src3",
+            (_.VT (OpNode MaskVT.RC:$src1, _.RC:$src2,
+                   (_.VT (bitconvert (_.LdFrag addr:$src3)))))>,
----------------
Did you meen  OpNode _.RC:$src1, MaskVT.RC:$src2   ?

================
Comment at: test/CodeGen/X86/avx512-intrinsics.ll:2
@@ -2,1 +1,3 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s
 
----------------
Could you please commit updated test file as separate commit. Very difficult to identify changes from this patch.


Repository:
  rL LLVM

http://reviews.llvm.org/D14942