[PATCH] D119469: [AArch64] Turn truncating buildvectors into truncates

[David Sherwood via Phabricator via llvm-commits]

david-arm added inline comments.
Herald added a project: All.


================
Comment at: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp:9304
+      V.getOperand(8).getOperand(0), V.getOperand(12).getOperand(0)};
+  for (int I = 0; I < 4; I++)
+    if (Trunc[I].getValueType() == MVT::v4i32)
----------------
I think you're missing some test cases here because your code permits the possibility of mixed types, i.e.

  %a0 = extractelement <4 x i16> %a, i32 0
  ...
  %b0 = extractelement <4 x i32> %b, i32 0
  ...
  %c0 = extractelement <4 x i16> %c, i32 0
  ...
  %d0 = extractelement <4 x i32> %d, i32 0
  ...
  %t0 = trunc i16 %a0 to i8
  ...
  %t4 = trunc i32 %b0 to i8
  ...
  %t8 = trunc i16 %c0 to i8
  ...
  %t12 = trunc i32 %d0 to i8
  ...
  %i0 = insertelement <16 x i8> undef, i8 %t0, i32 0
  ...
  %i4 = insertelement <16 x i8> %i3, i8 %t4, i32 4
  ...
  %i8 = insertelement <16 x i8> %i7, i8 %t8, i32 8
  ...
  %i12 = insertelement <16 x i8> %i11, i8 %t12, i32 12
  ...

Can you add at least one test for mixed types?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D119469/new/

https://reviews.llvm.org/D119469