[PATCH] D72931: AMDGPU: Add a16 feature to gfx10

[Sebastian Neubauer via Phabricator via llvm-commits]

sebastian-ne added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:3679
+      unsigned AddrWords = IsA16 ? (AddrComponents + 1) / 2 : AddrComponents;
+
       unsigned VAddrWords;
----------------
I think this should be the following because the extra arguments like bias are still f32 and should not be divided by two:

      unsigned AddrComponents =
                           (BaseOpcode->Gradients ? Dim->NumGradients : 0) +
                           (BaseOpcode->Coordinates ? Dim->NumCoords : 0) +
                           (BaseOpcode->LodOrClampOrMip ? 1 : 0);

      unsigned AddrWords = BaseOpcode->NumExtraArgs +
                           (IsA16 ? (AddrComponents + 1) / 2 : AddrComponents);


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D72931/new/

https://reviews.llvm.org/D72931