[PATCH] D121411: [AMDGPU] Support v_mov_b64 in dpp combine

Fri Mar 11 11:13:00 PST 2022

rampitec added a comment.

In D121411#3374893 <https://reviews.llvm.org/D121411#3374893>, @foad wrote:

> Looks OK.
>
> As an alternative, maybe we could remove V_MOV_B64_DPP_PSEUDO, use the pseudo V_MOV_B64_dpp on all targets, and expand it post-ra only on targets that don't support it?

This maybe a good idea at this point. As a separate change of course.

================
Comment at: llvm/lib/Target/AMDGPU/GCNDPPCombine.cpp:169
   case AMDGPU::COPY:
   case AMDGPU::V_MOV_B32_e32:
+  case AMDGPU::V_MOV_B64_PSEUDO:
----------------
foad wrote:
> I wonder why we don't include V_MOV_B32_e64? Maybe we never use it for an immediate copy?
Probably... Or perhaps it was just forgotten.

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D121411/new/

https://reviews.llvm.org/D121411