[PATCH] D127524: [AMDGPU] Make temp vgpr selection stable in indirectCopyToAGPR

[Mahesha S via Phabricator via llvm-commits]

hsmhsm added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/accvgpr-copy.mir:528
     ; GFX908-NEXT: {{  $}}
-    ; GFX908-NEXT: $vgpr1 = V_MOV_B32_e32 killed $sgpr0, implicit $exec
-    ; GFX908-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr1, implicit $exec, implicit $exec
+    ; GFX908-NEXT: $vgpr255 = V_MOV_B32_e32 killed $sgpr0, implicit $exec
+    ; GFX908-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr255, implicit $exec, implicit $exec
----------------
rampitec wrote:
> cdevadas wrote:
> > Unrelated to this patch; The highest VGPR register number should be shifted back to the lowest free register.
> > Was there any follow-up patch planned?
> Ping @hsmhsm for this question.
We need to see - for this testcase, the code which handles the finding and reserving lowest available free register will be hit or not. If the nature of this testcase is like that we won't hit that part of the code, then this is indeed the expected behavior.  Otherwise, there is an unhandled bug in the patch https://reviews.llvm.org/D123525 which need to be fixed.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D127524/new/

https://reviews.llvm.org/D127524