[PATCH] D130579: AMDGPU: Remove manual selection for atomic fadd

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/FLATInstructions.td:1031
+                                         ValueType data_vt = vt,
+                                         int complexity = 0, bit isIntr = 0> {
   defvar noRtnNode = !cast<PatFrags>(node # "_noret" # !if(isIntr, "", "_" # vt.Size));
----------------
abinavpp wrote:
> Since you're splitting the ret and noret patterns to their own multiclasses, I think you can remove the complexity argument here and from FlatSignedAtomicPatImplRtn and use `let AddedComplexity = ... in` in the defms.
Complexity argument is weird, should put in a let block around the instances


================
Comment at: llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.fadd.rtn_no-rtn.ll:6
 
+; GFX11-LABEL: {{^}}name:            raw_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_plus4095__sgpr_soffset
 ; GFX11: BUFFER_ATOMIC_ADD_F32_OFFEN
----------------
Why are these checks so sparse? Why isn't this test generated like the others?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D130579/new/

https://reviews.llvm.org/D130579