[AMDGPU] Add isAsCheapAsAMove for v_pk_mov_b32

[shiltian]

Co-authored-by: Matt Arsenault [email protected]

[shiltian]

• [AMDGPU] Add isAsCheapAsAMove for v_pk_mov_b32 #127632 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Shilei Tian (shiltian)

Changes

Credit: @arsenm

---

Full diff: https://github.com/llvm/llvm-project/pull/127632.diff

1 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/VOP3PInstructions.td (+1-1)

diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index 21898da1912f5..d5c6e8af109f4 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -1151,7 +1151,7 @@ let isCommutable = 1, isReMaterializable = 1 in {
     defm V_PK_ADD_F32 : VOP3PInst<"v_pk_add_f32", VOP3P_Profile<VOP_V2F32_V2F32_V2F32, VOP3_PACKED>, any_fadd>;
   } // End SubtargetPredicate = HasPackedFP32Ops
 
-  let SubtargetPredicate = HasPkMovB32 in
+  let SubtargetPredicate = HasPkMovB32, isAsCheapAsAMove = 1 in
   defm V_PK_MOV_B32 : VOP3PInst<"v_pk_mov_b32", VOP3P_Profile<VOP_V2I32_V2I32_V2I32, VOP3_PACKED>>;
 } // End isCommutable = 1, isReMaterializable = 1
 

[arsenm]

Best bet is likely a rematerialize test. e.g. add a case to remat-vop.mir that needs to handle v_pk_mov_b32. One in the coalescer might be better, like llvm/test/CodeGen/AMDGPU/vgpr-remat.mir

[shiltian]

For some reason, Graphite can't retarget this PR. It is supposed to be on top of #127715.