[PATCH] D62341: [DAGCombine][X86][AArch64][AMDGPU][MIPS][PPC] (sub x, c) -> (add x, -c) vector edition.

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: test/CodeGen/AMDGPU/sub.v2i16.ll:329
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_pk_sub_i16 v0, v0, 32
+; GFX9-NEXT:    v_pk_add_u16 v0, v0, s8
 ; GFX9-NEXT:    buffer_store_dword v0, off, s[4:7], 0
----------------
arsenm wrote:
> This is worse because it is no longer an inline immediate. We undo this already for the 32-bit case during selection:
> 
> 
> // Undo sub x, c -> add x, -c canonicalization since c is more likely
> // an inline immediate than -c.
> // TODO: Also do for 64-bit.
> def : GCNPat<
>   (add i32:$src0, (i32 NegSubInlineConst32:$src1)),
>   (S_SUB_I32 $src0, NegSubInlineConst32:$src1)
> >;
> 
I have a patch which should implement this for the v2i16 case


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62341/new/

https://reviews.llvm.org/D62341