[llvm] [AMDGPU] Promote uniform ops to i32 in GISel (PR #106557)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Tue Sep 24 04:00:59 PDT 2024
================
@@ -414,23 +418,25 @@ define amdgpu_ps { i16, i16 } @s_orn2_i16_multi_use(i16 inreg %src0, i16 inreg %
}
define amdgpu_ps { i16, i16 } @s_orn2_i16_multi_foldable_use(i16 inreg %src0, i16 inreg %src1, i16 inreg %src2) {
-; GCN-LABEL: s_orn2_i16_multi_foldable_use:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_orn2_b32 s0, s2, s4
-; GCN-NEXT: s_orn2_b32 s1, s3, s4
-; GCN-NEXT: ; return to shader part epilog
+; GFX6-LABEL: s_orn2_i16_multi_foldable_use:
+; GFX6: ; %bb.0:
+; GFX6-NEXT: s_orn2_b32 s0, s2, s4
+; GFX6-NEXT: s_orn2_b32 s1, s3, s4
+; GFX6-NEXT: ; return to shader part epilog
;
-; GFX10-LABEL: s_orn2_i16_multi_foldable_use:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_orn2_b32 s0, s2, s4
-; GFX10-NEXT: s_orn2_b32 s1, s3, s4
-; GFX10-NEXT: ; return to shader part epilog
+; GFX9-LABEL: s_orn2_i16_multi_foldable_use:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_xor_b32 s1, s4, -1
+; GFX9-NEXT: s_or_b32 s0, s2, s1
+; GFX9-NEXT: s_or_b32 s1, s3, s1
+; GFX9-NEXT: ; return to shader part epilog
----------------
arsenm wrote:
This is worse
https://github.com/llvm/llvm-project/pull/106557
More information about the llvm-commits
mailing list