[llvm] [AMDGPU] InstCombine llvm.amdgcn.ds.bpermute with uniform arguments (PR #129895)

[Jay Foad via llvm-commits]

================
@@ -6496,7 +6496,7 @@ define float @test_constant_fold_exp2_f32_1_strictfp() strictfp {
 
 define float @test_constant_fold_exp2_f32_neg1_strictfp() strictfp {
 ; CHECK-LABEL: @test_constant_fold_exp2_f32_neg1_strictfp(
-; CHECK-NEXT:    [[VAL:%.*]] = call float @llvm.amdgcn.exp2.f32(float -1.000000e+00) #[[ATTR16]]
+; CHECK-NEXT:    [[VAL:%.*]] = call float @llvm.amdgcn.exp2.f32(float -1.000000e+00) #[[ATTR19]]
----------------
jayfoad wrote:

> All the test attribute renumbering is annoying.

You're telling me! We should switch to printing attributes inline.

> Does it go away with a regenerate of the base checks?

Yes. Done.

https://github.com/llvm/llvm-project/pull/129895