aqjune added a comment.
Running Alive2 finds one failure, which is fixed by updating the shufflevector
semantics to return poison on undef/poison mask (D93818
<https://reviews.llvm.org/D93818>):
Transforms/InstCombine/shufflevector-div-rem-inseltpoison.ll
----------------------------------------
define <2 x float> @test_fdiv(float %a, float %b, i1 %cmp) {
%0:
%splatinsert = insertelement <2 x float> poison, float %a, i32 0
%denom = insertelement <2 x float> { 3.000000, poison }, float 1.000000,
i32 1
%t1 = fdiv <2 x float> %splatinsert, %denom
%splat.op = shufflevector <2 x float> %t1, <2 x float> poison, 4294967295,
0 ; THIS BLOCKS POISON
%t2 = select i1 %cmp, <2 x float> { 77.000000, 99.000000 }, <2 x float>
%splat.op
ret <2 x float> %t2
}
=>
define <2 x float> @test_fdiv(float %a, float %b, i1 %cmp) {
%0:
%1 = insertelement <2 x float> poison, float %a, i32 1
%2 = fdiv <2 x float> %1, { undef, 3.000000 }
%t2 = select i1 %cmp, <2 x float> { 77.000000, 99.000000 }, <2 x float> %2
ret <2 x float> %t2
}
Transformation doesn't verify!
ERROR: Target is more poisonous than source
For the diffs, I checked that all of these are replacements of undef with
poison, and there was no larger change.
After this patch, there are two places where undef still shows up as a
placeholder from llvm tests: I couldn't find where they are coming from. :(
If someone points me where they're likely to appear, I'll fix these.
$ ag "shufflevector <.*> .*, <.*> undef" | grep inseltpoison
Transforms/InstCombine/X86/x86-pshufb-inseltpoison.ll
$ ag "insertelement <.*> undef" | grep inseltpoison
Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts-inseltpoison.ll
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D93817/new/
https://reviews.llvm.org/D93817
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
