vpykhtin wrote:
Thanks Nicolai!
https://github.com/llvm/llvm-project/pull/68714
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
nhaehnle wrote:
I meant ballot.i64 in wave64. For some reason, I didn't see the changes you
made in those tests (perhaps I was looking at GitHub's display of your last
changes to the PR, instead of the full PR.) But since you did make those
changes, it's all good as far as I'm concerned.
http
nhaehnle wrote:
Though, on second thought, shouldn't there be some wave64 tests?
https://github.com/llvm/llvm-project/pull/68714
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
https://github.com/nhaehnle approved this pull request.
https://github.com/llvm/llvm-project/pull/68714
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
vpykhtin wrote:
With the latest commit I added a guard to skip _ballot.i64_ in wave32 mode
because it's lowered with _i64 AMDGPUISD::SETCC_ - this should be fixed with
i32 result similar to ICMP/FCMP intrinsics. After a fix that follows this PR
I'm going to turn the guard into assert - TODO is
@@ -396,3 +396,115 @@ true:
false:
ret i32 33
}
+
+declare i32 @llvm.amdgcn.icmp.i32(i1, i1, i32)
+
+define amdgpu_cs i32
@branch_divergent_simulated_negated_ballot_ne_zero_and(i32 %v1, i32 %v2) {
+; CHECK-LABEL: branch_divergent_simulated_negated_ballot_ne_zero_and:
+; CHEC