https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118696
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
--- gcc/config/s390/vector.md.jj 2025-01-24 17:37:48.987458141 +0100
+++ gcc/config/s390/vector.md 2025-01-30 09:10:53.413542300 +0100
@@ -2166,7 +2166,7 @@ (define_insn_and_split "*vec_cmpgt<mode>
(vec_duplicate:V2DI
(vec_select:DI
(match_dup 4)
- (parallel [(const_int 1)]))))
+ (parallel [(const_int 0)]))))
(set (match_dup 0)
(subreg:<MODE> (match_dup 4) 0))]
{
@@ -2198,7 +2198,7 @@ (define_insn_and_split "*vec_cmpgtu<mode
(vec_duplicate:V2DI
(vec_select:DI
(match_dup 4)
- (parallel [(const_int 1)]))))
+ (parallel [(const_int 0)]))))
(set (match_dup 0)
(subreg:<MODE> (match_dup 4) 0))]
{
fixes this testcase, but let me work on a testcase that tests all the different
eq/ne/gt/le/gtu/leu possibilities for hi vs. low parts.
