This is a (hopefully temporary) fix to PR96791. This will make the default be -mno-block-ops-vector-pair even on power10, so we will not hit the issue of DSE trying to truncate a POImode register. I am still concerned it will be possible to hit this because the MMA builtins will also generate POImode stores, but I think any example of that will be somewhat more contrived.
Bootstrap and regression tests passed on ppc64le power9. Pre-approved by Segher, posting after commit. gcc/ChangeLog: * config/rs6000/rs6000.c (rs6000_option_override_internal): Change default. --- gcc/config/rs6000/rs6000.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c index f6a3ff6f089..9908830b07a 100644 --- a/gcc/config/rs6000/rs6000.c +++ b/gcc/config/rs6000/rs6000.c @@ -4020,10 +4020,11 @@ rs6000_option_override_internal (bool global_init_p) if (!(rs6000_isa_flags_explicit & OPTION_MASK_BLOCK_OPS_VECTOR_PAIR)) { - if (TARGET_MMA && TARGET_EFFICIENT_UNALIGNED_VSX) - rs6000_isa_flags |= OPTION_MASK_BLOCK_OPS_VECTOR_PAIR; - else - rs6000_isa_flags &= ~OPTION_MASK_BLOCK_OPS_VECTOR_PAIR; + /* When the POImode issues of PR96791 are resolved, then we can + once again enable use of vector pair for memcpy/memmove on + P10 if we have TARGET_MMA. For now we make it disabled by + default for all targets. */ + rs6000_isa_flags &= ~OPTION_MASK_BLOCK_OPS_VECTOR_PAIR; } /* Use long double size to select the appropriate long double. We use -- 2.17.1