On Tue, Nov 14, 2017 at 09:45:12AM +0000, Peryt, Sebastian wrote: > Hi, > > This patch fixes PR82941 and PR82942 by adding vzeroupper generation on SKX. > Bootstrapped and tested. > > 14.11.2017 Sebastian Peryt <sebastian.pe...@intel.com> > > gcc/
In that case the ChangeLog entry should list the PRs, i.e. PR target/82941 PR target/82942 > * config/i386/i386.c (pass_insert_vzeroupper): Modify gate condition > to return true on Xeon and not on Xeon Phi. > (ix86_check_avx256_register): Changed to ... > (ix86_check_avx_upper_register): ... this. > (ix86_check_avx_upper_register): Add extra check for > VALID_AVX512F_REG_OR_XI_MODE. The way this is usually written is instead: (ix86_check_avx256_register): Changed to ... (ix86_check_avx_upper_register): ... this. Add extra check for VALID_AVX512F_REG_OR_XI_MODE. i.e. don't duplicate the function name, just continue mentioning further changes. > (ix86_avx_u128_mode_needed): Changed > ix86_check_avx256_register to ix86_check_avx_upper_register. > (ix86_check_avx256_stores): Changed to ... > (ix86_check_avx_upper_stores): ... this. > (ix86_check_avx_upper_stores): Changed > ix86_check_avx256_register to ix86_check_avx_upper_register. Likewise. > gcc/testsuite/ > * gcc.target/i386/pr82941.c: New test. > * gcc.target/i386/pr82942.c: New test. Shouldn't there be also a test that if using -march=knl and another one if using -mavx512f -mavx512er that we don't emit any vzeroupper? Jakub