diff options
author | Andy Polyakov <appro@openssl.org> | 2017-11-22 20:48:44 +0100 |
---|---|---|
committer | Andy Polyakov <appro@openssl.org> | 2017-11-23 21:05:44 +0100 |
commit | 88ac224cdaf65c3c3a08beddccb694706a311336 (patch) | |
tree | 4f9de52577d5c5fc2fcbbe986c3b6442646e1ddb /crypto | |
parent | 6df34091bac6f82b7d9553dc857123b2bd478c22 (diff) |
crypto/x86_64cpuid.pl: fix AVX512 capability masking.
Originally it was thought that it's possible to use AVX512VL+BW
instructions with XMM and YMM registers without kernel enabling
ZMM support, but it turned to be wrong assumption.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Diffstat (limited to 'crypto')
-rw-r--r-- | crypto/x86_64cpuid.pl | 9 |
1 files changed, 5 insertions, 4 deletions
diff --git a/crypto/x86_64cpuid.pl b/crypto/x86_64cpuid.pl index 7c8952e721..d30928efbd 100644 --- a/crypto/x86_64cpuid.pl +++ b/crypto/x86_64cpuid.pl @@ -197,10 +197,11 @@ OPENSSL_ia32_cpuid: and \$0xe6,%eax # isolate XMM, YMM and ZMM state support cmp \$0xe6,%eax je .Ldone - andl \$0xfffeffff,8(%rdi) # clear AVX512F, ~(1<<16) - # note that we don't touch other AVX512 - # extensions, because they can be used - # with YMM (without opmasking though) + andl \$0x3fdeffff,8(%rdi) # ~(1<<31|1<<30|1<<21|1<<16) + # clear AVX512F+BW+VL+FIMA, all of + # them are EVEX-encoded, which requires + # ZMM state support even if one uses + # only XMM and YMM :-( and \$6,%eax # isolate XMM and YMM state support cmp \$6,%eax je .Ldone |