diff options
| author | Andy Polyakov <appro@openssl.org> | 2007-05-14 21:35:25 +0000 |
|---|---|---|
| committer | Andy Polyakov <appro@openssl.org> | 2007-05-14 21:35:25 +0000 |
| commit | b2dba9bf1f8f73376b9c1f0904a86996c728b236 (patch) | |
| tree | 8848a4f5efdf02d841b1ebea4969d879d6e61eb5 /crypto/x86_64cpuid.pl | |
| parent | 932cc129ee61f5b72636eee6a7c3268e23967f7b (diff) | |
Profiling revealed that OPENSSL_cleanse consumes *more* CPU time than
sha1_block_data_order when hashing short messages. Move OPENSSL_cleanse
to "cpuid" assembler module and gain 2x.
Diffstat (limited to 'crypto/x86_64cpuid.pl')
| -rw-r--r-- | crypto/x86_64cpuid.pl | 32 |
1 files changed, 32 insertions, 0 deletions
diff --git a/crypto/x86_64cpuid.pl b/crypto/x86_64cpuid.pl index bc06e99cfb..2f657ca9d8 100644 --- a/crypto/x86_64cpuid.pl +++ b/crypto/x86_64cpuid.pl @@ -155,4 +155,36 @@ OPENSSL_ia32_cpuid: or %rcx,%rax ret .size OPENSSL_ia32_cpuid,.-OPENSSL_ia32_cpuid + +.globl OPENSSL_cleanse +.type OPENSSL_cleanse,\@function,2 +.align 16 +OPENSSL_cleanse: + xor %rax,%rax + cmp \$15,%rsi + jae .Lot +.Little: + mov %al,(%rdi) + sub \$1,%rsi + lea 1(%rdi),%rdi + jnz .Little + ret +.align 16 +.Lot: + test \$7,%rdi + jz .Laligned + mov %al,(%rdi) + lea -1(%rsi),%rsi + lea 1(%rdi),%rdi + jmp .Lot +.Laligned: + mov %rax,(%rdi) + lea -8(%rsi),%rsi + test \$-8,%rsi + lea 8(%rdi),%rdi + jnz .Laligned + cmp \$0,%rsi + jne .Little + ret +.size OPENSSL_cleanse,.-OPENSSL_cleanse ___ |