dect
/
linux-2.6
Archived
13
0
Fork 0

crypto: ghash-intel - Add PSHUFB macros

Add PSHUFB macros instead of repeating byte sequences, suggested
by Ingo.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Ingo Molnar <mingo@elte.hu>
This commit is contained in:
Herbert Xu 2009-11-03 09:11:15 -05:00
parent 2d06ef7f42
commit 3b0d65969b
2 changed files with 13 additions and 5 deletions

View File

@ -17,6 +17,7 @@
*/ */
#include <linux/linkage.h> #include <linux/linkage.h>
#include <asm/i387.h>
.align 16 .align 16
.Lbswap_mask: .Lbswap_mask:
@ -101,7 +102,7 @@ ENTRY(clmul_ghash_mul)
movups (%rsi), SHASH movups (%rsi), SHASH
movaps .Lbswap_mask, BSWAP movaps .Lbswap_mask, BSWAP
# pshufb BSWAP, DATA # pshufb BSWAP, DATA
.byte 0x66, 0x0f, 0x38, 0x00, 0xc5 PSHUFB_XMM5_XMM0
call __clmul_gf128mul_ble call __clmul_gf128mul_ble
# pshufb BSWAP, DATA # pshufb BSWAP, DATA
.byte 0x66, 0x0f, 0x38, 0x00, 0xc5 .byte 0x66, 0x0f, 0x38, 0x00, 0xc5
@ -119,12 +120,12 @@ ENTRY(clmul_ghash_update)
movups (%rdi), DATA movups (%rdi), DATA
movups (%rcx), SHASH movups (%rcx), SHASH
# pshufb BSWAP, DATA # pshufb BSWAP, DATA
.byte 0x66, 0x0f, 0x38, 0x00, 0xc5 PSHUFB_XMM5_XMM0
.align 4 .align 4
.Lupdate_loop: .Lupdate_loop:
movups (%rsi), IN1 movups (%rsi), IN1
# pshufb BSWAP, IN1 # pshufb BSWAP, IN1
.byte 0x66, 0x0f, 0x38, 0x00, 0xf5 PSHUFB_XMM5_XMM6
pxor IN1, DATA pxor IN1, DATA
call __clmul_gf128mul_ble call __clmul_gf128mul_ble
sub $16, %rdx sub $16, %rdx
@ -132,7 +133,7 @@ ENTRY(clmul_ghash_update)
cmp $16, %rdx cmp $16, %rdx
jge .Lupdate_loop jge .Lupdate_loop
# pshufb BSWAP, DATA # pshufb BSWAP, DATA
.byte 0x66, 0x0f, 0x38, 0x00, 0xc5 PSHUFB_XMM5_XMM0
movups DATA, (%rdi) movups DATA, (%rdi)
.Lupdate_just_ret: .Lupdate_just_ret:
ret ret
@ -146,7 +147,7 @@ ENTRY(clmul_ghash_setkey)
movaps .Lbswap_mask, BSWAP movaps .Lbswap_mask, BSWAP
movups (%rsi), %xmm0 movups (%rsi), %xmm0
# pshufb BSWAP, %xmm0 # pshufb BSWAP, %xmm0
.byte 0x66, 0x0f, 0x38, 0x00, 0xc5 PSHUFB_XMM5_XMM0
movaps %xmm0, %xmm1 movaps %xmm0, %xmm1
psllq $1, %xmm0 psllq $1, %xmm0
psrlq $63, %xmm1 psrlq $63, %xmm1

View File

@ -10,6 +10,8 @@
#ifndef _ASM_X86_I387_H #ifndef _ASM_X86_I387_H
#define _ASM_X86_I387_H #define _ASM_X86_I387_H
#ifndef __ASSEMBLY__
#include <linux/sched.h> #include <linux/sched.h>
#include <linux/kernel_stat.h> #include <linux/kernel_stat.h>
#include <linux/regset.h> #include <linux/regset.h>
@ -411,4 +413,9 @@ static inline unsigned short get_fpu_mxcsr(struct task_struct *tsk)
} }
} }
#endif /* __ASSEMBLY__ */
#define PSHUFB_XMM5_XMM0 .byte 0x66, 0x0f, 0x38, 0x00, 0xc5
#define PSHUFB_XMM5_XMM6 .byte 0x66, 0x0f, 0x38, 0x00, 0xf5
#endif /* _ASM_X86_I387_H */ #endif /* _ASM_X86_I387_H */