diff options
Diffstat (limited to 'src/crypto/rc4')
| -rw-r--r-- | src/crypto/rc4/asm/rc4-x86_64.pl | 26 | ||||
| -rw-r--r-- | src/crypto/rc4/rc4.c | 67 |
2 files changed, 21 insertions, 72 deletions
diff --git a/src/crypto/rc4/asm/rc4-x86_64.pl b/src/crypto/rc4/asm/rc4-x86_64.pl index 2c52ac0..db46242 100644 --- a/src/crypto/rc4/asm/rc4-x86_64.pl +++ b/src/crypto/rc4/asm/rc4-x86_64.pl @@ -502,32 +502,6 @@ asm_RC4_set_key: mov %eax,-4($dat) ret .size asm_RC4_set_key,.-asm_RC4_set_key - -.globl RC4_options -.type RC4_options,\@abi-omnipotent -.align 16 -RC4_options: - lea .Lopts(%rip),%rax - mov OPENSSL_ia32cap_P(%rip),%rdx - mov (%rdx),%edx - bt \$20,%edx - jc .L8xchar - bt \$30,%edx - jnc .Ldone - add \$25,%rax - ret -.L8xchar: - add \$12,%rax -.Ldone: - ret -.align 64 -.Lopts: -.asciz "rc4(8x,int)" -.asciz "rc4(8x,char)" -.asciz "rc4(16x,int)" -.asciz "RC4 for x86_64, CRYPTOGAMS by <appro\@openssl.org>" -.align 64 -.size RC4_options,.-RC4_options ___ # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, diff --git a/src/crypto/rc4/rc4.c b/src/crypto/rc4/rc4.c index 00b59c8..2a98fd0 100644 --- a/src/crypto/rc4/rc4.c +++ b/src/crypto/rc4/rc4.c @@ -67,8 +67,6 @@ #error "Unknown word size" #endif -#define RC4_INT uint32_t - /* RC4 as implemented from a posting from * Newsgroups: sci.crypt @@ -78,44 +76,14 @@ * Date: Wed, 14 Sep 1994 06:35:31 GMT */ void RC4(RC4_KEY *key, size_t len, const uint8_t *in, uint8_t *out) { - register RC4_INT *d; - register RC4_INT x, y, tx, ty; + uint32_t *d; + uint32_t x, y, tx, ty; size_t i; x = key->x; y = key->y; d = key->data; -#if defined(RC4_CHUNK) -/* The original reason for implementing this(*) was the fact that - * pre-21164a Alpha CPUs don't have byte load/store instructions - * and e.g. a byte store has to be done with 64-bit load, shift, - * and, or and finally 64-bit store. Peaking data and operating - * at natural word size made it possible to reduce amount of - * instructions as well as to perform early read-ahead without - * suffering from RAW (read-after-write) hazard. This resulted - * in ~40%(**) performance improvement on 21064 box with gcc. - * But it's not only Alpha users who win here:-) Thanks to the - * early-n-wide read-ahead this implementation also exhibits - * >40% speed-up on SPARC and 20-30% on 64-bit MIPS (depending - * on sizeof(RC4_INT)). - * - * (*) "this" means code which recognizes the case when input - * and output pointers appear to be aligned at natural CPU - * word boundary - * (**) i.e. according to 'apps/openssl speed rc4' benchmark, - * crypto/rc4/rc4speed.c exhibits almost 70% speed-up... - * - * Cavets. - * - * - RC4_CHUNK="unsigned long long" should be a #1 choice for - * UltraSPARC. Unfortunately gcc generates very slow code - * (2.5-3 times slower than one generated by Sun's WorkShop - * C) and therefore gcc (at least 2.95 and earlier) should - * always be told that RC4_CHUNK="unsigned long". - * - * <appro@fy.chalmers.se> */ - #define RC4_STEP \ (x = (x + 1) & 0xff, tx = d[x], y = (tx + y) & 0xff, ty = d[y], d[y] = tx, \ d[x] = ty, (RC4_CHUNK)d[(tx + ty) & 0xff]) @@ -255,7 +223,6 @@ void RC4(RC4_KEY *key, size_t len, const uint8_t *in, uint8_t *out) { return; } } -#endif #define LOOP(in, out) \ x = ((x + 1) & 0xff); \ tx = d[x]; \ @@ -285,34 +252,42 @@ void RC4(RC4_KEY *key, size_t len, const uint8_t *in, uint8_t *out) { in += 8; out += 8; #endif - if (--i == 0) + if (--i == 0) { break; + } } } i = len & 0x07; if (i) { for (;;) { RC4_LOOP(in, out, 0); - if (--i == 0) + if (--i == 0) { break; + } RC4_LOOP(in, out, 1); - if (--i == 0) + if (--i == 0) { break; + } RC4_LOOP(in, out, 2); - if (--i == 0) + if (--i == 0) { break; + } RC4_LOOP(in, out, 3); - if (--i == 0) + if (--i == 0) { break; + } RC4_LOOP(in, out, 4); - if (--i == 0) + if (--i == 0) { break; + } RC4_LOOP(in, out, 5); - if (--i == 0) + if (--i == 0) { break; + } RC4_LOOP(in, out, 6); - if (--i == 0) + if (--i == 0) { break; + } } } key->x = x; @@ -320,9 +295,9 @@ void RC4(RC4_KEY *key, size_t len, const uint8_t *in, uint8_t *out) { } void RC4_set_key(RC4_KEY *rc4key, unsigned len, const uint8_t *key) { - register RC4_INT tmp; - register int id1, id2; - register RC4_INT *d; + uint32_t tmp; + int id1, id2; + uint32_t *d; unsigned int i; d = &rc4key->data[0]; |