| 1 |
// This file is generated from a similarly-named Perl script in the BoringSSL |
| 2 |
// source tree. Do not edit by hand. |
| 3 |
|
| 4 |
#if !defined(__has_feature) |
| 5 |
#define __has_feature(x) 0 |
| 6 |
#endif |
| 7 |
#if __has_feature(memory_sanitizer) && !defined(OPENSSL_NO_ASM) |
| 8 |
#define OPENSSL_NO_ASM |
| 9 |
#endif |
| 10 |
|
| 11 |
#if !defined(OPENSSL_NO_ASM) && defined(__AARCH64EL__) && defined(__APPLE__) |
| 12 |
#if defined(BORINGSSL_PREFIX) |
| 13 |
#include <boringssl_prefix_symbols_asm.h> |
| 14 |
#endif |
| 15 |
#include <openssl/arm_arch.h> |
| 16 |
|
| 17 |
.text |
| 18 |
|
| 19 |
// BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, |
| 20 |
// size_t num); |
| 21 |
|
| 22 |
.globl _bn_add_words |
| 23 |
.private_extern _bn_add_words |
| 24 |
.align 4 |
| 25 |
_bn_add_words: |
| 26 |
AARCH64_VALID_CALL_TARGET |
| 27 |
# Clear the carry flag. |
| 28 |
cmn xzr, xzr |
| 29 |
|
| 30 |
# aarch64 can load two registers at a time, so we do two loop iterations at |
| 31 |
# at a time. Split x3 = 2 * x8 + x3. This allows loop |
| 32 |
# operations to use CBNZ without clobbering the carry flag. |
| 33 |
lsr x8, x3, #1 |
| 34 |
and x3, x3, #1 |
| 35 |
|
| 36 |
cbz x8, Ladd_tail |
| 37 |
Ladd_loop: |
| 38 |
ldp x4, x5, [x1], #16 |
| 39 |
ldp x6, x7, [x2], #16 |
| 40 |
sub x8, x8, #1 |
| 41 |
adcs x4, x4, x6 |
| 42 |
adcs x5, x5, x7 |
| 43 |
stp x4, x5, [x0], #16 |
| 44 |
cbnz x8, Ladd_loop |
| 45 |
|
| 46 |
Ladd_tail: |
| 47 |
cbz x3, Ladd_exit |
| 48 |
ldr x4, [x1], #8 |
| 49 |
ldr x6, [x2], #8 |
| 50 |
adcs x4, x4, x6 |
| 51 |
str x4, [x0], #8 |
| 52 |
|
| 53 |
Ladd_exit: |
| 54 |
cset x0, cs |
| 55 |
ret |
| 56 |
|
| 57 |
|
| 58 |
// BN_ULONG bn_sub_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp, |
| 59 |
// size_t num); |
| 60 |
|
| 61 |
.globl _bn_sub_words |
| 62 |
.private_extern _bn_sub_words |
| 63 |
.align 4 |
| 64 |
_bn_sub_words: |
| 65 |
AARCH64_VALID_CALL_TARGET |
| 66 |
# Set the carry flag. Arm's borrow bit is flipped from the carry flag, |
| 67 |
# so we want C = 1 here. |
| 68 |
cmp xzr, xzr |
| 69 |
|
| 70 |
# aarch64 can load two registers at a time, so we do two loop iterations at |
| 71 |
# at a time. Split x3 = 2 * x8 + x3. This allows loop |
| 72 |
# operations to use CBNZ without clobbering the carry flag. |
| 73 |
lsr x8, x3, #1 |
| 74 |
and x3, x3, #1 |
| 75 |
|
| 76 |
cbz x8, Lsub_tail |
| 77 |
Lsub_loop: |
| 78 |
ldp x4, x5, [x1], #16 |
| 79 |
ldp x6, x7, [x2], #16 |
| 80 |
sub x8, x8, #1 |
| 81 |
sbcs x4, x4, x6 |
| 82 |
sbcs x5, x5, x7 |
| 83 |
stp x4, x5, [x0], #16 |
| 84 |
cbnz x8, Lsub_loop |
| 85 |
|
| 86 |
Lsub_tail: |
| 87 |
cbz x3, Lsub_exit |
| 88 |
ldr x4, [x1], #8 |
| 89 |
ldr x6, [x2], #8 |
| 90 |
sbcs x4, x4, x6 |
| 91 |
str x4, [x0], #8 |
| 92 |
|
| 93 |
Lsub_exit: |
| 94 |
cset x0, cc |
| 95 |
ret |
| 96 |
|
| 97 |
#endif // !OPENSSL_NO_ASM && defined(__AARCH64EL__) && defined(__APPLE__) |
| 98 |
#if defined(__ELF__) |
| 99 |
// See https://www.airs.com/blog/archives/518. |
| 100 |
.section .note.GNU-stack,"",%progbits |
| 101 |
#endif |