mirrors/cosmopolitan
1
0
Fork 0
mirror of https://github.com/jart/cosmopolitan.git synced 2025-05-27 15:52:28 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Reduce build latency and fix old cpu bugs

Browse source
This commit is contained in:
Justine Tunney 2021-08-05 14:43:53 -07:00
parent df8ab0aa0c
commit 533f3d1ef1
69 changed files with 43069 additions and 43683 deletions
Download patch file Download diff file Expand all files Collapse all files

151
third_party/mbedtls/test/secp384r1_test.c vendored
View file

@ -48,96 +48,13 @@ TEST(secp384r1, testIsTheSame) {
for (i = 0; i < 12; ++i) {
printf("0x%016lx vs. 0x%016lx %d\n", A.p[i], B.p[i], A.p[i] == B.p[i]);
}
exit(1);
ASSERT_TRUE(false);
}
}
mbedtls_mpi_free(&B);
mbedtls_mpi_free(&A);
}
static inline bool mbedtls_p384_gte(uint64_t p[7]) {
return (((int64_t)p[6] > 0 ||
(p[5] > 0xffffffffffffffff ||
(p[5] == 0xffffffffffffffff &&
(p[4] > 0xffffffffffffffff ||
(p[4] == 0xffffffffffffffff &&
(p[3] > 0xffffffffffffffff ||
(p[3] == 0xffffffffffffffff &&
(p[2] > 0xfffffffffffffffe ||
(p[2] == 0xfffffffffffffffe &&
(p[1] > 0xffffffff00000000 ||
(p[1] == 0xffffffff00000000 &&
(p[0] > 0x00000000ffffffff ||
(p[0] == 0x00000000ffffffff))))))))))))));
}
static inline void mbedtls_p384_gro(uint64_t p[7]) {
#if defined(__x86_64__) && !defined(__STRICT_ANSI__)
asm("addq\t%1,%0\n\t"
"adcq\t%2,8+%0\n\t"
"adcq\t%3,16+%0\n\t"
"adcq\t%4,24+%0\n\t"
"adcq\t%4,32+%0\n\t"
"adcq\t%4,40+%0\n\t"
"adcq\t$0,48+%0"
: "+o"(*p)
: "r"(0x00000000ffffffffl), "r"(0xffffffff00000000),
"i"(0xfffffffffffffffel), "i"(0xffffffffffffffff)
: "memory", "cc");
#else
uint64_t c;
ADC(p[0], p[0], 0x00000000ffffffff, 0, c);
ADC(p[1], p[1], 0xffffffff00000000, c, c);
ADC(p[2], p[2], 0xfffffffffffffffe, c, c);
ADC(p[3], p[3], 0xffffffffffffffff, c, c);
ADC(p[4], p[4], 0xffffffffffffffff, c, c);
ADC(p[5], p[5], 0xffffffffffffffff, c, c);
ADC(p[6], p[6], 0, c, c);
#endif
}
static inline void mbedtls_p384_red(uint64_t p[7]) {
#if defined(__x86_64__) && !defined(__STRICT_ANSI__)
asm("subq\t%1,%0\n\t"
"sbbq\t%2,8+%0\n\t"
"sbbq\t%3,16+%0\n\t"
"sbbq\t%4,24+%0\n\t"
"sbbq\t%4,32+%0\n\t"
"sbbq\t%4,40+%0\n\t"
"sbbq\t$0,48+%0"
: "+o"(*p)
: "r"(0x00000000ffffffffl), "r"(0xffffffff00000000),
"i"(0xfffffffffffffffel), "i"(0xffffffffffffffff)
: "memory", "cc");
#else
uint64_t c;
SBB(p[0], p[0], 0x00000000ffffffff, 0, c);
SBB(p[1], p[1], 0xffffffff00000000, c, c);
SBB(p[2], p[2], 0xfffffffffffffffe, c, c);
SBB(p[3], p[3], 0xffffffffffffffff, c, c);
SBB(p[4], p[4], 0xffffffffffffffff, c, c);
SBB(p[5], p[5], 0xffffffffffffffff, c, c);
SBB(p[6], p[6], 0, c, c);
#endif
}
static inline void mbedtls_p384_rum(uint64_t p[7]) {
while (mbedtls_p384_gte(p)) mbedtls_p384_red(p);
}
static inline void mbedtls_p384_mod(uint64_t X[12]) {
secp384r1(X);
if ((int64_t)X[6] < 0) {
do {
mbedtls_p384_gro(X);
} while ((int64_t)X[6] < 0);
} else {
while (mbedtls_p384_gte(X)) {
mbedtls_p384_red(X);
}
}
}
TEST(secp384r1, needsDownwardCorrection) {
int i;
uint64_t P[6] = {
@ -174,7 +91,7 @@ TEST(secp384r1, needsDownwardCorrection) {
for (i = 0; i < 12; ++i) {
printf("0x%016lx vs. 0x%016lx %d\n", W[i], X[i], W[i] == X[i]);
}
exit(1);
ASSERT_TRUE(false);
}
}
@ -215,7 +132,7 @@ TEST(secp384r1, needsUpwardCorrection) {
for (i = 0; i < 12; ++i) {
printf("0x%016lx vs. 0x%016lx %d\n", W[i], X[i], W[i] == X[i]);
}
exit(1);
ASSERT_TRUE(false);
}
}
@ -229,66 +146,4 @@ BENCH(secp384r1, bench) {
mbedtls_mpi_free(&A);
}
void mbedtls_p384_shl_a(uint64_t p[7]) {
asm("shlq\t%0\n\t"
"rclq\t8+%0\n\t"
"rclq\t16+%0\n\t"
"rclq\t24+%0\n\t"
"rclq\t32+%0\n\t"
"rclq\t40+%0\n\t"
"rclq\t48+%0\n\t"
: "+o"(*p)
: /* no inputs */
: "memory", "cc");
mbedtls_p384_rum(p);
}
void mbedtls_p384_shl_b(uint64_t p[7]) {
p[6] = p[5] >> 63;
p[5] = p[5] << 1 | p[4] >> 63;
p[4] = p[4] << 1 | p[3] >> 63;
p[3] = p[3] << 1 | p[2] >> 63;
p[2] = p[2] << 1 | p[1] >> 63;
p[1] = p[1] << 1 | p[0] >> 63;
p[0] = p[0] << 1;
mbedtls_p384_rum(p);
}
BENCH(shl, bench) {
uint64_t A[7] = {0};
EZBENCH2("mbedtls_p384_shl_a", donothing, mbedtls_p384_shl_a(A));
EZBENCH2("mbedtls_p384_shl_b", donothing, mbedtls_p384_shl_b(A));
}
void mbedtls_p384_red_a(uint64_t p[7]) {
asm("subq\t%1,%0\n\t"
"sbbq\t%2,8+%0\n\t"
"sbbq\t%3,16+%0\n\t"
"sbbq\t%4,24+%0\n\t"
"sbbq\t%4,32+%0\n\t"
"sbbq\t%4,40+%0\n\t"
"sbbq\t$0,48+%0"
: "+o"(*p)
: "r"(0x00000000ffffffffl), "r"(0xffffffff00000000),
"i"(0xfffffffffffffffel), "i"(0xffffffffffffffff)
: "memory", "cc");
}
void mbedtls_p384_red_b(uint64_t p[7]) {
uint64_t c;
SBB(p[0], p[0], 0x00000000ffffffff, 0, c);
SBB(p[1], p[1], 0xffffffff00000000, c, c);
SBB(p[2], p[2], 0xfffffffffffffffe, c, c);
SBB(p[3], p[3], 0xffffffffffffffff, c, c);
SBB(p[4], p[4], 0xffffffffffffffff, c, c);
SBB(p[5], p[5], 0xffffffffffffffff, c, c);
SBB(p[6], p[6], 0, c, c);
}
BENCH(red, bench) {
uint64_t A[7] = {0};
EZBENCH2("mbedtls_p384_red_a", donothing, mbedtls_p384_red_a(A));
EZBENCH2("mbedtls_p384_red_b", donothing, mbedtls_p384_red_b(A));
}
#endif /* MBEDTLS_ECP_C */