mirrors/cosmopolitan
1
0
Fork 0
mirror of https://github.com/jart/cosmopolitan.git synced 2025-07-02 09:18:31 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Make fixes, improvements, and chibicc python bindings

Browse source 
- python now mixes audio 10x faster
- python octal notation is restored
- chibicc now builds code 3x faster
- chibicc now has help documentation
- chibicc can now generate basic python bindings
- linenoise now supports some paredit-like features

See #141
This commit is contained in:
Justine Tunney 2021-10-08 08:11:51 -07:00
parent 28997f3acb
commit 7061c79c22
121 changed files with 5272 additions and 1928 deletions
Download patch file Download diff file Expand all files Collapse all files

41
libc/intrin/memcmp.c
View file

@ -29,8 +29,8 @@ static noinline antiquity int memcmp_sse(const unsigned char *p,
unsigned u, u0, u1, u2, u3;
if (n > 32) {
while (n > 16 + 16) {
if (!(u = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
*(const xmm_t *)p, *(const xmm_t *)q)) -
if (!(u = __builtin_ia32_pmovmskb128(*(const xmm_t *)p ==
*(const xmm_t *)q) -
0xffff)) {
n -= 16;
p += 16;
@ -41,11 +41,10 @@ static noinline antiquity int memcmp_sse(const unsigned char *p,
}
}
}
if (!(u = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
*(const xmm_t *)p, *(const xmm_t *)q)) -
if (!(u = __builtin_ia32_pmovmskb128(*(const xmm_t *)p == *(const xmm_t *)q) -
0xffff)) {
if (!(u = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
*(const xmm_t *)(p + n - 16), *(const xmm_t *)(q + n - 16))) -
if (!(u = __builtin_ia32_pmovmskb128(*(const xmm_t *)(p + n - 16) ==
*(const xmm_t *)(q + n - 16)) -
0xffff)) {
return 0;
} else {
@ -65,14 +64,14 @@ microarchitecture("avx") static int memcmp_avx(const unsigned char *p,
unsigned u, u0, u1, u2, u3;
if (n > 32) {
while (n >= 16 + 64) {
u0 = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
((const xmm_t *)p)[0], ((const xmm_t *)q)[0]));
u1 = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
((const xmm_t *)p)[1], ((const xmm_t *)q)[1]));
u2 = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
((const xmm_t *)p)[2], ((const xmm_t *)q)[2]));
u3 = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
((const xmm_t *)p)[3], ((const xmm_t *)q)[3]));
u0 = __builtin_ia32_pmovmskb128(
(((const xmm_t *)p)[0] == ((const xmm_t *)q)[0]));
u1 = __builtin_ia32_pmovmskb128(
(((const xmm_t *)p)[1] == ((const xmm_t *)q)[1]));
u2 = __builtin_ia32_pmovmskb128(
(((const xmm_t *)p)[2] == ((const xmm_t *)q)[2]));
u3 = __builtin_ia32_pmovmskb128(
(((const xmm_t *)p)[3] == ((const xmm_t *)q)[3]));
w = (uint64_t)u0 | (uint64_t)u1 << 16 | (uint64_t)u2 << 32 |
(uint64_t)u3 << 48;
if (w == -1) {
@ -85,8 +84,8 @@ microarchitecture("avx") static int memcmp_avx(const unsigned char *p,
}
}
while (n > 16 + 16) {
if (!(u = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
*(const xmm_t *)p, *(const xmm_t *)q)) -
if (!(u = __builtin_ia32_pmovmskb128(*(const xmm_t *)p ==
*(const xmm_t *)q) -
0xffff)) {
n -= 16;
p += 16;
@ -97,11 +96,10 @@ microarchitecture("avx") static int memcmp_avx(const unsigned char *p,
}
}
}
if (!(u = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
*(const xmm_t *)p, *(const xmm_t *)q)) -
if (!(u = __builtin_ia32_pmovmskb128(*(const xmm_t *)p == *(const xmm_t *)q) -
0xffff)) {
if (!(u = __builtin_ia32_pmovmskb128(__builtin_ia32_pcmpeqb128(
*(const xmm_t *)(p + n - 16), *(const xmm_t *)(q + n - 16))) -
if (!(u = __builtin_ia32_pmovmskb128(*(const xmm_t *)(p + n - 16) ==
*(const xmm_t *)(q + n - 16)) -
0xffff)) {
return 0;
} else {
@ -149,7 +147,8 @@ int memcmp(const void *a, const void *b, size_t n) {
uint32_t k, i, j;
uint64_t w, x, y;
const unsigned char *p, *q;
if ((p = a) == (q = b)) return 0;
if ((p = a) == (q = b) || !n) return 0;
if ((c = *p - *q)) return c;
if (!IsTiny()) {
if (n <= 16) {
if (n >= 8) {