mirrors/cosmopolitan
1
0
Fork 0
mirror of https://github.com/jart/cosmopolitan.git synced 2025-07-06 03:08:31 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Make further progress on non-x86 support

Browse source
This commit is contained in:
Justine Tunney 2023-05-08 21:38:30 -07:00
parent aef9a69a60
commit 036b9a0002
No known key found for this signature in database
GPG key ID: BE714B4575D6E328
155 changed files with 2307 additions and 653 deletions
Download patch file Download diff file Expand all files Collapse all files

3
libc/str/bcmp.c
View file

@ -23,8 +23,7 @@
typedef uint64_t xmm_t __attribute__((__vector_size__(16), __aligned__(1)));
static dontinline antiquity int bcmp_sse(const char *p, const char *q,
size_t n) {
static int bcmp_sse(const char *p, const char *q, size_t n) {
xmm_t a;
while (n > 32) {
a = *(const xmm_t *)p ^ *(const xmm_t *)q;

27
libc/str/rawmemchr.c
View file

@ -22,8 +22,6 @@
#include "libc/nexgen32e/x86feature.h"
#include "libc/str/str.h"
typedef char xmm_t __attribute__((__vector_size__(16), __aligned__(16)));
static inline const unsigned char *rawmemchr_pure(const unsigned char *s,
unsigned char c) {
for (;; ++s) {
@ -34,6 +32,7 @@ static inline const unsigned char *rawmemchr_pure(const unsigned char *s,
}
#ifdef __x86_64__
typedef char xmm_t __attribute__((__vector_size__(16), __aligned__(16)));
noasan static inline const char *rawmemchr_sse(const char *s, unsigned char c) {
unsigned k;
unsigned m;
@ -54,6 +53,12 @@ noasan static inline const char *rawmemchr_sse(const char *s, unsigned char c) {
}
#endif
static inline noasan uint64_t UncheckedAlignedRead64(unsigned char *p) {
return (uint64_t)p[7] << 070 | (uint64_t)p[6] << 060 | (uint64_t)p[5] << 050 |
(uint64_t)p[4] << 040 | (uint64_t)p[3] << 030 | (uint64_t)p[2] << 020 |
(uint64_t)p[1] << 010 | (uint64_t)p[0] << 000;
}
/**
* Returns pointer to first instance of character.
*
@ -72,6 +77,22 @@ void *rawmemchr(const void *s, int c) {
}
return (void *)r;
#else
return rawmemchr_pure(s, c);
uint64_t v, w;
const unsigned char *p;
p = s;
c &= 255;
v = 0x0101010101010101ul * c;
for (; (uintptr_t)p & 7; ++p) {
if (*p == c) return p;
}
for (;; p += 8) {
w = UncheckedAlignedRead64(p);
if ((w = ~(w ^ v) & ((w ^ v) - 0x0101010101010101) & 0x8080808080808080)) {
p += (unsigned)__builtin_ctzll(w) >> 3;
break;
}
}
assert(*p == c);
return p;
#endif
}

3
libc/str/strchr.c
View file

@ -22,8 +22,6 @@
#include "libc/nexgen32e/x86feature.h"
#include "libc/str/str.h"
typedef char xmm_t __attribute__((__vector_size__(16), __aligned__(16)));
static inline const char *strchr_pure(const char *s, int c) {
for (;; ++s) {
if ((*s & 255) == (c & 255)) return s;
@ -32,6 +30,7 @@ static inline const char *strchr_pure(const char *s, int c) {
}
#ifdef __x86_64__
typedef char xmm_t __attribute__((__vector_size__(16), __aligned__(16)));
noasan static inline const char *strchr_sse(const char *s, unsigned char c) {
unsigned k;
unsigned m;

42
libc/str/strchrnul.c
View file

@ -22,8 +22,6 @@
#include "libc/nexgen32e/x86feature.h"
#include "libc/str/str.h"
typedef char xmm_t __attribute__((__vector_size__(16), __aligned__(16)));
static inline const char *strchrnul_pure(const char *s, int c) {
for (;; ++s) {
if ((*s & 255) == (c & 255)) return s;
@ -32,6 +30,7 @@ static inline const char *strchrnul_pure(const char *s, int c) {
}
#ifdef __x86_64__
typedef char xmm_t __attribute__((__vector_size__(16), __aligned__(16)));
noasan static inline const char *strchrnul_sse(const char *s, unsigned char c) {
unsigned k;
unsigned m;
@ -52,6 +51,36 @@ noasan static inline const char *strchrnul_sse(const char *s, unsigned char c) {
}
#endif
noasan static const char *strchrnul_x64(const char *p, uint64_t c) {
unsigned a, b;
uint64_t w, x, y;
for (c *= 0x0101010101010101;; p += 8) {
w = (uint64_t)(255 & p[7]) << 070 | (uint64_t)(255 & p[6]) << 060 |
(uint64_t)(255 & p[5]) << 050 | (uint64_t)(255 & p[4]) << 040 |
(uint64_t)(255 & p[3]) << 030 | (uint64_t)(255 & p[2]) << 020 |
(uint64_t)(255 & p[1]) << 010 | (uint64_t)(255 & p[0]) << 000;
if ((x = ~(w ^ c) & ((w ^ c) - 0x0101010101010101) & 0x8080808080808080) |
(y = ~w & (w - 0x0101010101010101) & 0x8080808080808080)) {
if (x) {
a = __builtin_ctzll(x);
if (y) {
b = __builtin_ctzll(y);
if (a <= b) {
return p + (a >> 3);
} else {
return p + (b >> 3);
}
} else {
return p + (a >> 3);
}
} else {
b = __builtin_ctzll(y);
return p + (b >> 3);
}
}
}
}
/**
* Returns pointer to first instance of character.
*
@ -74,6 +103,13 @@ char *strchrnul(const char *s, int c) {
_unassert((*r & 255) == (c & 255) || !*r);
return (char *)r;
#else
return strchrnul_pure(s, c);
char *r;
for (c &= 255; (uintptr_t)s & 7; ++s) {
if ((*s & 0xff) == c) return s;
if (!*s) return s;
}
r = strchrnul_x64(s, c);
assert((*r & 255) == c || !*r);
return r;
#endif
}

5
libc/str/timingsafe_bcmp.c
View file

@ -24,9 +24,8 @@
typedef uint64_t xmm_t __attribute__((__vector_size__(16), __aligned__(1)));
noasan static dontinline antiquity unsigned timingsafe_bcmp_sse(const char *p,
const char *q,
size_t n) {
noasan static unsigned timingsafe_bcmp_sse(const char *p, const char *q,
size_t n) {
uint64_t w;
xmm_t a = {0};
while (n > 16 + 16) {

14
libc/str/wcwidth.c
View file

@ -17,6 +17,7 @@
│ PERFORMANCE OF THIS SOFTWARE. │
╚─────────────────────────────────────────────────────────────────────────────*/
#include "libc/intrin/likely.h"
#include "libc/str/str.h"
#include "libc/str/unicode.h"
extern const uint8_t kEastAsianWidth[];
@ -28,6 +29,7 @@ extern const uint32_t kCombiningCharsBits;
* Returns cell width of monospace character.
*/
int wcwidth(wchar_t c) {
#ifdef __x86_64__
if (LIKELY(32 <= c && c < 127)) {
return 1;
} else if (!c) {
@ -42,4 +44,16 @@ int wcwidth(wchar_t c) {
} else {
return 1;
}
#else
if (!c) return 0;
if (c < 0 || iswcntrl(c)) return -1;
return 1 +
(c >= 0x1100 &&
(c <= 0x115f || c == 0x2329 || c == 0x232a ||
(c >= 0x2e80 && c <= 0xa4cf && c != 0x303f) ||
(c >= 0xac00 && c <= 0xd7a3) || (c >= 0xf900 && c <= 0xfaff) ||
(c >= 0xfe10 && c <= 0xfe19) || (c >= 0xfe30 && c <= 0xfe6f) ||
(c >= 0xff00 && c <= 0xff60) || (c >= 0xffe0 && c <= 0xffe6) ||
(c >= 0x20000 && c <= 0x2fffd) || (c >= 0x30000 && c <= 0x3fffd)));
#endif
}