mirrors/cosmopolitan
1
0
Fork 0
mirror of https://github.com/jart/cosmopolitan.git synced 2025-07-07 11:48:30 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Make minor improvements

Browse source
This commit is contained in:
Justine Tunney 2020-12-23 23:42:56 -08:00
parent 04caf6f9ad
commit 95b142e4e5
95 changed files with 3818 additions and 2760 deletions
Download patch file Download diff file Expand all files Collapse all files

5
libc/intrin/mpsadbw.h
View file

@ -10,11 +10,12 @@ void mpsadbw(uint16_t[8], const uint8_t[16], const uint8_t[16], uint8_t);
__intrin_xmm_t __mpsadbws(__intrin_xmm_t, __intrin_xmm_t);
#define mpsadbw(C, B, A, I) \
do { \
if (likely(!IsModeDbg() && X86_NEED(SSE) && X86_HAVE(SSE4_1))) { \
if (__builtin_expect(!IsModeDbg() && X86_NEED(SSE) && X86_HAVE(SSE4_1), \
1)) { \
__intrin_xmm_t *Xmm0 = (void *)(C); \
const __intrin_xmm_t *Xmm1 = (const __intrin_xmm_t *)(B); \
const __intrin_xmm_t *Xmm2 = (const __intrin_xmm_t *)(A); \
if (isconstant(I)) { \
if (__builtin_constant_p(I)) { \
if (!X86_NEED(AVX)) { \
asm("mpsadbw\t%2,%1,%0" \
: "=x"(*Xmm0) \

59
libc/intrin/palignr.h
View file

@ -9,35 +9,36 @@ void palignr(void *, const void *, const void *, unsigned long);
#if !defined(__STRICT_ANSI__) && !defined(__chibicc__)
__intrin_xmm_t __palignrs(__intrin_xmm_t, __intrin_xmm_t);
#define palignr(C, B, A, I) \
do { \
if (likely(!IsModeDbg() && X86_NEED(SSE) && X86_HAVE(SSSE3))) { \
__intrin_xmm_t *Xmm0 = (void *)(C); \
const __intrin_xmm_t *Xmm1 = (const __intrin_xmm_t *)(B); \
const __intrin_xmm_t *Xmm2 = (const __intrin_xmm_t *)(A); \
if (isconstant(I)) { \
if (!X86_NEED(AVX)) { \
asm("palignr\t%2,%1,%0" \
: "=x"(*Xmm0) \
: "x"(*Xmm2), "i"(I), "0"(*Xmm1)); \
} else { \
asm("vpalignr\t%3,%2,%1,%0" \
: "=x"(*Xmm0) \
: "x"(*Xmm1), "x"(*Xmm2), "i"(I)); \
} \
} else { \
unsigned long Vimm = (I); \
typeof(__palignrs) *Fn; \
if (likely(Vimm < 32)) { \
Fn = (typeof(__palignrs) *)((uintptr_t)&__palignrs + Vimm * 8); \
*Xmm0 = Fn(*Xmm1, *Xmm2); \
} else { \
memset(Xmm0, 0, 16); \
} \
} \
} else { \
palignr(C, B, A, I); \
} \
#define palignr(C, B, A, I) \
do { \
if (__builtin_expect(!IsModeDbg() && X86_NEED(SSE) && X86_HAVE(SSSE3), \
1)) { \
__intrin_xmm_t *Xmm0 = (void *)(C); \
const __intrin_xmm_t *Xmm1 = (const __intrin_xmm_t *)(B); \
const __intrin_xmm_t *Xmm2 = (const __intrin_xmm_t *)(A); \
if (__builtin_constant_p(I)) { \
if (!X86_NEED(AVX)) { \
asm("palignr\t%2,%1,%0" \
: "=x"(*Xmm0) \
: "x"(*Xmm2), "i"(I), "0"(*Xmm1)); \
} else { \
asm("vpalignr\t%3,%2,%1,%0" \
: "=x"(*Xmm0) \
: "x"(*Xmm1), "x"(*Xmm2), "i"(I)); \
} \
} else { \
unsigned long Vimm = (I); \
typeof(__palignrs) *Fn; \
if (__builtin_expect(Vimm < 32, 1)) { \
Fn = (typeof(__palignrs) *)((uintptr_t)&__palignrs + Vimm * 8); \
*Xmm0 = Fn(*Xmm1, *Xmm2); \
} else { \
memset(Xmm0, 0, 16); \
} \
} \
} else { \
palignr(C, B, A, I); \
} \
} while (0)
#endif

43
libc/intrin/pslldq.h
View file

@ -8,27 +8,28 @@ void pslldq(uint8_t[16], const uint8_t[16], unsigned long);
#ifndef __STRICT_ANSI__
__intrin_xmm_t __pslldqs(__intrin_xmm_t);
#define pslldq(B, A, I) \
do { \
if (likely(!IsModeDbg() && X86_NEED(SSE) && X86_HAVE(SSE2))) { \
__intrin_xmm_t *Xmm0 = (void *)(B); \
const __intrin_xmm_t *Xmm1 = (const __intrin_xmm_t *)(A); \
if (isconstant(I)) { \
if (!X86_NEED(AVX)) { \
asm("pslldq\t%1,%0" : "=x"(*Xmm0) : "i"(I), "0"(*Xmm1)); \
} else { \
asm("vpslldq\t%2,%1,%0" : "=x"(*Xmm0) : "x"(*Xmm1), "i"(I)); \
} \
} else { \
unsigned long Vimm = (I); \
typeof(__pslldqs) *Fn; \
if (Vimm > 16) Vimm = 16; \
Fn = (typeof(__pslldqs) *)((uintptr_t)&__pslldqs + Vimm * 8); \
*Xmm0 = Fn(*Xmm1); \
} \
} else { \
pslldq(B, A, I); \
} \
#define pslldq(B, A, I) \
do { \
if (__builtin_expect(!IsModeDbg() && X86_NEED(SSE) && X86_HAVE(SSE2), \
1)) { \
__intrin_xmm_t *Xmm0 = (void *)(B); \
const __intrin_xmm_t *Xmm1 = (const __intrin_xmm_t *)(A); \
if (__builtin_constant_p(I)) { \
if (!X86_NEED(AVX)) { \
asm("pslldq\t%1,%0" : "=x"(*Xmm0) : "i"(I), "0"(*Xmm1)); \
} else { \
asm("vpslldq\t%2,%1,%0" : "=x"(*Xmm0) : "x"(*Xmm1), "i"(I)); \
} \
} else { \
unsigned long Vimm = (I); \
typeof(__pslldqs) *Fn; \
if (Vimm > 16) Vimm = 16; \
Fn = (typeof(__pslldqs) *)((uintptr_t)&__pslldqs + Vimm * 8); \
*Xmm0 = Fn(*Xmm1); \
} \
} else { \
pslldq(B, A, I); \
} \
} while (0)
#endif

43
libc/intrin/psrldq.h
View file

@ -8,27 +8,28 @@ void psrldq(uint8_t[16], const uint8_t[16], unsigned long);
#ifndef __STRICT_ANSI__
__intrin_xmm_t __psrldqs(__intrin_xmm_t);
#define psrldq(B, A, I) \
do { \
if (likely(!IsModeDbg() && X86_NEED(SSE) && X86_HAVE(SSE2))) { \
__intrin_xmm_t *Xmm0 = (void *)(B); \
const __intrin_xmm_t *Xmm1 = (const __intrin_xmm_t *)(A); \
if (isconstant(I)) { \
if (!X86_NEED(AVX)) { \
asm("psrldq\t%1,%0" : "=x"(*Xmm0) : "i"(I), "0"(*Xmm1)); \
} else { \
asm("vpsrldq\t%2,%1,%0" : "=x"(*Xmm0) : "x"(*Xmm1), "i"(I)); \
} \
} else { \
unsigned long Vimm = (I); \
typeof(__psrldqs) *Fn; \
if (Vimm > 16) Vimm = 16; \
Fn = (typeof(__psrldqs) *)((uintptr_t)&__psrldqs + Vimm * 8); \
*Xmm0 = Fn(*Xmm1); \
} \
} else { \
psrldq(B, A, I); \
} \
#define psrldq(B, A, I) \
do { \
if (__builtin_expect(!IsModeDbg() && X86_NEED(SSE) && X86_HAVE(SSE2), \
1)) { \
__intrin_xmm_t *Xmm0 = (void *)(B); \
const __intrin_xmm_t *Xmm1 = (const __intrin_xmm_t *)(A); \
if (__builtin_constant_p(I)) { \
if (!X86_NEED(AVX)) { \
asm("psrldq\t%1,%0" : "=x"(*Xmm0) : "i"(I), "0"(*Xmm1)); \
} else { \
asm("vpsrldq\t%2,%1,%0" : "=x"(*Xmm0) : "x"(*Xmm1), "i"(I)); \
} \
} else { \
unsigned long Vimm = (I); \
typeof(__psrldqs) *Fn; \
if (Vimm > 16) Vimm = 16; \
Fn = (typeof(__psrldqs) *)((uintptr_t)&__psrldqs + Vimm * 8); \
*Xmm0 = Fn(*Xmm1); \
} \
} else { \
psrldq(B, A, I); \
} \
} while (0)
#endif