233 lines
12 KiB
C
233 lines
12 KiB
C
#ifndef COSMOPOLITAN_LIBC_BITS_XMMINTRIN_H_
|
||
#define COSMOPOLITAN_LIBC_BITS_XMMINTRIN_H_
|
||
#include "libc/bits/emmintrin.internal.h"
|
||
#include "libc/bits/progn.internal.h"
|
||
#include "libc/dce.h"
|
||
|
||
#define _MM_EXCEPT_MASK 0x003f
|
||
#define _MM_EXCEPT_INVALID 0x0001
|
||
#define _MM_EXCEPT_DENORM 0x0002
|
||
#define _MM_EXCEPT_DIV_ZERO 0x0004
|
||
#define _MM_EXCEPT_OVERFLOW 0x0008
|
||
#define _MM_EXCEPT_UNDERFLOW 0x0010
|
||
#define _MM_EXCEPT_INEXACT 0x0020
|
||
#define _MM_MASK_MASK 0x1f80
|
||
#define _MM_MASK_INVALID 0x0080
|
||
#define _MM_MASK_DENORM 0x0100
|
||
#define _MM_MASK_DIV_ZERO 0x0200
|
||
#define _MM_MASK_OVERFLOW 0x0400
|
||
#define _MM_MASK_UNDERFLOW 0x0800
|
||
#define _MM_MASK_INEXACT 0x1000
|
||
#define _MM_ROUND_MASK 0x6000
|
||
#define _MM_ROUND_NEAREST 0x0000
|
||
#define _MM_ROUND_DOWN 0x2000
|
||
#define _MM_ROUND_UP 0x4000
|
||
#define _MM_ROUND_TOWARD_ZERO 0x6000
|
||
#define _MM_FLUSH_ZERO_MASK 0x8000
|
||
#define _MM_FLUSH_ZERO_ON 0x8000
|
||
#define _MM_FLUSH_ZERO_OFF 0x0000
|
||
|
||
#define _MM_SHUFFLE(A, B, C, D) (((A) << 6) | ((B) << 4) | ((C) << 2) | (D))
|
||
|
||
#if !(__ASSEMBLER__ + __LINKER__ + 0)
|
||
|
||
typedef int __v4si _Vector_size(16);
|
||
typedef unsigned int __v4su _Vector_size(16);
|
||
typedef float __v4sf _Vector_size(16);
|
||
typedef float __m128 _Vector_size(16) forcealign(16) mayalias;
|
||
typedef float __m128_u _Vector_size(16) forcealign(1) mayalias;
|
||
|
||
/*───────────────────────────────────────────────────────────────────────────│─╗
|
||
│ cosmopolitan § it's a trap! » sse » simd ops ─╬─│┼
|
||
╚────────────────────────────────────────────────────────────────────────────│*/
|
||
|
||
#define _mm_add_ps(M128_0, M128_1) \
|
||
((__m128)((__v4sf)(M128_0) + (__v4sf)(M128_1)))
|
||
#define _mm_sub_ps(M128_0, M128_1) \
|
||
((__m128)((__v4sf)(M128_0) - (__v4sf)(M128_1)))
|
||
#define _mm_mul_ps(M128_0, M128_1) \
|
||
((__m128)((__v4sf)(M128_0) * (__v4sf)(M128_1)))
|
||
#define _mm_div_ps(M128_0, M128_1) \
|
||
((__m128)((__v4sf)(M128_0) / (__v4sf)(M128_1)))
|
||
#define _mm_and_ps(M128_0, M128_1) \
|
||
((__m128)((__v4su)(M128_0) & (__v4su)(M128_1)))
|
||
#define _mm_or_ps(M128_0, M128_1) \
|
||
((__m128)((__v4su)(M128_0) | (__v4su)(M128_1)))
|
||
#define _mm_xor_ps(M128_0, M128_1) /* XORPD [u32 simd xor] */ \
|
||
((__m128)((__v4su)(M128_0) ^ (__v4su)(M128_1)))
|
||
#define _mm_andnot_ps(M128_0, M128_1) /* ANDNPS [u32 simd nand] */ \
|
||
((__m128)(~(__v4su)(M128_0) & (__v4su)(M128_1)))
|
||
#define _mm_rcp_ps(M128) __builtin_ia32_rcpps((__v4sf)(M128))
|
||
#define _mm_sqrt_ps(M128) __builtin_ia32_sqrtps((__v4sf)(M128))
|
||
#define _mm_rsqrt_ps(M128) __builtin_ia32_rsqrtps((__v4sf)(M128))
|
||
|
||
#define _mm_min_ps(M128_0, M128_1) \
|
||
__builtin_ia32_minps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_max_ps(M128_0, M128_1) \
|
||
__builtin_ia32_maxps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_min_ss(M128_0, M128_1) \
|
||
__builtin_ia32_minss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_max_ss(M128_0, M128_1) \
|
||
__builtin_ia32_maxss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpeq_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpeqps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpneq_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpneqps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmplt_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpltps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpnlt_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpnltps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmple_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpleps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpnle_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpnleps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpgt_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpltps((__v4sf)(M128_1), (__v4sf)(M128_0))
|
||
#define _mm_cmpngt_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpnltps((__v4sf)(M128_1), (__v4sf)(M128_0))
|
||
#define _mm_cmpge_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpleps((__v4sf)(M128_1), (__v4sf)(M128_0))
|
||
#define _mm_cmpnge_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpnleps((__v4sf)(M128_1), (__v4sf)(M128_0))
|
||
#define _mm_cmpord_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpordps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpunord_ps(M128_0, M128_1) \
|
||
__builtin_ia32_cmpunordps((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
|
||
/*───────────────────────────────────────────────────────────────────────────│─╗
|
||
│ cosmopolitan § it's a trap! » sse » scalar ops ─╬─│┼
|
||
╚────────────────────────────────────────────────────────────────────────────│*/
|
||
|
||
forceinline __m128 _mm_add_ss(__m128 m128_0, __m128 m128_1) {
|
||
m128_0[0] += m128_1[0];
|
||
return m128_0;
|
||
}
|
||
|
||
forceinline __m128 _mm_sub_ss(__m128 m128_0, __m128 m128_1) {
|
||
m128_0[0] -= m128_1[0];
|
||
return m128_0;
|
||
}
|
||
|
||
forceinline __m128 _mm_mul_ss(__m128 m128_0, __m128 m128_1) {
|
||
m128_0[0] *= m128_1[0];
|
||
return m128_0;
|
||
}
|
||
|
||
forceinline __m128 _mm_div_ss(__m128 m128_0, __m128 m128_1) {
|
||
m128_0[0] /= m128_1[0];
|
||
return m128_0;
|
||
}
|
||
|
||
#define _mm_rcp_ss(M128) __builtin_ia32_rcpss((__v4sf)(M128)) /*~1/x*/
|
||
#define _mm_sqrt_ss(M128) __builtin_ia32_sqrtss((__v4sf)(M128)) /*sqrt𝑥*/
|
||
#define _mm_rsqrt_ss(M128) __builtin_ia32_rsqrtss((__v4sf)(M128)) /*~1/sqrt𝑥*/
|
||
|
||
#define _mm_min_ss(M128_0, M128_1) \
|
||
__builtin_ia32_minss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_max_ss(M128_0, M128_1) \
|
||
__builtin_ia32_maxss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpeq_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpeqss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpneq_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpneqss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmplt_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpltss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpnlt_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpnltss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmple_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpless((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpnle_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpnless((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpgt_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpltss((__v4sf)(M128_1), (__v4sf)(M128_0))
|
||
#define _mm_cmpngt_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpnltss((__v4sf)(M128_1), (__v4sf)(M128_0))
|
||
#define _mm_cmpge_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpless((__v4sf)(M128_1), (__v4sf)(M128_0))
|
||
#define _mm_cmpnge_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpnless((__v4sf)(M128_1), (__v4sf)(M128_0))
|
||
#define _mm_cmpord_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpordss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
#define _mm_cmpunord_ss(M128_0, M128_1) \
|
||
__builtin_ia32_cmpunordss((__v4sf)(M128_0), (__v4sf)(M128_1))
|
||
|
||
/*───────────────────────────────────────────────────────────────────────────│─╗
|
||
│ cosmopolitan § it's a trap! » sse » memory ops ─╬─│┼
|
||
╚────────────────────────────────────────────────────────────────────────────│*/
|
||
|
||
#define _mm_set1_ps(M128_0) ((__m128)(__v4sf){M128_0, M128_0, M128_0, M128_0})
|
||
#define _mm_setzero_ps() ((__m128)(__v4sf){0})
|
||
#define _mm_cvtss_f32(M128_0) (((__v4sf)(M128_0))[0])
|
||
#define _mm_load_ps(FLOATPTR) (*(__m128 *)(FLOATPTR))
|
||
#define _mm_loadu_ps(FLOATPTR) (*(__m128_u *)(FLOATPTR))
|
||
#define _mm_set_ps(WHO, DESIGNED, THIS, SHEESH) \
|
||
((__m128)(__v4sf){SHEESH, THIS, DESIGNED, WHO})
|
||
#define _mm_set_ss(FLOAT) ((__m128)(__v4sf){FLOAT, 0, 0, 0})
|
||
#define _mm_load_ss(FLOATPTR) _mm_set_ss(*(FLOATPTR))
|
||
#define _mm_store_ss(FLOATPTR, M128_0) ((FLOATPTR)[0] = ((__v4sf)(M128_0))[0])
|
||
#define _mm_store_ps(FLOATPTR, M128_0) (*(__m128 *)(FLOATPTR) = (M128_0))
|
||
#define _mm_storeu_ps(FLOATPTR, M128_0) (*(__m128_u *)(FLOATPTR) = (M128_0))
|
||
#define _mm_shuffle_ps(M128_0, M128_1, MASK) \
|
||
((__m128)__builtin_ia32_shufps((__v4sf)(M128_0), (__v4sf)(M128_1), (MASK)))
|
||
|
||
#ifdef __llvm__
|
||
#define _mm_movehl_ps(M128_0, M128_1) \
|
||
((__m128)__builtin_shufflevector((__v4sf)(__m128)(M128_0), \
|
||
(__v4sf)(__m128)(M128_1), 6, 7, 2, 3))
|
||
/* instrinsics unstable & constantly breaking, consider ansi c or asm. */
|
||
/* each version of llvm has a different incompatible impl for this one */
|
||
#else
|
||
#define _mm_movehl_ps(M128_0, M128_1) \
|
||
((__m128)__builtin_ia32_movhlps((__v4sf)(__m128)(M128_0), \
|
||
(__v4sf)(__m128)(M128_1)))
|
||
#define _mm_storel_pi(M64PTR, M128_0) \
|
||
__builtin_ia32_storelps((__v2sf *)(M64PTR), (__v4sf)(M128_0))
|
||
#endif
|
||
|
||
/*───────────────────────────────────────────────────────────────────────────│─╗
|
||
│ cosmopolitan § it's a trap! » sse » cast ops ─╬─│┼
|
||
╚────────────────────────────────────────────────────────────────────────────│*/
|
||
|
||
#define _mm_cvtps_epi32(M128_0) \
|
||
((__m128i)__builtin_ia32_cvtps2dq((__v4sf)(M128_0)))
|
||
|
||
#ifdef __llvm__
|
||
#define _mm_cvtepi32_ps(M128I_0) \
|
||
((__m128) __builtin_convertvector((__v4si)(__m128i)(M128I_0), __v4sf))
|
||
#else
|
||
#define _mm_cvtepi32_ps(M128I_0) \
|
||
((__m128)__builtin_ia32_cvtdq2ps((__v4si)(M128I_0)))
|
||
#endif
|
||
|
||
/*───────────────────────────────────────────────────────────────────────────│─╗
|
||
│ cosmopolitan § it's a trap! » sse » misc ─╬─│┼
|
||
╚────────────────────────────────────────────────────────────────────────────│*/
|
||
|
||
#define _mm_getcsr() (__builtin_ia32_stmxcsr())
|
||
#define _mm_setcsr(U32CONF) (__builtin_ia32_ldmxcsr(U32CONF))
|
||
|
||
#define _MM_GET_ROUNDING_MODE() (_mm_getcsr() & _MM_ROUND_MASK)
|
||
#define _MM_SET_ROUNDING_MODE(MODE) \
|
||
(_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | (MODE)))
|
||
|
||
#define XMM_DESTROY(VAR) \
|
||
do { \
|
||
if (!IsTrustworthy()) { \
|
||
asm volatile("xorps\t%1,%0" : "=x"(VAR) : "0"(VAR)); \
|
||
} \
|
||
} while (0)
|
||
|
||
/*
|
||
** Ternary:
|
||
**
|
||
** Integer: _mm_or_si128(_mm_and_si128(a, cond), _mm_andnot_si128(cond, b))
|
||
** 32-bit float: _mm_or_ps(_mm_and_ps(a, cond), _mm_andnot_ps(cond, b))
|
||
** 64-bit float: _mm_or_pd(_mm_and_pd(a, cond), _mm_andnot_pd(cond, b))
|
||
** Integer (SSE4.1+): _mm_blendv_epi8(a, b, cond)
|
||
** 32-bit float (SSE4.1+): _mm_blendv_ps(a, b, cond)
|
||
** 64-bit float (SSE4.1+): _mm_blendv_pd(a, b, cond)
|
||
*/
|
||
|
||
#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
|
||
#endif /* COSMOPOLITAN_LIBC_BITS_XMMINTRIN_H_ */
|