1193326Sed/*===---- pmmintrin.h - SSE3 intrinsics ------------------------------------=== 2193326Sed * 3193326Sed * Permission is hereby granted, free of charge, to any person obtaining a copy 4193326Sed * of this software and associated documentation files (the "Software"), to deal 5193326Sed * in the Software without restriction, including without limitation the rights 6193326Sed * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 7193326Sed * copies of the Software, and to permit persons to whom the Software is 8193326Sed * furnished to do so, subject to the following conditions: 9193326Sed * 10193326Sed * The above copyright notice and this permission notice shall be included in 11193326Sed * all copies or substantial portions of the Software. 12193326Sed * 13193326Sed * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 14193326Sed * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 15193326Sed * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 16193326Sed * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 17193326Sed * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 18193326Sed * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 19193326Sed * THE SOFTWARE. 20193326Sed * 21193326Sed *===-----------------------------------------------------------------------=== 22193326Sed */ 23193326Sed 24193326Sed#ifndef __PMMINTRIN_H 25193326Sed#define __PMMINTRIN_H 26193326Sed 27193326Sed#ifndef __SSE3__ 28193326Sed#error "SSE3 instruction set not enabled" 29193326Sed#else 30193326Sed 31193326Sed#include <emmintrin.h> 32193326Sed 33206084Srdivackystatic __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) 34249423Sdim_mm_lddqu_si128(__m128i const *__p) 35193326Sed{ 36249423Sdim return (__m128i)__builtin_ia32_lddqu((char const *)__p); 37193326Sed} 38193326Sed 39206084Srdivackystatic __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 40249423Sdim_mm_addsub_ps(__m128 __a, __m128 __b) 41193326Sed{ 42249423Sdim return __builtin_ia32_addsubps(__a, __b); 43193326Sed} 44193326Sed 45206084Srdivackystatic __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 46249423Sdim_mm_hadd_ps(__m128 __a, __m128 __b) 47193326Sed{ 48249423Sdim return __builtin_ia32_haddps(__a, __b); 49193326Sed} 50193326Sed 51206084Srdivackystatic __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 52249423Sdim_mm_hsub_ps(__m128 __a, __m128 __b) 53193326Sed{ 54249423Sdim return __builtin_ia32_hsubps(__a, __b); 55193326Sed} 56193326Sed 57206084Srdivackystatic __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 58249423Sdim_mm_movehdup_ps(__m128 __a) 59193326Sed{ 60249423Sdim return __builtin_shufflevector(__a, __a, 1, 1, 3, 3); 61193326Sed} 62193326Sed 63206084Srdivackystatic __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) 64249423Sdim_mm_moveldup_ps(__m128 __a) 65193326Sed{ 66249423Sdim return __builtin_shufflevector(__a, __a, 0, 0, 2, 2); 67193326Sed} 68193326Sed 69206084Srdivackystatic __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) 70249423Sdim_mm_addsub_pd(__m128d __a, __m128d __b) 71193326Sed{ 72249423Sdim return __builtin_ia32_addsubpd(__a, __b); 73193326Sed} 74193326Sed 75206084Srdivackystatic __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) 76249423Sdim_mm_hadd_pd(__m128d __a, __m128d __b) 77193326Sed{ 78249423Sdim return __builtin_ia32_haddpd(__a, __b); 79193326Sed} 80193326Sed 81206084Srdivackystatic __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) 82249423Sdim_mm_hsub_pd(__m128d __a, __m128d __b) 83193326Sed{ 84249423Sdim return __builtin_ia32_hsubpd(__a, __b); 85193326Sed} 86193326Sed 87226633Sdim#define _mm_loaddup_pd(dp) _mm_load1_pd(dp) 88193326Sed 89206084Srdivackystatic __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) 90249423Sdim_mm_movedup_pd(__m128d __a) 91193326Sed{ 92249423Sdim return __builtin_shufflevector(__a, __a, 0, 0); 93193326Sed} 94193326Sed 95193326Sed#define _MM_DENORMALS_ZERO_ON (0x0040) 96193326Sed#define _MM_DENORMALS_ZERO_OFF (0x0000) 97193326Sed 98193326Sed#define _MM_DENORMALS_ZERO_MASK (0x0040) 99193326Sed 100193326Sed#define _MM_GET_DENORMALS_ZERO_MODE() (_mm_getcsr() & _MM_DENORMALS_ZERO_MASK) 101193326Sed#define _MM_SET_DENORMALS_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_DENORMALS_ZERO_MASK) | (x))) 102193326Sed 103206084Srdivackystatic __inline__ void __attribute__((__always_inline__, __nodebug__)) 104249423Sdim_mm_monitor(void const *__p, unsigned __extensions, unsigned __hints) 105193326Sed{ 106249423Sdim __builtin_ia32_monitor((void *)__p, __extensions, __hints); 107193326Sed} 108193326Sed 109206084Srdivackystatic __inline__ void __attribute__((__always_inline__, __nodebug__)) 110249423Sdim_mm_mwait(unsigned __extensions, unsigned __hints) 111193326Sed{ 112249423Sdim __builtin_ia32_mwait(__extensions, __hints); 113193326Sed} 114193326Sed 115193326Sed#endif /* __SSE3__ */ 116193326Sed 117193326Sed#endif /* __PMMINTRIN_H */ 118