#include #include typedef float a2sf[2] __attribute__ ((aligned (16))); typedef float a4sf[4] __attribute__ ((aligned (16))); typedef double a2df[2] __attribute__ ((aligned (16))); typedef int a4si[4] __attribute__ ((aligned (16))); typedef int a8si[4] __attribute__ ((aligned (16))); typedef __m128 v4sf; typedef __m128d v2df; typedef __m128i v4si; typedef __m128i v8si; typedef __m128i v8hi; /* int */ #define SIMD_LOAD_V4SI(vec,arr) vec=_mm_loadu_si128((__m128i*)arr) #define SIMD_LOADA_V4SI(vec,arr) vec=_mm_load_si128((__m128i*)arr) #define SIMD_LOAD_BROADCAST_V4SI(vec,val) vec=_mm_set1_si128(val) #define SIMD_MULD(vec1,vec2,vec3) vec1=_mm_mul_epi32(vec2,vec3) #define SIMD_ADDD(vec1,vec2,vec3) vec1=_mm_add_epi32(vec2,vec3) #define SIMD_SUBD(vec1, vec2, vec3) vec1 = _mm_sub_epi32(vec2, vec3) #define SIMD_STORE_V4SI(vec,arr) _mm_storeu_si128((__m128i*)arr,vec) #define SIMD_STOREA_V4SI(vec,arr) _mm_store_si128((__m128i*)arr,vec) /* float */ #define SIMD_LOAD_V4SF(vec,arr) vec=_mm_loadu_ps(arr) #define SIMD_LOADA_V4SF(vec,arr) vec=_mm_load_ps(arr) #define SIMD_LOAD_BROADCAST_V4SF(vec,val) vec=_mm_set1_ps(val) #define SIMD_MULPS(vec1,vec2,vec3) vec1=_mm_mul_ps(vec2,vec3) #define SIMD_DIVPS(vec1,vec2,vec3) vec1=_mm_div_ps(vec2,vec3) #define SIMD_ADDPS(vec1,vec2,vec3) vec1=_mm_add_ps(vec2,vec3) #define SIMD_SUBPS(vec1, vec2, vec3) vec1 = _mm_sub_ps(vec2, vec3) #define SIMD_MULADDPS(vec1, vec2, vec3, vec4) \ do { \ __m128 __pips_tmp;\ SIMD_MULPS(__pips_tmp, vec3, vec4);\ SIMD_ADDPS(vec1, __pips_tmp, vec2); \ } while(0) #define SIMD_SHUFFLE_V4SF(dist,src,i0,i1,i2,i3) dist=_mm_shuffle_ps(src,src,_MM_SHUFFLE(i3,i2,i1,i0) /* umin as in unary minus */ #define SIMD_UMINPS(vec1, vec2) \ do { \ __m128 __pips_tmp; \ __pips_tmp = _mm_setzero_ps(); \ vec1 = _mm_sub_ps(__pips_tmp, vec2); \ } while(0) #define SIMD_STORE_V4SF(vec,arr) _mm_storeu_ps(arr,vec) #define SIMD_STOREA_V4SF(vec,arr) _mm_store_ps(arr,vec) #define SIMD_STORE_GENERIC_V4SF(vec,v0,v1,v2,v3) \ do { \ float __pips_tmp[4] __attribute__ ((aligned (16))); \ SIMD_STOREA_V4SF(vec,&__pips_tmp[0]); \ *(v0)=__pips_tmp[0]; \ *(v1)=__pips_tmp[1]; \ *(v2)=__pips_tmp[2]; \ *(v3)=__pips_tmp[3]; \ } while (0) #define SIMD_ZERO_V4SF(vec) vec = _mm_setzero_ps() #define SIMD_INVERT_V4SF(vec) vec = _mm_shuffle_ps(vec,vec,_MM_SHUFFLE(4,3,2,1)) #define SIMD_LOAD_GENERIC_V4SF(vec,v0,v1,v2,v3) \ do { \ float __pips_v[4] __attribute ((aligned (16)));\ __pips_v[0]=v0;\ __pips_v[1]=v1;\ __pips_v[2]=v2;\ __pips_v[3]=v3;\ SIMD_LOADA_V4SF(vec,&__pips_v[0]); \ } while(0) /* handle padded value, this is a very bad implementation ... */ #define SIMD_STORE_MASKED_V4SF(vec,arr) \ do { \ float __pips_tmp[4] __attribute__ ((aligned (16))); \ SIMD_STOREA_V4SF(vec,&__pips_tmp[0]); \ (arr)[0] = __pips_tmp[0]; \ (arr)[1] = __pips_tmp[1]; \ (arr)[2] = __pips_tmp[2]; \ } while(0) #define SIMD_LOAD_V4SI_TO_V4SF(v, f) \ do { \ float __pips_tmp[4]; \ __pips_tmp[0] = (f)[0]; \ __pips_tmp[1] = (f)[1]; \ __pips_tmp[2] = (f)[2]; \ __pips_tmp[3] = (f)[3]; \ SIMD_LOAD_V4SF(v, __pips_tmp); \ } while(0) /* double */ #define SIMD_LOAD_V2DF(vec,arr) vec=_mm_loadu_pd(arr) #define SIMD_MULPD(vec1,vec2,vec3) vec1=_mm_mul_pd(vec2,vec3) #define SIMD_ADDPD(vec1,vec2,vec3) vec1=_mm_add_pd(vec2,vec3) #define SIMD_MULADDPD(vec1, vec2, vec3, vec4) \ do { \ __m128 __pips_tmp;\ SIMD_MULPD(__pips_tmp, vec3, vec4);\ SIMD_ADDPD(vec1, __pips_tmp, vec2); \ } while(0) #define SIMD_UMINPD(vec1, vec2) \ do { \ __m128d __pips_tmp; \ __pips_tmp = _mm_setzero_pd(); \ vec1 = _mm_sub_pd(__pips_tmp, vec2); \ } while(0) #define SIMD_COSPD(vec1, vec2) \ do { \ double __pips_tmp[2] __attribute__ ((aligned (16))); \ SIMD_STORE_V2DF(vec2, __pips_tmp); \ __pips_tmp[0] = cos(__pips_tmp[0]); \ __pips_tmp[1] = cos(__pips_tmp[1]); \ SIMD_LOAD_V2DF(vec1, __pips_tmp); \ } while(0) #define SIMD_SINPD(vec1, vec2) \ do { \ double __pips_tmp[2] __attribute__ ((aligned (16))); \ SIMD_STORE_V2DF(vec2, __pips_tmp); \ __pips_tmp[0] = sin(__pips_tmp[0]); \ __pips_tmp[1] = sin(__pips_tmp[1]); \ SIMD_LOAD_V2DF(vec1, __pips_tmp); \ } while(0) #define SIMD_STORE_V2DF(vec,arr) _mm_storeu_pd(arr,vec) #define SIMD_STORE_GENERIC_V2DF(vec, v0, v1) \ do { \ double __pips_tmp[2]; \ SIMD_STORE_V2DF(vec,&__pips_tmp[0]); \ *(v0)=__pips_tmp[0]; \ *(v1)=__pips_tmp[1]; \ } while (0) #define SIMD_LOAD_GENERIC_V2DF(vec,v0,v1) \ do { \ double v[2] = { v0,v1}; \ SIMD_LOAD_V2DF(vec,&v[0]); \ } while(0) /* conversions */ #define SIMD_STORE_V2DF_TO_V2SF(vec,f) \ do { \ double __pips_tmp[2]; \ SIMD_STORE_V2DF(vec, __pips_tmp); \ (f)[0] = __pips_tmp[0]; \ (f)[1] = __pips_tmp[1]; \ } while(0) #define SIMD_LOAD_V2SF_TO_V2DF(vec,f) \ SIMD_LOAD_GENERIC_V2DF(vec,(f)[0],(f)[1]) /* char */ #define SIMD_LOAD_V8HI(vec,arr) \ vec = (__m128i*)(arr) #define SIMD_STORE_V8HI(vec,arr)\ *(__m128i *)(&(arr)[0]) = vec #define SIMD_STORE_V8HI_TO_V8SI(vec,arr)\ SIMD_STORE_V8HI(vec,arr) #define SIMD_LOAD_V8SI_TO_V8HI(vec,arr)\ SIMD_LOAD_V8HI(vec,arr)