double vNormalIntegral(double b)
{
__declspec(align(64)) __m512d vec_cf0, vec_cf1, vec_cf2, vec_s, vec_stp, vec_exp;
//NN/2-1 has to be the multiple of 8
//NN = (8*LV+1)*2, LV = 20 -> NN = 322
//const int NN = 322;
const int vecsize = 8;
const int nCal = (NN/2-1)/vecsize;
//const int left = NN%vecsize;
double a = 0.0f;
double s, h, sum = 0.0f;
h = (b-a)/NN;
// add in the first few terms
sum += exp(-a*a/2.0) + 4.0*exp(-(a+h)*(a+h)/2.0);
// and the last one
sum += exp(-b*b/2.0);
vec_cf0 = _mm512_set1_pd(a);
vec_cf1 = _mm512_set1_pd(2*h);
vec_cf2 = _mm512_set1_pd(-0.5);
vec_s = _mm512_set_pd(8,7,6,5,4,3,2,1);//vectorize
vec_s = _mm512_mul_pd(vec_s, vec_cf1);//(16h,14h,..,2h)
vec_s = _mm512_add_pd(vec_cf0, vec_s);//(a+16h,..,a+2h)
vec_stp = _mm512_set1_pd(2*h*vecsize-h);
vec_cf0 = _mm512_set1_pd(h);
for (int i = 0; i < nCal; ++i){
vec_exp = _mm512_mul_pd(vec_s, vec_s);
vec_exp = _mm512_mul_pd(vec_exp, vec_cf2);
vec_cf1 = _mm512_exp_pd(vec_exp);//vec_cf1->sum
sum += 2.0*_mm512_reduce_add_pd(vec_cf1);
vec_s = _mm512_add_pd(vec_s, vec_cf0);//s+=h
vec_exp = _mm512_mul_pd(vec_s, vec_s);
vec_exp = _mm512_mul_pd(vec_exp, vec_cf2);
vec_cf1 = _mm512_exp_pd(vec_exp);
sum += 4.0*_mm512_reduce_add_pd(vec_cf1);
vec_s = _mm512_add_pd(vec_s, vec_stp);
}
sum = 0.5*sqrt(2*PI) + h*sum/3.0;
return sum;
}
double const* b;
double complex const* e;
double complex * f;
int ix, iy, iz, n;
#ifdef ARTED_STENCIL_LOOP_BLOCKING
int bx, by;
#endif
__m512d at = _mm512_set1_pd(A);
__m512d HALF = _mm512_set1_pd(-0.5);
#ifdef TUNING_COMPLEX_MUL
__m512i INV = _mm512_set4_epi64(1LL << 63, 0, 1LL << 63, 0);
#else
__m512d ZI = _mm512_set_pd(-1, 0, -1, 0, -1, 0, -1, 0);
#endif
__declspec(align(64)) double G[12];
for(n = 0 ; n < 12 ; ++n)
G[n] = C[n] * -0.5;
__m512i nly = _mm512_set1_epi32(PNLy);
__m512i nlz = _mm512_set1_epi32(PNLz);
#ifdef ARTED_DOMAIN_POWER_OF_TWO
__m512i myx = _mm512_mask_blend_epi32(0xFF00, _mm512_set1_epi32(NLy - 1), _mm512_set1_epi32(NLx - 1));
__m512i nyx = _mm512_mask_blend_epi32(0xFF00, _mm512_set1_epi32(NLy ), _mm512_set1_epi32(NLx ));
#else
__m512i nyx = _mm512_mask_blend_epi32(0xFF00, _mm512_set1_epi32(PNLy ), _mm512_set1_epi32(PNLx ));
#endif
foo (double x)
{
return _mm512_set_pd (x, x, x, x, x, x, x, x);
}
foo (double *v)
{
return _mm512_set_pd (v[7], v[6], v[5], v[4], v[3], v[2], v[1], v[0]);
}
, double complex const* restrict E
, double * restrict F
, double * restrict G
, double * restrict H
) {
#if defined(__KNC__) || defined(__AVX512F__)
/* Hand-Code Vector processing for Xeon Phi */
double complex const* e;
int ix, iy, iz, n;
#ifdef ARTED_STENCIL_LOOP_BLOCKING
int bx, by;
#endif
__m512d CONJ = _mm512_set_pd(-1, 1, -1, 1, -1, 1, -1, 1);
__m512d tt[3];
for(n = 0 ; n < 3 ; ++n)
tt[n] = _mm512_setzero_pd();
__m512d wm[4];
__m512d wp[4];
__m512d v1, v2, v3;
__m512i nly = _mm512_set1_epi32(NLy);
__m512i nlz = _mm512_set1_epi32(NLz);
__m512i nyx = _mm512_mask_blend_epi32(0xFF00, _mm512_set1_epi32(NLy ), _mm512_set1_epi32(NLx ));
#ifdef ARTED_DOMAIN_POWER_OF_TWO
__m512i myx = _mm512_mask_blend_epi32(0xFF00, _mm512_set1_epi32(NLy - 1), _mm512_set1_epi32(NLx - 1));
#endif
