//int64
vec3l& vec3l::Sub(int64 v) {
__m256i vxmm = _mm256_set_epi64x(0, v, v, v);
__m256i xmm = _mm256_set_epi64x(0, z, y, x);
xmm = _mm256_sub_epi64(xmm, vxmm);
x = M256L(xmm, 0);
y = M256L(xmm, 1);
z = M256L(xmm, 2);
return *this;
}
vec3l& vec3l::Sub(const vec3l& v) {
__m256i vxmm = _mm256_set_epi64x(0, v.z, v.y, v.x);
__m256i xmm = _mm256_set_epi64x(0, z, y, x);
xmm = _mm256_sub_epi64(xmm, vxmm);
x = M256L(xmm, 0);
y = M256L(xmm, 1);
z = M256L(xmm, 2);
return *this;
}
void calculate_fma_double (unsigned char * out, double X0, double Y0, double scale, unsigned YSTART, unsigned SX, unsigned SY)
{
__m256d dd = _mm256_set1_pd (scale);
__m256d XX0 = _mm256_set1_pd (X0);
for (unsigned j = YSTART; j < SY; j++) {
__m256d y0 = _mm256_set1_pd (j*scale + Y0);
for (unsigned i = 0; i < SX; i += 4) {
__m128i ind = _mm_setr_epi32 (i, i + 1, i + 2, i + 3);
__m256d x0 = _mm256_fmadd_pd (dd, _mm256_cvtepi32_pd (ind), XX0);
__m256d x = x0;
__m256d y = y0;
__m256i counts = _mm256_setzero_si256 ();
__m256i cmp_mask = _mm256_set1_epi32 (0xFFFFFFFFu);
for (unsigned n = 0; n < 255; n++) {
__m256d x2 = _mm256_mul_pd (x, x);
__m256d y2 = _mm256_mul_pd (y, y);
__m256d abs = _mm256_add_pd (x2, y2);
__m256i cmp = _mm256_castpd_si256 (_mm256_cmp_pd (abs, _mm256_set1_pd (4), 1));
cmp_mask = _mm256_and_si256 (cmp_mask, cmp);
if (_mm256_testz_si256 (cmp_mask, cmp_mask)) {
break;
}
counts = _mm256_sub_epi64 (counts, cmp_mask);
__m256d t = _mm256_add_pd (x, x);
y = _mm256_fmadd_pd (t, y, y0);
x = _mm256_add_pd (_mm256_sub_pd (x2, y2), x0);
}
__m256i result = _mm256_shuffle_epi8 (counts, _mm256_setr_epi8 (0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8, 0, 8));
*(uint32_t*) out = _mm_extract_epi16 (_mm256_extracti128_si256 (result, 0), 0) | (_mm_extract_epi16 (_mm256_extracti128_si256 (result, 1), 0) << 16);
out += 4;
}
}
}
__m256i test_mm256_sub_epi64(__m256i a, __m256i b) {
// CHECK: sub <4 x i64>
return _mm256_sub_epi64(a, b);
}
/*!
* \brief Subtract the two given values and return the result.
*/
ETL_STATIC_INLINE(avx_simd_long) sub(avx_simd_long lhs, avx_simd_long rhs) {
return _mm256_sub_epi64(lhs.value, rhs.value);
}
int32_t end_ref = 0;
int64_t score = NEG_INF;
__m256i vNegInf = _mm256_set1_epi64x(NEG_INF);
__m256i vNegInf0 = _mm256_srli_si256_rpl(vNegInf, 8); /* shift in a 0 */
__m256i vOpen = _mm256_set1_epi64x(open);
__m256i vGap = _mm256_set1_epi64x(gap);
__m256i vOne = _mm256_set1_epi64x(1);
__m256i vN = _mm256_set1_epi64x(N);
__m256i vNegOne = _mm256_set1_epi64x(-1);
__m256i vI = _mm256_set_epi64x(0,1,2,3);
__m256i vJreset = _mm256_set_epi64x(0,-1,-2,-3);
__m256i vMaxScore = vNegInf;
__m256i vEndI = vNegInf;
__m256i vEndJ = vNegInf;
__m256i vILimit = _mm256_set1_epi64x(s1Len);
__m256i vILimit1 = _mm256_sub_epi64(vILimit, vOne);
__m256i vJLimit = _mm256_set1_epi64x(s2Len);
__m256i vJLimit1 = _mm256_sub_epi64(vJLimit, vOne);
/* convert _s1 from char to int in range 0-23 */
for (i=0; i<s1Len; ++i) {
s1[i] = matrix->mapper[(unsigned char)_s1[i]];
}
/* pad back of s1 with dummy values */
for (i=s1Len; i<s1Len_PAD; ++i) {
s1[i] = 0; /* point to first matrix row because we don't care */
}
/* convert _s2 from char to int in range 0-23 */
for (j=0; j<s2Len; ++j) {
