SHOW:
|
|
- or go back to the newest paste.
| 1 | #pragma GCC target("mmx,sse,sse2,ssse3,sse4.1,sse4.2,avx,avx2,fma,avx512f,avx512bw,avx512cd,avx512dq,avx512vl")
| |
| 2 | #include <immintrin.h> | |
| 3 | ||
| 4 | ||
| 5 | double xs[128] __attribute__((aligned (64))), ys[128] __attribute__((aligned (64))); | |
| 6 | ||
| 7 | double test_one(int i) {
| |
| 8 | double s = 1e9; | |
| 9 | static point vec[N]; | |
| 10 | int vsize = 0; | |
| 11 | for (int j = 0; j < N; ++j) {
| |
| 12 | if (i == j) {
| |
| 13 | continue; | |
| 14 | } | |
| 15 | vec[vsize] = p[i].to(p[j]); | |
| 16 | xs[vsize] = vec[vsize].x, ys[vsize] = vec[vsize].y; | |
| 17 | ++vsize; | |
| 18 | } | |
| 19 | for (int h = vsize; h < 128; ++h) | |
| 20 | xs[h] = 1e18, ys[h] = 1e17 + 50; | |
| 21 | for (int j = 0; j < vsize; ++j) {
| |
| 22 | __m512d xj = _mm512_set1_pd(xs[j]), yj = _mm512_set1_pd(ys[j]); | |
| 23 | __m512d mins = _mm512_set1_pd(1e50); | |
| 24 | for (int h = j + 1; h < vsize; h += 8) {
| |
| 25 | __m512d xc = _mm512_load_pd(xs + h), yc = _mm512_load_pd(ys + h); | |
| 26 | __m512d ans = _mm512_abs_pd(_mm512_fmsub_pd(xc, yj, _mm512_mul_pd(yc, xj))); | |
| 27 | mins = _mm512_min_pd(mins, ans); | |
| 28 | } | |
| 29 | s = min(s, _mm512_reduce_min_pd(mins)); | |
| 30 | } | |
| 31 | return s / 2; | |
| 32 | } |
