SHOW:
|
|
- or go back to the newest paste.
| 1 | #include <emmintrin.h> | |
| 2 | #include <stdlib.h> | |
| 3 | #include <stdio.h> | |
| 4 | #include <time.h> | |
| 5 | ||
| 6 | #define NEAR 10 | |
| 7 | #define T1 100 | |
| 8 | #define T2 1000 | |
| 9 | #define T3 10000 | |
| 10 | ||
| 11 | int func(int Di) {
| |
| 12 | if (Di <= -T3) return -4; | |
| 13 | if (Di <= -T2) return -3; | |
| 14 | if (Di <= -T1) return -2; | |
| 15 | if (Di < -NEAR) return -1; | |
| 16 | if (Di <= NEAR) return 0; | |
| 17 | if (Di < T1) return 1; | |
| 18 | if (Di < T2) return 2; | |
| 19 | if (Di < T3) return 3; | |
| 20 | return 4; | |
| 21 | } | |
| 22 | ||
| 23 | int funcX(int Di) {
| |
| 24 | int cmp_m3 = -(Di > -T3); | |
| 25 | int cmp_m2 = -(Di > -T2); | |
| 26 | int cmp_m1 = -(Di > -T1); | |
| 27 | int cmp_p0 = -(Di > NEAR); | |
| 28 | int reduce_true = cmp_m3 + cmp_m2 + cmp_m1 + cmp_p0; | |
| 29 | int cmp_m0 = -(Di < -NEAR); | |
| 30 | int cmp_p1 = -(Di < T1); | |
| 31 | int cmp_p2 = -(Di < T2); | |
| 32 | int cmp_p3 = -(Di < T3); | |
| 33 | int reduce_false = cmp_p3 + cmp_p2 + cmp_p1 + cmp_m0; | |
| 34 | return reduce_false - reduce_true; | |
| 35 | } | |
| 36 | ||
| 37 | __m128i func4(__m128i D) {
| |
| 38 | __m128i cmp_m3 = _mm_cmpgt_epi32(D, _mm_set1_epi32(-T3)); | |
| 39 | __m128i cmp_m2 = _mm_cmpgt_epi32(D, _mm_set1_epi32(-T2)); | |
| 40 | __m128i cmp_m1 = _mm_cmpgt_epi32(D, _mm_set1_epi32(-T1)); | |
| 41 | __m128i cmp_p0 = _mm_cmpgt_epi32(D, _mm_set1_epi32(NEAR)); | |
| 42 | __m128i reduce_true = _mm_add_epi32(_mm_add_epi32(cmp_m3, cmp_m2), _mm_add_epi32(cmp_m1, cmp_p0)); | |
| 43 | __m128i cmp_m0 = _mm_cmplt_epi32(D, _mm_set1_epi32(-NEAR)); | |
| 44 | __m128i cmp_p1 = _mm_cmplt_epi32(D, _mm_set1_epi32(T1)); | |
| 45 | __m128i cmp_p2 = _mm_cmplt_epi32(D, _mm_set1_epi32(T2)); | |
| 46 | __m128i cmp_p3 = _mm_cmplt_epi32(D, _mm_set1_epi32(T3)); | |
| 47 | __m128i reduce_false = _mm_add_epi32(_mm_add_epi32(cmp_p3, cmp_p2), _mm_add_epi32(cmp_p1, cmp_m0)); | |
| 48 | return _mm_sub_epi32(reduce_false, reduce_true); | |
| 49 | } | |
| 50 | ||
| 51 | - | data_i[i] = rand() - 16768; |
| 51 | + | |
| 52 | ||
| 53 | static union {
| |
| 54 | int data_i[4 * COUNT]; | |
| 55 | __m128i data_r[COUNT]; | |
| 56 | }; | |
| 57 | ||
| 58 | static union {
| |
| 59 | int res_i[4]; | |
| 60 | __m128i res_r; | |
| 61 | }; | |
| 62 | ||
| 63 | int main() {
| |
| 64 | for (int i = 0; i < 4*COUNT; i++) | |
| 65 | data_i[i] = rand() % 32768 - 16768; | |
| 66 | ||
| 67 | for (int i = 0; i < COUNT; i++) {
| |
| 68 | int corr[4]; | |
| 69 | for (int j = 0; j < 4; j++) | |
| 70 | corr[j] = func(data_i[4*i+j]); | |
| 71 | ||
| 72 | res_r = func4(data_r[i]); | |
| 73 | for (int j = 0; j < 4; j++) | |
| 74 | if (corr[j] != res_i[j]) | |
| 75 | printf("Wrong!");
| |
| 76 | ||
| 77 | for (int j = 0; j < 4; j++) | |
| 78 | res_i[j] = funcX(data_i[4*i+j]); | |
| 79 | for (int j = 0; j < 4; j++) | |
| 80 | if (corr[j] != res_i[j]) | |
| 81 | printf("Wrong!");
| |
| 82 | } | |
| 83 | ||
| 84 | {
| |
| 85 | int start = clock(); | |
| 86 | int sum = 0; | |
| 87 | for (int q = 0; q < 10000; q++) {
| |
| 88 | for (int i = 0; i < 4*COUNT; i++) | |
| 89 | sum += func(data_i[i]); | |
| 90 | } | |
| 91 | printf("Time = %0.3lf (%d)\n", double(clock() - start) / CLOCKS_PER_SEC, sum);
| |
| 92 | } | |
| 93 | ||
| 94 | {
| |
| 95 | int start = clock(); | |
| 96 | __m128i sum = _mm_setzero_si128(); | |
| 97 | for (int q = 0; q < 10000; q++) {
| |
| 98 | for (int i = 0; i < COUNT; i++) | |
| 99 | sum = _mm_add_epi32(sum, func4(data_r[i])); | |
| 100 | } | |
| 101 | res_r = sum; | |
| 102 | int rsum = res_i[0] + res_i[1] + res_i[2] + res_i[3]; | |
| 103 | printf("Time = %0.3lf (%d)\n", double(clock() - start) / CLOCKS_PER_SEC, rsum);
| |
| 104 | } | |
| 105 | ||
| 106 | return 0; | |
| 107 | } |
