| 1 | // RUN: %libomp-compile -O3 -ffast-math |
| 2 | // RUN: %libomp-run |
| 3 | #include <float.h> |
| 4 | #include <math.h> |
| 5 | #include <stdio.h> |
| 6 | #include <stdlib.h> |
| 7 | #include <time.h> |
| 8 | |
| 9 | int compare_float(float x1, float x2, float scalar) { |
| 10 | const float diff = fabsf(x: x1 - x2); |
| 11 | x1 = fabsf(x: x1); |
| 12 | x2 = fabsf(x: x2); |
| 13 | const float l = (x2 > x1) ? x2 : x1; |
| 14 | if (diff <= l * scalar * FLT_EPSILON) |
| 15 | return 1; |
| 16 | else |
| 17 | return 0; |
| 18 | } |
| 19 | |
| 20 | #define ARRAY_SIZE 256 |
| 21 | |
| 22 | __attribute__((noinline)) void |
| 23 | initialization_loop(float X[ARRAY_SIZE][ARRAY_SIZE], |
| 24 | float Y[ARRAY_SIZE][ARRAY_SIZE]) { |
| 25 | const float max = 1000.0; |
| 26 | srand(seed: time(NULL)); |
| 27 | for (int r = 0; r < ARRAY_SIZE; r++) { |
| 28 | for (int c = 0; c < ARRAY_SIZE; c++) { |
| 29 | X[r][c] = ((float)rand() / (float)(RAND_MAX)) * max; |
| 30 | Y[r][c] = X[r][c]; |
| 31 | } |
| 32 | } |
| 33 | } |
| 34 | |
| 35 | __attribute__((noinline)) void omp_simd_loop(float X[ARRAY_SIZE][ARRAY_SIZE]) { |
| 36 | for (int r = 1; r < ARRAY_SIZE; ++r) { |
| 37 | for (int c = 1; c < ARRAY_SIZE; ++c) { |
| 38 | #pragma omp simd |
| 39 | for (int k = 2; k < ARRAY_SIZE; ++k) { |
| 40 | #pragma omp ordered simd |
| 41 | X[r][k] = X[r][k - 2] + sinf(x: (float)(r / c)); |
| 42 | } |
| 43 | } |
| 44 | } |
| 45 | } |
| 46 | |
| 47 | __attribute__((noinline)) int comparison_loop(float X[ARRAY_SIZE][ARRAY_SIZE], |
| 48 | float Y[ARRAY_SIZE][ARRAY_SIZE]) { |
| 49 | int totalErrors_simd = 0; |
| 50 | const float scalar = 1.0; |
| 51 | for (int r = 1; r < ARRAY_SIZE; ++r) { |
| 52 | for (int c = 1; c < ARRAY_SIZE; ++c) { |
| 53 | for (int k = 2; k < ARRAY_SIZE; ++k) { |
| 54 | Y[r][k] = Y[r][k - 2] + sinf(x: (float)(r / c)); |
| 55 | } |
| 56 | } |
| 57 | // check row for simd update |
| 58 | for (int k = 0; k < ARRAY_SIZE; ++k) { |
| 59 | if (!compare_float(x1: X[r][k], x2: Y[r][k], scalar)) { |
| 60 | ++totalErrors_simd; |
| 61 | } |
| 62 | } |
| 63 | } |
| 64 | return totalErrors_simd; |
| 65 | } |
| 66 | |
| 67 | int main(void) { |
| 68 | float X[ARRAY_SIZE][ARRAY_SIZE]; |
| 69 | float Y[ARRAY_SIZE][ARRAY_SIZE]; |
| 70 | |
| 71 | initialization_loop(X, Y); |
| 72 | omp_simd_loop(X); |
| 73 | const int totalErrors_simd = comparison_loop(X, Y); |
| 74 | |
| 75 | if (totalErrors_simd) { |
| 76 | fprintf(stdout, format: "totalErrors_simd: %d \n" , totalErrors_simd); |
| 77 | fprintf(stdout, format: "%s : %d - FAIL: error in ordered simd computation.\n" , |
| 78 | __FILE__, __LINE__); |
| 79 | } else { |
| 80 | fprintf(stdout, format: "Success!\n" ); |
| 81 | } |
| 82 | |
| 83 | return totalErrors_simd; |
| 84 | } |
| 85 | |