From b1737698db5773ffde6a3a6c8586da4bfb991099 Mon Sep 17 00:00:00 2001 From: Chip Kerchner Date: Tue, 13 Aug 2024 07:01:21 -0500 Subject: [PATCH] Fix DEFAULTS in SBGEMM for POWER10. Also comparisons for SBGEMM unit test can be exactly due to epilison differences. --- param.h | 4 +-- test/compare_sgemm_sbgemm.c | 68 +++++++++++++++++++++++-------------- 2 files changed, 45 insertions(+), 27 deletions(-) diff --git a/param.h b/param.h index 2618e1f60..0e4d8965d 100644 --- a/param.h +++ b/param.h @@ -2637,8 +2637,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #undef SBGEMM_DEFAULT_Q #define SBGEMM_DEFAULT_UNROLL_M 16 #define SBGEMM_DEFAULT_UNROLL_N 8 -#define SBGEMM_DEFAULT_P 832 -#define SBGEMM_DEFAULT_Q 1026 +#define SBGEMM_DEFAULT_P 512 +#define SBGEMM_DEFAULT_Q 1024 #define SBGEMM_DEFAULT_R 4096 #endif diff --git a/test/compare_sgemm_sbgemm.c b/test/compare_sgemm_sbgemm.c index de589458b..4b546fb1f 100644 --- a/test/compare_sgemm_sbgemm.c +++ b/test/compare_sgemm_sbgemm.c @@ -81,6 +81,8 @@ float16to32 (bfloat16_bits f16) return f32.v; } +#define SBGEMM_LARGEST 256 + int main (int argc, char *argv[]) { @@ -88,12 +90,39 @@ main (int argc, char *argv[]) int i, j, l; blasint x, y; int ret = 0; - int loop = 100; + int loop = SBGEMM_LARGEST; char transA = 'N', transB = 'N'; float alpha = 1.0, beta = 0.0; for (x = 0; x <= loop; x++) { + if ((x > 100) && (x != SBGEMM_LARGEST)) continue; + m = k = n = x; + float *A = (float *)malloc(m * k * sizeof(FLOAT)); + float *B = (float *)malloc(k * n * sizeof(FLOAT)); + float *C = (float *)malloc(m * n * sizeof(FLOAT)); + bfloat16_bits *AA = (bfloat16_bits *)malloc(m * k * sizeof(bfloat16_bits)); + bfloat16_bits *BB = (bfloat16_bits *)malloc(k * n * sizeof(bfloat16_bits)); + float *DD = (float *)malloc(m * n * sizeof(FLOAT)); + float *CC = (float *)malloc(m * n * sizeof(FLOAT)); + if ((A == NULL) || (B == NULL) || (C == NULL) || (AA == NULL) || (BB == NULL) || + (DD == NULL) || (CC == NULL)) + return 1; + bfloat16 atmp,btmp; + blasint one=1; + + for (j = 0; j < m; j++) + { + for (i = 0; i < n; i++) + { + A[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5; + B[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5; + sbstobf16_(&one, &A[j*k+i], &one, &atmp, &one); + sbstobf16_(&one, &B[j*k+i], &one, &btmp, &one); + AA[j * k + i].v = atmp; + BB[j * k + i].v = btmp; + } + } for (y = 0; y < 4; y++) { if ((y == 0) || (y == 2)) { @@ -106,34 +135,16 @@ main (int argc, char *argv[]) } else { transB = 'T'; } - m = k = n = x; - float A[m * k]; - float B[k * n]; - float C[m * n]; - bfloat16_bits AA[m * k], BB[k * n]; - float DD[m * n], CC[m * n]; - bfloat16 atmp,btmp; - blasint one=1; - for (j = 0; j < m; j++) - { - for (i = 0; i < m; i++) - { - A[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5; - B[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5; - C[j * k + i] = 0; - sbstobf16_(&one, &A[j*k+i], &one, &atmp, &one); - sbstobf16_(&one, &B[j*k+i], &one, &btmp, &one); - AA[j * k + i].v = atmp; - BB[j * k + i].v = btmp; - CC[j * k + i] = 0; - DD[j * k + i] = 0; - } - } + memset(CC, 0, m * n * sizeof(FLOAT)); + memset(DD, 0, m * n * sizeof(FLOAT)); + memset(C, 0, m * n * sizeof(FLOAT)); + SGEMM (&transA, &transB, &m, &n, &k, &alpha, A, &m, B, &k, &beta, C, &m); SBGEMM (&transA, &transB, &m, &n, &k, &alpha, (bfloat16*) AA, &m, (bfloat16*)BB, &k, &beta, CC, &m); + for (i = 0; i < n; i++) for (j = 0; j < m; j++) if (fabs (CC[i * m + j] - C[i * m + j]) > 1.0) @@ -160,9 +171,16 @@ main (int argc, char *argv[]) } for (i = 0; i < n; i++) for (j = 0; j < m; j++) - if (CC[i * m + j] != DD[i * m + j]) + if (fabs (CC[i * m + j] - DD[i * m + j]) > 1.0) ret++; } + free(A); + free(B); + free(C); + free(AA); + free(BB); + free(DD); + free(CC); } if (ret != 0)