/***************************************************************************** Copyright (c) 2020, The OpenBLAS Project All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. 3. Neither the name of the OpenBLAS project nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. **********************************************************************************/ #include #include #include "common.h" void openblas_warning(int verbose, const char * msg); #ifndef COMPLEX #ifdef XDOUBLE #define ERROR_NAME "QGEMM_BATCH " #elif defined(DOUBLE) #define ERROR_NAME "DGEMM_BATCH " #define GEMM_BATCH_THREAD dgemm_batch_thread #else #define ERROR_NAME "SGEMM_BATCH " #define GEMM_BATCH_THREAD sgemm_batch_thread #endif #else #ifdef XDOUBLE #define ERROR_NAME "XGEMM_BATCH " #elif defined(DOUBLE) #define ERROR_NAME "ZGEMM_BATCH " #define GEMM_BATCH_THREAD zgemm_batch_thread #else #define ERROR_NAME "CGEMM_BATCH " #define GEMM_BATCH_THREAD cgemm_batch_thread #endif #endif static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, IFLOAT *, IFLOAT *, BLASLONG) = { GEMM_NN, GEMM_TN, GEMM_RN, GEMM_CN, GEMM_NT, GEMM_TT, GEMM_RT, GEMM_CT, GEMM_NR, GEMM_TR, GEMM_RR, GEMM_CR, GEMM_NC, GEMM_TC, GEMM_RC, GEMM_CC, }; #if defined(SMALL_MATRIX_OPT) && !defined(GEMM3M) && !defined(XDOUBLE) #define USE_SMALL_MATRIX_OPT 1 #else #define USE_SMALL_MATRIX_OPT 0 #endif #if USE_SMALL_MATRIX_OPT #ifndef DYNAMIC_ARCH #define SMALL_KERNEL_ADDR(table, idx) ((void *)(table[idx])) #else #define SMALL_KERNEL_ADDR(table, idx) ((void *)(*(uintptr_t *)((char *)gotoblas + (size_t)(table[idx])))) #endif #ifndef COMPLEX static size_t gemm_small_kernel[] = { GEMM_SMALL_KERNEL_NN, GEMM_SMALL_KERNEL_TN, 0, 0, GEMM_SMALL_KERNEL_NT, GEMM_SMALL_KERNEL_TT, 0, 0, }; static size_t gemm_small_kernel_b0[] = { GEMM_SMALL_KERNEL_B0_NN, GEMM_SMALL_KERNEL_B0_TN, 0, 0, GEMM_SMALL_KERNEL_B0_NT, GEMM_SMALL_KERNEL_B0_TT, 0, 0, }; #define GEMM_SMALL_KERNEL_B0(idx) (int (*)(BLASLONG, BLASLONG, BLASLONG, IFLOAT *, BLASLONG, FLOAT, IFLOAT *, BLASLONG, FLOAT *, BLASLONG)) SMALL_KERNEL_ADDR(gemm_small_kernel_b0, (idx)) #define GEMM_SMALL_KERNEL(idx) (int (*)(BLASLONG, BLASLONG, BLASLONG, IFLOAT *, BLASLONG, FLOAT, IFLOAT *, BLASLONG, FLOAT, FLOAT *, BLASLONG)) SMALL_KERNEL_ADDR(gemm_small_kernel, (idx)) #else static size_t zgemm_small_kernel[] = { GEMM_SMALL_KERNEL_NN, GEMM_SMALL_KERNEL_TN, GEMM_SMALL_KERNEL_RN, GEMM_SMALL_KERNEL_CN, GEMM_SMALL_KERNEL_NT, GEMM_SMALL_KERNEL_TT, GEMM_SMALL_KERNEL_RT, GEMM_SMALL_KERNEL_CT, GEMM_SMALL_KERNEL_NR, GEMM_SMALL_KERNEL_TR, GEMM_SMALL_KERNEL_RR, GEMM_SMALL_KERNEL_CR, GEMM_SMALL_KERNEL_NC, GEMM_SMALL_KERNEL_TC, GEMM_SMALL_KERNEL_RC, GEMM_SMALL_KERNEL_CC, }; static size_t zgemm_small_kernel_b0[] = { GEMM_SMALL_KERNEL_B0_NN, GEMM_SMALL_KERNEL_B0_TN, GEMM_SMALL_KERNEL_B0_RN, GEMM_SMALL_KERNEL_B0_CN, GEMM_SMALL_KERNEL_B0_NT, GEMM_SMALL_KERNEL_B0_TT, GEMM_SMALL_KERNEL_B0_RT, GEMM_SMALL_KERNEL_B0_CT, GEMM_SMALL_KERNEL_B0_NR, GEMM_SMALL_KERNEL_B0_TR, GEMM_SMALL_KERNEL_B0_RR, GEMM_SMALL_KERNEL_B0_CR, GEMM_SMALL_KERNEL_B0_NC, GEMM_SMALL_KERNEL_B0_TC, GEMM_SMALL_KERNEL_B0_RC, GEMM_SMALL_KERNEL_B0_CC, }; #define ZGEMM_SMALL_KERNEL(idx) (int (*)(BLASLONG, BLASLONG, BLASLONG, FLOAT *, BLASLONG, FLOAT , FLOAT, FLOAT *, BLASLONG, FLOAT , FLOAT, FLOAT *, BLASLONG)) SMALL_KERNEL_ADDR(zgemm_small_kernel, (idx)) #define ZGEMM_SMALL_KERNEL_B0(idx) (int (*)(BLASLONG, BLASLONG, BLASLONG, FLOAT *, BLASLONG, FLOAT , FLOAT, FLOAT *, BLASLONG, FLOAT *, BLASLONG)) SMALL_KERNEL_ADDR(zgemm_small_kernel_b0, (idx)) #endif #endif void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE * transa_array, enum CBLAS_TRANSPOSE * transb_array, blasint * m_array, blasint * n_array, blasint * k_array, #ifndef COMPLEX FLOAT * alpha_array, IFLOAT ** a_array, blasint * lda_array, IFLOAT ** b_array, blasint * ldb_array, FLOAT * beta_array, FLOAT ** c_array, blasint * ldc_array, blasint group_count, blasint * group_size) { #else void * valpha_array, void ** va_array, blasint * lda_array, void ** vb_array, blasint * ldb_array, void * vbeta_array, void ** vc_array, blasint * ldc_array, blasint group_count, blasint * group_size) { FLOAT * alpha_array=(FLOAT *)valpha_array; FLOAT * beta_array=(FLOAT *)vbeta_array; FLOAT ** a_array=(FLOAT**)va_array; FLOAT ** b_array=(FLOAT**)vb_array; FLOAT ** c_array=(FLOAT**)vc_array; #endif blas_arg_t * args_array=NULL; int mode=0, group_mode=0; blasint total_num=0; blasint i=0, j=0, matrix_idx=0, count=0; int group_transa, group_transb; BLASLONG group_nrowa, group_nrowb; blasint info; void * group_alpha, * group_beta; BLASLONG group_m, group_n, group_k; BLASLONG group_lda, group_ldb, group_ldc; void * group_routine=NULL; #ifdef SMALL_MATRIX_OPT void * group_small_matrix_opt_routine=NULL; #endif #if defined (SMP) || defined(SMALL_MATRIX_OPT) double MNK; #endif PRINT_DEBUG_CNAME; for(i=0; i= 0) { BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME)); free(args_array); return; } if (group_m == 0 || group_n == 0) continue; group_mode=mode; #if defined(SMP) || defined(SMALL_MATRIX_OPT) MNK = (double) group_m * (double) group_n * (double) group_k; #endif #ifdef SMALL_MATRIX_OPT if (MNK <= 100.0*100.0*100.0){ group_routine=NULL; #if !defined(COMPLEX) if(*(FLOAT *)(group_beta) == 0.0){ group_mode=mode | BLAS_SMALL_B0_OPT; group_small_matrix_opt_routine=(void *)(gemm_small_kernel_b0[(group_transb<<2)|group_transa]); }else{ group_mode=mode | BLAS_SMALL_OPT; group_small_matrix_opt_routine=(void *)(gemm_small_kernel[(group_transb<<2)|group_transa]); } #else if(((FLOAT *)(group_beta))[0] == 0.0 && ((FLOAT *)(group_beta))[1] == 0.0){ group_mode=mode | BLAS_SMALL_B0_OPT; group_small_matrix_opt_routine=(void *)(zgemm_small_kernel_b0[(group_transb<<2)|group_transa]); }else{ group_mode=mode | BLAS_SMALL_OPT; group_small_matrix_opt_routine=(void *)(zgemm_small_kernel[(group_transb<<2)|group_transa]); } #endif }else{ #endif group_routine=(void*)(gemm[(group_transb<<2)|group_transa]); #ifdef SMALL_MATRIX_OPT } #endif for(j=0; j0){ GEMM_BATCH_THREAD(args_array,count); } free(args_array); }