OpenBLAS/interface/gemm_batch.c

/*****************************************************************************
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 <stdio.h>
#include <stdlib.h>
#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<group_count; i++){
    total_num+=group_size[i];
  }

  args_array=(blas_arg_t *)malloc(total_num * sizeof(blas_arg_t));
  
  if(args_array == NULL){
    openblas_warning(0, "memory alloc failed!\n");
    return;
  }

#ifdef SMP
#ifndef COMPLEX
#ifdef XDOUBLE
  mode  =  BLAS_XDOUBLE | BLAS_REAL;
#elif defined(DOUBLE)
  mode  =  BLAS_DOUBLE  | BLAS_REAL;
#else
  mode  =  BLAS_SINGLE  | BLAS_REAL;
#endif
#else
#ifdef XDOUBLE
  mode  =  BLAS_XDOUBLE | BLAS_COMPLEX;
#elif defined(DOUBLE)
  mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
#else
  mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
#endif
#endif
#endif
  
  for(i=0; i<group_count; matrix_idx+=group_size[i], i++){
    group_alpha = (void *)&alpha_array[i * COMPSIZE];
    group_beta = (void *)&beta_array[i * COMPSIZE];

    group_m = group_n = group_k = 0;
    group_lda = group_ldb = group_ldc = 0;
    group_transa = -1;
    group_transb = -1;
    info   = 0;
    
    if (order == CblasColMajor) {
      group_m = m_array[i];
      group_n = n_array[i];
      group_k = k_array[i];

      group_lda = lda_array[i];
      group_ldb = ldb_array[i];
      group_ldc = ldc_array[i];
      
      if (transa_array[i] == CblasNoTrans)     group_transa = 0;
      if (transa_array[i] == CblasTrans)       group_transa = 1;
#ifndef COMPLEX
      if (transa_array[i] == CblasConjNoTrans) group_transa = 0;
      if (transa_array[i] == CblasConjTrans)   group_transa = 1;
#else
      if (transa_array[i] == CblasConjNoTrans) group_transa = 2;
      if (transa_array[i] == CblasConjTrans)   group_transa = 3;
#endif
      if (transb_array[i] == CblasNoTrans)     group_transb = 0;
      if (transb_array[i] == CblasTrans)       group_transb = 1;
#ifndef COMPLEX
      if (transb_array[i] == CblasConjNoTrans) group_transb = 0;
      if (transb_array[i] == CblasConjTrans)   group_transb = 1;
#else
      if (transb_array[i] == CblasConjNoTrans) group_transb = 2;
      if (transb_array[i] == CblasConjTrans)   group_transb = 3;
#endif
      group_nrowa = group_m;
      if (group_transa & 1) group_nrowa = group_k;
      group_nrowb = group_k;
      if (group_transb & 1) group_nrowb = group_n;

      info=-1;
	
      if (group_ldc < group_m) info = 13;
      if (group_ldb < group_nrowb)  info = 10;
      if (group_lda < group_nrowa)  info =  8;
      if (group_k < 0)        info =  5;
      if (group_n < 0)        info =  4;
      if (group_m < 0)        info =  3;
      if (group_transb < 0)        info =  2;
      if (group_transa < 0)        info =  1;

    }else if (order == CblasRowMajor) {
      
      group_m = n_array[i];
      group_n = m_array[i];
      group_k = k_array[i];

      group_lda = ldb_array[i];
      group_ldb = lda_array[i];
      group_ldc = ldc_array[i];
      
      if (transb_array[i] == CblasNoTrans)     group_transa = 0;
      if (transb_array[i] == CblasTrans)       group_transa = 1;
#ifndef COMPLEX
      if (transb_array[i] == CblasConjNoTrans) group_transa = 0;
      if (transb_array[i] == CblasConjTrans)   group_transa = 1;
#else
      if (transb_array[i] == CblasConjNoTrans) group_transa = 2;
      if (transb_array[i] == CblasConjTrans)   group_transa = 3;
#endif
      if (transa_array[i] == CblasNoTrans)     group_transb = 0;
      if (transa_array[i] == CblasTrans)       group_transb = 1;
#ifndef COMPLEX
      if (transa_array[i] == CblasConjNoTrans) group_transb = 0;
      if (transa_array[i] == CblasConjTrans)   group_transb = 1;
#else
      if (transa_array[i] == CblasConjNoTrans) group_transb = 2;
      if (transa_array[i] == CblasConjTrans)   group_transb = 3;
#endif
      group_nrowa = group_m;
      if (group_transa & 1) group_nrowa = group_k;
      group_nrowb = group_k;
      if (group_transb & 1) group_nrowb = group_n;

      info=-1;
	
      if (group_ldc < group_m) info = 13;
      if (group_ldb < group_nrowb)  info = 10;
      if (group_lda < group_nrowa)  info =  8;
      if (group_k < 0)        info =  5;
      if (group_n < 0)        info =  4;
      if (group_m < 0)        info =  3;
      if (group_transb < 0)        info =  2;
      if (group_transa < 0)        info =  1;      
    }

    if (info >= 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; j<group_size[i]; j++){
      args_array[count].m=group_m;
      args_array[count].n=group_n;
      args_array[count].k=group_k;
      args_array[count].lda=group_lda;
      args_array[count].ldb=group_ldb;
      args_array[count].ldc=group_ldc;
      args_array[count].alpha=group_alpha;
      args_array[count].beta=group_beta;

      if (order == CblasColMajor) {      
	args_array[count].a=(a_array[matrix_idx+j]);
	args_array[count].b=(b_array[matrix_idx+j]);
      }else if(order == CblasRowMajor){
	args_array[count].a=(b_array[matrix_idx+j]);
	args_array[count].b=(a_array[matrix_idx+j]);
      }
      
      args_array[count].c=(c_array[matrix_idx+j]);
      
      args_array[count].routine_mode=group_mode;
      args_array[count].routine=group_routine;
#ifdef SMALL_MATRIX_OPT
      if (!group_routine)
        args_array[count].routine=group_small_matrix_opt_routine;
#endif      
      count++;
    }
  }

  if(count>0){
    GEMM_BATCH_THREAD(args_array,count);
  }

  free(args_array);
}