OpenBLAS/driver/level3/level3_gemm3m_thread.c

/*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin.           */
/* Copyright 2023 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.                              */
/*                                                                   */
/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
/*    AUSTIN  ``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 UNIVERSITY  OF TEXAS AT    */
/*    AUSTIN 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.                                    */
/*                                                                   */
/* The views and conclusions contained in the software and           */
/* documentation are those of the authors and should not be          */
/* interpreted as representing official policies, either expressed   */
/* or implied, of The University of Texas at Austin.                 */
/*********************************************************************/

#ifndef CACHE_LINE_SIZE
#define CACHE_LINE_SIZE 8
#endif

#ifndef DIVIDE_RATE
#define DIVIDE_RATE 2
#endif

//The array of job_t may overflow the stack.
//Instead, use malloc to alloc job_t.
#if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD
#define USE_ALLOC_HEAP
#endif

#ifndef GEMM3M_LOCAL
#if   defined(NN)
#define GEMM3M_LOCAL    GEMM3M_NN
#elif defined(NT)
#define GEMM3M_LOCAL    GEMM3M_NT
#elif defined(NR)
#define GEMM3M_LOCAL    GEMM3M_NR
#elif defined(NC)
#define GEMM3M_LOCAL    GEMM3M_NC
#elif defined(TN)
#define GEMM3M_LOCAL    GEMM3M_TN
#elif defined(TT)
#define GEMM3M_LOCAL    GEMM3M_TT
#elif defined(TR)
#define GEMM3M_LOCAL    GEMM3M_TR
#elif defined(TC)
#define GEMM3M_LOCAL    GEMM3M_TC
#elif defined(RN)
#define GEMM3M_LOCAL    GEMM3M_RN
#elif defined(RT)
#define GEMM3M_LOCAL    GEMM3M_RT
#elif defined(RR)
#define GEMM3M_LOCAL    GEMM3M_RR
#elif defined(RC)
#define GEMM3M_LOCAL    GEMM3M_RC
#elif defined(CN)
#define GEMM3M_LOCAL    GEMM3M_CN
#elif defined(CT)
#define GEMM3M_LOCAL    GEMM3M_CT
#elif defined(CR)
#define GEMM3M_LOCAL    GEMM3M_CR
#elif defined(CC)
#define GEMM3M_LOCAL    GEMM3M_CC
#endif
#endif

typedef struct {
#ifdef HAVE_C11
  _Atomic
#else
  volatile
#endif  
   BLASLONG working[MAX_CPU_NUMBER][CACHE_LINE_SIZE * DIVIDE_RATE];
} job_t;


#ifndef BETA_OPERATION
#define BETA_OPERATION(M_FROM, M_TO, N_FROM, N_TO, BETA, C, LDC) \
	GEMM_BETA((M_TO) - (M_FROM), (N_TO - N_FROM), 0, \
		  BETA[0], BETA[1], NULL, 0, NULL, 0, \
		  (FLOAT *)(C) + ((M_FROM) + (N_FROM) * (LDC)) * COMPSIZE, LDC)
#endif

#ifndef ICOPYB_OPERATION
#if defined(NN) || defined(NT) || defined(NC) || defined(NR) || \
    defined(RN) || defined(RT) || defined(RC) || defined(RR)
#define ICOPYB_OPERATION(M, N, A, LDA, X, Y, BUFFER) \
	GEMM3M_ITCOPYB(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
#else
#define ICOPYB_OPERATION(M, N, A, LDA, X, Y, BUFFER) \
	GEMM3M_INCOPYB(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
#endif
#endif

#ifndef ICOPYR_OPERATION
#if defined(NN) || defined(NT) || defined(NC) || defined(NR) || \
    defined(RN) || defined(RT) || defined(RC) || defined(RR)
#define ICOPYR_OPERATION(M, N, A, LDA, X, Y, BUFFER) \
	GEMM3M_ITCOPYR(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
#else
#define ICOPYR_OPERATION(M, N, A, LDA, X, Y, BUFFER) \
	GEMM3M_INCOPYR(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
#endif
#endif

#ifndef ICOPYI_OPERATION
#if defined(NN) || defined(NT) || defined(NC) || defined(NR) || \
    defined(RN) || defined(RT) || defined(RC) || defined(RR)
#define ICOPYI_OPERATION(M, N, A, LDA, X, Y, BUFFER) \
	GEMM3M_ITCOPYI(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, BUFFER);
#else
#define ICOPYI_OPERATION(M, N, A, LDA, X, Y, BUFFER) \
	GEMM3M_INCOPYI(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, BUFFER);
#endif
#endif


#ifndef OCOPYB_OPERATION
#if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
    defined(NR) || defined(TR) || defined(CR) || defined(RR)
#define OCOPYB_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
	GEMM3M_ONCOPYB(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, ALPHA_R, ALPHA_I, BUFFER);
#else
#define OCOPYB_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
	GEMM3M_OTCOPYB(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, ALPHA_R, ALPHA_I, BUFFER);
#endif
#endif

#ifndef OCOPYR_OPERATION
#if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
    defined(NR) || defined(TR) || defined(CR) || defined(RR)
#define OCOPYR_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
	GEMM3M_ONCOPYR(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, ALPHA_R, ALPHA_I, BUFFER);
#else
#define OCOPYR_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
	GEMM3M_OTCOPYR(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, ALPHA_R, ALPHA_I, BUFFER);
#endif
#endif


#ifndef OCOPYI_OPERATION
#if defined(NN) || defined(TN) || defined(CN) || defined(RN) || \
    defined(NR) || defined(TR) || defined(CR) || defined(RR)
#define OCOPYI_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
	GEMM3M_ONCOPYI(M, N, (FLOAT *)(A) + ((X) + (Y) * (LDA)) * COMPSIZE, LDA, ALPHA_R, ALPHA_I, BUFFER);
#else
#define OCOPYI_OPERATION(M, N, A, LDA, ALPHA_R, ALPHA_I, X, Y, BUFFER) \
	GEMM3M_OTCOPYI(M, N, (FLOAT *)(A) + ((Y) + (X) * (LDA)) * COMPSIZE, LDA, ALPHA_R, ALPHA_I, BUFFER);
#endif
#endif

#ifndef KERNEL_FUNC
#define KERNEL_FUNC	GEMM3M_KERNEL
#endif

#ifndef KERNEL_OPERATION
#define KERNEL_OPERATION(M, N, K, ALPHA_R, ALPHA_I, SA, SB, C, LDC, X, Y) \
	KERNEL_FUNC(M, N, K, ALPHA_R, ALPHA_I, SA, SB, (FLOAT *)(C) + ((X) + (Y) * LDC) * COMPSIZE, LDC)
#endif

#ifndef A
#define A	args -> a
#endif
#ifndef LDA
#define LDA	args -> lda
#endif
#ifndef B
#define B	args -> b
#endif
#ifndef LDB
#define LDB	args -> ldb
#endif
#ifndef C
#define C	args -> c
#endif
#ifndef LDC
#define LDC	args -> ldc
#endif
#ifndef M
#define M	args -> m
#endif
#ifndef N
#define N	args -> n
#endif
#ifndef K
#define K	args -> k
#endif

#if defined(NN) || defined(NT) || defined(TN) || defined(TT)
#define ALPHA1	ONE
#define ALPHA2	ONE
#define ALPHA5	ZERO
#define ALPHA6	ONE

#define ALPHA7	ONE
#define ALPHA8	ZERO
#define ALPHA11	ONE
#define ALPHA12	-ONE

#define ALPHA13	ZERO
#define ALPHA14	ONE
#define ALPHA17	-ONE
#define ALPHA18	-ONE
#endif

#if defined(NR) || defined(NC) || defined(TR) || defined(TC)
#define ALPHA1	ONE
#define ALPHA2	ONE
#define ALPHA5	ONE
#define ALPHA6	ZERO

#define ALPHA7	ZERO
#define ALPHA8	ONE
#define ALPHA11	-ONE
#define ALPHA12	-ONE

#define ALPHA13	ONE
#define ALPHA14	ZERO
#define ALPHA17	-ONE
#define ALPHA18	ONE
#endif

#if defined(RN) || defined(RT) || defined(CN) || defined(CT)
#define ALPHA1	ONE
#define ALPHA2	ONE
#define ALPHA5	ONE
#define ALPHA6	ZERO

#define ALPHA7	ZERO
#define ALPHA8	ONE
#define ALPHA11	-ONE
#define ALPHA12	ONE

#define ALPHA13	ONE
#define ALPHA14	ZERO
#define ALPHA17	-ONE
#define ALPHA18	-ONE
#endif

#if defined(RR) || defined(RC) || defined(CR) || defined(CC)
#define ALPHA1	ONE
#define ALPHA2	ONE
#define ALPHA5	ZERO
#define ALPHA6	-ONE

#define ALPHA7	ONE
#define ALPHA8	ZERO
#define ALPHA11	ONE
#define ALPHA12	ONE

#define ALPHA13	ZERO
#define ALPHA14	ONE
#define ALPHA17	-ONE
#define ALPHA18	ONE
#endif

#ifdef TIMING
#define START_RPCC()		rpcc_counter = rpcc()
#define STOP_RPCC(COUNTER)	COUNTER  += rpcc() - rpcc_counter
#else
#define START_RPCC()
#define STOP_RPCC(COUNTER)
#endif

static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){

  BLASLONG k, lda, ldb, ldc;
  BLASLONG m_from, m_to, n_from, n_to, N_from, N_to;

  FLOAT *alpha, *beta;
  FLOAT *a, *b, *c;
  job_t *job = (job_t *)args -> common;
  BLASLONG xxx, bufferside;
  FLOAT *buffer[DIVIDE_RATE];

  BLASLONG ls, min_l, jjs, min_jj;
  BLASLONG is, min_i, div_n;
  BLASLONG i, current;

#ifdef TIMING
  BLASLONG rpcc_counter;
  BLASLONG copy_A = 0;
  BLASLONG copy_B = 0;
  BLASLONG kernel = 0;
  BLASLONG waiting1 = 0;
  BLASLONG waiting2 = 0;
  BLASLONG waiting3 = 0;
  BLASLONG waiting6[MAX_CPU_NUMBER];
  BLASLONG ops    = 0;

  for (i = 0; i < args -> nthreads; i++) waiting6[i] = 0;
#endif

  k = K;

  a = (FLOAT *)A;
  b = (FLOAT *)B;
  c = (FLOAT *)C;

  lda = LDA;
  ldb = LDB;
  ldc = LDC;

  alpha = (FLOAT *)args -> alpha;
  beta  = (FLOAT *)args -> beta;

  m_from = 0;
  m_to   = M;

  if (range_m) {
    m_from = range_m[0];
    m_to   = range_m[1];
  }

  n_from = 0;
  n_to   = N;

  N_from = 0;
  N_to   = N;

  if (range_n) {
    n_from = range_n[mypos + 0];
    n_to   = range_n[mypos + 1];

    N_from = range_n[0];
    N_to   = range_n[args -> nthreads];
  }

  if (beta) {
    if ((beta[0] != ONE) || (beta[1] != ZERO))
      BETA_OPERATION(m_from, m_to, N_from, N_to, beta, c, ldc);
  }

  if ((k == 0) || (alpha == NULL)) return 0;

  if ((alpha[0] == ZERO) && (alpha[1] == ZERO)) return 0;

#if 0
  fprintf(stderr, "Thread[%ld]  m_from : %ld m_to : %ld n_from : %ld n_to : %ld N_from : %ld N_to : %ld\n",
	  mypos, m_from, m_to, n_from, n_to, N_from, N_to);
#endif

  div_n = (n_to - n_from + DIVIDE_RATE - 1) / DIVIDE_RATE;

  buffer[0] = sb;
  for (i = 1; i < DIVIDE_RATE; i++) {
    buffer[i] = buffer[i - 1] + GEMM3M_Q * (((div_n + GEMM3M_UNROLL_N - 1)/GEMM3M_UNROLL_N) * GEMM3M_UNROLL_N);
  }

  for(ls = 0; ls < k; ls += min_l){
    min_l = k - ls;
    if (min_l >= GEMM3M_Q * 2) {
      min_l = GEMM3M_Q;
    } else {
      if (min_l > GEMM3M_Q) {
	min_l = (min_l + 1) / 2;
      }
    }

    min_i = m_to - m_from;

    if (min_i >= GEMM3M_P * 2) {
      min_i = GEMM3M_P;
    } else {
      if (min_i > GEMM3M_P) {
	min_i = ((min_i / 2 + GEMM3M_UNROLL_M - 1)/GEMM3M_UNROLL_M) * GEMM3M_UNROLL_M;
      }
    }


    START_RPCC();

    ICOPYB_OPERATION(min_l, min_i, a, lda, ls, m_from, sa);

    STOP_RPCC(copy_A);

    div_n = (n_to - n_from + DIVIDE_RATE - 1) / DIVIDE_RATE;

    for (xxx = n_from, bufferside = 0; xxx < n_to; xxx += div_n, bufferside ++) {

      START_RPCC();

      /* Make sure if no one is using another buffer */
      for (i = 0; i < args -> nthreads; i++)
	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;MB;};

      STOP_RPCC(waiting1);

      for(jjs = xxx; jjs < MIN(n_to, xxx + div_n); jjs += min_jj){
	min_jj = MIN(n_to, xxx + div_n) - jjs;
	if (min_jj > GEMM3M_UNROLL_N*3) min_jj = GEMM3M_UNROLL_N*3;

	START_RPCC();

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || defined(RN) || defined(RT) || defined(CN) || defined(CT)
	OCOPYB_OPERATION(min_l, min_jj, b, ldb, alpha[0],  alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#else
	OCOPYB_OPERATION(min_l, min_jj, b, ldb, alpha[0], -alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#endif

	STOP_RPCC(copy_B);

	START_RPCC();

	  KERNEL_OPERATION(min_i, min_jj, min_l, ALPHA5, ALPHA6,
			   sa, buffer[bufferside] + min_l * (jjs - xxx),
			   c, ldc, m_from, jjs);

	STOP_RPCC(kernel);
#ifdef TIMING
	  ops += 2 * min_i * min_jj * min_l;
#endif

      }

      for (i = 0; i < args -> nthreads; i++)
	job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
      WMB;
	}

    current = mypos;

    do {
      current ++;
      if (current >= args -> nthreads) current = 0;

      div_n = (range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE;

      for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {

	if (current != mypos) {

	  START_RPCC();

	  /* thread has to wait */
	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;MB;};

	  STOP_RPCC(waiting2);

	  START_RPCC();


	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1]  - xxx,  div_n), min_l, ALPHA5, ALPHA6,
			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
			   c, ldc, m_from, xxx);

	STOP_RPCC(kernel);
#ifdef TIMING
	  ops += 2 * min_i * MIN(range_n[current + 1]  - xxx,  div_n) * min_l;
#endif
	}

	if (m_to - m_from == min_i) {
	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] = 0;
	WMB;
	}
      }
    } while (current != mypos);

    for(is = m_from + min_i; is < m_to; is += min_i){
      min_i = m_to - is;
      if (min_i >= GEMM3M_P * 2) {
	min_i = GEMM3M_P;
      } else
	if (min_i > GEMM3M_P) {
	  min_i = (((min_i + 1) / 2 + GEMM3M_UNROLL_M - 1)/GEMM3M_UNROLL_M) * GEMM3M_UNROLL_M;
	}

      START_RPCC();

      ICOPYB_OPERATION(min_l, min_i, a, lda, ls, is, sa);

      STOP_RPCC(copy_A);

      current = mypos;
      do {

	div_n = (range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE;

	for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {

	  START_RPCC();


	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - xxx, div_n), min_l, ALPHA5, ALPHA6,
			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
			   c, ldc, is, xxx);

	STOP_RPCC(kernel);
#ifdef TIMING
	ops += 2 * min_i * (range_n[current + 1]  - range_n[current] - div_n) * min_l;
#endif
	if (is + min_i >= m_to) {
	  /* Thread doesn't need this buffer any more */
	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] = 0;
	WMB;
	}
	}

	current ++;
	if (current >= args -> nthreads) current = 0;

      } while (current != mypos);

    } /* end of is */

    START_RPCC();

    ICOPYR_OPERATION(min_l, min_i, a, lda, ls, m_from, sa);

    STOP_RPCC(copy_A);

    div_n = (n_to - n_from + DIVIDE_RATE - 1) / DIVIDE_RATE;

    for (xxx = n_from, bufferside = 0; xxx < n_to; xxx += div_n, bufferside ++) {

      START_RPCC();

      /* Make sure if no one is using another buffer */
      for (i = 0; i < args -> nthreads; i++)
	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;MB;};

      STOP_RPCC(waiting1);

      for(jjs = xxx; jjs < MIN(n_to, xxx + div_n); jjs += min_jj){
	min_jj = MIN(n_to, xxx + div_n) - jjs;
	if (min_jj > GEMM3M_UNROLL_N*3) min_jj = GEMM3M_UNROLL_N*3;

	START_RPCC();

#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
	OCOPYR_OPERATION(min_l, min_jj, b, ldb, alpha[0],  alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
	OCOPYR_OPERATION(min_l, min_jj, b, ldb, alpha[0], -alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#elif defined(RN) || defined(RT) || defined(CN) || defined(CT)
	OCOPYI_OPERATION(min_l, min_jj, b, ldb, alpha[0],  alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#else
	OCOPYI_OPERATION(min_l, min_jj, b, ldb, alpha[0], -alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#endif

	STOP_RPCC(copy_B);

	START_RPCC();

	  KERNEL_OPERATION(min_i, min_jj, min_l, ALPHA11, ALPHA12,
			   sa, buffer[bufferside] + min_l * (jjs - xxx),
			   c, ldc, m_from, jjs);

	STOP_RPCC(kernel);
#ifdef TIMING
	  ops += 2 * min_i * min_jj * min_l;
#endif

      }

      for (i = 0; i < args -> nthreads; i++)
	job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
      }

    current = mypos;

    do {
      current ++;
      if (current >= args -> nthreads) current = 0;

      div_n = (range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE;

      for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {

	if (current != mypos) {

	  START_RPCC();

	  /* thread has to wait */
	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;MB;};

	  STOP_RPCC(waiting2);

	  START_RPCC();

	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1]  - xxx,  div_n), min_l, ALPHA11, ALPHA12,
			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
			   c, ldc, m_from, xxx);

	STOP_RPCC(kernel);
#ifdef TIMING
	  ops += 2 * min_i * MIN(range_n[current + 1]  - xxx,  div_n) * min_l;
#endif
	}

	if (m_to - m_from == min_i) {
	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] = 0;
	WMB;
	}
      }
    } while (current != mypos);

    for(is = m_from + min_i; is < m_to; is += min_i){
      min_i = m_to - is;
      if (min_i >= GEMM3M_P * 2) {
	min_i = GEMM3M_P;
      } else
	if (min_i > GEMM3M_P) {
	  min_i = (((min_i + 1) / 2 + GEMM3M_UNROLL_M - 1)/GEMM3M_UNROLL_M) * GEMM3M_UNROLL_M;
	}

      START_RPCC();

      ICOPYR_OPERATION(min_l, min_i, a, lda, ls, is, sa);

      STOP_RPCC(copy_A);

      current = mypos;
      do {

	div_n = (range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE;

	for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {

	  START_RPCC();

	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - xxx, div_n), min_l, ALPHA11, ALPHA12,
			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
			   c, ldc, is, xxx);

	STOP_RPCC(kernel);
#ifdef TIMING
	ops += 2 * min_i * (range_n[current + 1]  - range_n[current] - div_n) * min_l;
#endif
	if (is + min_i >= m_to) {
	  /* Thread doesn't need this buffer any more */
	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] = 0;
	}
	}

	current ++;
	if (current >= args -> nthreads) current = 0;

      } while (current != mypos);

    } /* end of is */


    START_RPCC();

    ICOPYI_OPERATION(min_l, min_i, a, lda, ls, m_from, sa);

    STOP_RPCC(copy_A);

    div_n = (n_to - n_from + DIVIDE_RATE - 1) / DIVIDE_RATE;

    for (xxx = n_from, bufferside = 0; xxx < n_to; xxx += div_n, bufferside ++) {

      START_RPCC();

      /* Make sure if no one is using another buffer */
      for (i = 0; i < args -> nthreads; i++)
	while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;MB;};

      STOP_RPCC(waiting1);

      for(jjs = xxx; jjs < MIN(n_to, xxx + div_n); jjs += min_jj){
	min_jj = MIN(n_to, xxx + div_n) - jjs;
	if (min_jj > GEMM3M_UNROLL_N*3) min_jj = GEMM3M_UNROLL_N*3;

	START_RPCC();

#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
	OCOPYI_OPERATION(min_l, min_jj, b, ldb, alpha[0],  alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
	OCOPYI_OPERATION(min_l, min_jj, b, ldb, alpha[0], -alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#elif defined(RN) || defined(RT) || defined(CN) || defined(CT)
	OCOPYR_OPERATION(min_l, min_jj, b, ldb, alpha[0],  alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#else
	OCOPYR_OPERATION(min_l, min_jj, b, ldb, alpha[0], -alpha[1], ls, jjs, buffer[bufferside] + min_l * (jjs - xxx));
#endif

	STOP_RPCC(copy_B);

	START_RPCC();

	KERNEL_OPERATION(min_i, min_jj, min_l, ALPHA17, ALPHA18,
			 sa, buffer[bufferside] + min_l * (jjs - xxx),
			 c, ldc, m_from, jjs);

	STOP_RPCC(kernel);
#ifdef TIMING
	  ops += 2 * min_i * min_jj * min_l;
#endif

      }

      for (i = 0; i < args -> nthreads; i++)
	job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
      }

    current = mypos;

    do {
      current ++;
      if (current >= args -> nthreads) current = 0;

      div_n = (range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE;

      for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {

	if (current != mypos) {

	  START_RPCC();

	  /* thread has to wait */
	  while(job[current].working[mypos][CACHE_LINE_SIZE * bufferside] == 0) {YIELDING;MB;};

	  STOP_RPCC(waiting2);

	  START_RPCC();

	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1]  - xxx,  div_n), min_l, ALPHA17, ALPHA18,
			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
			   c, ldc, m_from, xxx);

	STOP_RPCC(kernel);
#ifdef TIMING
	  ops += 2 * min_i * MIN(range_n[current + 1]  - xxx,  div_n) * min_l;
#endif
	}

	if (m_to - m_from == min_i) {
	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
	WMB;
}
      }
    } while (current != mypos);

    for(is = m_from + min_i; is < m_to; is += min_i){
      min_i = m_to - is;
      if (min_i >= GEMM3M_P * 2) {
	min_i = GEMM3M_P;
      } else
	if (min_i > GEMM3M_P) {
	  min_i = (((min_i + 1) / 2 + GEMM3M_UNROLL_M - 1)/GEMM3M_UNROLL_M) * GEMM3M_UNROLL_M;
	}

      START_RPCC();

      ICOPYI_OPERATION(min_l, min_i, a, lda, ls, is, sa);

      STOP_RPCC(copy_A);

      current = mypos;
      do {

	div_n = (range_n[current + 1]  - range_n[current] + DIVIDE_RATE - 1) / DIVIDE_RATE;

	for (xxx = range_n[current], bufferside = 0; xxx < range_n[current + 1]; xxx += div_n, bufferside ++) {

	  START_RPCC();

	  KERNEL_OPERATION(min_i, MIN(range_n[current + 1] - xxx, div_n), min_l, ALPHA17, ALPHA18,
			   sa, (FLOAT *)job[current].working[mypos][CACHE_LINE_SIZE * bufferside],
			   c, ldc, is, xxx);

	STOP_RPCC(kernel);
#ifdef TIMING
	ops += 2 * min_i * (range_n[current + 1]  - range_n[current] - div_n) * min_l;
#endif
	if (is + min_i >= m_to) {
	  /* Thread doesn't need this buffer any more */
	  job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
	  WMB;
	}
	}

	current ++;
	if (current >= args -> nthreads) current = 0;

      } while (current != mypos);

    } /* end of is */

  }

  START_RPCC();

  for (i = 0; i < args -> nthreads; i++) {
    for (xxx = 0; xxx < DIVIDE_RATE; xxx++) {
      while (job[mypos].working[i][CACHE_LINE_SIZE * xxx] ) {YIELDING;MB;};
    }
  }

  STOP_RPCC(waiting3);

#ifdef TIMING
  BLASLONG waiting = waiting1 + waiting2 + waiting3;
  BLASLONG total = copy_A + copy_B + kernel + waiting;

  fprintf(stderr, "GEMM   [%2ld] Copy_A : %6.2f  Copy_B : %6.2f  Wait : %6.2f Kernel : %6.2f\n",
	  mypos, (double)copy_A /(double)total * 100., (double)copy_B /(double)total * 100.,
	  (double)waiting /(double)total * 100.,
	  (double)ops/(double)kernel / 2. * 100.);

  fprintf(stderr, "GEMM   [%2ld] Copy_A : %6.2ld  Copy_B : %6.2ld  Wait : %6.2ld\n",
	  mypos, copy_A, copy_B, waiting);

#if 0
  fprintf(stderr, "Waiting[%2ld] %6.2f %6.2f %6.2f\n",
	  mypos,
	  (double)waiting1/(double)waiting * 100.,
	  (double)waiting2/(double)waiting * 100.,
	  (double)waiting3/(double)waiting * 100.);
#endif
  fprintf(stderr, "\n");
#endif



  return 0;
}

static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
		       *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){

#ifndef USE_OPENMP
#ifndef OS_WINDOWS
static pthread_mutex_t  level3_lock    = PTHREAD_MUTEX_INITIALIZER;
#else
CRITICAL_SECTION level3_lock;
InitializeCriticalSection((PCRITICAL_SECTION)&level3_lock);
#endif
#endif

  blas_arg_t newarg;

  blas_queue_t queue[MAX_CPU_NUMBER];

  BLASLONG range_M[MAX_CPU_NUMBER + 1];
  BLASLONG range_N[MAX_CPU_NUMBER + 1];

#ifndef USE_ALLOC_HEAP
  job_t          job[MAX_CPU_NUMBER];
#else
  job_t *        job = NULL;
#endif

  BLASLONG num_cpu_m, num_cpu_n;

  BLASLONG nthreads = args -> nthreads;

  BLASLONG width, i, j, k, js;
  BLASLONG m, n, n_from, n_to;
  int  mode;

#ifdef XDOUBLE
  mode  =  BLAS_XDOUBLE | BLAS_REAL | BLAS_NODE;
#elif defined(DOUBLE)
  mode  =  BLAS_DOUBLE  | BLAS_REAL | BLAS_NODE;
#else
  mode  =  BLAS_SINGLE  | BLAS_REAL | BLAS_NODE;
#endif

#ifndef USE_OPENMP
#ifndef OS_WINDOWS
pthread_mutex_lock(&level3_lock);
#else
EnterCriticalSection((PCRITICAL_SECTION)&level3_lock);
#endif
#endif

  newarg.m        = args -> m;
  newarg.n        = args -> n;
  newarg.k        = args -> k;
  newarg.a        = args -> a;
  newarg.b        = args -> b;
  newarg.c        = args -> c;
  newarg.lda      = args -> lda;
  newarg.ldb      = args -> ldb;
  newarg.ldc      = args -> ldc;
  newarg.alpha    = args -> alpha;
  newarg.beta     = args -> beta;
  newarg.nthreads = args -> nthreads;

#ifdef USE_ALLOC_HEAP
  job = (job_t*)malloc(MAX_CPU_NUMBER * sizeof(job_t));
  if(job==NULL){
    fprintf(stderr, "OpenBLAS: malloc failed in %s\n", __func__);
    exit(1);
  }
#endif

  newarg.common   = (void *)job;

  if (!range_m) {
    range_M[0] = 0;
    m          = args -> m;
  } else {
    range_M[0] = range_m[0];
    m          = range_m[1] - range_m[0];
  }

  num_cpu_m  = 0;

  while (m > 0){

    width  = blas_quickdivide(m + nthreads - num_cpu_m - 1, nthreads - num_cpu_m);

    m -= width;
    if (m < 0) width = width + m;

    range_M[num_cpu_m + 1] = range_M[num_cpu_m] + width;

    num_cpu_m ++;
  }

  for (i = 0; i < num_cpu_m; i++) {
    queue[i].mode    = mode;
    queue[i].routine = inner_thread;
    queue[i].args    = &newarg;
    queue[i].range_m = &range_M[i];
    queue[i].range_n = &range_N[0];
    queue[i].sa      = NULL;
    queue[i].sb      = NULL;
    queue[i].next    = &queue[i + 1];
  }

  queue[0].sa = sa;
  queue[0].sb = sb;

  if (!range_n) {
    n_from = 0;
    n_to   = args -> n;
  } else {
    n_from = range_n[0];
    n_to   = range_n[1];
  }

  for(js = n_from; js < n_to; js += GEMM_R * nthreads){
    n = n_to - js;
    if (n > GEMM_R * nthreads) n = GEMM_R * nthreads;

    range_N[0] = js;

    num_cpu_n  = 0;

    while (n > 0){

      width  = blas_quickdivide(n + nthreads - num_cpu_n - 1, nthreads - num_cpu_n);

      n -= width;
      if (n < 0) width = width + n;

      range_N[num_cpu_n + 1] = range_N[num_cpu_n] + width;

      num_cpu_n ++;
    }

    for (j = 0; j < num_cpu_m; j++) {
      for (i = 0; i < num_cpu_m; i++) {
	for (k = 0; k < DIVIDE_RATE; k++) {
	  job[j].working[i][CACHE_LINE_SIZE * k] = 0;
	}
      }
    }

    queue[num_cpu_m - 1].next = NULL;

    exec_blas(num_cpu_m, queue);
  }

#ifdef USE_ALLOC_HEAP
  free(job);
#endif

#ifndef USE_OPENMP
#ifndef OS_WINDOWS
  pthread_mutex_unlock(&level3_lock);
#else
  LeaveCriticalSection((PCRITICAL_SECTION)&level3_lock);
#endif
#endif

  return 0;
}

int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLOAT *sb, BLASLONG mypos){

  BLASLONG m = args -> m;
  // BLASLONG n = args -> n;
  BLASLONG nthreads = args -> nthreads;
  BLASLONG divN, divT;
  int mode;

#if defined(DYNAMIC_ARCH)
  int switch_ratio = gotoblas->switch_ratio;
#else
  int switch_ratio = SWITCH_RATIO;
#endif

  if (range_m) {
    BLASLONG m_from = *(((BLASLONG *)range_m) + 0);
    BLASLONG m_to   = *(((BLASLONG *)range_m) + 1);

    m = m_to - m_from;
  }
/*
  if (range_n) {
    BLASLONG n_from = *(((BLASLONG *)range_n) + 0);
    BLASLONG n_to   = *(((BLASLONG *)range_n) + 1);

    n = n_to - n_from;
  }
*/

  if ((args -> m < nthreads * switch_ratio) || (args -> n < nthreads * switch_ratio)) {
    GEMM3M_LOCAL(args, range_m, range_n, sa, sb, 0);
    return 0;
  }

  divT = nthreads;
  divN = 1;

  while ((GEMM3M_P * divT > m * switch_ratio) && (divT > 1)) {
    do {
      divT --;
      divN = 1;
      while (divT * divN < nthreads) divN ++;
    } while ((divT * divN != nthreads) && (divT > 1));
  }

  args -> nthreads = divT;

  if (divN == 1){
    gemm_driver(args, range_m, range_n, sa, sb, 0);
  } else {
#ifdef XDOUBLE
    mode  =  BLAS_XDOUBLE | BLAS_COMPLEX;
#elif defined(DOUBLE)
    mode  =  BLAS_DOUBLE  | BLAS_COMPLEX;
#else
    mode  =  BLAS_SINGLE  | BLAS_COMPLEX;
#endif

#if defined(TN) || defined(TT) || defined(TR) || defined(TC) || \
    defined(CN) || defined(CT) || defined(CR) || defined(CC)
    mode |= (BLAS_TRANSA_T);
#endif
#if defined(NT) || defined(TT) || defined(RT) || defined(CT) || \
    defined(NC) || defined(TC) || defined(RC) || defined(CC)
    mode |= (BLAS_TRANSB_T);
#endif

    gemm_thread_n(mode, args, range_m, range_n, gemm_driver, sa, sb, divN);
  }

  return 0;
}