Home | History | Annotate | Line # | Download | only in m4
matmulavx128.m4 revision 1.1.1.2
      1      1.1  mrg `/* Implementation of the MATMUL intrinsic
      2  1.1.1.2  mrg    Copyright (C) 2002-2020 Free Software Foundation, Inc.
      3      1.1  mrg    Contributed by Thomas Koenig <tkoenig (a] gcc.gnu.org>.
      4      1.1  mrg 
      5      1.1  mrg This file is part of the GNU Fortran runtime library (libgfortran).
      6      1.1  mrg 
      7      1.1  mrg Libgfortran is free software; you can redistribute it and/or
      8      1.1  mrg modify it under the terms of the GNU General Public
      9      1.1  mrg License as published by the Free Software Foundation; either
     10      1.1  mrg version 3 of the License, or (at your option) any later version.
     11      1.1  mrg 
     12      1.1  mrg Libgfortran is distributed in the hope that it will be useful,
     13      1.1  mrg but WITHOUT ANY WARRANTY; without even the implied warranty of
     14      1.1  mrg MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
     15      1.1  mrg GNU General Public License for more details.
     16      1.1  mrg 
     17      1.1  mrg Under Section 7 of GPL version 3, you are granted additional
     18      1.1  mrg permissions described in the GCC Runtime Library Exception, version
     19      1.1  mrg 3.1, as published by the Free Software Foundation.
     20      1.1  mrg 
     21      1.1  mrg You should have received a copy of the GNU General Public License and
     22      1.1  mrg a copy of the GCC Runtime Library Exception along with this program;
     23      1.1  mrg see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
     24      1.1  mrg <http://www.gnu.org/licenses/>.  */
     25      1.1  mrg 
     26      1.1  mrg #include "libgfortran.h"
     27      1.1  mrg #include <string.h>
     28      1.1  mrg #include <assert.h>'
     29      1.1  mrg 
     30      1.1  mrg include(iparm.m4)dnl
     31      1.1  mrg 
     32      1.1  mrg /* These are the specific versions of matmul with -mprefer-avx128.  */
     33      1.1  mrg 
     34      1.1  mrg `#if defined (HAVE_'rtype_name`)
     35      1.1  mrg 
     36      1.1  mrg /* Prototype for the BLAS ?gemm subroutine, a pointer to which can be
     37      1.1  mrg    passed to us by the front-end, in which case we call it for large
     38      1.1  mrg    matrices.  */
     39      1.1  mrg 
     40      1.1  mrg typedef void (*blas_call)(const char *, const char *, const int *, const int *,
     41      1.1  mrg                           const int *, const 'rtype_name` *, const 'rtype_name` *,
     42      1.1  mrg                           const int *, const 'rtype_name` *, const int *,
     43      1.1  mrg                           const 'rtype_name` *, 'rtype_name` *, const int *,
     44      1.1  mrg                           int, int);
     45      1.1  mrg 
     46      1.1  mrg #if defined(HAVE_AVX) && defined(HAVE_FMA3) && defined(HAVE_AVX128)
     47      1.1  mrg 'define(`matmul_name',`matmul_'rtype_code`_avx128_fma3')dnl
     48      1.1  mrg `void
     49      1.1  mrg 'matmul_name` ('rtype` * const restrict retarray, 
     50      1.1  mrg 	'rtype` * const restrict a, 'rtype` * const restrict b, int try_blas,
     51      1.1  mrg 	int blas_limit, blas_call gemm) __attribute__((__target__("avx,fma")));
     52      1.1  mrg internal_proto('matmul_name`);
     53      1.1  mrg 'include(matmul_internal.m4)dnl
     54      1.1  mrg `#endif
     55      1.1  mrg 
     56      1.1  mrg #if defined(HAVE_AVX) && defined(HAVE_FMA4) && defined(HAVE_AVX128)
     57      1.1  mrg 'define(`matmul_name',`matmul_'rtype_code`_avx128_fma4')dnl
     58      1.1  mrg `void
     59      1.1  mrg 'matmul_name` ('rtype` * const restrict retarray, 
     60      1.1  mrg 	'rtype` * const restrict a, 'rtype` * const restrict b, int try_blas,
     61      1.1  mrg 	int blas_limit, blas_call gemm) __attribute__((__target__("avx,fma4")));
     62      1.1  mrg internal_proto('matmul_name`);
     63      1.1  mrg 'include(matmul_internal.m4)dnl
     64      1.1  mrg `#endif
     65      1.1  mrg 
     66      1.1  mrg #endif
     67      1.1  mrg '
     68