1 |
733 |
jeremybenn |
`/* Implementation of the MATMUL intrinsic
|
2 |
|
|
Copyright 2002, 2005, 2006, 2007, 2009 Free Software Foundation, Inc.
|
3 |
|
|
Contributed by Paul Brook <paul@nowt.org>
|
4 |
|
|
|
5 |
|
|
This file is part of the GNU Fortran 95 runtime library (libgfortran).
|
6 |
|
|
|
7 |
|
|
Libgfortran is free software; you can redistribute it and/or
|
8 |
|
|
modify it under the terms of the GNU General Public
|
9 |
|
|
License as published by the Free Software Foundation; either
|
10 |
|
|
version 3 of the License, or (at your option) any later version.
|
11 |
|
|
|
12 |
|
|
Libgfortran is distributed in the hope that it will be useful,
|
13 |
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
14 |
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
15 |
|
|
GNU General Public License for more details.
|
16 |
|
|
|
17 |
|
|
Under Section 7 of GPL version 3, you are granted additional
|
18 |
|
|
permissions described in the GCC Runtime Library Exception, version
|
19 |
|
|
3.1, as published by the Free Software Foundation.
|
20 |
|
|
|
21 |
|
|
You should have received a copy of the GNU General Public License and
|
22 |
|
|
a copy of the GCC Runtime Library Exception along with this program;
|
23 |
|
|
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
24 |
|
|
<http://www.gnu.org/licenses/>. */
|
25 |
|
|
|
26 |
|
|
#include "libgfortran.h"
|
27 |
|
|
#include <stdlib.h>
|
28 |
|
|
#include <string.h>
|
29 |
|
|
#include <assert.h>'
|
30 |
|
|
|
31 |
|
|
include(iparm.m4)dnl
|
32 |
|
|
|
33 |
|
|
`#if defined (HAVE_'rtype_name`)
|
34 |
|
|
|
35 |
|
|
/* Prototype for the BLAS ?gemm subroutine, a pointer to which can be
|
36 |
|
|
passed to us by the front-end, in which case we''`ll call it for large
|
37 |
|
|
matrices. */
|
38 |
|
|
|
39 |
|
|
typedef void (*blas_call)(const char *, const char *, const int *, const int *,
|
40 |
|
|
const int *, const 'rtype_name` *, const 'rtype_name` *,
|
41 |
|
|
const int *, const 'rtype_name` *, const int *,
|
42 |
|
|
const 'rtype_name` *, 'rtype_name` *, const int *,
|
43 |
|
|
int, int);
|
44 |
|
|
|
45 |
|
|
/* The order of loops is different in the case of plain matrix
|
46 |
|
|
multiplication C=MATMUL(A,B), and in the frequent special case where
|
47 |
|
|
the argument A is the temporary result of a TRANSPOSE intrinsic:
|
48 |
|
|
C=MATMUL(TRANSPOSE(A),B). Transposed temporaries are detected by
|
49 |
|
|
looking at their strides.
|
50 |
|
|
|
51 |
|
|
The equivalent Fortran pseudo-code is:
|
52 |
|
|
|
53 |
|
|
DIMENSION A(M,COUNT), B(COUNT,N), C(M,N)
|
54 |
|
|
IF (.NOT.IS_TRANSPOSED(A)) THEN
|
55 |
|
|
C = 0
|
56 |
|
|
DO J=1,N
|
57 |
|
|
DO K=1,COUNT
|
58 |
|
|
DO I=1,M
|
59 |
|
|
C(I,J) = C(I,J)+A(I,K)*B(K,J)
|
60 |
|
|
ELSE
|
61 |
|
|
DO J=1,N
|
62 |
|
|
DO I=1,M
|
63 |
|
|
S = 0
|
64 |
|
|
DO K=1,COUNT
|
65 |
|
|
S = S+A(I,K)*B(K,J)
|
66 |
|
|
C(I,J) = S
|
67 |
|
|
ENDIF
|
68 |
|
|
*/
|
69 |
|
|
|
70 |
|
|
/* If try_blas is set to a nonzero value, then the matmul function will
|
71 |
|
|
see if there is a way to perform the matrix multiplication by a call
|
72 |
|
|
to the BLAS gemm function. */
|
73 |
|
|
|
74 |
|
|
extern void matmul_'rtype_code` ('rtype` * const restrict retarray,
|
75 |
|
|
'rtype` * const restrict a, 'rtype` * const restrict b, int try_blas,
|
76 |
|
|
int blas_limit, blas_call gemm);
|
77 |
|
|
export_proto(matmul_'rtype_code`);
|
78 |
|
|
|
79 |
|
|
void
|
80 |
|
|
matmul_'rtype_code` ('rtype` * const restrict retarray,
|
81 |
|
|
'rtype` * const restrict a, 'rtype` * const restrict b, int try_blas,
|
82 |
|
|
int blas_limit, blas_call gemm)
|
83 |
|
|
{
|
84 |
|
|
const 'rtype_name` * restrict abase;
|
85 |
|
|
const 'rtype_name` * restrict bbase;
|
86 |
|
|
'rtype_name` * restrict dest;
|
87 |
|
|
|
88 |
|
|
index_type rxstride, rystride, axstride, aystride, bxstride, bystride;
|
89 |
|
|
index_type x, y, n, count, xcount, ycount;
|
90 |
|
|
|
91 |
|
|
assert (GFC_DESCRIPTOR_RANK (a) == 2
|
92 |
|
|
|| GFC_DESCRIPTOR_RANK (b) == 2);
|
93 |
|
|
|
94 |
|
|
/* C[xcount,ycount] = A[xcount, count] * B[count,ycount]
|
95 |
|
|
|
96 |
|
|
Either A or B (but not both) can be rank 1:
|
97 |
|
|
|
98 |
|
|
o One-dimensional argument A is implicitly treated as a row matrix
|
99 |
|
|
dimensioned [1,count], so xcount=1.
|
100 |
|
|
|
101 |
|
|
o One-dimensional argument B is implicitly treated as a column matrix
|
102 |
|
|
dimensioned [count, 1], so ycount=1.
|
103 |
|
|
*/
|
104 |
|
|
|
105 |
|
|
if (retarray->data == NULL)
|
106 |
|
|
{
|
107 |
|
|
if (GFC_DESCRIPTOR_RANK (a) == 1)
|
108 |
|
|
{
|
109 |
|
|
GFC_DIMENSION_SET(retarray->dim[0], 0,
|
110 |
|
|
GFC_DESCRIPTOR_EXTENT(b,1) - 1, 1);
|
111 |
|
|
}
|
112 |
|
|
else if (GFC_DESCRIPTOR_RANK (b) == 1)
|
113 |
|
|
{
|
114 |
|
|
GFC_DIMENSION_SET(retarray->dim[0], 0,
|
115 |
|
|
GFC_DESCRIPTOR_EXTENT(a,0) - 1, 1);
|
116 |
|
|
}
|
117 |
|
|
else
|
118 |
|
|
{
|
119 |
|
|
GFC_DIMENSION_SET(retarray->dim[0], 0,
|
120 |
|
|
GFC_DESCRIPTOR_EXTENT(a,0) - 1, 1);
|
121 |
|
|
|
122 |
|
|
GFC_DIMENSION_SET(retarray->dim[1], 0,
|
123 |
|
|
GFC_DESCRIPTOR_EXTENT(b,1) - 1,
|
124 |
|
|
GFC_DESCRIPTOR_EXTENT(retarray,0));
|
125 |
|
|
}
|
126 |
|
|
|
127 |
|
|
retarray->data
|
128 |
|
|
= internal_malloc_size (sizeof ('rtype_name`) * size0 ((array_t *) retarray));
|
129 |
|
|
retarray->offset = 0;
|
130 |
|
|
}
|
131 |
|
|
else if (unlikely (compile_options.bounds_check))
|
132 |
|
|
{
|
133 |
|
|
index_type ret_extent, arg_extent;
|
134 |
|
|
|
135 |
|
|
if (GFC_DESCRIPTOR_RANK (a) == 1)
|
136 |
|
|
{
|
137 |
|
|
arg_extent = GFC_DESCRIPTOR_EXTENT(b,1);
|
138 |
|
|
ret_extent = GFC_DESCRIPTOR_EXTENT(retarray,0);
|
139 |
|
|
if (arg_extent != ret_extent)
|
140 |
|
|
runtime_error ("Incorrect extent in return array in"
|
141 |
|
|
" MATMUL intrinsic: is %ld, should be %ld",
|
142 |
|
|
(long int) ret_extent, (long int) arg_extent);
|
143 |
|
|
}
|
144 |
|
|
else if (GFC_DESCRIPTOR_RANK (b) == 1)
|
145 |
|
|
{
|
146 |
|
|
arg_extent = GFC_DESCRIPTOR_EXTENT(a,0);
|
147 |
|
|
ret_extent = GFC_DESCRIPTOR_EXTENT(retarray,0);
|
148 |
|
|
if (arg_extent != ret_extent)
|
149 |
|
|
runtime_error ("Incorrect extent in return array in"
|
150 |
|
|
" MATMUL intrinsic: is %ld, should be %ld",
|
151 |
|
|
(long int) ret_extent, (long int) arg_extent);
|
152 |
|
|
}
|
153 |
|
|
else
|
154 |
|
|
{
|
155 |
|
|
arg_extent = GFC_DESCRIPTOR_EXTENT(a,0);
|
156 |
|
|
ret_extent = GFC_DESCRIPTOR_EXTENT(retarray,0);
|
157 |
|
|
if (arg_extent != ret_extent)
|
158 |
|
|
runtime_error ("Incorrect extent in return array in"
|
159 |
|
|
" MATMUL intrinsic for dimension 1:"
|
160 |
|
|
" is %ld, should be %ld",
|
161 |
|
|
(long int) ret_extent, (long int) arg_extent);
|
162 |
|
|
|
163 |
|
|
arg_extent = GFC_DESCRIPTOR_EXTENT(b,1);
|
164 |
|
|
ret_extent = GFC_DESCRIPTOR_EXTENT(retarray,1);
|
165 |
|
|
if (arg_extent != ret_extent)
|
166 |
|
|
runtime_error ("Incorrect extent in return array in"
|
167 |
|
|
" MATMUL intrinsic for dimension 2:"
|
168 |
|
|
" is %ld, should be %ld",
|
169 |
|
|
(long int) ret_extent, (long int) arg_extent);
|
170 |
|
|
}
|
171 |
|
|
}
|
172 |
|
|
'
|
173 |
|
|
sinclude(`matmul_asm_'rtype_code`.m4')dnl
|
174 |
|
|
`
|
175 |
|
|
if (GFC_DESCRIPTOR_RANK (retarray) == 1)
|
176 |
|
|
{
|
177 |
|
|
/* One-dimensional result may be addressed in the code below
|
178 |
|
|
either as a row or a column matrix. We want both cases to
|
179 |
|
|
work. */
|
180 |
|
|
rxstride = rystride = GFC_DESCRIPTOR_STRIDE(retarray,0);
|
181 |
|
|
}
|
182 |
|
|
else
|
183 |
|
|
{
|
184 |
|
|
rxstride = GFC_DESCRIPTOR_STRIDE(retarray,0);
|
185 |
|
|
rystride = GFC_DESCRIPTOR_STRIDE(retarray,1);
|
186 |
|
|
}
|
187 |
|
|
|
188 |
|
|
|
189 |
|
|
if (GFC_DESCRIPTOR_RANK (a) == 1)
|
190 |
|
|
{
|
191 |
|
|
/* Treat it as a a row matrix A[1,count]. */
|
192 |
|
|
axstride = GFC_DESCRIPTOR_STRIDE(a,0);
|
193 |
|
|
aystride = 1;
|
194 |
|
|
|
195 |
|
|
xcount = 1;
|
196 |
|
|
count = GFC_DESCRIPTOR_EXTENT(a,0);
|
197 |
|
|
}
|
198 |
|
|
else
|
199 |
|
|
{
|
200 |
|
|
axstride = GFC_DESCRIPTOR_STRIDE(a,0);
|
201 |
|
|
aystride = GFC_DESCRIPTOR_STRIDE(a,1);
|
202 |
|
|
|
203 |
|
|
count = GFC_DESCRIPTOR_EXTENT(a,1);
|
204 |
|
|
xcount = GFC_DESCRIPTOR_EXTENT(a,0);
|
205 |
|
|
}
|
206 |
|
|
|
207 |
|
|
if (count != GFC_DESCRIPTOR_EXTENT(b,0))
|
208 |
|
|
{
|
209 |
|
|
if (count > 0 || GFC_DESCRIPTOR_EXTENT(b,0) > 0)
|
210 |
|
|
runtime_error ("dimension of array B incorrect in MATMUL intrinsic");
|
211 |
|
|
}
|
212 |
|
|
|
213 |
|
|
if (GFC_DESCRIPTOR_RANK (b) == 1)
|
214 |
|
|
{
|
215 |
|
|
/* Treat it as a column matrix B[count,1] */
|
216 |
|
|
bxstride = GFC_DESCRIPTOR_STRIDE(b,0);
|
217 |
|
|
|
218 |
|
|
/* bystride should never be used for 1-dimensional b.
|
219 |
|
|
in case it is we want it to cause a segfault, rather than
|
220 |
|
|
an incorrect result. */
|
221 |
|
|
bystride = 0xDEADBEEF;
|
222 |
|
|
ycount = 1;
|
223 |
|
|
}
|
224 |
|
|
else
|
225 |
|
|
{
|
226 |
|
|
bxstride = GFC_DESCRIPTOR_STRIDE(b,0);
|
227 |
|
|
bystride = GFC_DESCRIPTOR_STRIDE(b,1);
|
228 |
|
|
ycount = GFC_DESCRIPTOR_EXTENT(b,1);
|
229 |
|
|
}
|
230 |
|
|
|
231 |
|
|
abase = a->data;
|
232 |
|
|
bbase = b->data;
|
233 |
|
|
dest = retarray->data;
|
234 |
|
|
|
235 |
|
|
|
236 |
|
|
/* Now that everything is set up, we''`re performing the multiplication
|
237 |
|
|
itself. */
|
238 |
|
|
|
239 |
|
|
#define POW3(x) (((float) (x)) * ((float) (x)) * ((float) (x)))
|
240 |
|
|
|
241 |
|
|
if (try_blas && rxstride == 1 && (axstride == 1 || aystride == 1)
|
242 |
|
|
&& (bxstride == 1 || bystride == 1)
|
243 |
|
|
&& (((float) xcount) * ((float) ycount) * ((float) count)
|
244 |
|
|
> POW3(blas_limit)))
|
245 |
|
|
{
|
246 |
|
|
const int m = xcount, n = ycount, k = count, ldc = rystride;
|
247 |
|
|
const 'rtype_name` one = 1, zero = 0;
|
248 |
|
|
const int lda = (axstride == 1) ? aystride : axstride,
|
249 |
|
|
ldb = (bxstride == 1) ? bystride : bxstride;
|
250 |
|
|
|
251 |
|
|
if (lda > 0 && ldb > 0 && ldc > 0 && m > 1 && n > 1 && k > 1)
|
252 |
|
|
{
|
253 |
|
|
assert (gemm != NULL);
|
254 |
|
|
gemm (axstride == 1 ? "N" : "T", bxstride == 1 ? "N" : "T", &m, &n, &k,
|
255 |
|
|
&one, abase, &lda, bbase, &ldb, &zero, dest, &ldc, 1, 1);
|
256 |
|
|
return;
|
257 |
|
|
}
|
258 |
|
|
}
|
259 |
|
|
|
260 |
|
|
if (rxstride == 1 && axstride == 1 && bxstride == 1)
|
261 |
|
|
{
|
262 |
|
|
const 'rtype_name` * restrict bbase_y;
|
263 |
|
|
'rtype_name` * restrict dest_y;
|
264 |
|
|
const 'rtype_name` * restrict abase_n;
|
265 |
|
|
'rtype_name` bbase_yn;
|
266 |
|
|
|
267 |
|
|
if (rystride == xcount)
|
268 |
|
|
memset (dest, 0, (sizeof ('rtype_name`) * xcount * ycount));
|
269 |
|
|
else
|
270 |
|
|
{
|
271 |
|
|
for (y = 0; y < ycount; y++)
|
272 |
|
|
for (x = 0; x < xcount; x++)
|
273 |
|
|
dest[x + y*rystride] = ('rtype_name`)0;
|
274 |
|
|
}
|
275 |
|
|
|
276 |
|
|
for (y = 0; y < ycount; y++)
|
277 |
|
|
{
|
278 |
|
|
bbase_y = bbase + y*bystride;
|
279 |
|
|
dest_y = dest + y*rystride;
|
280 |
|
|
for (n = 0; n < count; n++)
|
281 |
|
|
{
|
282 |
|
|
abase_n = abase + n*aystride;
|
283 |
|
|
bbase_yn = bbase_y[n];
|
284 |
|
|
for (x = 0; x < xcount; x++)
|
285 |
|
|
{
|
286 |
|
|
dest_y[x] += abase_n[x] * bbase_yn;
|
287 |
|
|
}
|
288 |
|
|
}
|
289 |
|
|
}
|
290 |
|
|
}
|
291 |
|
|
else if (rxstride == 1 && aystride == 1 && bxstride == 1)
|
292 |
|
|
{
|
293 |
|
|
if (GFC_DESCRIPTOR_RANK (a) != 1)
|
294 |
|
|
{
|
295 |
|
|
const 'rtype_name` *restrict abase_x;
|
296 |
|
|
const 'rtype_name` *restrict bbase_y;
|
297 |
|
|
'rtype_name` *restrict dest_y;
|
298 |
|
|
'rtype_name` s;
|
299 |
|
|
|
300 |
|
|
for (y = 0; y < ycount; y++)
|
301 |
|
|
{
|
302 |
|
|
bbase_y = &bbase[y*bystride];
|
303 |
|
|
dest_y = &dest[y*rystride];
|
304 |
|
|
for (x = 0; x < xcount; x++)
|
305 |
|
|
{
|
306 |
|
|
abase_x = &abase[x*axstride];
|
307 |
|
|
s = ('rtype_name`) 0;
|
308 |
|
|
for (n = 0; n < count; n++)
|
309 |
|
|
s += abase_x[n] * bbase_y[n];
|
310 |
|
|
dest_y[x] = s;
|
311 |
|
|
}
|
312 |
|
|
}
|
313 |
|
|
}
|
314 |
|
|
else
|
315 |
|
|
{
|
316 |
|
|
const 'rtype_name` *restrict bbase_y;
|
317 |
|
|
'rtype_name` s;
|
318 |
|
|
|
319 |
|
|
for (y = 0; y < ycount; y++)
|
320 |
|
|
{
|
321 |
|
|
bbase_y = &bbase[y*bystride];
|
322 |
|
|
s = ('rtype_name`) 0;
|
323 |
|
|
for (n = 0; n < count; n++)
|
324 |
|
|
s += abase[n*axstride] * bbase_y[n];
|
325 |
|
|
dest[y*rystride] = s;
|
326 |
|
|
}
|
327 |
|
|
}
|
328 |
|
|
}
|
329 |
|
|
else if (axstride < aystride)
|
330 |
|
|
{
|
331 |
|
|
for (y = 0; y < ycount; y++)
|
332 |
|
|
for (x = 0; x < xcount; x++)
|
333 |
|
|
dest[x*rxstride + y*rystride] = ('rtype_name`)0;
|
334 |
|
|
|
335 |
|
|
for (y = 0; y < ycount; y++)
|
336 |
|
|
for (n = 0; n < count; n++)
|
337 |
|
|
for (x = 0; x < xcount; x++)
|
338 |
|
|
/* dest[x,y] += a[x,n] * b[n,y] */
|
339 |
|
|
dest[x*rxstride + y*rystride] += abase[x*axstride + n*aystride] * bbase[n*bxstride + y*bystride];
|
340 |
|
|
}
|
341 |
|
|
else if (GFC_DESCRIPTOR_RANK (a) == 1)
|
342 |
|
|
{
|
343 |
|
|
const 'rtype_name` *restrict bbase_y;
|
344 |
|
|
'rtype_name` s;
|
345 |
|
|
|
346 |
|
|
for (y = 0; y < ycount; y++)
|
347 |
|
|
{
|
348 |
|
|
bbase_y = &bbase[y*bystride];
|
349 |
|
|
s = ('rtype_name`) 0;
|
350 |
|
|
for (n = 0; n < count; n++)
|
351 |
|
|
s += abase[n*axstride] * bbase_y[n*bxstride];
|
352 |
|
|
dest[y*rxstride] = s;
|
353 |
|
|
}
|
354 |
|
|
}
|
355 |
|
|
else
|
356 |
|
|
{
|
357 |
|
|
const 'rtype_name` *restrict abase_x;
|
358 |
|
|
const 'rtype_name` *restrict bbase_y;
|
359 |
|
|
'rtype_name` *restrict dest_y;
|
360 |
|
|
'rtype_name` s;
|
361 |
|
|
|
362 |
|
|
for (y = 0; y < ycount; y++)
|
363 |
|
|
{
|
364 |
|
|
bbase_y = &bbase[y*bystride];
|
365 |
|
|
dest_y = &dest[y*rystride];
|
366 |
|
|
for (x = 0; x < xcount; x++)
|
367 |
|
|
{
|
368 |
|
|
abase_x = &abase[x*axstride];
|
369 |
|
|
s = ('rtype_name`) 0;
|
370 |
|
|
for (n = 0; n < count; n++)
|
371 |
|
|
s += abase_x[n*aystride] * bbase_y[n*bxstride];
|
372 |
|
|
dest_y[x*rxstride] = s;
|
373 |
|
|
}
|
374 |
|
|
}
|
375 |
|
|
}
|
376 |
|
|
}
|
377 |
|
|
|
378 |
|
|
#endif'
|