1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
|
dnl
dnl
dnl @author: Michele Martone
dnl
dnl The generated code will expand from here
dnl Take care of compiling this code without loop unrolling optimizations (-fno-unroll-loops, or -ON with N<=2 on gcc)
dnl
include(`rsb_misc.m4')dnl
include(`do_unroll.m4')dnl
dnl
dnl
dnl
dnl FIXME : THERE ARE NO TRANSPOSED KERNELS
dnl
dnl
dnl
dnl RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_NAME(mtype,matrix_storage,transposition,symmetry,b_rows,b_columns,unrolling,mop) dnl -----------------------------------------------------------------------------------------------
dnl
define(`RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_NAME',`dnl
dnl
dnl
pushdef(`mtype',$1)dnl
pushdef(`matrix_storage',$2)dnl
pushdef(`transposition',$3)dnl
pushdef(`symmetry',$4)dnl
pushdef(`b_rows',$5)dnl block rows
pushdef(`b_columns',$6)dnl block columns
pushdef(`itype',`rsb_coo_idx_t')dnl integer type (for indices)
pushdef(`unrolling',$7)dnl
pushdef(`mop',$8)dnl
pushdef(`citype',$9)dnl
dnl
dnl pushdef(`diagonal',ifelse($10,`',`e',diagonal,uplo))dnl FIXME: for now this prameter is optional
pushdef(`diagonal',$10)dnl FIXME: new
pushdef(`uplo',$11)dnl
dnl
RSB_M4_PREFIX`'matrix_storage`_'mop`_'RSB_M4_TYPE_CODE(mtype)`_'dnl
RSB_M4_MATRIX_INDEX_COORDINATE_TYPE_CHARCODE(citype)`_'dnl
ifelse(matrix_storage,`fixed_block',`_r'b_rows`_c'b_columns)`'dnl
`_t'touppercase(RSB_M4_TRANSPOSITION_CODE(transposition))`'dnl
ifelse(RSB_M4_IS_FORMAT_BCSS(matrix_storage),1,`_r'b_rows`_c'b_columns)`'dnl
ifelse(RSB_M4_IS_FORMAT_BCOO(matrix_storage),1,`_r'b_rows`_c'b_columns)`'dnl
dnl ifelse(matrix_storage,`VBR',`_r'b_rows`_c'b_columns)`'dnl
dnl ifelse(matrix_storage,`VBC',`_r'b_rows`_c'b_columns)`'dnl
`_u'unrolling`_s'touppercase(symmetry)`'dnl
`_d'touppercase(diagonal)`'dnl
`_u'touppercase(uplo)`'dnl
popdef(`uplo')dnl
popdef(`citype')dnl
popdef(`mop')dnl
popdef(`mtype')dnl
popdef(`matrix_storage')dnl
popdef(`symmetry')dnl
popdef(`transposition')dnl
popdef(`b_rows')dnl
popdef(`b_columns')dnl
popdef(`itype')dnl
popdef(`unrolling')dnl
popdef(`diagonal')dnl
')dnl
dnl
dnl
dnl
dnl RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_ARGS(mtype,matrix_storage,b_rows,b_columns,unrolling,mop)
dnl -----------------------------------------------------------------------------------------------
dnl
define(`RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_ARGS',`dnl
dnl
dnl
pushdef(`mtype',$1)dnl
pushdef(`matrix_storage',$2)dnl
pushdef(`b_rows',$3)dnl block rows
pushdef(`b_columns',$4)dnl block columns
pushdef(`itype',`rsb_coo_idx_t')dnl integer type (for indices)
pushdef(`unrolling',$5)dnl
pushdef(`mop',$6)dnl
pushdef(`matrix_structs',`const itype Mdim, const itype mdim, const rsb_nnz_idx_t * RSB_M4_RESTRICT bindx, const rsb_nnz_idx_t * RSB_M4_RESTRICT bpntr, const rsb_nnz_idx_t *RSB_M4_RESTRICT indptr, const rsb_coo_idx_t * RSB_M4_RESTRICT rpntr, const rsb_coo_idx_t * RSB_M4_RESTRICT cpntr, const rsb_coo_idx_t dummy_br, const rsb_coo_idx_t dummy_bc')dnl
(
ifelse(mop,`scale',`dnl
mtype * VA,
matrix_structs,
const mtype *scale_factors
')dnl
ifelse(RSB_M4_MEMBER(mop,`spmv_uauz',`spmv_uaua',`spmv_unua'),1,`dnl
const mtype * VA, const mtype * rhs, mtype * out,
matrix_structs
')dnl
ifelse(mop,`spmm_az',`dnl
const mtype * VA, const mtype * rhs, mtype * out,
matrix_structs,
const itype bstride, const itype cstride, const itype nrhs
')dnl
ifelse(mop,`spmv_uxux',`dnl
const mtype * VA, const mtype * rhs, mtype * out,
matrix_structs,
const mtype * alphap, const mtype * betap
')dnl
ifelse(mop,`infty_norm',`dnl
ifelse(matrix_storage,`_fixed_block',`dnl
const mtype * VA, mtype * local_row_sums,
matrix_structs
',`dnl
const mtype * VA, mtype * global_row_sums,
matrix_structs
')dnl
')dnl
ifelse(mop,`negation',`dnl
ifelse(matrix_storage,`_fixed_block',`dnl
mtype * VA,
matrix_structs
',`dnl
mtype * VA,
matrix_structs
')dnl
')dnl
)dnl
popdef(`matrix_structs')dnl
popdef(`mop')dnl
popdef(`mtype')dnl
popdef(`matrix_storage')dnl
popdef(`b_rows')dnl
popdef(`b_columns')dnl
popdef(`itype')dnl
popdef(`unrolling')dnl
')dnl
dnl
dnl
dnl
dnl
dnl RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_PROTOTYPE(mtype,matrix_storage,transposition,b_rows,b_columns,unrolling,mop)
dnl ----------------------------------------------------------------------------------------------------
dnl
define(`RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_PROTOTYPE',`dnl
dnl
dnl
pushdef(`mtype',$1)dnl
pushdef(`matrix_storage',$2)dnl
pushdef(`transposition',$3)dnl
pushdef(`b_rows',$4)dnl block rows
pushdef(`b_columns',$5)dnl block columns
pushdef(`itype',`rsb_coo_idx_t')dnl integer type (for indices)
pushdef(`unrolling',$6)dnl
pushdef(`mop',$7)dnl
rsb_err_t RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_NAME(mtype,matrix_storage,transposition,symmetry,b_rows,b_columns,unrolling,mop)dnl
RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_ARGS(mtype,matrix_storage,b_rows,b_columns,unrolling,mop)dnl
popdef(`mop')dnl
popdef(`mtype')dnl
popdef(`matrix_storage')dnl
popdef(`transposition')dnl
popdef(`b_rows')dnl
popdef(`b_columns')dnl
popdef(`itype')dnl
popdef(`unrolling')dnl
')dnl
dnl
dnl
dnl
dnl RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_BODY(mtype,matrix_storage,b_rows,b_columns,unrolling,mop)
dnl -----------------------------------------------------------------------------------------------
dnl
define(`RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_BODY',`dnl
dnl
dnl
pushdef(`mtype',$1)dnl
pushdef(`matrix_storage',$2)dnl
pushdef(`b_rows',$3)dnl block rows
pushdef(`b_columns',$4)dnl block columns
pushdef(`itype',`rsb_coo_idx_t')dnl integer type (for indices)
pushdef(`unrolling',$5)dnl
pushdef(`mop',$6)dnl
{
ifelse(RSB_M4_IS_FORMAT_COLUMN_MAJOR(matrix_storage),1,`dnl
pushdef(`out_dim',rpntr[Mdim])dnl
pushdef(`mi',`i')dnl
pushdef(`Mi',`j')dnl
')dnl
ifelse(RSB_M4_IS_FORMAT_ROW_MAJOR(matrix_storage),1,`dnl
pushdef(`out_dim',rpntr[mdim])dnl
pushdef(`mi',`j')dnl
pushdef(`Mi',`i')dnl
')dnl
dnl
/**
* \ingroup rsb_doc_kernels
ifelse(matrix_storage,`VBR',`dnl
* This code is intended for a pure VBR partitioned matrix.
* It does not dispatch the kernel function for each block,
* but employ explicitly inlined kernels.
')dnl
*
* \return \rsb_errval_inp_param_msg
*/
register rsb_coo_idx_t i,j;
register rsb_nnz_idx_t k;
rsb_coo_idx_t columns;
rsb_coo_idx_t rows ;
ifelse(mop,`spmv_uxux',`dnl
/* this is slow, however this is column based scanning. FIXME : optimize spmv_uxux */
rsb__cblas_Xscal(RSB_M4_NUMERICAL_TYPE_PREPROCESSOR_SYMBOL(type), out_dim, betap, out, 1);/* we scale the destination vector */
')dnl
ifelse(mop,`spmv_uauz',`dnl
rsb__cblas_Xscal(RSB_M4_NUMERICAL_TYPE_PREPROCESSOR_SYMBOL(type), out_dim, NULL, out, 1);// FIXME: stride is bad
')dnl
for(Mi=0;Mi<Mdim;++Mi)
{
/* Mi is the working block row */
if(bpntr[Mi]==bpntr[Mi+1])continue;/* empty block row (or column) */
for(k=bpntr[Mi];k<bpntr[Mi+1];++k) /* k is the index of the block */
{
/* mi is the working block column (or row) */
mi=bindx[k];
ifelse(matrix_storage,`VBR',`dnl
columns=cpntr[j+1]-cpntr[j];
rows =rpntr[i+1]-rpntr[i];
')dnl
ifelse(1,0,`dnl
ifelse(RSB_M4_MEMBER(matrix_storage,`VBR',`VBC'),1,`dnl
/* matrix_storage ! mop */
DOUBLE_LINEAR_KERNEL_SEARCH_MACRO_IDENTIFIER(mop,mtype,unrolling)(rows,columns)
ifelse(mop,`scale',`dnl
(
VA+indptr[(k)],
scale_factors+rpntr[i],
rows, columns
);
')dnl
ifelse(mop,`spmv_uauz',`dnl
(
VA+indptr[(k)],
rhs+cpntr[j],
out+rpntr[i],
rows, columns
);
')dnl
ifelse(mop,`spmv_unua',`dnl
(
VA+indptr[(k)],
rhs+cpntr[j],
out+rpntr[i],
rows, columns
);
')dnl
ifelse(mop,`spmv_uaua',`dnl
(
VA+indptr[(k)],
rhs+cpntr[j],
out+rpntr[i],
rows, columns
);
')dnl
ifelse(mop,`spmm_az',`dnl
(
VA+indptr[(k)],
rhs+cpntr[j],
out+rpntr[i],
rows, columns,
bstride, cstride, nrhs
);
')dnl
ifelse(mop,`spmv_uxux',`dnl
(
VA+indptr[(k)],
rhs+cpntr[j],
out+rpntr[i],
rows, columns,
alphap, betap
);
')dnl
ifelse(mop,`infty_norm',`dnl
(
VA+indptr[(k)],
global_row_sums+cpntr[j],
rows, columns
);
')dnl
ifelse(mop,`negation',`dnl
(
VA+indptr[(k)],
rows, columns
);
')dnl
')dnl
')dnl
ifelse(RSB_M4_MEMBER(matrix_storage,`VBR',`VBC'),1,`dnl
ifelse(mop,`scale',`dnl
mtype *a = VA+indptr[(k)];
const mtype *d = scale_factors+rpntr[i];
')dnl
ifelse(RSB_M4_MEMBER(mop,`spmv_uaua',`spmv_unua',`spmv_uauz',`spmm_az',`spmv_uxux'),1,`dnl
const mtype *a = VA+indptr[(k)];
const mtype *b = rhs+cpntr[j];
mtype *c = out+rpntr[i];
')dnl
ifelse(mop,`infty_norm',`dnl
const mtype *a = VA+indptr[(k)];
mtype *local_row_sums = global_row_sums+rpntr[i];
')dnl
ifelse(mop,`negation',`dnl
mtype *a = VA+indptr[(k)];
')dnl
columns=cpntr[j+1]-cpntr[j];
rows =rpntr[i+1]-rpntr[i];
/* we jump to the right point */
dnl
ifelse(1,0,`
#if 0
/* it will not happen */
#ifdef RSB_WANT_BLOCK_TRAILING_STRUCT /* EXPERIMENTAL */
ifelse(mop,`scale',`dnl
a = (mtype*) (((char*)a) + (RSB_BLOCK_EXTRA_BYTES)*(k+1)) ;
')dnl
ifelse(mop,`spmv_uauz',`dnl
a = (const mtype*) (((const char*)a) + (RSB_BLOCK_EXTRA_BYTES)*(k+1)) ;
')dnl
ifelse(mop,`spmm_az',`dnl
a = (const mtype*) (((const char*)a) + (RSB_BLOCK_EXTRA_BYTES)*(k+1)) ;
')dnl
ifelse(mop,`spmv_uxux',`dnl
a = (const mtype*) (((const char*)a) + (RSB_BLOCK_EXTRA_BYTES)*(k+1)) ;
')dnl
ifelse(mop,`infty_norm',`dnl
a = (const mtype*) (((const char*)a) + (RSB_BLOCK_EXTRA_BYTES)*(k+1)) ;
')dnl
ifelse(mop,`negation',`dnl
a = (mtype*) (((char*)a) + (RSB_BLOCK_EXTRA_BYTES)*(k+1)) ;
')dnl
dnl
#endif /* RSB_WANT_BLOCK_TRAILING_STRUCT */
#endif /* 0 */
')dnl
switch(rows)
{
foreach(`rowsu',RSB_M4_ROWS_UNROLL,`dnl
case rowsu:
switch(columns)
{
foreach(`colsu',RSB_M4_COLUMNS_UNROLL,`dnl
case colsu:
dnl goto RSB_M4_KERNEL_FUNCTION_NAME(mtype,rowsu,colsu,`',mop,transposition,RSB_M4_DEFAULT_COORDINATE_INDEX_TYPE);
{RSB_M4_KERNEL_FUNCTION_BODY(`row',`rows',rowsu,`column',`columns',colsu,mtype,,mop,`')}
break;
')dnl
default: goto RSB_M4_KERNEL_FUNCTION_NAME(mtype,RSB_M4_ROWS_FALLBACK_UNROLL,RSB_M4_COLUMNS_FALLBACK_UNROLL,`l',mop,transposition,RSB_M4_DEFAULT_COORDINATE_INDEX_TYPE);
}
break;
')dnl
default: goto RSB_M4_KERNEL_FUNCTION_NAME(mtype,RSB_M4_ROWS_FALLBACK_UNROLL,RSB_M4_COLUMNS_FALLBACK_UNROLL,`l',mop,transposition,RSB_M4_DEFAULT_COORDINATE_INDEX_TYPE);
RSB_M4_KERNEL_FUNCTION_NAME(mtype,RSB_M4_ROWS_FALLBACK_UNROLL,RSB_M4_COLUMNS_FALLBACK_UNROLL,`l',mop,transposition,RSB_M4_DEFAULT_COORDINATE_INDEX_TYPE):
{RSB_M4_KERNEL_FUNCTION_BODY(`row',`rows',RSB_M4_ROWS_FALLBACK_UNROLL,`column',`columns',RSB_M4_COLUMNS_FALLBACK_UNROLL,mtype,,mop,`l')}
};
')dnl end VBR
}
}
popdef(`mi')dnl
popdef(`Mi')dnl
dnl
dnl
dnl
popdef(`out_dim')dnl
return RSB_ERR_NO_ERROR;
}
popdef(`mop')dnl
popdef(`mtype')dnl
popdef(`matrix_storage')dnl
popdef(`b_rows')dnl
popdef(`b_columns')dnl
popdef(`itype')dnl
popdef(`unrolling')dnl
')dnl
dnl
dnl
dnl
dnl RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION(mtype,matrix_storage,transposition,b_rows,b_columns,unrolling,mop)
dnl ------------------------------------------------------------------------------------------
dnl
dnl These functions will perform their operations on fixed block matrices.
dnl
define(`RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION',`dnl
dnl
dnl
pushdef(`mtype',$1)dnl
pushdef(`matrix_storage',$2)dnl
pushdef(`transposition',$3)dnl
pushdef(`b_rows',$4)dnl block rows
pushdef(`b_columns',$5)dnl block columns
pushdef(`itype',`rsb_coo_idx_t')dnl integer type (for indices)
pushdef(`unrolling',$6)dnl
pushdef(`mop',$7)dnl
ifelse(matrix_storage,`fixed_block',dnl
`/* This code is intended for a purely blocked matrix. */',dnl
`/* This code is intended for a pure VBR partitioned matrix. */')
RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_PROTOTYPE(mtype,matrix_storage,transposition,b_rows,b_columns,unrolling,mop)dnl
ifdef(`ONLY_WANT_HEADERS',`;
',`
RSB_M4_KERNEL_DIRECT_DISPATCHER_FUNCTION_BODY(mtype,matrix_storage,b_rows,b_columns,unrolling,mop)
')dnl
popdef(`mop')dnl
popdef(`matrix_storage')dnl
popdef(`transposition')dnl
popdef(`mtype')dnl
popdef(`itype')dnl
popdef(`unrolling')dnl
')dnl
dnl
dnl
dnl
|
