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
|
//------------------------------------------------------------------------------
// GB_convert_b2s: construct triplets or CSC/CSR from bitmap
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2025, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Constructs a triplet or CSC/CSR form (in Cp, Ci, Cj, and Cx_new) from the
// bitmap input matrix A. If A is iso or Cx_new is NULL then no values are
// extracted. The iso case is handled by the caller.
// Ci, Cj, and Cx_new may be NULL.
// FUTURE: make a separate function for constructing triplets
#include "GB.h"
#include "jitifyer/GB_stringify.h"
#include "unaryop/GB_unop.h"
#define GB_FREE_ALL GB_FREE_MEMORY (&W, W_size) ;
GrB_Info GB_convert_b2s // extract CSC/CSR or triplets from bitmap
(
// outputs:
void *Cp, // vector pointers for CSC/CSR form
void *Ci, // indices for CSC/CSR or triplet form
void *Cj, // vector indices for triplet form
void *Cx_new, // values for CSC/CSR or triplet form
int64_t *cnvec_nonempty, // # of non-empty vectors
// inputs: not modified
const bool Cp_is_32, // if true, Cp is uint32_t; otherwise uint64_t
const bool Ci_is_32, // if true, Ci is uint32_t; otherwise uint64_t
const bool Cj_is_32, // if true, Cj is uint32_t; otherwise uint64_t
const GrB_Type ctype, // type of Cx
const GrB_Matrix A, // matrix to extract; not modified
GB_Werk Werk
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
ASSERT (GB_IS_BITMAP (A)) ;
ASSERT (Cp != NULL) ; // must be provided on input, size avdim+1
ASSERT_MATRIX_OK (A, "A for b2s", GB0) ;
ASSERT_TYPE_OK (ctype, "ctype for b2s", GB0) ;
ASSERT ((Cp_is_32 && A->nvals < UINT32_MAX) || !Cp_is_32) ;
//--------------------------------------------------------------------------
// get inputs and determine tasks
//--------------------------------------------------------------------------
void *W = NULL ; size_t W_size = 0 ;
GB_IDECL (W , , u) ;
GB_IDECL (Cp, , u) ; GB_IPTR (Cp, Cp_is_32) ;
GB_IDECL (Ci, , u) ; GB_IPTR (Ci, Ci_is_32) ;
GB_IDECL (Cj, , u) ; GB_IPTR (Cj, Cj_is_32) ;
size_t psize = Cp_is_32 ? sizeof (uint32_t) : sizeof (uint64_t) ;
const int64_t avdim = A->vdim ;
const int64_t avlen = A->vlen ;
const size_t asize = A->type->size ;
const int8_t *restrict Ab = A->b ;
int nthreads_max = GB_Context_nthreads_max ( ) ;
double chunk = GB_Context_chunk ( ) ;
int nthreads = GB_nthreads (avlen*avdim, chunk, nthreads_max) ;
bool by_vector = (nthreads <= avdim) ;
//--------------------------------------------------------------------------
// count the entries in each vector
//--------------------------------------------------------------------------
if (by_vector)
{
//----------------------------------------------------------------------
// compute all vectors in parallel (no workspace)
//----------------------------------------------------------------------
int64_t j ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (j = 0 ; j < avdim ; j++)
{
// ajnz = nnz (A (:,j))
uint64_t ajnz = 0 ;
int64_t pA_start = j * avlen ;
for (int64_t i = 0 ; i < avlen ; i++)
{
// see if A(i,j) is present in the bitmap
int64_t p = i + pA_start ;
ajnz += Ab [p] ;
ASSERT (Ab [p] == 0 || Ab [p] == 1) ;
}
// Cp [j] = ajnz ;
GB_ISET (Cp, j, ajnz) ;
}
}
else
{
//----------------------------------------------------------------------
// compute blocks of rows in parallel
//----------------------------------------------------------------------
// allocate one row of W per thread, each row of length avdim
W = GB_MALLOC_MEMORY (nthreads * avdim, psize, &W_size) ;
if (W == NULL)
{
// out of memory
return (GrB_OUT_OF_MEMORY) ;
}
GB_IPTR (W, Cp_is_32) ;
//----------------------------------------------------------------------
// count each block
//----------------------------------------------------------------------
int tid ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (tid = 0 ; tid < nthreads ; tid++)
{
uint32_t *restrict Wtask32 = Cp_is_32 ? (W32 + tid * avdim) : NULL ;
uint64_t *restrict Wtask64 = Cp_is_32 ? NULL : (W64 + tid * avdim) ;
int64_t istart, iend ;
GB_PARTITION (istart, iend, avlen, tid, nthreads) ;
for (int64_t j = 0 ; j < avdim ; j++)
{
// ajnz = nnz (A (istart:iend-1,j))
uint64_t ajnz = 0 ;
int64_t pA_start = j * avlen ;
for (int64_t i = istart ; i < iend ; i++)
{
// see if A(i,j) is present in the bitmap
int64_t p = i + pA_start ;
ajnz += Ab [p] ;
ASSERT (Ab [p] == 0 || Ab [p] == 1) ;
}
// Wtask [j] = ajnz ;
GB_ISET (Wtask, j, ajnz) ;
}
}
//----------------------------------------------------------------------
// cumulative sum to compute nnz(A(:,j)) for each vector j
//----------------------------------------------------------------------
int64_t j ;
if (Cp_is_32)
{
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (j = 0 ; j < avdim ; j++)
{
uint32_t ajnz = 0 ;
for (int tid = 0 ; tid < nthreads ; tid++)
{
uint32_t *restrict Wtask32 = W32 + tid * avdim ;
uint32_t c = Wtask32 [j] ;
Wtask32 [j] = ajnz ;
ajnz += c ;
}
Cp32 [j] = ajnz ;
}
}
else
{
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (j = 0 ; j < avdim ; j++)
{
uint64_t ajnz = 0 ;
for (int tid = 0 ; tid < nthreads ; tid++)
{
uint64_t *restrict Wtask64 = W64 + tid * avdim ;
uint64_t c = Wtask64 [j] ;
Wtask64 [j] = ajnz ;
ajnz += c ;
}
Cp64 [j] = ajnz ;
}
}
}
//--------------------------------------------------------------------------
// cumulative sum of Cp
//--------------------------------------------------------------------------
// This cannot overflow if Cp is uint32_t, because in that case A->nvals
// is < UINT32_MAX (see assertion above).
int nth = GB_nthreads (avdim, chunk, nthreads_max) ;
GB_cumsum (Cp, Cp_is_32, avdim, cnvec_nonempty, nth, Werk) ;
ASSERT (GB_IGET (Cp, avdim) == A->nvals) ;
//--------------------------------------------------------------------------
// gather the pattern and values from the bitmap
//--------------------------------------------------------------------------
GrB_Info info = GrB_NO_VALUE ;
if (Cx_new == NULL || A->x == NULL || A->iso)
{
//----------------------------------------------------------------------
// via the symbolic kernel
//----------------------------------------------------------------------
#undef GB_COPY
#define GB_COPY(Cx,pC,Ax,pA)
#include "convert/template/GB_convert_b2s_template.c"
info = GrB_SUCCESS ;
}
else
{
//----------------------------------------------------------------------
// via an inline kernel for types of size 1, 2, 4, 8, or 16
//----------------------------------------------------------------------
if (ctype == A->type)
{
#undef GB_COPY
#define GB_COPY(Cx,pC,Ax,pA) Cx [pC] = Ax [pA] ;
#ifndef GBCOMPACT
GB_IF_FACTORY_KERNELS_ENABLED
{
switch (asize)
{
case GB_1BYTE : // uint8, int8, bool, or 1-byte user
#define GB_C_TYPE uint8_t
#define GB_A_TYPE uint8_t
#include "convert/template/GB_convert_b2s_template.c"
info = GrB_SUCCESS ;
break ;
case GB_2BYTE : // uint16, int16, or 2-byte user-defined
#define GB_C_TYPE uint16_t
#define GB_A_TYPE uint16_t
#include "convert/template/GB_convert_b2s_template.c"
info = GrB_SUCCESS ;
break ;
case GB_4BYTE : // uint32, int32, float, or 4-byte user
#define GB_C_TYPE uint32_t
#define GB_A_TYPE uint32_t
#include "convert/template/GB_convert_b2s_template.c"
info = GrB_SUCCESS ;
break ;
case GB_8BYTE : // uint64, int64, double, float complex,
// or 8-byte user defined
#define GB_C_TYPE uint64_t
#define GB_A_TYPE uint64_t
#include "convert/template/GB_convert_b2s_template.c"
info = GrB_SUCCESS ;
break ;
case GB_16BYTE : // double complex or 16-byte user-defined
#define GB_C_TYPE GB_blob16
#define GB_A_TYPE GB_blob16
#include "convert/template/GB_convert_b2s_template.c"
info = GrB_SUCCESS ;
break ;
default:;
}
}
#endif
}
//----------------------------------------------------------------------
// via the JIT kernel
//----------------------------------------------------------------------
if (info == GrB_NO_VALUE)
{
struct GB_UnaryOp_opaque op_header ;
GB_Operator op = GB_unop_identity (ctype, &op_header) ;
info = GB_convert_b2s_jit (Cp, Ci, Cj, Cx_new,
Cp_is_32, Ci_is_32, Cj_is_32, ctype, op, A, W, nthreads) ;
}
//----------------------------------------------------------------------
// via the generic kernel
//----------------------------------------------------------------------
if (info == GrB_NO_VALUE)
{
GB_Type_code ccode = ctype->code ;
GB_Type_code acode = A->type->code ;
const size_t csize = ctype->size ;
GB_cast_function cast_A_to_C = GB_cast_factory (ccode, acode) ;
#define GB_C_TYPE GB_void
#define GB_A_TYPE GB_void
#undef GB_COPY
#define GB_COPY(Cx,pC,Ax,pA) \
cast_A_to_C (Cx +(pC)*csize, Ax +(pA)*asize, asize)
#include "convert/template/GB_convert_b2s_template.c"
info = GrB_SUCCESS ;
}
}
//--------------------------------------------------------------------------
// free workspace return result
//--------------------------------------------------------------------------
GB_FREE_ALL ;
return (info) ;
}
|
