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
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
|
/* ************************************************************************
* Copyright (C) 2021-2024 Advanced Micro Devices, Inc. All rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*
* ************************************************************************ */
#include "testing.hpp"
#include <tuple>
template <typename I, typename J, typename A, typename B, typename C, typename T>
void testing_spmm_csr_bad_arg(const Arguments& arg)
{
static const size_t safe_size = 100;
// Create rocsparse handle
rocsparse_local_handle local_handle;
rocsparse_handle handle = local_handle;
J m = safe_size;
J n = safe_size;
J k = safe_size;
I nnz = safe_size;
const T* alpha = (const T*)0x4;
const T* beta = (const T*)0x4;
void* csr_val = (void*)0x4;
void* csr_row_ptr = (void*)0x4;
void* csr_col_ind = (void*)0x4;
void* dB = (void*)0x4;
void* dC = (void*)0x4;
rocsparse_operation trans_A = rocsparse_operation_none;
rocsparse_operation trans_B = rocsparse_operation_none;
rocsparse_index_base base = rocsparse_index_base_zero;
rocsparse_order order_B = rocsparse_order_column;
rocsparse_order order_C = rocsparse_order_column;
rocsparse_spmm_alg alg = rocsparse_spmm_alg_default;
rocsparse_spmm_stage stage = rocsparse_spmm_stage_compute;
rocsparse_indextype itype = get_indextype<I>();
rocsparse_indextype jtype = get_indextype<J>();
rocsparse_datatype atype = get_datatype<A>();
rocsparse_datatype btype = get_datatype<B>();
rocsparse_datatype ctype = get_datatype<C>();
rocsparse_datatype compute_type = get_datatype<T>();
// SpMM structures
rocsparse_local_spmat local_mat_A(
m, k, nnz, csr_row_ptr, csr_col_ind, csr_val, itype, jtype, base, atype);
rocsparse_local_dnmat local_mat_B(k, n, k, dB, btype, order_B);
rocsparse_local_dnmat local_mat_C(m, n, m, dC, ctype, order_C);
rocsparse_spmat_descr mat_A = local_mat_A;
rocsparse_dnmat_descr mat_B = local_mat_B;
rocsparse_dnmat_descr mat_C = local_mat_C;
int nargs_to_exclude = 2;
const int args_to_exclude[2] = {11, 12};
#define PARAMS \
handle, trans_A, trans_B, alpha, mat_A, mat_B, beta, mat_C, compute_type, alg, stage, \
buffer_size, temp_buffer
{
size_t* buffer_size = (size_t*)0x4;
void* temp_buffer = (void*)0x4;
select_bad_arg_analysis(rocsparse_spmm, nargs_to_exclude, args_to_exclude, PARAMS);
}
{
size_t* buffer_size = (size_t*)0x4;
void* temp_buffer = nullptr;
select_bad_arg_analysis(rocsparse_spmm, nargs_to_exclude, args_to_exclude, PARAMS);
}
{
size_t* buffer_size = nullptr;
void* temp_buffer = (void*)0x4;
select_bad_arg_analysis(rocsparse_spmm, nargs_to_exclude, args_to_exclude, PARAMS);
}
{
size_t* buffer_size = nullptr;
void* temp_buffer = nullptr;
select_bad_arg_analysis(rocsparse_spmm, nargs_to_exclude, args_to_exclude, PARAMS);
}
#undef PARAMS
}
template <typename I, typename J, typename A, typename B, typename C, typename T>
void testing_spmm_csr(const Arguments& arg)
{
J M = arg.M;
J N = arg.N;
J K = arg.K;
rocsparse_operation trans_A = arg.transA;
rocsparse_operation trans_B = arg.transB;
rocsparse_index_base base = arg.baseA;
rocsparse_spmm_alg alg = arg.spmm_alg;
rocsparse_order order_B = arg.orderB;
rocsparse_order order_C = arg.orderC;
rocsparse_int ld_multiplier_B = arg.ld_multiplier_B;
rocsparse_int ld_multiplier_C = arg.ld_multiplier_C;
T halpha = arg.get_alpha<T>();
T hbeta = arg.get_beta<T>();
// Index and data type
rocsparse_indextype itype = get_indextype<I>();
rocsparse_indextype jtype = get_indextype<J>();
rocsparse_datatype atype = get_datatype<A>();
rocsparse_datatype btype = get_datatype<B>();
rocsparse_datatype ctype = get_datatype<C>();
rocsparse_datatype ttype = get_datatype<T>();
// Create rocsparse handle
rocsparse_local_handle handle(arg);
// Allocate host memory for matrix
host_vector<I> hcsr_row_ptr;
host_vector<J> hcsr_col_ind;
host_vector<A> hcsr_val;
// Allocate host memory for matrix
rocsparse_matrix_factory<A, I, J> matrix_factory(arg);
I nnz_A;
matrix_factory.init_csr(hcsr_row_ptr,
hcsr_col_ind,
hcsr_val,
(trans_A == rocsparse_operation_none) ? M : K,
(trans_A == rocsparse_operation_none) ? K : M,
nnz_A,
base);
// Some matrix properties
J A_m = (trans_A == rocsparse_operation_none) ? M : K;
J A_n = (trans_A == rocsparse_operation_none) ? K : M;
J B_m = (trans_B == rocsparse_operation_none) ? K : N;
J B_n = (trans_B == rocsparse_operation_none) ? N : K;
J C_m = M;
J C_n = N;
int64_t ldb = (order_B == rocsparse_order_column)
? ((trans_B == rocsparse_operation_none) ? (int64_t(ld_multiplier_B) * K)
: (int64_t(ld_multiplier_B) * N))
: ((trans_B == rocsparse_operation_none) ? (int64_t(ld_multiplier_B) * N)
: (int64_t(ld_multiplier_B) * K));
int64_t ldc = (order_C == rocsparse_order_column) ? (int64_t(ld_multiplier_C) * M)
: (int64_t(ld_multiplier_C) * N);
int64_t nrowB = (order_B == rocsparse_order_column) ? ldb : B_m;
int64_t ncolB = (order_B == rocsparse_order_column) ? B_n : ldb;
int64_t nrowC = (order_C == rocsparse_order_column) ? ldc : C_m;
int64_t ncolC = (order_C == rocsparse_order_column) ? C_n : ldc;
int64_t nnz_B = nrowB * ncolB;
int64_t nnz_C = nrowC * ncolC;
// Allocate host memory for vectors
host_vector<B> hB(nnz_B);
host_vector<C> hC_1(nnz_C);
host_vector<C> hC_2(nnz_C);
host_vector<C> hC_gold(nnz_C);
// Initialize data on CPU
rocsparse_init<B>(hB, nnz_B, 1, 1);
rocsparse_init<C>(hC_1, nnz_C, 1, 1);
hC_2 = hC_1;
hC_gold = hC_1;
// Allocate device memory
device_vector<I> dcsr_row_ptr(A_m + 1);
device_vector<J> dcsr_col_ind(nnz_A);
device_vector<A> dcsr_val(nnz_A);
device_vector<B> dB(nnz_B);
device_vector<C> dC_1(nnz_C);
device_vector<C> dC_2(nnz_C);
device_vector<T> dalpha(1);
device_vector<T> dbeta(1);
// Copy data from CPU to device
CHECK_HIP_ERROR(
hipMemcpy(dcsr_row_ptr, hcsr_row_ptr.data(), sizeof(I) * (A_m + 1), hipMemcpyHostToDevice));
CHECK_HIP_ERROR(
hipMemcpy(dcsr_col_ind, hcsr_col_ind.data(), sizeof(J) * nnz_A, hipMemcpyHostToDevice));
CHECK_HIP_ERROR(hipMemcpy(dcsr_val, hcsr_val.data(), sizeof(A) * nnz_A, hipMemcpyHostToDevice));
CHECK_HIP_ERROR(hipMemcpy(dB, hB, sizeof(B) * nnz_B, hipMemcpyHostToDevice));
CHECK_HIP_ERROR(hipMemcpy(dC_1, hC_1, sizeof(C) * nnz_C, hipMemcpyHostToDevice));
CHECK_HIP_ERROR(hipMemcpy(dC_2, hC_2, sizeof(C) * nnz_C, hipMemcpyHostToDevice));
CHECK_HIP_ERROR(hipMemcpy(dalpha, &halpha, sizeof(T), hipMemcpyHostToDevice));
CHECK_HIP_ERROR(hipMemcpy(dbeta, &hbeta, sizeof(T), hipMemcpyHostToDevice));
// Create descriptors
rocsparse_local_spmat mat_A(
A_m, A_n, nnz_A, dcsr_row_ptr, dcsr_col_ind, dcsr_val, itype, jtype, base, atype);
ldb = std::max(int64_t(1), ldb);
ldc = std::max(int64_t(1), ldc);
rocsparse_local_dnmat mat_B(B_m, B_n, ldb, dB, btype, order_B);
rocsparse_local_dnmat mat_C1(C_m, C_n, ldc, dC_1, ctype, order_C);
rocsparse_local_dnmat mat_C2(C_m, C_n, ldc, dC_2, ctype, order_C);
// Query SpMM buffer
size_t buffer_size;
CHECK_ROCSPARSE_ERROR(rocsparse_spmm(handle,
trans_A,
trans_B,
&halpha,
mat_A,
mat_B,
&hbeta,
mat_C1,
ttype,
alg,
rocsparse_spmm_stage_buffer_size,
&buffer_size,
nullptr));
// Allocate buffer
void* dbuffer;
CHECK_HIP_ERROR(rocsparse_hipMalloc(&dbuffer, buffer_size));
CHECK_ROCSPARSE_ERROR(rocsparse_spmm(handle,
trans_A,
trans_B,
&halpha,
mat_A,
mat_B,
&hbeta,
mat_C1,
ttype,
alg,
rocsparse_spmm_stage_preprocess,
&buffer_size,
dbuffer));
if(arg.unit_check)
{
// SpMM
// Pointer mode host
CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_host));
CHECK_ROCSPARSE_ERROR(testing::rocsparse_spmm(handle,
trans_A,
trans_B,
&halpha,
mat_A,
mat_B,
&hbeta,
mat_C1,
ttype,
alg,
rocsparse_spmm_stage_compute,
&buffer_size,
dbuffer));
if(ROCSPARSE_REPRODUCIBILITY)
{
rocsparse_reproducibility::save("dC_1", dC_1);
}
// Pointer mode device
CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_device));
CHECK_ROCSPARSE_ERROR(testing::rocsparse_spmm(handle,
trans_A,
trans_B,
dalpha,
mat_A,
mat_B,
dbeta,
mat_C2,
ttype,
alg,
rocsparse_spmm_stage_compute,
&buffer_size,
dbuffer));
if(ROCSPARSE_REPRODUCIBILITY)
{
rocsparse_reproducibility::save("dC_2", dC_2);
}
// Copy output to host
CHECK_HIP_ERROR(hipMemcpy(hC_1, dC_1, sizeof(C) * nnz_C, hipMemcpyDeviceToHost));
CHECK_HIP_ERROR(hipMemcpy(hC_2, dC_2, sizeof(C) * nnz_C, hipMemcpyDeviceToHost));
// CPU csrmm
host_csrmm<T, I, J, A, B, C>(A_m,
N,
A_n,
trans_A,
trans_B,
halpha,
hcsr_row_ptr,
hcsr_col_ind,
hcsr_val,
hB,
ldb,
order_B,
hbeta,
hC_gold,
ldc,
order_C,
base,
false);
hC_gold.near_check(hC_1, get_near_check_tol<C>(arg));
hC_gold.near_check(hC_2, get_near_check_tol<C>(arg));
}
if(arg.timing)
{
int number_cold_calls = 2;
int number_hot_calls = arg.iters;
CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_host));
// Warm up
for(int iter = 0; iter < number_cold_calls; ++iter)
{
CHECK_ROCSPARSE_ERROR(rocsparse_spmm(handle,
trans_A,
trans_B,
&halpha,
mat_A,
mat_B,
&hbeta,
mat_C1,
ttype,
alg,
rocsparse_spmm_stage_compute,
&buffer_size,
dbuffer));
}
double gpu_time_used = get_time_us();
// Performance run
for(int iter = 0; iter < number_hot_calls; ++iter)
{
CHECK_ROCSPARSE_ERROR(rocsparse_spmm(handle,
trans_A,
trans_B,
&halpha,
mat_A,
mat_B,
&hbeta,
mat_C1,
ttype,
alg,
rocsparse_spmm_stage_compute,
&buffer_size,
dbuffer));
}
gpu_time_used = (get_time_us() - gpu_time_used) / number_hot_calls;
double gflop_count
= spmm_gflop_count(N, nnz_A, (I)C_m * (I)C_n, hbeta != static_cast<T>(0));
double gpu_gflops = get_gpu_gflops(gpu_time_used, gflop_count);
double gbyte_count = csrmm_gbyte_count<T>(
A_m, nnz_A, (I)B_m * (I)B_n, (I)C_m * (I)C_n, hbeta != static_cast<T>(0));
double gpu_gbyte = get_gpu_gbyte(gpu_time_used, gbyte_count);
display_timing_info(display_key_t::M,
M,
display_key_t::N,
N,
display_key_t::K,
K,
display_key_t::nnz_A,
nnz_A,
display_key_t::alpha,
halpha,
display_key_t::beta,
hbeta,
display_key_t::algorithm,
rocsparse_spmmalg2string(alg),
display_key_t::gflops,
gpu_gflops,
display_key_t::bandwidth,
gpu_gbyte,
display_key_t::time_ms,
get_gpu_time_msec(gpu_time_used));
}
CHECK_HIP_ERROR(rocsparse_hipFree(dbuffer));
}
#define INSTANTIATE(ITYPE, JTYPE, TTYPE) \
template void testing_spmm_csr_bad_arg<ITYPE, JTYPE, TTYPE, TTYPE, TTYPE, TTYPE>( \
const Arguments& arg); \
template void testing_spmm_csr<ITYPE, JTYPE, TTYPE, TTYPE, TTYPE, TTYPE>(const Arguments& arg)
#define INSTANTIATE_MIXED(ITYPE, JTYPE, ATYPE, XTYPE, YTYPE, TTYPE) \
template void testing_spmm_csr_bad_arg<ITYPE, JTYPE, ATYPE, XTYPE, YTYPE, TTYPE>( \
const Arguments& arg); \
template void testing_spmm_csr<ITYPE, JTYPE, ATYPE, XTYPE, YTYPE, TTYPE>(const Arguments& arg)
INSTANTIATE(int32_t, int32_t, float);
INSTANTIATE(int32_t, int32_t, double);
INSTANTIATE(int32_t, int32_t, rocsparse_float_complex);
INSTANTIATE(int32_t, int32_t, rocsparse_double_complex);
INSTANTIATE(int64_t, int32_t, float);
INSTANTIATE(int64_t, int32_t, double);
INSTANTIATE(int64_t, int32_t, rocsparse_float_complex);
INSTANTIATE(int64_t, int32_t, rocsparse_double_complex);
INSTANTIATE(int64_t, int64_t, float);
INSTANTIATE(int64_t, int64_t, double);
INSTANTIATE(int64_t, int64_t, rocsparse_float_complex);
INSTANTIATE(int64_t, int64_t, rocsparse_double_complex);
INSTANTIATE_MIXED(int32_t, int32_t, int8_t, int8_t, int32_t, int32_t);
INSTANTIATE_MIXED(int64_t, int32_t, int8_t, int8_t, int32_t, int32_t);
INSTANTIATE_MIXED(int64_t, int64_t, int8_t, int8_t, int32_t, int32_t);
INSTANTIATE_MIXED(int32_t, int32_t, int8_t, int8_t, float, float);
INSTANTIATE_MIXED(int64_t, int32_t, int8_t, int8_t, float, float);
INSTANTIATE_MIXED(int64_t, int64_t, int8_t, int8_t, float, float);
void testing_spmm_csr_extra(const Arguments& arg) {}
|
