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
|
/* StarPU --- Runtime system for heterogeneous multicore architectures.
*
* Copyright (C) 2009, 2010, 2011 Université de Bordeaux 1
* Copyright (C) 2010 Mehdi Juhoor <mjuhoor@gmail.com>
* Copyright (C) 2010, 2011, 2012 Centre National de la Recherche Scientifique
*
* StarPU is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation; either version 2.1 of the License, or (at
* your option) any later version.
*
* StarPU is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
*
* See the GNU Lesser General Public License in COPYING.LGPL for more details.
*/
#include "spmv.h"
unsigned nblocks = 4;
uint32_t size = 4*1024*1024;
starpu_data_handle_t sparse_matrix;
starpu_data_handle_t vector_in, vector_out;
static void parse_args(int argc, char **argv)
{
int i;
for (i = 1; i < argc; i++)
{
if (strcmp(argv[i], "-size") == 0)
{
char *argptr;
size = strtol(argv[++i], &argptr, 10);
}
if (strcmp(argv[i], "-nblocks") == 0)
{
char *argptr;
nblocks = strtol(argv[++i], &argptr, 10);
}
}
}
/* This filter function takes a CSR matrix, and divides it into nparts with the
* same number of non-zero entries. */
static void csr_filter_func(void *father_interface, void *child_interface, struct starpu_data_filter *f, unsigned id, unsigned nparts)
{
struct starpu_csr_interface *csr_father = (struct starpu_csr_interface *) father_interface;
struct starpu_csr_interface *csr_child = (struct starpu_csr_interface *) child_interface;
uint32_t nrow = csr_father->nrow;
size_t elemsize = csr_father->elemsize;
uint32_t firstentry = csr_father->firstentry;
/* Every sub-parts should contain the same number of non-zero entries */
uint32_t chunk_size = (nrow + nparts - 1)/nparts;
uint32_t *rowptr = csr_father->rowptr;
uint32_t first_index = id*chunk_size - firstentry;
uint32_t local_firstentry = rowptr[first_index];
uint32_t child_nrow = STARPU_MIN(chunk_size, nrow - id*chunk_size);
uint32_t local_nnz = rowptr[first_index + child_nrow] - rowptr[first_index];
csr_child->nnz = local_nnz;
csr_child->nrow = child_nrow;
csr_child->firstentry = local_firstentry;
csr_child->elemsize = elemsize;
if (csr_father->nzval)
{
csr_child->rowptr = &csr_father->rowptr[first_index];
csr_child->colind = &csr_father->colind[local_firstentry];
csr_child->nzval = csr_father->nzval + local_firstentry * elemsize;
}
}
/* partition the CSR matrix along a block distribution */
static struct starpu_data_filter csr_f =
{
.filter_func = csr_filter_func,
/* This value is defined later on */
.nchildren = -1,
/* the children also use a csr interface */
};
static struct starpu_data_filter vector_f =
{
.filter_func = starpu_block_filter_func_vector,
/* This value is defined later on */
.nchildren = -1,
};
static struct starpu_codelet spmv_cl =
{
.where = STARPU_CPU|STARPU_CUDA|STARPU_OPENCL,
.cpu_funcs = {spmv_kernel_cpu, NULL},
#ifdef STARPU_USE_CUDA
.cuda_funcs = {spmv_kernel_cuda, NULL},
#endif
#ifdef STARPU_USE_OPENCL
.opencl_funcs = {spmv_kernel_opencl, NULL},
#endif
.nbuffers = 3,
.modes = {STARPU_R, STARPU_R, STARPU_W},
.model = NULL
};
int main(int argc, char **argv)
{
int ret;
unsigned part;
double timing;
struct timeval start, end;
unsigned row, pos;
unsigned ind;
/* CSR matrix description */
float *nzval;
uint32_t nnz;
uint32_t *colind;
uint32_t *rowptr;
/* Input and Output vectors */
float *vector_in_ptr;
float *vector_out_ptr;
/*
* Parse command-line arguments
*/
parse_args(argc, argv);
/*
* Launch StarPU
*/
ret = starpu_init(NULL);
if (ret == -ENODEV)
return 77;
STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
/*
* Create a 3-band sparse matrix as input example
*/
nnz = 3*size-2;
starpu_malloc((void **)&nzval, nnz*sizeof(float));
starpu_malloc((void **)&colind, nnz*sizeof(uint32_t));
starpu_malloc((void **)&rowptr, (size+1)*sizeof(uint32_t));
assert(nzval && colind && rowptr);
/* fill the matrix */
for (row = 0, pos = 0; row < size; row++)
{
rowptr[row] = pos;
if (row > 0)
{
nzval[pos] = 1.0f;
colind[pos] = row-1;
pos++;
}
nzval[pos] = 5.0f;
colind[pos] = row;
pos++;
if (row < size - 1)
{
nzval[pos] = 1.0f;
colind[pos] = row+1;
pos++;
}
}
STARPU_ASSERT(pos == nnz);
rowptr[size] = nnz;
/* initiate the 2 vectors */
starpu_malloc((void **)&vector_in_ptr, size*sizeof(float));
starpu_malloc((void **)&vector_out_ptr, size*sizeof(float));
assert(vector_in_ptr && vector_out_ptr);
/* fill them */
for (ind = 0; ind < size; ind++)
{
vector_in_ptr[ind] = 2.0f;
vector_out_ptr[ind] = 0.0f;
}
/*
* Register the CSR matrix and the 2 vectors
*/
starpu_csr_data_register(&sparse_matrix, 0, nnz, size, (uintptr_t)nzval, colind, rowptr, 0, sizeof(float));
starpu_vector_data_register(&vector_in, 0, (uintptr_t)vector_in_ptr, size, sizeof(float));
starpu_vector_data_register(&vector_out, 0, (uintptr_t)vector_out_ptr, size, sizeof(float));
/*
* Partition the CSR matrix and the output vector
*/
csr_f.nchildren = nblocks;
vector_f.nchildren = nblocks;
starpu_data_partition(sparse_matrix, &csr_f);
starpu_data_partition(vector_out, &vector_f);
/*
* If we use OpenCL, we need to compile the SpMV kernel
*/
#ifdef STARPU_USE_OPENCL
compile_spmv_opencl_kernel();
#endif
gettimeofday(&start, NULL);
/*
* Create and submit StarPU tasks
*/
for (part = 0; part < nblocks; part++)
{
struct starpu_task *task = starpu_task_create();
task->cl = &spmv_cl;
task->handles[0] = starpu_data_get_sub_data(sparse_matrix, 1, part);
task->handles[1] = vector_in;
task->handles[2] = starpu_data_get_sub_data(vector_out, 1, part);
ret = starpu_task_submit(task);
if (STARPU_UNLIKELY(ret == -ENODEV))
{
FPRINTF(stderr, "No worker may execute this task\n");
exit(0);
}
}
starpu_task_wait_for_all();
gettimeofday(&end, NULL);
/*
* Unregister the CSR matrix and the output vector
*/
starpu_data_unpartition(sparse_matrix, 0);
starpu_data_unpartition(vector_out, 0);
/*
* Unregister data
*/
starpu_data_unregister(sparse_matrix);
starpu_data_unregister(vector_in);
starpu_data_unregister(vector_out);
/*
* Display the result
*/
for (row = 0; row < STARPU_MIN(size, 16); row++)
{
FPRINTF(stdout, "%2.2f\t%2.2f\n", vector_in_ptr[row], vector_out_ptr[row]);
}
starpu_free(nzval);
starpu_free(colind);
starpu_free(rowptr);
starpu_free(vector_in_ptr);
starpu_free(vector_out_ptr);
/*
* Stop StarPU
*/
starpu_shutdown();
timing = (double)((end.tv_sec - start.tv_sec)*1000000 + (end.tv_usec - start.tv_usec));
FPRINTF(stderr, "Computation took (in ms)\n");
FPRINTF(stdout, "%2.2f\n", timing/1000);
return 0;
}
|