3 * Copyright 2007,2008,2010 Free Software Foundation, Inc.
5 * This file is part of GNU Radio
7 * GNU Radio is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 3, or (at your option)
12 * GNU Radio 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.
17 * You should have received a copy of the GNU General Public License along
18 * with this program; if not, write to the Free Software Foundation, Inc.,
19 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
22 #if defined(HAVE_CONFIG_H)
25 #include <gcell/gc_job_manager.h>
26 #include <boost/date_time/posix_time/posix_time_types.hpp>
30 #include <boost/scoped_array.hpp>
33 // handle to embedded SPU executable that contains benchmark routines
34 // (The name of the variable (benchmark_procs) is the name of the spu executable.)
35 extern spe_program_handle_t benchmark_procs;
37 static gc_proc_id_t gcp_benchmark_udelay = GCP_UNKNOWN_PROC;
39 #define BENCHMARK_PUT 0x1
40 #define BENCHMARK_GET 0x2
41 #define BENCHMARK_GET_PUT (BENCHMARK_PUT|BENCHMARK_GET)
46 power_of_2_p(unsigned long x)
48 int nbits = sizeof(x) * 8;
49 for (int i = 0; i < nbits; i++)
58 init_jd(gc_job_desc *jd, unsigned int usecs,
59 unsigned char *getbuf, unsigned char *putbuf, size_t buflen,
62 jd->proc_id = gcp_benchmark_udelay;
64 jd->input.arg[0].u32 = usecs;
67 switch(getput_mask & BENCHMARK_GET_PUT){
71 jd->eaa.arg[0].direction = GCJD_DMA_GET;
72 jd->eaa.arg[0].ea_addr = ptr_to_ea(getbuf);
73 jd->eaa.arg[0].get_size = buflen;
78 jd->eaa.arg[0].direction = GCJD_DMA_PUT;
79 jd->eaa.arg[0].ea_addr = ptr_to_ea(putbuf);
80 jd->eaa.arg[0].put_size = buflen;
83 case BENCHMARK_GET_PUT:
85 jd->eaa.arg[0].direction = GCJD_DMA_GET;
86 jd->eaa.arg[0].ea_addr = ptr_to_ea(getbuf);
87 jd->eaa.arg[0].get_size = buflen;
88 jd->eaa.arg[1].direction = GCJD_DMA_PUT;
89 jd->eaa.arg[1].ea_addr = ptr_to_ea(putbuf);
90 jd->eaa.arg[1].put_size = buflen;
96 run_test(unsigned int nspes, unsigned int usecs, unsigned int dma_size, int getput_mask)
98 using namespace boost::posix_time;
100 static const int64_t TOTAL_SIZE_DMA = 5LL << 30;
101 static const int NJDS = 64;
102 unsigned int njobs = (unsigned int)(TOTAL_SIZE_DMA / dma_size);
103 //unsigned int njobs = NJDS * 16;
104 unsigned int nsubmitted = 0;
105 unsigned int ncompleted = 0;
106 gc_job_desc *all_jds[NJDS];
107 gc_job_desc *jds[2][NJDS];
108 unsigned int njds[2];
109 unsigned int ci; // current index
112 static const unsigned int BUFSIZE = (32 << 10) * NJDS;
113 unsigned char *getbuf = new unsigned char[BUFSIZE];
114 boost::scoped_array<unsigned char> _getbuf(getbuf);
115 unsigned char *putbuf = new unsigned char[BUFSIZE];
116 boost::scoped_array<unsigned char> _putbuf(putbuf);
119 // touch all pages to force allocation now
120 for (unsigned int i = 0; i < BUFSIZE; i += 4096){
126 opts.program_handle = gc_program_handle_from_address(&benchmark_procs);
128 //opts.enable_logging = true;
129 //opts.log2_nlog_entries = 13;
130 gc_job_manager_sptr mgr = gc_make_job_manager(&opts);
132 if ((gcp_benchmark_udelay = mgr->lookup_proc("benchmark_udelay")) == GCP_UNKNOWN_PROC){
133 fprintf(stderr, "lookup_proc: failed to find \"benchmark_udelay\"\n");
137 // allocate and init all job descriptors
138 for (int i = 0; i < NJDS; i++){
139 if (gbi + dma_size > BUFSIZE)
142 all_jds[i] = mgr->alloc_job_desc();
143 if (all_jds[i] == 0){
144 fprintf(stderr, "alloc_job_desc() returned 0\n");
147 init_jd(all_jds[i], usecs, &getbuf[gbi], &putbuf[gbi], dma_size, getput_mask);
151 for (int iter = 0; iter < 1; iter++){
153 ptime t_start(microsec_clock::universal_time());
162 // submit the first batch
163 for (int i = 0; i < NJDS; i++){
164 if (mgr->submit_job(all_jds[i])){
165 jds[ci][njds[ci]++] = all_jds[i];
169 printf("submit_job(jds[%d]) failed, status = %d\n",
170 i, all_jds[i]->status);
174 while (ncompleted < njobs){
176 int n = mgr->wait_jobs(njds[ci], jds[ci], done, GC_WAIT_ANY);
177 // printf("%2d\n", n);
179 fprintf(stderr, "mgr->wait_jobs failed\n");
182 for (unsigned int i = 0; i < njds[ci]; i++){
183 if (!done[i]){ // remember for next iteration
184 jds[ci^1][njds[ci^1]++] = jds[ci][i];
188 if (jds[ci][i]->status != JS_OK){
189 printf("js_status = %d, job_id = %d, ncompleted = %d\n",
190 jds[ci][i]->status, jds[ci][i]->sys.job_id, ncompleted);
192 if (nsubmitted < njobs){ // submit another one
193 if (mgr->submit_job(jds[ci][i])){
194 jds[ci^1][njds[ci^1]++] = jds[ci][i]; // remember for next iter
198 printf("submit_job(jds[%d]) failed, status = %d\n",
199 i, jds[ci][i]->status);
204 ci ^= 1; // toggle current
208 ptime t_stop(microsec_clock::universal_time());
210 double delta = (t_stop - t_start).total_microseconds() * 1e-6;
211 printf("nspes: %2d udelay: %4d elapsed_time: %7.3f dma_size: %5d dma_throughput: %7.3e\n",
212 mgr->nspes(), usecs, delta, dma_size,
213 (double) njobs * dma_size / delta * (getput_mask == BENCHMARK_GET_PUT ? 2.0 : 1.0));
221 fprintf(stderr, "usage: benchmark_dma [-p] [-g] [-n <nspes>] [-u <udelay>] [-s <dma_size>]\n");
222 fprintf(stderr, " you must specify one or both of -p (put) and -g (get)\n");
227 main(int argc, char **argv)
229 unsigned int nspes = 0;
230 unsigned int usecs = 0;
231 unsigned int dma_size = 32 << 10;
235 while ((ch = getopt(argc, argv, "n:u:s:pg")) != EOF){
238 nspes = strtol(optarg, 0, 0);
242 usecs = strtol(optarg, 0, 0);
246 dma_size = strtol(optarg, 0, 0);
248 fprintf(stderr, "-s <dma_size> must be > 0\n");
254 getput_mask |= BENCHMARK_PUT;
258 getput_mask |= BENCHMARK_GET;
268 if (getput_mask == 0){
273 run_test(nspes, usecs, dma_size, getput_mask);