/* -*- c++ -*- */ /* * Copyright 2007,2008,2009 Free Software Foundation, Inc. * * This file is part of GNU Radio * * GNU Radio is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 3, or (at your option) * any later version. * * GNU Radio 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 General Public License for more details. * * You should have received a copy of the GNU General Public License along * with this program; if not, write to the Free Software Foundation, Inc., * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ #if defined(HAVE_CONFIG_H) #include #endif #include #include #include #include #include #include #include // handle to embedded SPU executable that contains benchmark routines // (The name of the variable (benchmark_procs) is the name of the spu executable.) extern spe_program_handle_t benchmark_procs; static gc_proc_id_t gcp_benchmark_udelay = GCP_UNKNOWN_PROC; #define BENCHMARK_PUT 0x1 #define BENCHMARK_GET 0x2 #define BENCHMARK_GET_PUT (BENCHMARK_PUT|BENCHMARK_GET) #if 0 static bool power_of_2_p(unsigned long x) { int nbits = sizeof(x) * 8; for (int i = 0; i < nbits; i++) if (x == (1UL << i)) return true; return false; } #endif static void init_jd(gc_job_desc *jd, unsigned int usecs, unsigned char *getbuf, unsigned char *putbuf, size_t buflen, int getput_mask) { jd->proc_id = gcp_benchmark_udelay; jd->input.nargs = 1; jd->input.arg[0].u32 = usecs; jd->output.nargs = 0; switch(getput_mask & BENCHMARK_GET_PUT){ case BENCHMARK_GET: jd->eaa.nargs = 1; jd->eaa.arg[0].direction = GCJD_DMA_GET; jd->eaa.arg[0].ea_addr = ptr_to_ea(getbuf); jd->eaa.arg[0].get_size = buflen; break; case BENCHMARK_PUT: jd->eaa.nargs = 1; jd->eaa.arg[0].direction = GCJD_DMA_PUT; jd->eaa.arg[0].ea_addr = ptr_to_ea(putbuf); jd->eaa.arg[0].put_size = buflen; break; case BENCHMARK_GET_PUT: jd->eaa.nargs = 2; jd->eaa.arg[0].direction = GCJD_DMA_GET; jd->eaa.arg[0].ea_addr = ptr_to_ea(getbuf); jd->eaa.arg[0].get_size = buflen; jd->eaa.arg[1].direction = GCJD_DMA_PUT; jd->eaa.arg[1].ea_addr = ptr_to_ea(putbuf); jd->eaa.arg[1].put_size = buflen; break; } } static void run_test(unsigned int nspes, unsigned int usecs, unsigned int dma_size, int getput_mask, int njobs_at_once) { int NJDS = njobs_at_once; gc_job_desc *all_jds[NJDS]; bool done[NJDS]; static const unsigned int BUFSIZE = (32 << 10) * NJDS; unsigned char *getbuf = new unsigned char[BUFSIZE]; boost::scoped_array _getbuf(getbuf); unsigned char *putbuf = new unsigned char[BUFSIZE]; boost::scoped_array _putbuf(putbuf); int gbi = 0; // touch all pages to force allocation now for (unsigned int i = 0; i < BUFSIZE; i += 4096){ getbuf[i] = 0; putbuf[i] = 0; } gc_jm_options opts; opts.program_handle = gc_program_handle_from_address(&benchmark_procs); opts.nspes = nspes; //opts.enable_logging = true; //opts.log2_nlog_entries = 13; gc_job_manager_sptr mgr = gc_make_job_manager(&opts); if ((gcp_benchmark_udelay = mgr->lookup_proc("benchmark_udelay")) == GCP_UNKNOWN_PROC){ fprintf(stderr, "lookup_proc: failed to find \"benchmark_udelay\"\n"); return; } // allocate and init all job descriptors for (int i = 0; i < NJDS; i++){ if (gbi + dma_size > BUFSIZE) gbi = 0; all_jds[i] = mgr->alloc_job_desc(); if (all_jds[i] == 0){ fprintf(stderr, "alloc_job_desc() returned 0\n"); return; } init_jd(all_jds[i], usecs, &getbuf[gbi], &putbuf[gbi], dma_size, getput_mask); gbi += dma_size; } int niter = 100000; omni_time t_start = omni_time::time(); for (int iter = 0; iter < niter; iter++){ // submit the jobs for (int i = 0; i < NJDS; i++){ if (!mgr->submit_job(all_jds[i])){ printf("submit_job(jds[%d]) failed, status = %d\n", i, all_jds[i]->status); } } int n = mgr->wait_jobs(NJDS, all_jds, done, GC_WAIT_ALL); if (n < 0){ fprintf(stderr, "mgr->wait_jobs failed\n"); break; } if (n != NJDS){ fprintf(stderr, "mgr->wait_jobs returned short count. Expected %d, got %d\n", NJDS, n); } } // stop timing omni_time t_stop = omni_time::time(); double delta = (t_stop - t_start).double_time(); printf("nspes: %2d udelay: %4d elapsed_time: %7.3f dma_size: %5d dma_throughput: %7.3e round_trip: %gus\n", mgr->nspes(), usecs, delta, dma_size, (double) NJDS * niter * dma_size / delta * (getput_mask == BENCHMARK_GET_PUT ? 2.0 : 1.0), delta / niter * 1e6); } static void usage() { fprintf(stderr, "usage: benchmark_dma [-p] [-g] [-n ] [-u ] [-s ] [-N ]\n"); fprintf(stderr, " you must specify one or both of -p (put) and -g (get)\n"); } int main(int argc, char **argv) { unsigned int nspes = 0; unsigned int usecs = 0; unsigned int dma_size = 32 << 10; int njobs_at_once = -1; int getput_mask = 0; int ch; while ((ch = getopt(argc, argv, "n:u:s:pgN:")) != EOF){ switch(ch){ case 'n': nspes = strtol(optarg, 0, 0); break; case 'u': usecs = strtol(optarg, 0, 0); break; case 'N': njobs_at_once = strtol(optarg, 0, 0); break; case 's': dma_size = strtol(optarg, 0, 0); if (dma_size == 0){ fprintf(stderr, "-s must be > 0\n"); return 1; } break; case 'p': getput_mask |= BENCHMARK_PUT; break; case 'g': getput_mask |= BENCHMARK_GET; break; case '?': default: usage(); return 1; } } if (njobs_at_once < 0) njobs_at_once = nspes; if (getput_mask == 0){ usage(); return 1; } run_test(nspes, usecs, dma_size, getput_mask, njobs_at_once); return 0; }