/* $Source: bitbucket.org:berkeleylab/gasnet.git/tests/testlarge.c $ * Description: GASNet bulk (EVENT_DEFER) get/put performance test * measures the ping-pong average round-trip time and * average flood throughput of GASNet gets and puts * over varying payload size and synchronization mechanisms * Copyright 2002, Jaein Jeong and Dan Bonachea * Terms of use are as specified in license.txt */ /************************************************************ testlarge.c: measures the bandwidth of get and put for large messages with payload size (512 .. limit bytes) also measures the barrier time. *************************************************************/ #include #if GASNET_HAVE_MK_CLASS_MULTIPLE #include #endif int numprocs; size_t maxsz = 0; #ifndef TEST_SEGSZ #define TEST_SEGSZ_EXPR ((numprocs==1?2:1)*(uintptr_t)alignup(maxsz,PAGESZ)) #endif #include "test.h" #if GASNET_HAVE_MK_CLASS_ZE #include "zekind.h" #endif #define GASNET_HEADNODE 0 #define PRINT_LATENCY 0 #define PRINT_THROUGHPUT 1 static gex_Client_t myclient; static gex_EP_t myep; static gex_TM_t myteam; static gex_Segment_t mysegment; typedef struct { size_t datasize; int iters; uint64_t time; double max_throughput; } stat_struct_t; gex_AM_Entry_t handler_table[2]; int insegment = 1; int myproc; int peerproc = -1; int iamsender = 0; int unitsMB = 0; int doputs = 1; int dogets = 1; int lbufs = 0; #if GASNET_HAVE_MK_CLASS_MULTIPLE static int use_loc_gpu = 0; static int use_rem_gpu = 1; #else #define use_loc_gpu 0 #define use_rem_gpu 0 #endif size_t min_payload = 0; size_t max_payload; size_t max_step = 0; char *tgtmem; void *msgbuf; #define init_stat \ GASNETT_TRACE_SETSOURCELINE(__FILE__,__LINE__), _init_stat #define update_stat \ GASNETT_TRACE_SETSOURCELINE(__FILE__,__LINE__), _update_stat #define print_stat \ GASNETT_TRACE_SETSOURCELINE(__FILE__,__LINE__), _print_stat void _init_stat(stat_struct_t *st, size_t sz) { st->iters = 0; st->datasize = sz; st->time = 0; } void _update_stat(stat_struct_t *st, uint64_t temptime, int iters) { st->iters += iters; st->time += temptime; } void _print_stat(int myproc, stat_struct_t *st, const char *name, int operation) { switch (operation) { case PRINT_LATENCY: printf("%c: %3i - %10li byte : %7i iters," " latency %12i us total, %9.3f us ave. (%s)\n", TEST_SECTION_NAME(), myproc, (long) st->datasize, st->iters, (int) st->time, ((double)st->time) / st->iters, name); fflush(stdout); break; case PRINT_THROUGHPUT: printf((unitsMB ? "%c: %3i - %10li byte : %7i iters, throughput %11.6f MB/sec (%s)\n": "%c: %3i - %10li byte : %7i iters, throughput %11.3f KB/sec (%s)\n"), TEST_SECTION_NAME(), myproc, (long) st->datasize, st->iters, ((int)st->time == 0 ? 0.0 : (1000000.0 * st->datasize * st->iters / (unitsMB?(1024.0*1024.0):1024.0)) / ((int)st->time)), name); fflush(stdout); break; default: break; } } // Double payload at each iter, subject to max_step // but include max_payload which may not otherwise be visited #define ADVANCE(sz) do { \ size_t step = MIN(max_step, sz); \ if (!sz) { \ sz = 1; \ } else if (sz < max_payload && sz+step > max_payload) { \ sz = max_payload; \ } else { \ sz += step; \ } \ } while (0) void bulk_test(int iters) {GASNET_BEGIN_FUNCTION(); int i; int64_t begin, end; stat_struct_t stget, stput; size_t payload; for (payload = min_payload; payload <= max_payload && payload > 0; ) { init_stat(&stput, payload); BARRIER(); if (iamsender && doputs) { /* measure the throughput of sending a message */ begin = TIME(); if (lbufs > 1) { for (i = 0; i < iters; i++) { void *loc_addr = (void*)((uintptr_t)msgbuf + max_payload * (i%lbufs)); gex_RMA_PutBlocking(myteam, peerproc, tgtmem, loc_addr, payload, 0); } } else { for (i = 0; i < iters; i++) { gex_RMA_PutBlocking(myteam, peerproc, tgtmem, msgbuf, payload, 0); } } end = TIME(); update_stat(&stput, (end - begin), iters); } BARRIER(); if (iamsender && doputs) { print_stat(myproc, &stput, "PutBlocking throughput", PRINT_THROUGHPUT); } init_stat(&stget, payload); if (iamsender && dogets) { /* measure the throughput of receiving a message */ begin = TIME(); if (lbufs > 1) { for (i = 0; i < iters; i++) { void *loc_addr = (void*)((uintptr_t)msgbuf + max_payload * (i%lbufs)); gex_RMA_GetBlocking(myteam, loc_addr, peerproc, tgtmem, payload, 0); } } else { for (i = 0; i < iters; i++) { gex_RMA_GetBlocking(myteam, msgbuf, peerproc, tgtmem, payload, 0); } } end = TIME(); update_stat(&stget, (end - begin), iters); } BARRIER(); if (iamsender && dogets) { print_stat(myproc, &stget, "GetBlocking throughput", PRINT_THROUGHPUT); } ADVANCE(payload); } } void bulk_test_nbi(int iters, int doalc) {GASNET_BEGIN_FUNCTION(); int i; int64_t begin, end; stat_struct_t stget, stput; size_t payload; gex_Event_t *lc_opt = doalc ? GEX_EVENT_GROUP : GEX_EVENT_DEFER; for (payload = min_payload; payload <= max_payload && payload > 0; ) { init_stat(&stput, payload); BARRIER(); if (iamsender && doputs) { /* measure the throughput of sending a message */ begin = TIME(); if (lbufs > 1) { for (i = 0; i < iters; i++) { void *loc_addr = (void*)((uintptr_t)msgbuf + max_payload * (i%lbufs)); gex_RMA_PutNBI(myteam, peerproc, tgtmem, loc_addr, payload, lc_opt, 0); } } else { for (i = 0; i < iters; i++) { gex_RMA_PutNBI(myteam, peerproc, tgtmem, msgbuf, payload, lc_opt, 0); } } gex_NBI_Wait(doalc ? GEX_EC_ALL : GEX_EC_PUT, 0); end = TIME(); update_stat(&stput, (end - begin), iters); } BARRIER(); if (iamsender && doputs) { const char *name = doalc ? "PutNBI+GROUP throughput" : "PutNBI+DEFER throughput"; print_stat(myproc, &stput, name, PRINT_THROUGHPUT); } init_stat(&stget, payload); if (iamsender && dogets && !doalc) { /* measure the throughput of receiving a message */ begin = TIME(); if (lbufs > 1) { for (i = 0; i < iters; i++) { void *loc_addr = (void*)((uintptr_t)msgbuf + max_payload * (i%lbufs)); gex_RMA_GetNBI(myteam, loc_addr, peerproc, tgtmem, payload, 0); } } else { for (i = 0; i < iters; i++) { gex_RMA_GetNBI(myteam, msgbuf, peerproc, tgtmem, payload, 0); } } gex_NBI_Wait(GEX_EC_GET,0); end = TIME(); update_stat(&stget, (end - begin), iters); } BARRIER(); if (iamsender && dogets && !doalc) { print_stat(myproc, &stget, "GetNBI throughput", PRINT_THROUGHPUT); } ADVANCE(payload); } } void bulk_test_nb(int iters, int doalc) {GASNET_BEGIN_FUNCTION(); int i; int64_t begin, end; stat_struct_t stget, stput; gex_Event_t *events; size_t payload; int nevents = iters * (doalc ? 2 : 1); events = (gex_Event_t *) test_malloc(sizeof(gex_Event_t) * nevents); for (payload = min_payload; payload <= max_payload && payload > 0; ) { init_stat(&stput, payload); BARRIER(); if (iamsender && doputs) { /* measure the throughput of sending a message */ gex_Event_t *lc_opt = doalc ? (events+iters) : GEX_EVENT_DEFER; begin = TIME(); if (lbufs > 1) { for (i = 0; i < iters; i++, lc_opt += doalc) { void *loc_addr = (void*)((uintptr_t)msgbuf + max_payload * (i%lbufs)); events[i] = gex_RMA_PutNB(myteam, peerproc, tgtmem, loc_addr, payload, lc_opt, 0); } } else { for (i = 0; i < iters; i++, lc_opt += doalc) { events[i] = gex_RMA_PutNB(myteam, peerproc, tgtmem, msgbuf, payload, lc_opt, 0); } } gex_Event_WaitAll(events, nevents, 0); end = TIME(); update_stat(&stput, (end - begin), iters); } BARRIER(); if (iamsender && doputs) { const char *name = doalc ? "PutNB+event throughput" : "PutNB+DEFER throughput"; print_stat(myproc, &stput, name, PRINT_THROUGHPUT); } init_stat(&stget, payload); if (iamsender && dogets && !doalc) { /* measure the throughput of receiving a message */ begin = TIME(); if (lbufs > 1) { for (i = 0; i < iters; i++) { void *loc_addr = (void*)((uintptr_t)msgbuf + max_payload * (i%lbufs)); events[i] = gex_RMA_GetNB(myteam, loc_addr, peerproc, tgtmem, payload, 0); } } else { for (i = 0; i < iters; i++) { events[i] = gex_RMA_GetNB(myteam, msgbuf, peerproc, tgtmem, payload, 0); } } gex_Event_WaitAll(events, iters, 0); end = TIME(); update_stat(&stget, (end - begin), iters); } BARRIER(); if (iamsender && dogets && !doalc) { print_stat(myproc, &stget, "GetNB throughput", PRINT_THROUGHPUT); } ADVANCE(payload); } test_free(events); } int main(int argc, char **argv) { int iters = 0; int arg; void *myseg; void *alloc = NULL; int firstlastmode = 0; int fullduplexmode = 0; int crossmachinemode = 0; int skipwarmup = 0; size_t segsz = 0; #if GASNET_HAVE_MK_CLASS_MULTIPLE int use_cuda_uva = 0; int use_hip = 0; int use_ze = 0; #endif int use_host = 1; int help = 0; /* call startup */ GASNET_Safe(gex_Client_Init(&myclient, &myep, &myteam, "testlarge", &argc, &argv, 0)); /* parse arguments */ arg = 1; while (argc > arg) { if (!strcmp(argv[arg], "-in")) { insegment = 1; ++arg; } else if (!strcmp(argv[arg], "-out")) { insegment = 0; ++arg; } else if (!strcmp(argv[arg], "-f")) { firstlastmode = 1; ++arg; } else if (!strcmp(argv[arg], "-c")) { crossmachinemode = 1; ++arg; } else if (!strcmp(argv[arg], "-a")) { fullduplexmode = 1; ++arg; } else if (!strcmp(argv[arg], "-m")) { unitsMB = 1; ++arg; } else if (!strcmp(argv[arg], "-p")) { dogets = 0; doputs = 1; ++arg; } else if (!strcmp(argv[arg], "-g")) { dogets = 1; doputs = 0; ++arg; } else if (!strcmp(argv[arg], "-s")) { skipwarmup = 1; ++arg; } else if (!strcmp(argv[arg], "-minsz")) { ++arg; if (argc > arg) { min_payload = gasnett_parse_int(argv[arg], 1); arg++; } else help = 1; } else if (!strcmp(argv[arg], "-max-step")) { ++arg; if (argc > arg) { max_step = gasnett_parse_int(argv[arg], 1); arg++; } else help = 1; } else if (!strcmp(argv[arg], "-lbufs")) { ++arg; if (argc > arg) { lbufs = atoi(argv[arg]); arg++; } else help = 1; } else if (!strcmp(argv[arg], "-segsz")) { ++arg; if (argc > arg) { segsz = gasnett_parse_int(argv[arg], 1024*1024); arg++; } else help = 1; #if GASNET_HAVE_MK_CLASS_MULTIPLE #if GASNET_HAVE_MK_CLASS_CUDA_UVA } else if (!strcmp(argv[arg], "-cuda-uva")) { use_cuda_uva = 1; use_hip = 0; use_ze = 0; use_host = 0; ++arg; #endif #if GASNET_HAVE_MK_CLASS_HIP } else if (!strcmp(argv[arg], "-hip")) { use_hip = 1; use_cuda_uva = 0; use_ze = 0; use_host = 0; ++arg; #endif #if GASNET_HAVE_MK_CLASS_ZE } else if (!strcmp(argv[arg], "-ze")) { use_cuda_uva = 0; use_hip = 0; use_ze = 1; use_host = 0; ++arg; #endif } else if (!strcmp(argv[arg], "-host")) { use_cuda_uva = 0; use_hip = 0; use_ze = 0; use_host = 1; ++arg; } else if (!strcmp(argv[arg], "-local-gpu")) { use_loc_gpu = 1; ++arg; } else if (!strcmp(argv[arg], "-local-host")) { use_loc_gpu = 0; ++arg; } else if (!strcmp(argv[arg], "-remote-gpu")) { use_rem_gpu = 1; ++arg; } else if (!strcmp(argv[arg], "-remote-host")) { use_rem_gpu = 0; ++arg; #endif } else if (argv[arg][0] == '-') { help = 1; ++arg; } else break; } if (argc > arg) { iters = atoi(argv[arg]); arg++; } if (!iters) iters = 1000; if (argc > arg) { maxsz = gasnett_parse_int(argv[arg], 1); arg++; } if (!maxsz) maxsz = 2*1024*1024; /* 2 MB default */ if (argc > arg) { TEST_SECTION_PARSE(argv[arg]); arg++; } #if GASNET_HAVE_MK_CLASS_MULTIPLE if (use_host) { use_loc_gpu = use_rem_gpu = 0; } #endif if (!max_step) max_step = maxsz; if (!min_payload) min_payload = 16; if (!insegment) { // silently map out-of-range to iters if ((lbufs < 0) || (lbufs > iters)) lbufs = iters; } else if (lbufs) { MSG0("WARNING: Ignoring '-lbufs N' without '-out'."); lbufs = 1; } if (!lbufs) lbufs = 1; // default /* get SPMD info (needed for segment size) */ myproc = gex_TM_QueryRank(myteam); numprocs = gex_TM_QuerySize(myteam); #ifdef GASNET_SEGMENT_EVERYTHING if (maxsz > TEST_SEGSZ) { ERR("maxsz must be <= %"PRIuPTR" on GASNET_SEGMENT_EVERYTHING",(uintptr_t)TEST_SEGSZ); gasnet_exit(1); } #endif if (segsz == 0) { segsz = TEST_SEGSZ_REQUEST; } else if (segsz < TEST_SEGSZ_REQUEST) { ERR("Command line -segsz %"PRIuPTR" is less than %"PRIuPTR, (uintptr_t)segsz, (uintptr_t)(TEST_SEGSZ_REQUEST)); gasnet_exit(1); } GASNET_Safe(gex_Segment_Attach(&mysegment, myteam, segsz)); #if GASNET_HAVE_MK_CLASS_MULTIPLE #define KIND_USAGE_BEGIN \ "\n" \ " Memory kind selection (last-used has precedence):\n" \ " -host Test host memory, aka GEX_MK_CLASS_HOST (default)\n" #define KIND_USAGE_END \ " Memory kind buffer location (ignored with -host):\n" \ " Local buffer location (last-used has precedence):\n" \ " -local-host Local buffer is in host memory (default)\n" \ " -local-gpu Local buffer is in GPU memory\n" \ " Remote buffer location (last-used has precedence):\n" \ " -remote-host Remote buffer is in host memory\n" \ " -remote-gpu Remote buffer is in GPU memory (default)" #else #define KIND_USAGE_BEGIN // empty #define KIND_USAGE_END // empty #endif #if GASNET_HAVE_MK_CLASS_CUDA_UVA #define KIND_USAGE_CUDA_UVA " -cuda-uva Test GEX_MK_CLASS_CUDA_UVA\n" #else #define KIND_USAGE_CUDA_UVA // empty #endif #if GASNET_HAVE_MK_CLASS_HIP #define KIND_USAGE_HIP " -hip Test GEX_MK_CLASS_HIP\n" #else #define KIND_USAGE_HIP // empty #endif #if GASNET_HAVE_MK_CLASS_ZE #define KIND_USAGE_ZE " -ze Test GEX_MK_CLASS_ZE\n" #else #define KIND_USAGE_ZE // empty #endif test_init("testlarge",1, "[options] (iters) (maxsz) (test_sections)\n" " The '-in' or '-out' option selects whether the initiator-side\n" " memory is in the GASNet segment or not (default is 'in').\n" " The -p/-g option selects puts only or gets only (default is both).\n" " The -s option skips warm-up iterations\n" " The -m option enables MB/sec units for bandwidth output (MB=2^20 bytes).\n" " The -a option enables full-duplex mode, where all nodes send.\n" " The -c option enables cross-machine pairing, default is nearest neighbor.\n" " The -f option enables 'first/last' mode, where the first/last\n" " nodes communicate with each other, while all other nodes sit idle.\n" " The '-minsz N' option sets the minimum transfer size tested (default is 16).\n" " The '-max-step N' option selects the maximum step between transfer sizes,\n" " which by default advance by doubling until maxsz is reached.\n" " The '-segsz N' option sets the segment space to use (in MB).\n" " The default is the minimum necessary to complete the tests.\n" " The '-lbufs N' option is valid only with '-out' and sets the number of\n" " distinct local (initiator-side) buffers to cycle through (default is 1)." KIND_USAGE_BEGIN KIND_USAGE_CUDA_UVA KIND_USAGE_HIP KIND_USAGE_ZE KIND_USAGE_END ); if (help || argc > arg) test_usage(); max_payload = maxsz; if (max_payload < min_payload) { ERR("maxsz (%"PRIuPTR") must be >= %"PRIuPTR"\n",(uintptr_t)max_payload,(uintptr_t)min_payload); test_usage(); } if (!firstlastmode) { /* Only allow 1 or even number for numprocs */ if (numprocs > 1 && numprocs % 2 != 0) { MSG0("WARNING: This test requires a unary or even number of nodes. Test skipped.\n"); gasnet_exit(0); /* exit 0 to prevent false negatives in test harnesses for smp-conduit */ } } /* Setting peer thread rank */ if (firstlastmode) { peerproc = (myproc == 0 ? numprocs-1 : 0); iamsender = (fullduplexmode ? myproc == 0 || myproc == numprocs-1 : myproc == 0); } else if (numprocs == 1) { peerproc = 0; iamsender = 1; } else if (crossmachinemode) { if (myproc < numprocs / 2) { peerproc = myproc + numprocs/2; iamsender = 1; } else { peerproc = myproc - numprocs/2; iamsender = fullduplexmode; } } else { peerproc = (myproc % 2) ? (myproc - 1) : (myproc + 1); iamsender = (fullduplexmode || myproc % 2 == 0); } myseg = TEST_SEG(myproc); tgtmem = (numprocs > 1) ? TEST_SEG(peerproc) : (void*)(alignup(maxsz,PAGESZ) + (uintptr_t)myseg); #if GASNET_HAVE_MK_CLASS_MULTIPLE test_static_assert(GASNET_MAXEPS >= 2); gex_EP_t gpu_ep; gex_MK_t kind; gex_Segment_t d_segment = GEX_SEGMENT_INVALID; gex_MK_Create_args_t args; args.gex_flags = 0; int use_device = 0; if (use_cuda_uva) { MSG0("***NOTICE***: Using EXPERIMENTAL/UNTUNED support for CUDA UVA memory kind (local %s, remote %s)", (use_loc_gpu ? "GPU" : "host"), (use_rem_gpu ? "GPU" : "host")); args.gex_class = GEX_MK_CLASS_CUDA_UVA; args.gex_args.gex_class_cuda_uva.gex_CUdevice = 0; use_device = 1; } if (use_hip) { MSG0("***NOTICE***: Using EXPERIMENTAL/UNTUNED support for HIP memory kind (local %s, remote %s)", (use_loc_gpu ? "GPU" : "host"), (use_rem_gpu ? "GPU" : "host")); args.gex_class = GEX_MK_CLASS_HIP; args.gex_args.gex_class_hip.gex_hipDevice = 0; use_device = 1; } if (use_ze) { MSG0("***NOTICE***: Using EXPERIMENTAL/UNTUNED support for ZE memory kind (local %s, remote %s)", (use_loc_gpu ? "GPU" : "host"), (use_rem_gpu ? "GPU" : "host")); #if GASNET_HAVE_MK_CLASS_ZE if (! test_open_ze_device(0, &args)) { FATALERR("GEX_MK_CLASS_ZE: could not find a GPU device"); } #endif use_device = 1; } if (use_device) { // Due to bug 4149, single-process + use_device is currently unsupported if (numprocs == 1) { MSG0("WARNING: Device memory mode requires more than one process. Test skipped.\n"); gasnet_exit(0); } GASNET_Safe( gex_MK_Create(&kind, myclient, &args, 0) ); GASNET_Safe( gex_Segment_Create(&d_segment, myclient, NULL, segsz, kind, 0) ); GASNET_Safe( gex_EP_Create(&gpu_ep, myclient, GEX_EP_CAPABILITY_RMA, 0) ); GASNET_Safe( gex_EP_BindSegment(gpu_ep, d_segment, 0) ); gex_EP_PublishBoundSegment(myteam, &gpu_ep, 1, 0); // The "trick" to diverting RMA operation to local and/or remote GPU memory myteam = gex_TM_Pair(use_loc_gpu ? gpu_ep : myep, use_rem_gpu ? gex_EP_QueryIndex(gpu_ep) : 0); gex_Event_Wait( gex_EP_QueryBoundSegmentNB(myteam, peerproc, (void**)&tgtmem, NULL, NULL, 0) ); } #endif if (insegment) { #if GASNET_HAVE_MK_CLASS_MULTIPLE msgbuf = use_loc_gpu ? gex_Segment_QueryAddr(d_segment) : (void *) myseg; #else msgbuf = (void *) myseg; #endif } else { alloc = (void *) test_calloc(maxsz*lbufs+PAGESZ,1); // calloc prevents valgrind warnings msgbuf = (void *) alignup(((uintptr_t)alloc), PAGESZ); /* ensure page alignment of base */ } assert(((uintptr_t)msgbuf) % PAGESZ == 0); if (myproc == 0) { MSG("Running %i iterations of %s%s%sbulk %s%s%s with %d local address%s %sside the segment for sizes: %"PRIuPTR"...%"PRIuPTR"\n", iters, (firstlastmode ? "first/last " : ""), (fullduplexmode ? "full-duplex ": ""), (crossmachinemode ? "cross-machine ": ""), doputs?"put":"", (doputs&&dogets)?"/":"", dogets?"get":"", lbufs, (lbufs>1)?"s":"", insegment ? "in" : "out", (uintptr_t)min_payload, (uintptr_t)max_payload); if (segsz > TEST_SEGSZ_REQUEST) { MSG("Using non-default segment size %"PRIuPTR, (uintptr_t)segsz); } } BARRIER(); if (iamsender && !skipwarmup) { /* pay some warm-up costs */ int i; int warm_iters = MIN(iters, 32767); /* avoid hitting 65535-handle limit */ gex_Event_t *h = test_malloc(2*sizeof(gex_Event_t)*warm_iters); for (i = 0; i < warm_iters; i++) { gex_RMA_PutBlocking(myteam, peerproc, tgtmem, msgbuf, 8, 0); gex_RMA_GetBlocking(myteam, msgbuf, peerproc, tgtmem, 8, 0); gex_RMA_PutNBI(myteam, peerproc, tgtmem, msgbuf, 8, GEX_EVENT_DEFER, 0); gex_RMA_GetNBI(myteam, msgbuf, peerproc, tgtmem, 8, 0); h[i] = gex_RMA_PutNB(myteam, peerproc, tgtmem, msgbuf, 8, GEX_EVENT_DEFER, 0); h[i+warm_iters] = gex_RMA_GetNB(myteam, msgbuf, peerproc, tgtmem, 8, 0); } if (lbufs > 1) { for (i = 0; i < lbufs; i++) { void * loc_addr = (void*)((uintptr_t)msgbuf + max_payload * i); gex_RMA_PutBlocking(myteam, peerproc, tgtmem, loc_addr, max_payload, 0); gex_RMA_GetBlocking(myteam, loc_addr, peerproc, tgtmem, max_payload, 0); } } else { gex_RMA_PutBlocking(myteam, peerproc, tgtmem, msgbuf, max_payload, 0); gex_RMA_GetBlocking(myteam, msgbuf, peerproc, tgtmem, max_payload, 0); } gex_Event_WaitAll(h, warm_iters*2, 0); gex_NBI_Wait(GEX_EC_ALL,0); test_free(h); } BARRIER(); if (TEST_SECTION_BEGIN_ENABLED()) bulk_test(iters); if (TEST_SECTION_BEGIN_ENABLED()) bulk_test_nbi(iters,0); if (TEST_SECTION_BEGIN_ENABLED()) bulk_test_nb(iters,0); if (TEST_SECTION_BEGIN_ENABLED()) bulk_test_nbi(iters,1); if (TEST_SECTION_BEGIN_ENABLED()) bulk_test_nb(iters,1); BARRIER(); if (alloc) test_free(alloc); MSG0("done."); gasnet_exit(0); return 0; } /* ------------------------------------------------------------------------------------ */