int main( int argc, char *argv[ ] ){
allocator_init();
int testNumber = 0;
if( argc != 2 ){
printf("Usage: %s | 1 | 2 | 3 | 4 | 5 | 6 |\n", argv[ 0 ] );
return 0;
}
testNumber = atoi(argv[1]);
if ( ( testNumber < 1 ) || ( testNumber > 6 ) ) {
printf("\nError.\nOption %d is an invalid token!.\n", testNumber );
printf( "Please choose a test number between 1 and 6 \n\n" );
return 0;
}
if ( testNumber == 1 ) { runTime( test1 ); }
else if ( testNumber == 2 ){ runTime( test2 ); }
else if ( testNumber == 3 ){ runTime( test3 ); }
else if ( testNumber == 4 ){ runTime( test4 ); }
else if ( testNumber == 5 ){ runTime( test5 ); }
else if ( testNumber == 6 ){ runTime( test6 ); }
allocator_finalize();
return 0;
}
static void signctx_init(struct signctx *self)
{
allocator_init(&self->mAllocator, 256);
vector_init(&self->signatureParams, sizeof(struct signature_param));
string_init(&self->signatureParamsBuf);
string_init(&self->signatureSeed);
}
struct compiler *compiler_create(const char* input, const char *filename)
{
struct compiler* result = malloc(sizeof(struct compiler));
result->filename = filename;
result->err_count = 0;
result->current_block = 0;
allocator_init(&result->allocator);
lexer_create(result, input);
return result;
}
double frageval(int alg, int seed, int requests){
srand(seed);
int r = 0;
int s = 0;
void *p = NULL;
allocator_init(10240);
while (r < requests){
s = (rand() % 900) + 100;
p = allocate(alg, s);
deallocate(p);
}
double frag = average_frag();
return frag;
}
/** @fn DataBuffer_t::DataBuffer_t(size_t allocSize )
* Constructor.
* Allocate ION (other allocator) mapped memory.
*
* @param size_t allocSize - required size in bytes.
*/
DataBuffer_t::DataBuffer_t(size_t allocSize )
{
#ifndef DVP_USE_ION
pData = calloc(1, allocSize);
if ( pData)
nBuffSize = allocSize;
bytesWritten = 0;
#else // DVP_USE_ION
mutex_init(&mBuffLock);
pAllocator = allocator_init();
if( pAllocator )
{
dims[0].img.bpp = 1;
dims[0].img.width = allocSize;
dims[0].img.height = 1;
dims[0].dim.x = allocSize;
dims[0].dim.y = 1;
dims[0].dim.z = 1;
memset((void*)handles, 0, sizeof(handles));
if( false_e == allocator_calloc( pAllocator, ALLOCATOR_MEMORY_TYPE_TILED_1D_UNCACHED, 1,
1, dims, &pData, handles, &strides ) )
{
pData = NULL;
}
if ( pData )
nBuffSize = allocSize;
else
nBuffSize = 0;
bytesWritten = 0;
}
#endif // DVP_USE_ION
}
int
main(int argc, char* argv[])
{
int* arr[ALLOC_TIMES];
int i, counter;
clock_t beg, end;
allocator_init();
counter = 0;
beg = clock();
while (counter++ < LOOP_TIMES) {
for (i = 0; i < ALLOC_TIMES; ++i)
arr[i] = (int*)malloc(sizeof(int));
for (i = 0; i < ALLOC_TIMES; ++i)
free(arr[i]);
}
end = clock();
fprintf(stdout, "default use : %lu\n", end - beg);
counter = 0;
beg = clock();
while (counter++ < LOOP_TIMES) {
for (i = 0; i < ALLOC_TIMES; ++i)
arr[i] = (int*)al_malloc(sizeof(int));
for (i = 0; i < ALLOC_TIMES; ++i)
al_free(arr[i]);
}
end = clock();
fprintf(stdout, "allocator use : %lu\n", end - beg);
allocator_destroy();
return 0;
}
int EPLIB_init()
{
process_env_vars();
init_sig_handlers();
PMPI_Comm_rank(MPI_COMM_WORLD, &taskid);
PMPI_Comm_size(MPI_COMM_WORLD, &num_tasks);
set_local_uuid(taskid);
allocator_init();
if (max_ep == 0) return MPI_SUCCESS;
/* Initialize client */
client_init(taskid, num_tasks);
/* Register MPI type and MPI Op before any other cqueue commands */
cqueue_mpi_type_register();
cqueue_mpi_op_register();
/* Initialize communicator handles table */
handle_init();
/* Initialize window object table */
window_init();
/* Create server world and peer comm for MPI_COMM_WORLD */
EPLIB_split_comm(MPI_COMM_WORLD, 0, taskid, MPI_COMM_WORLD);
if (std_mpi_mode == STD_MPI_MODE_IMPLICIT)
block_coll_request = malloc(max_ep*sizeof(MPI_Request));
return MPI_SUCCESS;
}
int main(int argc, char *argv[]) {
int policy;
unsigned int seed;
int requests;
/* temp pointer to the memory */
void *p = NULL;
int counter=0;
int s; /* temp size of the allocated memory */
double frag; /* Average fragmentation size */
if(argc!=4){
printf("usage: frag-eval [algorithm] [seed] [int requests]\n");
exit(-1);
}else{
policy=atoi(argv[1]);
seed=atoi(argv[2]);
requests=atoi(argv[3]);
}
#ifdef DEBUG
printf("policy: %d, seed: %u, requests: %d\n", policy, seed, requests);
#endif /* DEBUG */
srand(seed);
if(allocator_init(SIZE, policy)){
printf("init failed");
exit(-1);
}
#ifdef DEBUG
printf("successful init\n");
#endif /* DEBUG */
while(counter<requests) {
s = rand()%900+100;
printf("#################### Allocating size: %d#######################\n", s);
p = allocate(policy, s);
if(NULL==p) printf("\n");
else{
deallocate(p);
printf("------------------ Deallocated size: %d------------------\n\n",s);
}
counter++;
}
frag = average_frag();
printf("Average fragmentation size: %f\n", frag);
/*
printf("**********allocating blk1: 128************\n");
void *ptr1;
ptr1=allocate(128);
if(NULL==ptr1)
printf("allocation failed.\n");
else printf("blk1 is allocated: %p\n", ptr1);
printf("**********allocating blk2: 512************\n");
void *ptr2;
ptr2=allocate(512);
if(NULL==ptr2)
printf("allocation failed.\n");
else printf("blk2 is allocated: %p\n", ptr2);
printf("**********allocating blk3: 64************\n");
void *ptr3;
ptr3=allocate(64);
if(NULL==ptr3)
printf("allocation failed.\n");
else printf("blk3 is allocated: %p\n", ptr3);
printf("**********Deallocating blk2: 512************\n");
if(deallocate(ptr2)){
printf("failed deallocation");
exit(-1);
}else printf("blk2 is deallocated.\n");
printf("**********allocating blk4: 256************\n");
void *ptr4;
ptr4=allocate(256);
if(NULL==ptr4)
printf("allocation failed.\n");
else printf("blk4 is allocated: %p\n", ptr4);
printf("**********Deallocating blk3: 64************\n");
if(deallocate(ptr3)){
printf("failed deallocation");
exit(-1);
}else printf("blk3 is deallocated.\n");
printf("**********allocating blk5: 64************\n");
void *ptr5;
ptr5=allocate(64);
if(NULL==ptr5)
printf("allocation failed.\n");
else printf("blk5 is allocated: %p\n", ptr5);
printf("**********allocating blk6: 512************\n");
void *ptr6;
ptr6=allocate(512);
if(NULL==ptr6)
printf("allocation failed.\n");
else printf("blk6 is allocated: %p\n", ptr6);
*/
return 0;
}
int main(int argc, char** argv)
{
int iter = 100;
uint32_t i;
uint32_t j;
int globalNumberOfThreads = 0;
int optPrintDomains = 0;
int c;
ThreadUserData myData;
bstring testcase = bfromcstr("none");
uint32_t numberOfWorkgroups = 0;
int tmp = 0;
double time;
const TestCase* test = NULL;
Workgroup* currentWorkgroup = NULL;
Workgroup* groups = NULL;
cpuid_init();
numa_init();
affinity_init();
/* Handling of command line options */
if (argc == 1) { HELP_MSG; }
while ((c = getopt (argc, argv, "g:w:t:i:l:aphv")) != -1) {
switch (c)
{
case 'h':
HELP_MSG;
exit (EXIT_SUCCESS);
case 'v':
VERSION_MSG;
exit (EXIT_SUCCESS);
case 'a':
printf(TESTS"\n");
exit (EXIT_SUCCESS);
case 'w':
tmp--;
if (tmp == -1)
{
fprintf (stderr, "More workgroups configured than allocated!\n");
return EXIT_FAILURE;
}
if (!test)
{
fprintf (stderr, "You need to specify a test case first!\n");
return EXIT_FAILURE;
}
testcase = bfromcstr(optarg);
currentWorkgroup = groups+tmp; /*FIXME*/
bstr_to_workgroup(currentWorkgroup, testcase, test->type, test->streams);
bdestroy(testcase);
for (i=0; i< test->streams; i++)
{
if (currentWorkgroup->streams[i].offset%test->stride)
{
fprintf (stderr, "Stream %d: offset is not a multiple of stride!\n",i);
return EXIT_FAILURE;
}
allocator_allocateVector(&(currentWorkgroup->streams[i].ptr),
PAGE_ALIGNMENT,
currentWorkgroup->size,
currentWorkgroup->streams[i].offset,
test->type,
currentWorkgroup->streams[i].domain);
}
break;
case 'i':
iter = atoi(optarg);
break;
case 'l':
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (biseqcstr(testcase,"none"))
{
fprintf (stderr, "Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
else
{
printf("Name: %s\n",test->name);
printf("Number of streams: %d\n",test->streams);
printf("Loop stride: %d\n",test->stride);
printf("Flops: %d\n",test->flops);
printf("Bytes: %d\n",test->bytes);
switch (test->type)
{
case SINGLE:
printf("Data Type: Single precision float\n");
break;
case DOUBLE:
printf("Data Type: Double precision float\n");
break;
}
}
bdestroy(testcase);
exit (EXIT_SUCCESS);
break;
case 'p':
optPrintDomains = 1;
break;
case 'g':
numberOfWorkgroups = atoi(optarg);
allocator_init(numberOfWorkgroups * MAX_STREAMS);
tmp = numberOfWorkgroups;
groups = (Workgroup*) malloc(numberOfWorkgroups*sizeof(Workgroup));
break;
case 't':
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (biseqcstr(testcase,"none"))
{
fprintf (stderr, "Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
bdestroy(testcase);
break;
case '?':
if (isprint (optopt))
fprintf (stderr, "Unknown option `-%c'.\n", optopt);
else
fprintf (stderr,
"Unknown option character `\\x%x'.\n",
optopt);
return EXIT_FAILURE;
default:
HELP_MSG;
}
}
if (optPrintDomains)
{
affinity_printDomains();
exit (EXIT_SUCCESS);
}
timer_init();
/* :WARNING:05/04/2010 08:58:05 AM:jt: At the moment the thread
* module only allows equally sized thread groups*/
for (i=0; i<numberOfWorkgroups; i++)
{
globalNumberOfThreads += groups[i].numberOfThreads;
}
threads_init(globalNumberOfThreads);
threads_createGroups(numberOfWorkgroups);
/* we configure global barriers only */
barrier_init(1);
barrier_registerGroup(globalNumberOfThreads);
#ifdef PERFMON
printf("Using likwid\n");
likwid_markerInit();
#endif
/* initialize data structures for threads */
for (i=0; i<numberOfWorkgroups; i++)
{
myData.iter = iter;
myData.size = groups[i].size;
myData.test = test;
myData.numberOfThreads = groups[i].numberOfThreads;
myData.processors = (int*) malloc(myData.numberOfThreads * sizeof(int));
myData.streams = (void**) malloc(test->streams * sizeof(void*));
for (j=0; j<groups[i].numberOfThreads; j++)
{
myData.processors[j] = groups[i].processorIds[j];
}
for (j=0; j< test->streams; j++)
{
myData.streams[j] = groups[i].streams[j].ptr;
}
threads_registerDataGroup(i, &myData, copyThreadData);
free(myData.processors);
free(myData.streams);
}
printf(HLINE);
printf("LIKWID MICRO BENCHMARK\n");
printf("Test: %s\n",test->name);
printf(HLINE);
printf("Using %d work groups\n",numberOfWorkgroups);
printf("Using %d threads\n",globalNumberOfThreads);
printf(HLINE);
threads_create(runTest);
threads_destroy();
allocator_finalize();
time = (double) threads_data[0].cycles / (double) timer_getCpuClock();
printf("Cycles: %llu \n", LLU_CAST threads_data[0].cycles);
printf("Iterations: %llu \n", LLU_CAST iter);
printf("Size: %d \n", currentWorkgroup->size );
printf("Vectorlength: %d \n", threads_data[0].data.size);
printf("Time: %e sec\n", time);
printf("MFlops/s:\t%.2f\n",
1.0E-06 * ((double) numberOfWorkgroups * iter * currentWorkgroup->size * test->flops/ time));
printf("MByte/s:\t%.2f\n",
1.0E-06 * ( (double) numberOfWorkgroups * iter * currentWorkgroup->size * test->bytes/ time));
printf("Cycles per update:\t%f\n",
((double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
switch ( test->type )
{
case SINGLE:
printf("Cycles per cacheline:\t%f\n",
(16.0 * (double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
break;
case DOUBLE:
printf("Cycles per cacheline:\t%f\n",
(8.0 * (double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
break;
}
printf(HLINE);
#ifdef PERFMON
likwid_markerClose();
#endif
return EXIT_SUCCESS;
}
int main(int argc, char** argv)
{
uint64_t iter = 100;
uint32_t i;
uint32_t j;
int globalNumberOfThreads = 0;
int optPrintDomains = 0;
int c;
ThreadUserData myData;
bstring testcase = bfromcstr("none");
uint64_t numberOfWorkgroups = 0;
int tmp = 0;
double time;
double cycPerUp = 0.0;
const TestCase* test = NULL;
uint64_t realSize = 0;
uint64_t realIter = 0;
uint64_t maxCycles = 0;
uint64_t minCycles = UINT64_MAX;
uint64_t cyclesClock = 0;
uint64_t demandIter = 0;
TimerData itertime;
Workgroup* currentWorkgroup = NULL;
Workgroup* groups = NULL;
uint32_t min_runtime = 1; /* 1s */
bstring HLINE = bfromcstr("");
binsertch(HLINE, 0, 80, '-');
binsertch(HLINE, 80, 1, '\n');
int (*ownprintf)(const char *format, ...);
ownprintf = &printf;
/* Handling of command line options */
if (argc == 1)
{
HELP_MSG;
exit(EXIT_SUCCESS);
}
while ((c = getopt (argc, argv, "w:t:s:l:aphvi:")) != -1) {
switch (c)
{
case 'h':
HELP_MSG;
exit (EXIT_SUCCESS);
case 'v':
VERSION_MSG;
exit (EXIT_SUCCESS);
case 'a':
ownprintf(TESTS"\n");
exit (EXIT_SUCCESS);
case 'w':
numberOfWorkgroups++;
break;
case 's':
min_runtime = atoi(optarg);
break;
case 'i':
demandIter = strtoul(optarg, NULL, 10);
if (demandIter <= 0)
{
fprintf (stderr, "Error: Iterations must be greater than 0\n");
return EXIT_FAILURE;
}
break;
case 'l':
bdestroy(testcase);
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (test == NULL)
{
fprintf (stderr, "Error: Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
else
{
ownprintf("Name: %s\n",test->name);
ownprintf("Number of streams: %d\n",test->streams);
ownprintf("Loop stride: %d\n",test->stride);
ownprintf("Flops: %d\n",test->flops);
ownprintf("Bytes: %d\n",test->bytes);
switch (test->type)
{
case INT:
ownprintf("Data Type: Integer\n");
break;
case SINGLE:
ownprintf("Data Type: Single precision float\n");
break;
case DOUBLE:
ownprintf("Data Type: Double precision float\n");
break;
}
if (test->loads >= 0)
{
ownprintf("Load Ops: %d\n",test->loads);
}
if (test->stores >= 0)
{
ownprintf("Store Ops: %d\n",test->stores);
}
if (test->branches >= 0)
{
ownprintf("Branches: %d\n",test->branches);
}
if (test->instr_const >= 0)
{
ownprintf("Constant instructions: %d\n",test->instr_const);
}
if (test->instr_loop >= 0)
{
ownprintf("Loop instructions: %d\n",test->instr_loop);
}
}
bdestroy(testcase);
exit (EXIT_SUCCESS);
break;
case 'p':
optPrintDomains = 1;
break;
case 'g':
numberOfWorkgroups = LLU_CAST atol(optarg);
tmp = numberOfWorkgroups;
break;
case 't':
bdestroy(testcase);
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (test == NULL)
{
fprintf (stderr, "Error: Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
bdestroy(testcase);
break;
case '?':
if (isprint (optopt))
fprintf (stderr, "Unknown option `-%c'.\n", optopt);
else
fprintf (stderr,
"Unknown option character `\\x%x'.\n",
optopt);
return EXIT_FAILURE;
default:
HELP_MSG;
}
}
if ((numberOfWorkgroups == 0) && (!optPrintDomains))
{
fprintf(stderr, "Error: At least one workgroup (-w) must be set on commandline\n");
exit (EXIT_FAILURE);
}
if (topology_init() != EXIT_SUCCESS)
{
fprintf(stderr, "Error: Unsupported processor!\n");
exit(EXIT_FAILURE);
}
if ((test == NULL) && (!optPrintDomains))
{
fprintf(stderr, "Unknown test case. Please check likwid-bench -a for available tests\n");
fprintf(stderr, "and select one using the -t commandline option\n");
exit(EXIT_FAILURE);
}
numa_init();
affinity_init();
timer_init();
if (optPrintDomains)
{
bdestroy(testcase);
AffinityDomains_t affinity = get_affinityDomains();
ownprintf("Number of Domains %d\n",affinity->numberOfAffinityDomains);
for (i=0; i < affinity->numberOfAffinityDomains; i++ )
{
ownprintf("Domain %d:\n",i);
ownprintf("\tTag %s:",bdata(affinity->domains[i].tag));
for ( uint32_t j=0; j < affinity->domains[i].numberOfProcessors; j++ )
{
ownprintf(" %d",affinity->domains[i].processorList[j]);
}
ownprintf("\n");
}
exit (EXIT_SUCCESS);
}
allocator_init(numberOfWorkgroups * MAX_STREAMS);
groups = (Workgroup*) malloc(numberOfWorkgroups*sizeof(Workgroup));
tmp = 0;
optind = 0;
while ((c = getopt (argc, argv, "w:t:s:l:i:aphv")) != -1)
{
switch (c)
{
case 'w':
currentWorkgroup = groups+tmp;
bstring groupstr = bfromcstr(optarg);
i = bstr_to_workgroup(currentWorkgroup, groupstr, test->type, test->streams);
bdestroy(groupstr);
if (i == 0)
{
for (i=0; i< test->streams; i++)
{
if (currentWorkgroup->streams[i].offset%test->stride)
{
fprintf (stderr, "Error: Stream %d: offset is not a multiple of stride!\n",i);
return EXIT_FAILURE;
}
allocator_allocateVector(&(currentWorkgroup->streams[i].ptr),
PAGE_ALIGNMENT,
currentWorkgroup->size,
currentWorkgroup->streams[i].offset,
test->type,
currentWorkgroup->streams[i].domain);
}
tmp++;
}
else
{
exit(EXIT_FAILURE);
}
break;
default:
continue;
break;
}
}
/* :WARNING:05/04/2010 08:58:05 AM:jt: At the moment the thread
* module only allows equally sized thread groups*/
for (i=0; i<numberOfWorkgroups; i++)
{
globalNumberOfThreads += groups[i].numberOfThreads;
}
ownprintf(bdata(HLINE));
ownprintf("LIKWID MICRO BENCHMARK\n");
ownprintf("Test: %s\n",test->name);
ownprintf(bdata(HLINE));
ownprintf("Using %" PRIu64 " work groups\n",numberOfWorkgroups);
ownprintf("Using %d threads\n",globalNumberOfThreads);
ownprintf(bdata(HLINE));
threads_init(globalNumberOfThreads);
threads_createGroups(numberOfWorkgroups);
/* we configure global barriers only */
barrier_init(1);
barrier_registerGroup(globalNumberOfThreads);
cyclesClock = timer_getCycleClock();
#ifdef LIKWID_PERFMON
if (getenv("LIKWID_FILEPATH") != NULL)
{
ownprintf("Using Likwid Marker API\n");
}
LIKWID_MARKER_INIT;
ownprintf(bdata(HLINE));
#endif
/* initialize data structures for threads */
for (i=0; i<numberOfWorkgroups; i++)
{
myData.iter = iter;
if (demandIter > 0)
{
myData.iter = demandIter;
}
myData.min_runtime = min_runtime;
myData.size = groups[i].size;
myData.test = test;
myData.cycles = 0;
myData.numberOfThreads = groups[i].numberOfThreads;
myData.processors = (int*) malloc(myData.numberOfThreads * sizeof(int));
myData.streams = (void**) malloc(test->streams * sizeof(void*));
for (j=0; j<groups[i].numberOfThreads; j++)
{
myData.processors[j] = groups[i].processorIds[j];
}
for (j=0; j< test->streams; j++)
{
myData.streams[j] = groups[i].streams[j].ptr;
}
threads_registerDataGroup(i, &myData, copyThreadData);
free(myData.processors);
free(myData.streams);
}
if (demandIter == 0)
{
getIterSingle((void*) &threads_data[0]);
for (i=0; i<numberOfWorkgroups; i++)
{
iter = threads_updateIterations(i, demandIter);
}
}
#ifdef DEBUG_LIKWID
else
{
ownprintf("Using manually selected iterations per thread\n");
}
#endif
threads_create(runTest);
threads_join();
for (int i=0; i<globalNumberOfThreads; i++)
{
realSize += threads_data[i].data.size;
realIter += threads_data[i].data.iter;
if (threads_data[i].cycles > maxCycles)
{
maxCycles = threads_data[i].cycles;
}
if (threads_data[i].cycles < minCycles)
{
minCycles = threads_data[i].cycles;
}
}
time = (double) maxCycles / (double) cyclesClock;
ownprintf(bdata(HLINE));
ownprintf("Cycles:\t\t\t%" PRIu64 "\n", maxCycles);
ownprintf("CPU Clock:\t\t%" PRIu64 "\n", timer_getCpuClock());
ownprintf("Cycle Clock:\t\t%" PRIu64 "\n", cyclesClock);
ownprintf("Time:\t\t\t%e sec\n", time);
ownprintf("Iterations:\t\t%" PRIu64 "\n", realIter);
ownprintf("Iterations per thread:\t%" PRIu64 "\n",threads_data[0].data.iter);
ownprintf("Inner loop executions:\t%.0f\n", ((double)realSize)/((double)test->stride));
ownprintf("Size:\t\t\t%" PRIu64 "\n", realSize*test->bytes );
ownprintf("Size per thread:\t%" PRIu64 "\n", threads_data[0].data.size*test->bytes);
ownprintf("Number of Flops:\t%" PRIu64 "\n", (threads_data[0].data.iter * realSize * test->flops));
ownprintf("MFlops/s:\t\t%.2f\n",
1.0E-06 * ((double) threads_data[0].data.iter * realSize * test->flops/ time));
ownprintf("Data volume (Byte):\t%llu\n", LLU_CAST (threads_data[0].data.iter * realSize * test->bytes));
ownprintf("MByte/s:\t\t%.2f\n",
1.0E-06 * ( (double) threads_data[0].data.iter * realSize * test->bytes/ time));
cycPerUp = ((double) maxCycles / (double) (threads_data[0].data.iter * realSize));
ownprintf("Cycles per update:\t%f\n", cycPerUp);
switch ( test->type )
{
case INT:
case SINGLE:
ownprintf("Cycles per cacheline:\t%f\n", (16.0 * cycPerUp));
break;
case DOUBLE:
ownprintf("Cycles per cacheline:\t%f\n", (8.0 * cycPerUp));
break;
}
ownprintf("Loads per update:\t%ld\n", test->loads );
ownprintf("Stores per update:\t%ld\n", test->stores );
if ((test->loads > 0) && (test->stores > 0))
{
ownprintf("Load/store ratio:\t%.2f\n", ((double)test->loads)/((double)test->stores) );
}
if ((test->instr_loop > 0) && (test->instr_const > 0))
{
ownprintf("Instructions:\t\t%" PRIu64 "\n", LLU_CAST ((double)realSize/test->stride)*test->instr_loop*threads_data[0].data.iter + test->instr_const );
}
if (test->uops > 0)
{
ownprintf("UOPs:\t\t\t%" PRIu64 "\n", LLU_CAST ((double)realSize/test->stride)*test->uops*threads_data[0].data.iter);
}
ownprintf(bdata(HLINE));
threads_destroy(numberOfWorkgroups, test->streams);
allocator_finalize();
workgroups_destroy(&groups, numberOfWorkgroups, test->streams);
#ifdef LIKWID_PERFMON
if (getenv("LIKWID_FILEPATH") != NULL)
{
ownprintf("Writing Likwid Marker API results to file %s\n", getenv("LIKWID_FILEPATH"));
}
LIKWID_MARKER_CLOSE;
#endif
bdestroy(HLINE);
return EXIT_SUCCESS;
}