static void
printActivity(CUpti_Activity *record)
{
switch (record->kind) {
case CUPTI_ACTIVITY_KIND_DEVICE:
{
CUpti_ActivityDevice *device = (CUpti_ActivityDevice *)record;
printf("DEVICE %s (%u), capability %u.%u, global memory (bandwidth %u GB/s, size %u MB), "
"multiprocessors %u, clock %u MHz\n",
device->name, device->id,
device->computeCapabilityMajor, device->computeCapabilityMinor,
(unsigned int)(device->globalMemoryBandwidth / 1024 / 1024),
(unsigned int)(device->globalMemorySize / 1024 / 1024),
device->numMultiprocessors, (unsigned int)(device->coreClockRate / 1000));
break;
}
case CUPTI_ACTIVITY_KIND_CONTEXT:
{
CUpti_ActivityContext *context = (CUpti_ActivityContext *)record;
printf("CONTEXT %u, device %u, compute API %s\n",
context->contextId, context->deviceId,
(context->computeApiKind == CUPTI_ACTIVITY_COMPUTE_API_CUDA) ? "CUDA" : "unknown");
break;
}
case CUPTI_ACTIVITY_KIND_MEMCPY:
{
CUpti_ActivityMemcpy *memcpy = (CUpti_ActivityMemcpy *)record;
printf("MEMCPY %s [ %llu - %llu ] device %u, context %u, stream %u, correlation %u/r%u\n",
getMemcpyKindString((CUpti_ActivityMemcpyKind)memcpy->copyKind),
(unsigned long long)(memcpy->start - startTimestamp),
(unsigned long long)(memcpy->end - startTimestamp),
memcpy->deviceId, memcpy->contextId, memcpy->streamId,
memcpy->correlationId, memcpy->runtimeCorrelationId);
break;
}
case CUPTI_ACTIVITY_KIND_MEMSET:
{
CUpti_ActivityMemset *memset = (CUpti_ActivityMemset *)record;
printf("MEMSET value=%u [ %llu - %llu ] device %u, context %u, stream %u, correlation %u/r%u\n",
memset->value,
(unsigned long long)(memset->start - startTimestamp),
(unsigned long long)(memset->end - startTimestamp),
memset->deviceId, memset->contextId, memset->streamId,
memset->correlationId, memset->runtimeCorrelationId);
break;
}
case CUPTI_ACTIVITY_KIND_KERNEL:
{
CUpti_ActivityKernel *kernel = (CUpti_ActivityKernel *)record;
printf("KERNEL \"%s\" [ %llu - %llu ] device %u, context %u, stream %u, correlation %u/r%u\n",
kernel->name,
(unsigned long long)(kernel->start - startTimestamp),
(unsigned long long)(kernel->end - startTimestamp),
kernel->deviceId, kernel->contextId, kernel->streamId,
kernel->correlationId, kernel->runtimeCorrelationId);
printf(" grid [%u,%u,%u], block [%u,%u,%u], shared memory (static %u, dynamic %u)\n",
kernel->gridX, kernel->gridY, kernel->gridZ,
kernel->blockX, kernel->blockY, kernel->blockZ,
kernel->staticSharedMemory, kernel->dynamicSharedMemory);
break;
}
case CUPTI_ACTIVITY_KIND_DRIVER:
{
CUpti_ActivityAPI *api = (CUpti_ActivityAPI *)record;
printf("DRIVER cbid=%u [ %llu - %llu ] process %u, thread %u, correlation %u\n",
api->cbid,
(unsigned long long)(api->start - startTimestamp),
(unsigned long long)(api->end - startTimestamp),
api->processId, api->threadId, api->correlationId);
break;
}
case CUPTI_ACTIVITY_KIND_RUNTIME:
{
CUpti_ActivityAPI *api = (CUpti_ActivityAPI *)record;
printf("RUNTIME cbid=%u [ %llu - %llu ] process %u, thread %u, correlation %u\n",
api->cbid,
(unsigned long long)(api->start - startTimestamp),
(unsigned long long)(api->end - startTimestamp),
api->processId, api->threadId, api->correlationId);
break;
}
default:
printf(" <unknown>\n");
break;
}
}
static void
printActivity(CUpti_Activity *record)
{
printf ("<--------------------------------print-------------------------->\n");
switch (record->kind) {
case CUPTI_ACTIVITY_KIND_DEVICE:
{
CUpti_ActivityDevice *device = (CUpti_ActivityDevice *)record;
printf("DEVICE %s (%u), capability %u.%u, global memory (bandwidth %u GB/s, size %u MB), "
"multiprocessors %u, clock %u MHz\n",
device->name, device->id,
device->computeCapabilityMajor, device->computeCapabilityMinor,
(unsigned int)(device->globalMemoryBandwidth / 1024 / 1024),
(unsigned int)(device->globalMemorySize / 1024 / 1024),
device->numMultiprocessors, (unsigned int)(device->coreClockRate / 1000));
break;
}
case CUPTI_ACTIVITY_KIND_CONTEXT:
{
CUpti_ActivityContext *context = (CUpti_ActivityContext *)record;
printf("CONTEXT %u, device %u, compute API %s\n",
context->contextId, context->deviceId,
(context->computeApiKind == CUPTI_ACTIVITY_COMPUTE_API_CUDA) ? "CUDA" :
(context->computeApiKind == CUPTI_ACTIVITY_COMPUTE_API_OPENCL) ? "OpenCL" : "unknown");
break;
}
case CUPTI_ACTIVITY_KIND_MEMCPY:
{
CUpti_ActivityMemcpy *memcpy = (CUpti_ActivityMemcpy *)record;
/* CUPTI_Event Event;
Event.pid = getpid ();
Event.device_id = memcpy->deviceId;
Event.context_id = memcpy->contextId;
Event.stream_id = memcpy->streamId;
Event.type = CUDA_MEM;
Event.name = getMemcpyKindString((CUpti_ActivityMemcpyKind)memcpy->copyKind);
snprintf (Event.seg1, 80, "%llu Bytes", memcpy->bytes);
snprintf (Event.seg2, 80, "%s", getTypeString(memcpy->srcKind));
snprintf (Event.seg3, 80, "%s", getTypeString(memcpy->dstKind));
Event.starttime = (long long unsigned)(memcpy->start);
Event.endtime = (long long unsigned)(memcpy->end);
send_CUPTI_info (Event);*/
printf("MEMCPY %s [ %llu - %llu ] device %u, context %u, stream %u, correlation %u/r%u\n",
getMemcpyKindString((CUpti_ActivityMemcpyKind)memcpy->copyKind),
(unsigned long long)(memcpy->start ),
(unsigned long long)(memcpy->start - startTimestamp),
(unsigned long long)(memcpy->end ),
(unsigned long long)(memcpy->end - startTimestamp),
memcpy->deviceId, memcpy->contextId, memcpy->streamId,
memcpy->correlationId, memcpy->runtimeCorrelationId);
break;
}
case CUPTI_ACTIVITY_KIND_MEMSET:
{
CUpti_ActivityMemset *memset = (CUpti_ActivityMemset *)record;
printf("MEMSET value=%u [ %llu - %llu ] device %u, context %u, stream %u, correlation %u/r%u\n",
memset->value,
(unsigned long long)(memset->start - startTimestamp),
(unsigned long long)(memset->end - startTimestamp),
memset->deviceId, memset->contextId, memset->streamId,
memset->correlationId, memset->runtimeCorrelationId);
break;
}
case CUPTI_ACTIVITY_KIND_KERNEL:
{
CUpti_ActivityKernel *kernel = (CUpti_ActivityKernel *)record;
/* CUPTI_Event Event;
Event.pid = getpid ();
Event.device_id = kernel->deviceId;
Event.context_id = kernel->contextId;
Event.stream_id = kernel->streamId;
Event.type = CUDA_KER;
Event.name = kernel->name;
snprintf (Event.seg1, 80, "Grid (%d,%d,%d), Block (%d,%d,%d)", kernel->gridX, kernel->gridY, kernel->gridZ, kernel->blockX, kernel->blockY, kernel->blockZ);
snprintf (Event.seg2, 80, "dynamicSharedMemory %d, staticSharedMemory %d", kernel->dynamicSharedMemory, kernel->staticSharedMemory);
snprintf (Event.seg3, 80, "localMemoryPerThread %u, localMemoryTotal %u", kernel->localMemoryPerThread, kernel->localMemoryTotal);
Event.starttime = (long long unsigned)(kernel->start);
Event.endtime = (long long unsigned)(kernel->end);
send_CUPTI_info (Event);*/
printf("KERNEL \"%s\" [ %llu - %llu ] device %u, context %u, stream %u, correlation %u/r%u\n",
kernel->name,
(unsigned long long)(kernel->start - startTimestamp),
(unsigned long long)(kernel->end - startTimestamp),
kernel->deviceId, kernel->contextId, kernel->streamId,
kernel->correlationId, kernel->runtimeCorrelationId);
printf(" grid [%u,%u,%u], block [%u,%u,%u], shared memory (static %u, dynamic %u)\n",
kernel->gridX, kernel->gridY, kernel->gridZ,
kernel->blockX, kernel->blockY, kernel->blockZ,
kernel->staticSharedMemory, kernel->dynamicSharedMemory);
break;
}
case CUPTI_ACTIVITY_KIND_DRIVER:
{
CUpti_ActivityAPI *api = (CUpti_ActivityAPI *)record;
printf("DRIVER cbid=%u [ %llu - %llu ] process %u, thread %u, correlation %u\n",
api->cbid,
(unsigned long long)(api->start - startTimestamp),
(unsigned long long)(api->end - startTimestamp),
api->processId, api->threadId, api->correlationId);
break;
}
case CUPTI_ACTIVITY_KIND_RUNTIME:
{
CUpti_ActivityAPI *api = (CUpti_ActivityAPI *)record;
printf("RUNTIME cbid=%u [ %llu - %llu ] process %u, thread %u, correlation %u\n",
api->cbid,
(unsigned long long)(api->start - startTimestamp),
(unsigned long long)(api->end - startTimestamp),
api->processId, api->threadId, api->correlationId);
break;
}
default:
printf(" <unknown>\n");
break;
}
}