Question

运行以下示例时出现上述错误。我在几台机器上尝试了这个但是我得到了同样的错误。我从activity_trace_async复制了这个。这里有什么想法错了吗？

#include <cuda.h>
#include <cuda_runtime.h> 
#include <cupti.h>
#include <stdio.h>
#include <string.h>

#define BUF_SIZE (32 * 1024)
#define ALIGN_SIZE (8)
#define ALIGN_BUFFER(buffer, align)                                            \
(((uintptr_t) (buffer) & ((align)-1)) ? ((buffer) + (align) - ((uintptr_t) (buffer)  &
((align)-1))) : (buffer))
#define CUPTI_CALL(call)                                                \
do {                                                                  \
CUptiResult _status = call;                                         \
if (_status != CUPTI_SUCCESS) {                                     \
  const char *errstr;                                               \
  cuptiGetResultString(_status, &errstr);                           \
  fprintf(stderr, "%s:%d: error: function %s failed with error %s.\n", \
          __FILE__, __LINE__, #call, errstr);                       \
  exit(-1);                                                         \
}                                                                   \
} while (0)


const char *
getActivityPower(CUpti_ActivityEnvironmentKind kind)
{
    switch (kind) {
    case CUPTI_ACTIVITY_ENVIRONMENT_SPEED:
         return "SPEED";
    case CUPTI_ACTIVITY_ENVIRONMENT_TEMPERATURE:
         return "TEMPERATURE";
    case CUPTI_ACTIVITY_ENVIRONMENT_POWER:
         return "POWER";
    case CUPTI_ACTIVITY_ENVIRONMENT_COOLING:
         return "COOLING";
    default:
         break;
    }

    return "<unknown>";
}

static void
printActivity(CUpti_Activity *record)
{
    switch (record->kind)
    {
         case CUPTI_ACTIVITY_KIND_ENVIRONMENT:
         {
              CUpti_ActivityEnvironment *env= ( CUpti_ActivityEnvironment *) record;
              printf("POWER %s \n",
              getActivityPower(env->environmentKind)) ;
              break;
         }

         default:
              printf("  <unknown>\n");
              break;
    }
}


void CUPTIAPI bufferRequested(uint8_t **buffer, size_t *size, size_t *maxNumRecords)
{
     uint8_t *bfr = (uint8_t *) malloc(BUF_SIZE + ALIGN_SIZE);
     if (bfr == NULL) {
         printf("Error: out of memory\n");
         exit(-1);
     }

     *size = BUF_SIZE;
     *buffer = ALIGN_BUFFER(bfr, ALIGN_SIZE);
     *maxNumRecords = 0;
}

void CUPTIAPI bufferCompleted(CUcontext ctx, uint32_t streamId, uint8_t *buffer, size_t size, size_t validSize)
{
     CUptiResult status;
     CUpti_Activity *record = NULL;

     if (validSize > 0) {
       if (ctx == NULL) {
             printf("==== Starting dump for global ====\n");
       }
       else {
             printf("==== Starting dump for context %p, stream %u ====\n", ctx, streamId);
       }

       do {
          status = cuptiActivityGetNextRecord(buffer, validSize, &record);
          if (status == CUPTI_SUCCESS) {
              printActivity(record);
          }
          else if (status == CUPTI_ERROR_MAX_LIMIT_REACHED)
              break;
          else {
              CUPTI_CALL(status);
          }
       } while (1);

       // report any records dropped from the queue
       size_t dropped;
       CUPTI_CALL(cuptiActivityGetNumDroppedRecords(ctx, streamId, &dropped));
       if (dropped != 0) {
           printf("Dropped %u activity records\n", (unsigned int) dropped);
       }

       if (ctx == NULL) {
          printf("==== Finished dump for global ====\n");
       }
       else {
          printf("==== Finished dump for context %p, stream %u ====\n", ctx, streamId);
       }
    }

    free(buffer);
}


int
main(int argc, char *argv[])
{
     size_t attrValue = 0, attrValueSize = sizeof(size_t);

     CUPTI_CALL(cuptiActivityEnable(CUPTI_ACTIVITY_KIND_ENVIRONMENT));
     CUPTI_CALL(cuptiActivityRegisterCallbacks(bufferRequested, bufferCompleted));

     CUPTI_CALL(cuptiActivityGetAttribute(CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_SIZE, &attrValueSize, &attrValue));
     printf("%s = %llu\n", "CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_SIZE", (long long unsigned)attrValue);
     attrValue *= 2;
     CUPTI_CALL(cuptiActivitySetAttribute(CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_SIZE, &attrValueSize, &attrValue));

     CUPTI_CALL(cuptiActivityGetAttribute(CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_POOL_LIMIT, &attrValueSize, &attrValue));
     printf("%s = %llu\n", "CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_POOL_LIMIT", (long long unsigned)attrValue);
     attrValue *= 2;
     CUPTI_CALL(cuptiActivitySetAttribute(CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_POOL_LIMIT, &attrValueSize, &attrValue));


     return 0;
}

Answer 1

您需要创建CUDA设备上下文。只需添加

int count = 0;
cudaGetDeviceCount(&count);

在main（）的开头，重新编译，错误应该消失。

尝试启用CUPTI_ACTIVITY_KIND_ENVIRONMENT时出现CUPTI_ERROR_NOT_COMPATIBLE错误

1 个答案: