cuda

https://developer.nvidia.com/cuda-gpus#compute 查询gpu及其Compute Capability

Q: ERROR: INVALID_CONFIG: The engine plan file is generated on an incompatible device, expecting compute 7.5 got compute 6.1, please rebuild.

A: tensorrt 的序列化和反序列化操作只能在特定硬件上做，两个操作需配套；

cuda gdb:

#include <iostream>
#include <stdio.h>
#include "book.h"

__global__ void add(int a, int b, int *c) {
  *c = a + b;
}

__global__ void kernel() {
  printf("tttt
");
}

int main(void) {
  int c;
  int *dev_c;
  HANDLE_ERROR( cudaMalloc((void**)&dev_c, sizeof(int) ) );
  int count;
  HANDLE_ERROR( cudaGetDeviceCount(&count) );
  printf("count: %d
", count);
  add<<<1,1>>>(2,7,dev_c);
  HANDLE_ERROR( cudaMemcpy( &c, dev_c, sizeof(int),
                            cudaMemcpyDeviceToHost ) );
  printf("2 + 7 = %d
", c);
  cudaFree(dev_c);
  return 0;
}

-g 表示将CPU代码(host)编译成可调式版本，-G表示将GPU代码(kernel)编译成可调式版本。

Breakpoint 2, add<<<(1,1,1),(1,1,1)>>> (a=2, b=7, c=0x7fffc7800000)
    at fircuda.cu:6
6      *c = a + b;

(cuda-gdb) p a
Python Exception <type 'exceptions.AttributeError'> 'gdb.Type' object has no attribute 'name':
Python Exception <type 'exceptions.AttributeError'> 'gdb.Type' object has no attribute 'name':
$1 = 2

...

(cuda-gdb) p *c

Python Exception <type 'exceptions.AttributeError'> 'gdb.Type' object has no attribute 'name':
Python Exception <type 'exceptions.AttributeError'> 'gdb.Type' object has no attribute 'name':
$6 = 9

apollo@in_dev_docker:/apollo/ttt/caffe/build$ cat detect_cuda_archs.cu 
#include <cstdio>
int main()
{
  int count = 0;
  if (cudaSuccess != cudaGetDeviceCount(&count)) return -1;
  if (count == 0) return -1;
  for (int device = 0; device < count; ++device)
  {
    cudaDeviceProp prop;
    if (cudaSuccess == cudaGetDeviceProperties(&prop, device))
      std::printf("%d.%d ", prop.major, prop.minor);
  }
  return 0;
}

apollo@in_dev_docker:/apollo/ttt/caffe/build$ ./a.out 
5.0

HELP: https://docs.nvidia.com/cuda/cuda-gdb/#what-is-cuda-gdb