cuda

https://developer.nvidia.com/cuda-gpus#compute 查詢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做只能在特定硬件上作，兩個操做需配套；

cuda gdb:

 1 #include <iostream>
 2 #include <stdio.h>
 3 #include "book.h"
 4 
 5 __global__ void add(int a, int b, int *c) {  6   *c = a + b;  7 }  8 
 9 __global__ void kernel() { 10   printf("tttt\n"); 11 } 12 
13 int main(void) { 14   int c; 15   int *dev_c; 16   HANDLE_ERROR( cudaMalloc((void**)&dev_c, sizeof(int) ) ); 17   int count; 18   HANDLE_ERROR( cudaGetDeviceCount(&count) ); 19   printf("count: %d\n", count); 20   add<<<1,1>>>(2,7,dev_c); 21   HANDLE_ERROR( cudaMemcpy( &c, dev_c, sizeof(int), 22  cudaMemcpyDeviceToHost ) ); 23   printf("2 + 7 = %d\n", c); 24  cudaFree(dev_c); 25   return 0; 26 }

fircuda.cu  nvcc -g -G fircuda.cu

-g 表示將CPU代碼(host)編譯成可調式版本，-G表示將GPU代碼(kernel)編譯成可調式版本。

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

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