Skip to content

Instantly share code, notes, and snippets.

@maleadt
Created January 23, 2021 20:18
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save maleadt/d84401fe402fe86f90f018c3a4747b78 to your computer and use it in GitHub Desktop.
Save maleadt/d84401fe402fe86f90f018c3a4747b78 to your computer and use it in GitHub Desktop.
Stream-ordered memory allocator + device reset = launch failure
#include <stdio.h>
#include <cuda.h>
#define check(ans) { _check((ans), __FILE__, __LINE__); }
inline void _check(CUresult code, const char *file, int line)
{
if (code != CUDA_SUCCESS)
{
const char *name;
cuGetErrorName(code, &name);
fprintf(stderr,"ERROR %s at %s:%d\n", name, file, line);
exit(code);
}
}
int main() {
check(cuInit(0));
CUdevice dev;
check(cuDeviceGet(&dev, 0));
CUcontext ctx;
check(cuDevicePrimaryCtxRetain(&ctx, dev));
check(cuCtxSetCurrent(ctx));
CUstream stream;
check(cuStreamCreate(&stream, 0));
CUdeviceptr ptr;
check(cuMemAllocAsync(&ptr, 8, stream));
// reset
check(cuDevicePrimaryCtxReset_v2(dev));
check(cuDevicePrimaryCtxRetain(&ctx, dev));
check(cuCtxSetCurrent(ctx));
check(cuStreamCreate(&stream, 0));
check(cuMemAllocAsync(&ptr, 8, stream));
void* cpu_ptr = malloc(8);
check(cuMemcpyHtoDAsync_v2(ptr, cpu_ptr, 8, stream));
check(cuStreamSynchronize(stream));
return 0;
}
#include <stdio.h>
#define check(ans) { _check((ans), __FILE__, __LINE__); }
inline void _check(cudaError_t code, const char *file, int line)
{
if (code != cudaSuccess)
{
fprintf(stderr,"ERROR %s at %s:%d\n", cudaGetErrorName(code), file, line);
exit(code);
}
}
int main() {
cudaStream_t stream;
void *cpu, *gpu;
cpu = malloc(8);
check(cudaStreamCreate(&stream));
check(cudaMallocAsync(&gpu, 8, stream));
printf("ptr: 0x%p\n", gpu);
check(cudaMemcpyAsync(gpu, cpu, 8, cudaMemcpyHostToDevice, stream));
check(cudaStreamSynchronize(stream));
check(cudaStreamDestroy(stream));
check(cudaDeviceReset());
check(cudaStreamCreate(&stream));
check(cudaMallocAsync(&gpu, 8, stream));
printf("ptr: 0x%p\n", gpu);
check(cudaMemcpyAsync(gpu, cpu, 8, cudaMemcpyHostToDevice, stream));
check(cudaStreamSynchronize(stream));
check(cudaStreamDestroy(stream));
return 0;
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment