Shuenhoy/vtk-m
1
0
Fork 0
mirror of https://gitlab.kitware.com/vtk/vtk-m synced 2024-09-16 17:22:55 +00:00
Code Issues 2 Actions Packages Projects Releases Wiki Activity

Only use cuda hints for CUDA 8.0+.

Browse Source
This commit is contained in:
Allison Vacanti 2017-10-24 11:55:07 -04:00
parent b5b56cd891
commit 5a99dd761b
1 changed files with 8 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
vtkm/cont/cuda/internal/CudaAllocator.cu
View File

@ -120,12 +120,14 @@ void CudaAllocator::PrepareForControl(const void* ptr, std::size_t numBytes)
if (ManagedMemorySupported)
{
#if CUDART_VERSION >= 8000
// TODO these hints need to be benchmarked and adjusted once we start
// sharing the pointers between cont/exec
VTKM_CUDA_CALL(
cudaMemAdvise(ptr, numBytes, cudaMemAdviseSetPreferredLocation, cudaCpuDeviceId));
VTKM_CUDA_CALL(cudaMemAdvise(ptr, numBytes, cudaMemAdviseUnsetReadMostly, cudaCpuDeviceId));
VTKM_CUDA_CALL(cudaMemPrefetchAsync(ptr, numBytes, cudaCpuDeviceId, cudaStreamPerThread));
#endif // CUDA >= 8.0
}
}
@ -135,11 +137,13 @@ void CudaAllocator::PrepareForInput(const void* ptr, std::size_t numBytes)
if (ManagedMemorySupported)
{
#if CUDART_VERSION >= 8000
int dev;
VTKM_CUDA_CALL(cudaGetDevice(&dev));
VTKM_CUDA_CALL(cudaMemAdvise(ptr, numBytes, cudaMemAdviseSetPreferredLocation, dev));
VTKM_CUDA_CALL(cudaMemAdvise(ptr, numBytes, cudaMemAdviseSetReadMostly, dev));
VTKM_CUDA_CALL(cudaMemPrefetchAsync(ptr, numBytes, dev, cudaStreamPerThread));
#endif // CUDA >= 8.0
}
}
@ -149,11 +153,13 @@ void CudaAllocator::PrepareForOutput(const void* ptr, std::size_t numBytes)
if (ManagedMemorySupported)
{
#if CUDART_VERSION >= 8000
int dev;
VTKM_CUDA_CALL(cudaGetDevice(&dev));
VTKM_CUDA_CALL(cudaMemAdvise(ptr, numBytes, cudaMemAdviseSetPreferredLocation, dev));
VTKM_CUDA_CALL(cudaMemAdvise(ptr, numBytes, cudaMemAdviseUnsetReadMostly, dev));
VTKM_CUDA_CALL(cudaMemPrefetchAsync(ptr, numBytes, dev, cudaStreamPerThread));
#endif // CUDA >= 8.0
}
}
@ -163,11 +169,13 @@ void CudaAllocator::PrepareForInPlace(const void* ptr, std::size_t numBytes)
if (ManagedMemorySupported)
{
#if CUDART_VERSION >= 8000
int dev;
VTKM_CUDA_CALL(cudaGetDevice(&dev));
VTKM_CUDA_CALL(cudaMemAdvise(ptr, numBytes, cudaMemAdviseSetPreferredLocation, dev));
VTKM_CUDA_CALL(cudaMemAdvise(ptr, numBytes, cudaMemAdviseUnsetReadMostly, dev));
VTKM_CUDA_CALL(cudaMemPrefetchAsync(ptr, numBytes, dev, cudaStreamPerThread));
#endif // CUDA >= 8.0
}
}