merge with torch #1
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
reduce and BLAS work
Pass CMAKE_CXX_COMPILER to new select_compute_arch.cmake
This allows to compile THC without cutorch
Move CMake files to THC
scatter / gather to new types
Fix sub and div for integer types
Index* and sort for new types
Use TH_INDEX_BASE in THC
Fixing bug in regex not accepting 2.1(2.0) notation
adding multiple types to squeeze
Magma functions to generic
Implement fmod, remainder, equal in Cutorch
fix memory leak in (equal)
Add half support for addmv and addr.
guard random functions for half
Previously, cutorch would initialize every CUDA device and enable P2P access between all pairs. This slows down start-up, especially with 8 devices. Now, THCudaInit does not initialize any devices and P2P access is enabled lazily. Setting the random number generator seed also does not initialize the device until random numbers are actually used.
* Implemented cudaMemGetInfo for caching allocator
Lazily initialize CUDA devices
Revert "Lazily initialize CUDA devices"
Previously, cutorch would initialize every CUDA device and enable P2P access between all pairs. This slows down start-up, especially with 8 devices. Now, THCudaInit does not initialize any devices and P2P access is enabled lazily. Setting the random number generator seed also does not initialize the device until random numbers are actually used.
Lazily initialize CUDA devices (take 2)
use local modified select_compute_arch.cmake for msvc
Adds a CUDA "sleep" kernel which spins for the given number of iterations. This is useful for testing correct synchronization with streams.
Adds a caching allocator for CUDA pinned (page-locked) memory. This avoid synchronization due to cudaFreeHost or cudaHostUnregister at the expense of potentially higher host memory usage. Correctness is preserved by recording CUDA events after each cudaMemcpyAsync involving the pinned memory. The pinned memory allocations are not reused until all events associated with it have completed.
Add caching allocator for pinned (page-locked) memory
Without this, the cuda_events could continuously grow from calls to cudaMemcpyAsync, but would never be processed if there were no new pinned memory allocations. For example: t1 = cutorch.createCudaHostTensor(10) t2 = torch.CudaTensor(10) while true do t2:copyAsync(t1) end
Process outstanding CUDA events in recordEvent
TensorInfo related code documentation
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.