Refactoring of THCNumerics #678
Closed
Conversation
This file contains hidden or 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
Conflicts: CMakeLists.txt Tensor.lua generic/CStorage.c generic/CTensor.c init.c lib/THC/THCCachingAllocator.cpp lib/THC/THCGeneral.c lib/THC/THCGenerateHalfType.h lib/THC/THCHalf.cu lib/THC/THCStorageCopy.c lib/THC/THCStorageCopy.cu lib/THC/THCTensorCopy.c lib/THC/generic/THCStorageCopy.c lib/THC/generic/THCStorageCopy.cu lib/THC/generic/THCStorageCopy.h lib/THC/generic/THCTensorCopy.h test/test_shutdown.lua
Contributor
Author
|
Replaced by #679 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
Here, this refactoring was initially driven by idea of supporting both pseudo and native fp16 for a single arch. That support proved to be unfeasible, but the refactoring made the code much cleaner and shorter, take a look.
Also, CUDA_HALF_INSTRUCTIONS compile-time switch is turned on for 6.0 and 6.2+ architectures in this version - if you encounter any new issues with accuracy on Pascal, try turning it off.