Refactoring of THCNumerics #679
Open
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
commit ec4653659f0a0301da5c8aa533d7592ffdb79329 Author: Boris Fomitchev <[email protected]> Date: Mon Jan 16 18:35:45 2017 -0800 cleanup commit b6771be Author: Boris Fomitchev <[email protected]> Date: Sun Jan 15 16:03:40 2017 -0800 Enabling half instructions for selected Pascal archs commit 18f5ce1 Author: Boris Fomitchev <[email protected]> Date: Sun Jan 15 15:31:38 2017 -0800 cleanup commit 0114d65 Author: Boris Fomitchev <[email protected]> Date: Sun Jan 15 02:16:14 2017 -0800 Build working commit 5b7ed04 Merge: 03a77af 16fb59f Author: Boris Fomitchev <[email protected]> Date: Fri Jan 13 18:44:52 2017 -0800 Merge remote-tracking branch 'upstream/master' into spec_refactor 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 commit 03a77af Merge: ed0f6b4 2876894 Author: Boris Fomitchev <[email protected]> Date: Wed Nov 23 00:59:51 2016 -0800 Merge branch 'android' into spec_refactor commit ed0f6b4 Merge: b572cf9 2d75d41 Author: Boris Fomitchev <[email protected]> Date: Mon Nov 21 17:28:04 2016 -0800 Merge branch 'upstream' into spec_refactor commit 2876894 Author: Boris Fomitchev <[email protected]> Date: Thu Nov 17 03:08:21 2016 -0800 Added C++ flags commit 9328d7d Merge: 80df9d9 10ef056 Author: Boris Fomitchev <[email protected]> Date: Wed Nov 16 21:49:32 2016 -0800 Merge branch 'cpu_half' into android commit 80df9d9 Merge: 0afffe1 9a6ba41 Author: Boris Fomitchev <[email protected]> Date: Wed Nov 16 21:25:06 2016 -0800 Merge branch 'getmeminfo-fix' commit 9a6ba41 Author: Boris Fomitchev <[email protected]> Date: Tue Nov 15 14:48:52 2016 -0800 Bugfix, test extended commit 9aef731 Author: Boris Fomitchev <[email protected]> Date: Tue Nov 15 02:46:49 2016 -0800 Added some memory allocations to test_shutdown to test acching allocator commit 838ec70 Author: Boris Fomitchev <[email protected]> Date: Tue Nov 15 02:29:53 2016 -0800 Implemented cudaMemGetInfo for caching allocator commit b572cf9 Merge: 38882e6 10ef056 Author: Boris Fomitchev <[email protected]> Date: Tue Nov 1 13:02:47 2016 -0700 Merge branch 'cpu_half' into spec_refactor commit 10ef056 Author: Boris Fomitchev <[email protected]> Date: Sat Oct 29 15:54:48 2016 -0700 Using half from TH commit 1838317 Author: Boris Fomitchev <[email protected]> Date: Fri Oct 28 17:13:04 2016 -0700 checkpoint commit 38882e6 Merge: acaf1a6 21ad069 Author: Boris Fomitchev <[email protected]> Date: Fri Oct 21 00:42:28 2016 -0700 cutorch refactoring commit acaf1a6 Author: Boris Fomitchev <[email protected]> Date: Sun Oct 16 22:27:04 2016 -0700 Checkpoint commit ac964ee Author: Boris Fomitchev <[email protected]> Date: Fri Oct 14 01:03:47 2016 -0700 checkpoint commit 8a75e34 Merge: a80d28b ada3cbd Author: Boris Fomitchev <[email protected]> Date: Wed Oct 12 17:43:51 2016 -0700 Merge remote-tracking branch 'upstream/master' into spec_refactor commit a80d28b Author: Boris Fomitchev <[email protected]> Date: Wed Oct 12 17:41:43 2016 -0700 checkpoint commit 79a9006 Merge: 6933617 9efd392 Author: Boris Fomitchev <[email protected]> Date: Mon Oct 10 12:42:21 2016 -0700 Merge remote-tracking branch 'upstream/master' commit 6933617 Merge: b928ca0 afd74e7 Author: Boris Fomitchev <[email protected]> Date: Sun Oct 2 21:35:17 2016 -0700 Merge remote-tracking branch 'upstream/master' commit b928ca0 Author: Boris Fomitchev <[email protected]> Date: Fri Sep 30 02:03:56 2016 -0700 temporarily force pseudo-fp16 mode (HAS_HALF_INSTRUCTIONS=FALSE) for Pascal: torch#520
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.
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.