[REVIEW] FEA subgraph modifications #1
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
Moves the JSON interface from using the Writer / SAX version of RapipJSON's interface to the Document version. This allows sorting of the document and throwing assertions for invalid JSON. Previously we could write incomplete files when a job was interrupted. Now incomplete jobs will have an elapsed time of -1, but otherwise be valid JSON. Similarly some keys could have values of NaN or INF, which caused the previous interface to write empty string for their value. This is because RapidJSON disallows NaN and INF (the JSON spec doesn't include them, but Javascript and other interfaces are more permissive). Now these invalid values will be written as null to the resulting document.
Initial JSON writer improvements
The shared_libs examples need to be linked with a C++ compiler since they link against a C++ library. Previously, cmake properly used g++ as expected. It's not clear why, but CMake decided to start using gcc for these files because they end in ".c". This may be due to a change to one of the parent CMakeLists.txt files, but it's difficult to find why this behavior changed. This change explicitly tells CMake to use the C++ linker with the set_target_properties command.
Workaround for the PageRank Validation SegFault (issue #684 and #683)
* added mem calculation * update google test for sm * don't use cpu reference when using quick mode * add compression * revert two files * revert one more files
RepeatFor0 uses whole grid sync operations. These are only supported on Pascal and Volta architectures - SM 6.0 or greater. See https://stackoverflow.com/a/48436451 Prevents an issue when compiling Google tests (unittests subdirectory) for other architectures.
My previous commit broke when SM = 7.0. I think that when compiling both host and device side, the host side version no longer had the kernels defined but the device side did. The fix is to keep the kernel defined in all compilations but #ifdef out the use of this_grid.
* Prevent duplication of key "load-time" in output json * Deleting commented out code for setting loadtime in json
Updating dev with master's bug-fixes and changes.
After updating moderngpu from 1.0 -> 2.0 we forgot to make sure that the mgpu calls were on the same stream as our other kernels. Operators using mgpu's SortedSearch and Scan functions expected them to be complete before the next kernel call, but since the mgpu context was on the default stream it executed in parallel with the other kernels. This caused the output_offsets array used in the following kernel to be incorrect, and out of bounds memory accesses would occur. This fix ensures that the mgpu::standard_context_t is setup to use the same stream as the other kernels. To do so, we keep a context pointer, allocate it on the heap when the operator parameters are intialized, and free it when they are released.
Add target device option in subgraph matching
Add the option to print to a file
Reviewer Request Scripts & Docs updated
Updating dev (docs, broken links, and scripts)
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.