Skip to content

WIP: Added performance based Kernels along with general kernels to ha… - #67

Open
nicolin wants to merge 5 commits into
mainfrom
WIP_CUDA_KernelDev
Open

WIP: Added performance based Kernels along with general kernels to ha…#67
nicolin wants to merge 5 commits into
mainfrom
WIP_CUDA_KernelDev

Conversation

@nicolin

@nicolin nicolin commented Jan 12, 2026

Copy link
Copy Markdown
Collaborator

…ndle any node/edge connection size, along with WL2 matrix coloring and started with def iso check based on WL2 matrix coloring

…ndle any node/edge connection size, along with WL2 matrix coloring and started with def iso check based on WL2 matrix coloring
@nicolin nicolin self-assigned this Jan 12, 2026
@nicolin nicolin added the enhancement New feature or request label Jan 12, 2026
blaze added 3 commits January 15, 2026 16:10
… is still WIP.  The new WP2 test that colors the adjacency matrix seems to work, just tested that it did not crash and converged. Also finished the CompareHisto between two graphs for both WL class tests using the same function. Have to update Cuda free flags to be outside the checks due to early exit when it is production
… and working on coloring using WL1-2-3 to distinguish nodes into color bins via a conanical mapping to color intergers from hash space, selecting optmial pivots to split bins, it uses a direct compare of node profiles to check if splitting is valid followed by a full check of expanded edges using a hybrid cpu/gpu check when bins are split into nodes along with single step backtracking in the case of hashing or structural weakness of the color from the WL2/3 test leading to an incorrect node mapping for a pivot. The full method is in a word document, It has been tested with a number of cases in the input folder along with WL defeating graphs such as rook, rings, Shrikhande and Miyazaki, along with typical proof checker graphs. Work still remains on checking against isomorphic to non-isomorphic that will pass beyond the WL3 and enter the IR depth search where it will fail. Along with checks on the HostVerify and tests that may break the hash and color or other tests from proof checker input.
@nicolin

nicolin commented Feb 11, 2026

Copy link
Copy Markdown
Collaborator Author

Final push as I move off the project.
Work done for the past 16 working days, solver is fully GPU optimized and working on coloring using WL1-2-3 to distinguish nodes into color bins via a canonical mapping to color integers from hash space, selecting optimal pivots to split bins, it uses a direct compare of node profiles to check if splitting is valid followed by a full check of expanded edges using a hybrid cpu/gpu check when bins are split into nodes along with single step backtracking in the case of hashing or structural weakness of the color from the WL2/3 test leading to an incorrect node mapping for a pivot. The full method is in a word document, It has been tested with a number of cases in the input folder along with WL defeating graphs such as rook, rings, Shrikhande and Miyazaki, along with typical proof checker graphs. Work still remains on checking against isomorphic to non-isomorphic that will pass beyond the WL3 and enter the IR depth search where it will fail. Along with checks on the HostVerify and tests that may break the hash and color or other tests from proof checker input.

@nicolin
nicolin marked this pull request as ready for review February 11, 2026 21:53
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant