tqchen opened a new pull request, #14674: URL: https://github.com/apache/tvm/pull/14674
This PR improves webgpu codegen to handle large launch grid. Background: webgpu do not allow grid size bigger than 65535 so we have to factorize the gridDim.x when it is too big and spread it across gridDim.x and gridDim.z. This approach however is not always possible. This PR pass in extra parameter packDimX which records the original requested dim overpad if factorization is not possible and immediately returns if the index is out of bound -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
