viiccwen opened a new issue, #940: URL: https://github.com/apache/mahout/issues/940
### Description The current batched L2 norm CUDA kernel performs vectorized loads starting from input_batch + base, where: https://github.com/apache/mahout/blob/d5fc18bd01bd4fbaa6c947252b142f38c9d8b720/qdp/qdp-kernels/src/amplitude.cu#L367C5-L367C49 When sample_len is odd and sample_idx is also odd, base becomes odd, meaning input_batch + base is only 8-byte aligned. This violates the alignment requirements of vectorized loads: double2 requires 16-byte alignment. Although this does not cause incorrect results, it introduces avoidable performance penalties and makes the kernel’s memory access behavior architecture-dependent. reference: [comment](https://github.com/apache/mahout/pull/918#discussion_r2724245973) -- 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]
