[GitHub] [incubator-mxnet] larroy commented on issue #14522: mx.nd.Custom conflicts with memory management

[GitBox <gi...@apache.org>]

larroy commented on issue #14522: mx.nd.Custom conflicts with memory management
URL: https://github.com/apache/incubator-mxnet/issues/14522#issuecomment-477697395
 
 
   Hi
   
   I'm investigating this one. With the provided example I believe the cause of the error is the following at least in CPU context (changing gpu to cpu context in the example above).
   
   ```
   I believe the current example crashes in Gemm due to an integer overflow in GEMM in the following code in dot_engine-inl.h
   
      
       for (int i = 0; i < batch_count; ++i) {
         gemm(stream, transa, transb, m, n, k, alpha,
              A + i * m * k, lda, B + i * k * n, ldb,
              beta, C + i * m * n, ldc);
       }
   
   When having a gemm with dimension 6000 * 7000 * 100 which is > 2^32 we have an overflow which leads to the crash.
    I would suggest checking matrix dimensions for overflow before dispatching to GEMM.
   ```

----------------------------------------------------------------
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.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services