https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/Packed-GEMM-APIs-and-dynamic-batch-size/m-p/1137822#M26153 <P>Some answers here: <A href="https://github.com/intel/mkl-dnn/issues/537">https://github.com/intel/mkl-dnn/issues/537</A></P> Mon, 02 Sep 2019 15:11:59 GMT guillaumekln 2019-09-02T15:11:59Z https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/Packed-GEMM-APIs-and-dynamic-batch-size/m-p/1137821#M26152 <P>Hi,</P><P>I'm interested in further&nbsp;optimizing my application using the packed GEMM API. However, I'm unclear how it behaves in the case of dynamic batch sizes. For example,</P><UL><LI>X, the input of shape [M, K] where M is the batch size</LI><LI>W, the weight of shape [N, K]</LI></UL><P>The GEMM function should compute X*W<SUP>T</SUP> where W can be packed as it remains constant.</P><P>How does a change in M affect the packed representation of W? Do cblas_gemm_*_compute functions silently repack W if any of M, N, K is different? Or should it be done manually?</P><P>Thanks,</P><P>Guillaume</P> Fri, 02 Aug 2019 14:45:54 GMT https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/Packed-GEMM-APIs-and-dynamic-batch-size/m-p/1137821#M26152 guillaumekln 2019-08-02T14:45:54Z https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/Packed-GEMM-APIs-and-dynamic-batch-size/m-p/1137822#M26153 <P>Some answers here: <A href="https://github.com/intel/mkl-dnn/issues/537">https://github.com/intel/mkl-dnn/issues/537</A></P> Mon, 02 Sep 2019 15:11:59 GMT https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/Packed-GEMM-APIs-and-dynamic-batch-size/m-p/1137822#M26153 guillaumekln 2019-09-02T15:11:59Z