Pointer Array batch mode unlike the traditionally supported strided batch mode, allows individual matrices in non-contiguous memory locations to be part of batch by passing an array of device pointers. But what happens for the scale A pointer and other scaling factor pointers for other matrices. Does each matrix in the batch have a unique scaling factor and hence even this becomes a pointer array or does a single scaling factor is shared across the batch. Please advise.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| New cuBlas function "getrfBatched" | 5 | 8664 | January 21, 2014 | |
| CUDA Pro Tip: How to Call Batched cuBLAS routines from CUDA Fortran | 4 | 443 | April 27, 2015 | |
| How can I call cublasSgemmBatched on pointer device arrays without allocating them twice? | 3 | 1618 | August 2, 2018 | |
| Way to covert pointer (*d_A) to array (*d_Array []) | 4 | 659 | April 3, 2019 | |
| Confusion with cublas pointer mode | 0 | 914 | May 15, 2015 | |
| Try calling cublasgemmBatch in OpenACC | 2 | 711 | June 26, 2023 | |
| CUBLAS Batched Matrix Multiplies? | 4 | 6826 | March 12, 2012 | |
| Solve one dense linear system in one thread block | 4 | 771 | January 20, 2017 | |
| CUBLAS operating on different parts of an array CUBLAS based code development | 2 | 8132 | August 8, 2010 | |
| Batch Matrix Multiplication using CuBLAS | 17 | 4041 | March 2, 2021 |