Artem Kroviakov

Element-wise multiplication

1

**Describe the bug** Element-wise multiplication of SparseTensor is not consistent with documentation. Documentation states: _If you add two sparse tensors, this will add two features. In case where there is...

[L0] Asynchronous data fetching

1

This PR introduces an asynchronous (batched) data fetching for L0 GPUs. Its purpose is to reduce end-to-end execution time of a workload. ____________ ### Why? We have recursive materializations (from...

This PR introduces an index structure for free buffers of a slab, this allows keep data fetching time pretty much constant. Example: 1000 fragments, 15 columns, (we observe GPU as...

device-linear-multifrag execution mode

16

As of now, HDK's heterogeneity looks like this: - We have X fragments, when we schedule them on a GPU, it will receive X kernels, X fragments and execute kernels...

The code in this branch is supposed to link into separately built shared library and currently represents the HDK side of talking to a shared library. The current (ugly) workflow...