Vukasin Milovanovic comments

Results 139 comments of


                                            Vukasin Milovanovic

C++ Test

CC @bdice @kingcrimsontianyu

[Minor] Extend the Parquet writer's dictionary encoding benchmark.

what's the reason for this change?

WIP: Improve cudf::io::datasource::create().

Apologies for the slow response; we've been thoroughly reevaluating our IO options, in part prompted by your PR :) We're planning to drastically change the available IO options to avoid...

WIP: Improve cudf::io::datasource::create().

closing as the datasource code has changed quite a bit and we're not even quite done with changes to enable this feature.

[FEA] Improve cudf::io::datasource::create()

Thank you for the detailed proposal! One question to better understand the WekaFS situation - what is the preferred method to perform IO in this case? GDS?

[FEA] Improve Parquet decoding throughput in libcudf

There's an ongoing effort to remove purging of non-empty nulls from the column factories - https://github.com/rapidsai/cudf/pull/12873. Should further improve throughput for lists and structs columns.

Make random data in Python tests deterministic

CC @galipremsagar

Multi-column null sanitization for struct columns

> > /ok to test [19c37f2](https://github.com/rapidsai/cudf/commit/19c37f23ad02cf5c1c9a1cf26c12100ed5fc7f2f) > > Apparently your commits are signed. I wonder if you already are `cpp-codeowner`? There's one that's not verified :(

[FEA] Cache schema tree in `chunked_parquet_writer`

We would still want to check that the schemas match, right? Is that much faster than building the schema tree?

Improve parallel POSIX read performance

> It may be worth having a page-locked memory pool, or at least pre-allocating nthreads * task_size page-locked memory block. How much memory would be required for this? Can we...