Use cuFile for Parquet IO when available #7444
Conversation
kkraus14
requested review from
devavret
and removed request for
trxcllnt and
jrhemstad
|
reassigned review to @devavret since it's cuIO related |
cpp/src/io/parquet/writer_impl.cu
Outdated
| cudaFreeHost}; | ||
| } | ||
| }(); | ||
| auto host_bfr = pinned_buffer<uint8_t>{[](size_t size) { |
There was a problem hiding this comment.
cudaMallocHost is expensive. There was an optimization to not do it when using device write. Why was it removed?
There was a problem hiding this comment.
Not intentional, this is probably because this change was here before the optimization and I messed up a merge. Will restore the behavior.
There was a problem hiding this comment.
Fixed, now the buffer is lazily allocated on first use.
|
Updating the |
|
@kkraus14 @vuule FYI I just merged rapidsai/gpuci-build-environment#166 - it should take ~1hr for the images to build and be available for use in gpuCI |
|
Rerun tests |
|
@gpucibot merge |
|
@vuule awesome!!! So we should be able to try on nightly's tmw+? |
|
Nightly's available now should be able to be tested. You'll need to install GDS from here: https://developer.nvidia.com/gpudirect-storage Then there's the environment variable of
By default cuFile / GDS will not be used and our existing IO code will be used. |
|
Amazing, thanks all :) Starting on the build process here, and will hopefully have our first benchmarks over the weekend / early next. Will share on Slack when up :) |
Adds optional cuFile integration:
cufile.his included in the build when available.libcufile.sois loaded at runtime ifLIBCUDF_CUFILE_POLICYenvironment variable is set to "ALWAYS" or "GDS".device_read.