[
https://issues.apache.org/jira/browse/ARROW-8516?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Raphael Taylor-Davies updated ARROW-8516:
-----------------------------------------
Description:
{color:#000000}BufferBuilder{color}{color:#0073bf}<{color}{color:#000000}BooleanType>::append_slice
is called by PrimitiveBuilder{color}{color:#000000}::append_slice with a
constructed vector of true values. {color}
{color:#000000}Even in release builds the associated allocations and iterations
are not optimised out, resulting in a third of the time to parse a parquet file
containing single integers being spent in PrimitiveBuilder::append_slice.{color}
{color:#000000}This PR adds an append_n method to the BufferBuilderTrait that
allows this to be handled more efficiently. My rather unscientific testing
shows it to halve the amount of time spent in this method yielding an ~20%
speedup for my particular workload.{color}
was:
{color:#000000}BufferBuilder{color}{color:#0073bf}<{color}{color:#000000}BooleanType>::append_slice
is called by ArrayBuilder::append_slice with a constructed vector of true
values. {color}
{color:#000000}Even in release builds the associated allocations and iterations
are not optimised out, resulting in a third of the time to parse a parquet file
containing single integers being spent in PrimitiveBuilder::append_slice.{color}
{color:#000000}This PR adds an append_n method to the BufferBuilderTrait that
allows this to be handled more efficiently. My rather unscientific testing
shows it to halve the amount of time spent in this method yielding an ~20%
speedup for my particular workload.
{color}
> Slow BufferBuilder<BooleanType> inserts
> ---------------------------------------
>
> Key: ARROW-8516
> URL: https://issues.apache.org/jira/browse/ARROW-8516
> Project: Apache Arrow
> Issue Type: Improvement
> Reporter: Raphael Taylor-Davies
> Priority: Trivial
>
> {color:#000000}BufferBuilder{color}{color:#0073bf}<{color}{color:#000000}BooleanType>::append_slice
> is called by PrimitiveBuilder{color}{color:#000000}::append_slice with a
> constructed vector of true values. {color}
> {color:#000000}Even in release builds the associated allocations and
> iterations are not optimised out, resulting in a third of the time to parse a
> parquet file containing single integers being spent in
> PrimitiveBuilder::append_slice.{color}
> {color:#000000}This PR adds an append_n method to the BufferBuilderTrait that
> allows this to be handled more efficiently. My rather unscientific testing
> shows it to halve the amount of time spent in this method yielding an ~20%
> speedup for my particular workload.{color}
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
