You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@arrow.apache.org by GitBox <gi...@apache.org> on 2020/05/27 04:07:51 UTC
[GitHub] [arrow] wesm edited a comment on pull request #7267: ARROW-8772: [C++] Unrolled aggregate dense for better speculative execution
wesm edited a comment on pull request #7267:
URL: https://github.com/apache/arrow/pull/7267#issuecomment-633600133
This shows a small benefit on ARM64 architecture (ThunderX1):
before:
```
-----------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
-----------------------------------------------------------------------------
SumKernel/32768/0 19.5 us 19.5 us 35883 bytes_per_second=1.56853G/s null_percent=0 size=32.768k
SumKernel/32768/1 19.2 us 19.2 us 36172 bytes_per_second=1.58548G/s null_percent=1 size=32.768k
SumKernel/32768/10 21.7 us 21.7 us 32230 bytes_per_second=1.40723G/s null_percent=10 size=32.768k
SumKernel/32768/50 22.4 us 22.4 us 31172 bytes_per_second=1.36246G/s null_percent=50 size=32.768k
```
after (see the Int64 benchmarks):
```
-----------------------------------------------------------------------------------
Benchmark Time CPU Iterations UserCounters...
-----------------------------------------------------------------------------------
SumKernelFloat/32768/0 25.4 us 25.4 us 27590 bytes_per_second=1.20019G/s null_percent=0 size=32.768k
SumKernelFloat/32768/1 41.2 us 41.2 us 16973 bytes_per_second=758.715M/s null_percent=1 size=32.768k
SumKernelFloat/32768/10 49.0 us 49.0 us 14285 bytes_per_second=638.254M/s null_percent=10 size=32.768k
SumKernelFloat/32768/50 51.5 us 51.5 us 13625 bytes_per_second=606.657M/s null_percent=50 size=32.768k
SumKernelDouble/32768/0 18.5 us 18.5 us 37953 bytes_per_second=1.65038G/s null_percent=0 size=32.768k
SumKernelDouble/32768/1 26.2 us 26.2 us 26852 bytes_per_second=1.16631G/s null_percent=1 size=32.768k
SumKernelDouble/32768/10 27.3 us 27.3 us 25685 bytes_per_second=1.11778G/s null_percent=10 size=32.768k
SumKernelDouble/32768/50 27.7 us 27.7 us 25298 bytes_per_second=1.10219G/s null_percent=50 size=32.768k
SumKernelInt8/32768/0 64.4 us 64.4 us 10857 bytes_per_second=485.21M/s null_percent=0 size=32.768k
SumKernelInt8/32768/1 69.8 us 69.8 us 10025 bytes_per_second=448.008M/s null_percent=1 size=32.768k
SumKernelInt8/32768/10 92.7 us 92.7 us 7554 bytes_per_second=337.084M/s null_percent=10 size=32.768k
SumKernelInt8/32768/50 100 us 100 us 6974 bytes_per_second=311.424M/s null_percent=50 size=32.768k
SumKernelInt16/32768/0 41.0 us 41.0 us 17076 bytes_per_second=762.469M/s null_percent=0 size=32.768k
SumKernelInt16/32768/1 42.5 us 42.5 us 16513 bytes_per_second=735.553M/s null_percent=1 size=32.768k
SumKernelInt16/32768/10 53.7 us 53.7 us 12977 bytes_per_second=581.673M/s null_percent=10 size=32.768k
SumKernelInt16/32768/50 57.7 us 57.7 us 12103 bytes_per_second=541.264M/s null_percent=50 size=32.768k
SumKernelInt32/32768/0 26.0 us 26.0 us 26859 bytes_per_second=1.17154G/s null_percent=0 size=32.768k
SumKernelInt32/32768/1 26.5 us 26.5 us 26555 bytes_per_second=1.15362G/s null_percent=1 size=32.768k
SumKernelInt32/32768/10 32.3 us 32.3 us 21653 bytes_per_second=967.29M/s null_percent=10 size=32.768k
SumKernelInt32/32768/50 34.2 us 34.2 us 20513 bytes_per_second=914.616M/s null_percent=50 size=32.768k
SumKernelInt64/32768/0 18.0 us 18.0 us 39059 bytes_per_second=1.69949G/s null_percent=0 size=32.768k
SumKernelInt64/32768/1 19.3 us 19.3 us 36220 bytes_per_second=1.58067G/s null_percent=1 size=32.768k
SumKernelInt64/32768/10 21.8 us 21.8 us 32171 bytes_per_second=1.40112G/s null_percent=10 size=32.768k
SumKernelInt64/32768/50 22.5 us 22.5 us 31062 bytes_per_second=1.35471G/s null_percent=50 size=32.768k
```
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org