rapidsai/cudf

[FEA] Add `bytes_per_second` to all libcudf benchmarks

Open

#13 735 ouverte le 23 juil. 2023

Voir sur GitHub
 (8 commentaires) (0 réactions) (1 assigné)C++ (735 forks)batch import
0 - Backlogfeature requestgood first issuelibcudf

Métriques du dépôt

Stars
 (6 000 stars)
Métriques de merge PR
 (Merge moyen 17j 21h) (230 PRs mergées en 30 j)

Description

Many libcudf benchmarks report the bytes_per_second processed as part of the output data. This value is useful for comparing benchmarks because it normalizes the increasing execution time from processing more data. Also, bytes_per_second and real_time_s together let us compute bytes_processed values which serve as a useful Y-axis.

As of the end of 23.12 development, many benchmarks still do not report bytes_per_second in the output data. Here is a figure summarizing the metrics reported by the benchmark suite.

image

benchmark status notes
APPLY_BOOLEAN_MASK see #13937
BINARYOP see #13938
COPY_IF_ELSE see #13960
GROUPBY see #13984
GROUPBY_NV
HASHING see #13967, #13965
JOIN
JOIN_NV
QUANTILES
REDUCTION see #16126
REPLACE
SEARCH
SEARCH_NV
SET_OPS_NV
SHIFT see #13950
SORT
SORT_NV
STREAM_COMPACTION_NV see #14172
TRANSPOSE see #14170

Note: For this tracking list, cuIO benchmarks are omitted because "encoded file size" serves a similar purpose.

Guide contributeur