diff options
| author | Florian Fischer <florian.fl.fischer@fau.de> | 2019-10-16 18:31:53 +0200 |
|---|---|---|
| committer | Florian Fischer <florian.fl.fischer@fau.de> | 2019-10-16 18:31:53 +0200 |
| commit | 451632863ec751c66ecf989975ab32bdbb8cb750 (patch) | |
| tree | eca1b682efec8a5496739940a83234600ddfe2c8 /src/benchmarks/espresso.py | |
| parent | f075d76ed72f1f3472e997456fa5287b344059d3 (diff) | |
| download | allocbench-451632863ec751c66ecf989975ab32bdbb8cb750.tar.gz allocbench-451632863ec751c66ecf989975ab32bdbb8cb750.zip | |
improve analysis: add cycles portions and fix top10 percentages
Diffstat (limited to 'src/benchmarks/espresso.py')
| -rw-r--r-- | src/benchmarks/espresso.py | 12 |
1 files changed, 8 insertions, 4 deletions
diff --git a/src/benchmarks/espresso.py b/src/benchmarks/espresso.py index 4ffd843..68e474c 100644 --- a/src/benchmarks/espresso.py +++ b/src/benchmarks/espresso.py @@ -22,7 +22,11 @@ The file "largest.espresso" shipped with mimalloc-bench and allocbench generates a workload with 3367364 allocator calls (malloc: 1659385, free: 1691851, realloc: 16128). About 87% of all allocations are smaller than 64 Byte, the common cache line size. -Top 10 allocation sizes 0.91% of all allocations +Allocator portion of total cycles measured using perf record/report: +malloc 8.64% +free 5.04% + +Top 10 allocation sizes 90.73% of all allocations 1. 48 B occurred 615622 times 2. 16 B occurred 533267 times 3. 56 B occurred 235944 times @@ -34,9 +38,9 @@ Top 10 allocation sizes 0.91% of all allocations 9. 272 B occurred 14644 times 10. 96 B occurred 13175 times -allocations <= 64 1442648 -allocations <= 1024 1657509 -allocations <= 4096 1667112 +allocations <= 64 1442648 86.10% +allocations <= 1024 1657509 98.93% +allocations <= 4096 1667112 99.50% The relevant non functional allocator properties are the raw speed of the API function as well as memory placement strategies with good data locality. |