| Summary: | bmalloc: Use List<T> instead of Vector<T> in some places | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Product: | WebKit | Reporter: | Geoffrey Garen <ggaren> | ||||||
| Component: | New Bugs | Assignee: | Geoffrey Garen <ggaren> | ||||||
| Status: | RESOLVED FIXED | ||||||||
| Severity: | Normal | CC: | barraclough, cgarcia, clopez, commit-queue, ggaren, gyuyoung.kim, jh718.park, kling, oliver, ossy, ryanhaddad, sam | ||||||
| Priority: | P2 | ||||||||
| Version: | WebKit Nightly Build | ||||||||
| Hardware: | Unspecified | ||||||||
| OS: | Unspecified | ||||||||
| Bug Depends on: | 155171 | ||||||||
| Bug Blocks: | |||||||||
| Attachments: |
|
||||||||
|
Description
Geoffrey Garen
2016-03-07 17:04:48 PST
Created attachment 273251 [details]
Patch
~/OpenSource/PerformanceTests/MallocBench> ./run-malloc-benchmarks Baseline:~/OpenSource/WebKitBuildBaseline/Release/ Patch:~/OpenSource/WebKitBuild/Release/
Baseline Patch Δ
Execution Time:
churn 84ms 83ms ^ 1.01x faster
list_allocate 74ms 73ms ^ 1.01x faster
tree_allocate 83ms 80ms ^ 1.04x faster
tree_churn 85ms 86ms ! 1.01x slower
fragment 76ms 74ms ^ 1.03x faster
fragment_iterate 61ms 62ms ! 1.02x slower
medium 197ms 191ms ^ 1.03x faster
big 145ms 145ms
facebook 126ms 122ms ^ 1.03x faster
reddit 103ms 101ms ^ 1.02x faster
flickr 103ms 104ms ! 1.01x slower
theverge 137ms 135ms ^ 1.01x faster
nimlang 14ms 13ms ^ 1.08x faster
message_one 221ms 212ms ^ 1.04x faster
message_many 120ms 115ms ^ 1.04x faster
churn --parallel 39ms 38ms ^ 1.03x faster
list_allocate --parallel 73ms 66ms ^ 1.11x faster
tree_allocate --parallel 85ms 87ms ! 1.02x slower
tree_churn --parallel 87ms 86ms ^ 1.01x faster
fragment --parallel 53ms 49ms ^ 1.08x faster
fragment_iterate --parallel 32ms 31ms ^ 1.03x faster
medium --parallel 187ms 189ms ! 1.01x slower
big --parallel 135ms 134ms ^ 1.01x faster
<geometric mean> 87ms 85ms ^ 1.02x faster
<arithmetic mean> 101ms 99ms ^ 1.02x faster
<harmonic mean> 69ms 66ms ^ 1.04x faster
Peak Memory:
churn 804kB 788kB ^ 1.02x smaller
list_allocate 2,120kB 2,088kB ^ 1.02x smaller
tree_allocate 5,492kB 5,448kB ^ 1.01x smaller
tree_churn 4,800kB 4,756kB ^ 1.01x smaller
fragment 7,004kB 6,996kB ^ 1.0x smaller
fragment_iterate 25,596kB 25,576kB ^ 1.0x smaller
medium 1,068,420kB 1,068,416kB ^ 1.0x smaller
big 1,060,464kB 1,060,456kB ^ 1.0x smaller
facebook 19,616kB 19,212kB ^ 1.02x smaller
reddit 9,944kB 9,460kB ^ 1.05x smaller
flickr 15,492kB 15,028kB ^ 1.03x smaller
theverge 18,696kB 18,280kB ^ 1.02x smaller
nimlang 18,872kB 18,340kB ^ 1.03x smaller
message_one 4,776kB 4,608kB ^ 1.04x smaller
message_many 4,904kB 4,980kB ! 1.02x bigger
churn --parallel 968kB 936kB ^ 1.03x smaller
list_allocate --parallel 2,176kB 2,164kB ^ 1.01x smaller
tree_allocate --parallel 3,360kB 3,268kB ^ 1.03x smaller
tree_churn --parallel 3,004kB 2,992kB ^ 1.0x smaller
fragment --parallel 7,168kB 7,164kB ^ 1.0x smaller
fragment_iterate --parallel 25,736kB 25,736kB
medium --parallel 1,066,128kB 1,067,544kB ! 1.0x bigger
big --parallel 1,058,660kB 1,054,484kB ^ 1.0x smaller
<geometric mean> 14,931kB 14,736kB ^ 1.01x smaller
<arithmetic mean> 192,791kB 192,553kB ^ 1.0x smaller
<harmonic mean> 4,316kB 4,237kB ^ 1.02x smaller
Memory at End:
churn 392kB 380kB ^ 1.03x smaller
list_allocate 424kB 396kB ^ 1.07x smaller
tree_allocate 468kB 432kB ^ 1.08x smaller
tree_churn 452kB 420kB ^ 1.08x smaller
fragment 456kB 448kB ^ 1.02x smaller
fragment_iterate 640kB 628kB ^ 1.02x smaller
medium 4,612kB 4,612kB
big 4,596kB 4,592kB ^ 1.0x smaller
facebook 3,028kB 2,640kB ^ 1.15x smaller
reddit 2,164kB 1,684kB ^ 1.29x smaller
flickr 3,148kB 2,692kB ^ 1.17x smaller
theverge 3,168kB 2,764kB ^ 1.15x smaller
nimlang 18,208kB 17,684kB ^ 1.03x smaller
message_one 684kB 768kB ! 1.12x bigger
message_many 1,148kB 1,104kB ^ 1.04x smaller
churn --parallel 556kB 528kB ^ 1.05x smaller
list_allocate --parallel 604kB 576kB ^ 1.05x smaller
tree_allocate --parallel 700kB 672kB ^ 1.04x smaller
tree_churn --parallel 1,136kB 1,096kB ^ 1.04x smaller
fragment --parallel 1,276kB 1,280kB ! 1.0x bigger
fragment_iterate --parallel 844kB 852kB ! 1.01x bigger
medium --parallel 4,748kB 4,752kB ! 1.0x bigger
big --parallel 4,724kB 4,712kB ^ 1.0x smaller
<geometric mean> 1,366kB 1,303kB ^ 1.05x smaller
<arithmetic mean> 2,529kB 2,422kB ^ 1.04x smaller
<harmonic mean> 905kB 868kB ^ 1.04x smaller
=====
~/OpenSource/PerformanceTests/MallocBench> ./run-malloc-benchmarks Baseline:~/OpenSource/WebKitBuildBaseline/Release/ Patch:~/OpenSource/WebKitBuild/Release/
Baseline Patch Δ
Peak Memory:
reddit_memory_warning --runs 0 9,868kB 9,340kB ^ 1.06x smaller
flickr_memory_warning --runs 0 15,428kB 14,892kB ^ 1.04x smaller
theverge_memory_warning --runs 0 18,612kB 18,200kB ^ 1.02x smaller
<geometric mean> 14,151kB 13,629kB ^ 1.04x smaller
<arithmetic mean> 14,636kB 14,144kB ^ 1.03x smaller
<harmonic mean> 13,644kB 13,091kB ^ 1.04x smaller
Memory at End:
reddit_memory_warning --runs 0 7,828kB 7,304kB ^ 1.07x smaller
flickr_memory_warning --runs 0 10,312kB 9,796kB ^ 1.05x smaller
theverge_memory_warning --runs 0 16,920kB 16,520kB ^ 1.02x smaller
<geometric mean> 11,095kB 10,573kB ^ 1.05x smaller
<arithmetic mean> 11,687kB 11,207kB ^ 1.04x smaller
<harmonic mean> 10,570kB 10,016kB ^ 1.06x smaller
Created attachment 273252 [details]
Patch
The performance numbers posted above disable large allocations in recorded benchmarks.
Here are the results with large allocations enabled:
~/OpenSource/PerformanceTests/MallocBench> ./run-malloc-benchmarks Baseline:~/OpenSource/WebKitBuildBaseline/Release/ Patch:~/OpenSource/WebKitBuild/Release/
Baseline Patch Δ
Execution Time:
churn 85ms 82ms ^ 1.04x faster
list_allocate 73ms 72ms ^ 1.01x faster
tree_allocate 83ms 81ms ^ 1.02x faster
tree_churn 87ms 84ms ^ 1.04x faster
fragment 76ms 71ms ^ 1.07x faster
fragment_iterate 60ms 60ms
medium 206ms 188ms ^ 1.1x faster
big 147ms 141ms ^ 1.04x faster
facebook 248ms 251ms ! 1.01x slower
reddit 120ms 119ms ^ 1.01x faster
flickr 126ms 122ms ^ 1.03x faster
theverge 156ms 153ms ^ 1.02x faster
nimlang 135ms 136ms ! 1.01x slower
message_one 223ms 212ms ^ 1.05x faster
message_many 127ms 115ms ^ 1.1x faster
churn --parallel 38ms 38ms
list_allocate --parallel 77ms 65ms ^ 1.18x faster
tree_allocate --parallel 86ms 84ms ^ 1.02x faster
tree_churn --parallel 85ms 84ms ^ 1.01x faster
fragment --parallel 51ms 49ms ^ 1.04x faster
fragment_iterate --parallel 32ms 32ms
medium --parallel 191ms 188ms ^ 1.02x faster
big --parallel 133ms 133ms
<geometric mean> 101ms 98ms ^ 1.03x faster
<arithmetic mean> 115ms 111ms ^ 1.03x faster
<harmonic mean> 87ms 85ms ^ 1.03x faster
Peak Memory:
churn 804kB 792kB ^ 1.02x smaller
list_allocate 2,104kB 2,092kB ^ 1.01x smaller
tree_allocate 5,480kB 5,456kB ^ 1.0x smaller
tree_churn 4,800kB 4,760kB ^ 1.01x smaller
fragment 7,004kB 7,000kB ^ 1.0x smaller
fragment_iterate 25,596kB 25,580kB ^ 1.0x smaller
medium 1,068,420kB 1,068,416kB ^ 1.0x smaller
big 1,060,464kB 1,060,460kB ^ 1.0x smaller
facebook 73,948kB 73,428kB ^ 1.01x smaller
reddit 13,836kB 13,368kB ^ 1.04x smaller
flickr 25,848kB 25,316kB ^ 1.02x smaller
theverge 27,016kB 26,436kB ^ 1.02x smaller
nimlang 217,332kB 216,900kB ^ 1.0x smaller
message_one 4,876kB 4,636kB ^ 1.05x smaller
message_many 14,288kB 4,956kB ^ 2.88x smaller
churn --parallel 956kB 940kB ^ 1.02x smaller
list_allocate --parallel 2,200kB 2,200kB
tree_allocate --parallel 3,352kB 3,296kB ^ 1.02x smaller
tree_churn --parallel 3,004kB 3,000kB ^ 1.0x smaller
fragment --parallel 7,168kB 7,148kB ^ 1.0x smaller
fragment_iterate --parallel 25,760kB 25,720kB ^ 1.0x smaller
medium --parallel 1,065,664kB 1,064,996kB ^ 1.0x smaller
big --parallel 1,051,744kB 1,058,760kB ! 1.01x bigger
<geometric mean> 19,424kB 18,386kB ^ 1.06x smaller
<arithmetic mean> 204,855kB 204,594kB ^ 1.0x smaller
<harmonic mean> 4,557kB 4,388kB ^ 1.04x smaller
Memory at End:
churn 392kB 380kB ^ 1.03x smaller
list_allocate 408kB 396kB ^ 1.03x smaller
tree_allocate 456kB 436kB ^ 1.05x smaller
tree_churn 452kB 420kB ^ 1.08x smaller
fragment 456kB 448kB ^ 1.02x smaller
fragment_iterate 640kB 628kB ^ 1.02x smaller
medium 4,608kB 4,608kB
big 4,596kB 4,592kB ^ 1.0x smaller
facebook 3,320kB 2,804kB ^ 1.18x smaller
reddit 2,212kB 1,740kB ^ 1.27x smaller
flickr 3,248kB 2,724kB ^ 1.19x smaller
theverge 3,324kB 2,752kB ^ 1.21x smaller
nimlang 34,188kB 33,788kB ^ 1.01x smaller
message_one 684kB 672kB ^ 1.02x smaller
message_many 1,256kB 1,080kB ^ 1.16x smaller
churn --parallel 544kB 528kB ^ 1.03x smaller
list_allocate --parallel 612kB 564kB ^ 1.09x smaller
tree_allocate --parallel 712kB 660kB ^ 1.08x smaller
tree_churn --parallel 1,136kB 1,172kB ! 1.03x bigger
fragment --parallel 1,332kB 1,328kB ^ 1.0x smaller
fragment_iterate --parallel 864kB 836kB ^ 1.03x smaller
medium --parallel 4,736kB 4,736kB
big --parallel 4,684kB 4,716kB ! 1.01x bigger
<geometric mean> 1,422kB 1,340kB ^ 1.06x smaller
<arithmetic mean> 3,255kB 3,131kB ^ 1.04x smaller
<harmonic mean> 908kB 865kB ^ 1.05x smaller
=====
~/OpenSource/PerformanceTests/MallocBench> ./run-malloc-benchmarks Baseline:~/OpenSource/WebKitBuildBaseline/Release/ Patch:~/OpenSource/WebKitBuild/Release/
Baseline Patch Δ
Peak Memory:
reddit_memory_warning --runs 0 13,804kB 13,360kB ^ 1.03x smaller
flickr_memory_warning --runs 0 26,032kB 25,560kB ^ 1.02x smaller
theverge_memory_warning --runs 0 26,920kB 26,468kB ^ 1.02x smaller
<geometric mean> 21,307kB 20,830kB ^ 1.02x smaller
<arithmetic mean> 22,252kB 21,796kB ^ 1.02x smaller
<harmonic mean> 20,270kB 19,769kB ^ 1.03x smaller
Memory at End:
reddit_memory_warning --runs 0 9,284kB 8,828kB ^ 1.05x smaller
flickr_memory_warning --runs 0 12,320kB 11,848kB ^ 1.04x smaller
theverge_memory_warning --runs 0 21,212kB 20,740kB ^ 1.02x smaller
<geometric mean> 13,437kB 12,945kB ^ 1.04x smaller
<arithmetic mean> 14,272kB 13,805kB ^ 1.03x smaller
<harmonic mean> 12,711kB 12,200kB ^ 1.04x smaller
Comment on attachment 273252 [details]
Patch
r=me!
Committed r197722: <http://trac.webkit.org/changeset/197722> (In reply to comment #6) > Committed r197722: <http://trac.webkit.org/changeset/197722> It made many JSC stress tests crash on 32 bit Linux platforms, see the X86 GTK and the EFL ARMv7 bots on build.webkit.org for details. (In reply to comment #7) > (In reply to comment #6) > > Committed r197722: <http://trac.webkit.org/changeset/197722> > > It made many JSC stress tests crash on 32 bit Linux platforms, see the X86 > GTK and the EFL ARMv7 bots on build.webkit.org for details. and on the 32 bit x86 Apple Mac bots too - Apple Yosemite 32-bit JSC (BuildAndTest) https://build.webkit.org/builders/Apple%20Yosemite%2032-bit%20JSC%20%28BuildAndTest%29/builds/8322 - Apple El Capitan 32-bit JSC (BuildAndTest) https://build.webkit.org/builders/Apple%20El%20Capitan%2032-bit%20JSC%20%28BuildAndTest%29/builds/1705 - GTK Linux 32-bit Release https://build.webkit.org/builders/GTK%20Linux%2032-bit%20Release/builds/59651 - EFL Linux ARMv7 Thumb2 Release https://build.webkit.org/builders/EFL%20Linux%20ARMv7%20Thumb2%20Release/builds/17406 Please fix this serious regression. ping Re-opened since this is blocked by bug 155171 Committed r197797: <http://trac.webkit.org/changeset/197797> |