[Webkit-unassigned] [Bug 90057] New: llint regresses performance on some v8 benchmarks

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=90057

           Summary: llint regresses performance on some v8 benchmarks
           Product: WebKit
           Version: 528+ (Nightly build)
          Platform: Unspecified
        OS/Version: Unspecified
            Status: NEW
          Severity: Normal
          Priority: P2
         Component: JavaScriptCore
        AssignedTo: webkit-unassigned at lists.webkit.org
        ReportedBy: wingo at igalia.com


The LLInt is faster than the classic interpreter, overall, but it is not faster on all benchmarks.

I configured a build with runtime heuristics and the classic interpreter.  Like this:

CPPFLAGS='-DENABLE_RUN_TIME_HEURISTICS=1 -DENABLE_LLINT=0 -DENABLE_CLASSIC_INTERPRETER=1' Tools/Scripts/build-jsc --gtk

I configured another build with runtime heuristics and the LLInt:

CPPFLAGS='-DENABLE_RUN_TIME_HEURISTICS=1 -DENABLE_LLINT=1 -DENABLE_CLASSIC_INTERPRETER=0' Tools/Scripts/build-jsc --gtk

For the first build (classic interpreter):

~/src/v8/benchmarks$ JSC_useJIT=false ~/src/WebKit/WebKitBuild/Release/Programs/jsc-3 run.js 
Richards: 969
DeltaBlue: 1004
Crypto: 762
RayTrace: 2821
EarleyBoyer: 3342
RegExp: 396
Splay: 4458
NavierStokes: 1248
----
Score (version 7): 1407

For the second (llint):

~/src/v8/benchmarks$ JSC_useJIT=false ~/src/WebKit/WebKitBuild/Release/Programs/jsc-3 run.js 
Richards: 783
DeltaBlue: 713
Crypto: 1397
RayTrace: 2415
EarleyBoyer: 3344
RegExp: 389
Splay: 3936
NavierStokes: 2015
----
Score (version 7): 1449

You can see that the LLInt does really well with math-intensive benchmarks (crypto, navier-stokes) but does not do so well on the benchmarks that test property access (deltablue, richards, splay).

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.