Changeset 203329 in webkit


Ignore:
Timestamp:
Jul 16, 2016 5:19:41 PM (8 years ago)
Author:
Chris Dumez
Message:

Unreviewed, rolling out r203318.

Regressed most JS Benchmarks on MacBook Air by ~2% (7% on
SunSpider)

Reverted changeset:

"[JSC] Change some parameters based on a random search"
https://bugs.webkit.org/show_bug.cgi?id=158514
http://trac.webkit.org/changeset/203318

Location:
trunk
Files:
5 edited

Legend:

Unmodified
Added
Removed
  • trunk/Source/JavaScriptCore/ChangeLog

    r203320 r203329  
     12016-07-16  Chris Dumez  <cdumez@apple.com>
     2
     3        Unreviewed, rolling out r203318.
     4
     5        Regressed most JS Benchmarks on MacBook Air by ~2% (7% on
     6        SunSpider)
     7
     8        Reverted changeset:
     9
     10        "[JSC] Change some parameters based on a random search"
     11        https://bugs.webkit.org/show_bug.cgi?id=158514
     12        http://trac.webkit.org/changeset/203318
     13
    1142016-07-15  Benjamin Poulain  <bpoulain@apple.com>
    215
  • trunk/Source/JavaScriptCore/bytecode/CodeBlock.cpp

    r203318 r203329  
    36943694double CodeBlock::optimizationThresholdScalingFactor()
    36953695{
    3696     // We want a good threshold based on the instruction count.
    3697     // Here, we are trying to optimize the following formula:
    3698     //     F[x_] =: a * Sqrt[x + b] + Abs[c * x] + d
    3699     // The parameters were chosen by testing random values
    3700     // between 1 and 2 and keeping the best combination.
    3701     const double a = Options::optimizationThresholdScalingFactorA();
    3702     const double b = Options::optimizationThresholdScalingFactorB();
    3703     const double c = Options::optimizationThresholdScalingFactorC();
    3704     const double d = Options::optimizationThresholdScalingFactorD();
    3705 
     3696    // This expression arises from doing a least-squares fit of
     3697    //
     3698    // F[x_] =: a * Sqrt[x + b] + Abs[c * x] + d
     3699    //
     3700    // against the data points:
     3701    //
     3702    //    x       F[x_]
     3703    //    10       0.9          (smallest reasonable code block)
     3704    //   200       1.0          (typical small-ish code block)
     3705    //   320       1.2          (something I saw in 3d-cube that I wanted to optimize)
     3706    //  1268       5.0          (something I saw in 3d-cube that I didn't want to optimize)
     3707    //  4000       5.5          (random large size, used to cause the function to converge to a shallow curve of some sort)
     3708    // 10000       6.0          (similar to above)
     3709    //
     3710    // I achieve the minimization using the following Mathematica code:
     3711    //
     3712    // MyFunctionTemplate[x_, a_, b_, c_, d_] := a*Sqrt[x + b] + Abs[c*x] + d
     3713    //
     3714    // samples = {{10, 0.9}, {200, 1}, {320, 1.2}, {1268, 5}, {4000, 5.5}, {10000, 6}}
     3715    //
     3716    // solution =
     3717    //     Minimize[Plus @@ ((MyFunctionTemplate[#[[1]], a, b, c, d] - #[[2]])^2 & /@ samples),
     3718    //         {a, b, c, d}][[2]]
     3719    //
     3720    // And the code below (to initialize a, b, c, d) is generated by:
     3721    //
     3722    // Print["const double " <> ToString[#[[1]]] <> " = " <>
     3723    //     If[#[[2]] < 0.00001, "0.0", ToString[#[[2]]]] <> ";"] & /@ solution
     3724    //
     3725    // We've long known the following to be true:
     3726    // - Small code blocks are cheap to optimize and so we should do it sooner rather
     3727    //   than later.
     3728    // - Large code blocks are expensive to optimize and so we should postpone doing so,
     3729    //   and sometimes have a large enough threshold that we never optimize them.
     3730    // - The difference in cost is not totally linear because (a) just invoking the
     3731    //   DFG incurs some base cost and (b) for large code blocks there is enough slop
     3732    //   in the correlation between instruction count and the actual compilation cost
     3733    //   that for those large blocks, the instruction count should not have a strong
     3734    //   influence on our threshold.
     3735    //
     3736    // I knew the goals but I didn't know how to achieve them; so I picked an interesting
     3737    // example where the heuristics were right (code block in 3d-cube with instruction
     3738    // count 320, which got compiled early as it should have been) and one where they were
     3739    // totally wrong (code block in 3d-cube with instruction count 1268, which was expensive
     3740    // to compile and didn't run often enough to warrant compilation in my opinion), and
     3741    // then threw in additional data points that represented my own guess of what our
     3742    // heuristics should do for some round-numbered examples.
     3743    //
     3744    // The expression to which I decided to fit the data arose because I started with an
     3745    // affine function, and then did two things: put the linear part in an Abs to ensure
     3746    // that the fit didn't end up choosing a negative value of c (which would result in
     3747    // the function turning over and going negative for large x) and I threw in a Sqrt
     3748    // term because Sqrt represents my intution that the function should be more sensitive
     3749    // to small changes in small values of x, but less sensitive when x gets large.
     3750   
     3751    // Note that the current fit essentially eliminates the linear portion of the
     3752    // expression (c == 0.0).
     3753    const double a = 0.061504;
     3754    const double b = 1.02406;
     3755    const double c = 0.0;
     3756    const double d = 0.825914;
     3757   
    37063758    double instructionCount = this->instructionCount();
    37073759   
  • trunk/Source/JavaScriptCore/runtime/Options.h

    r203318 r203329  
    9898    unsigned m_highLimit;
    9999};
    100 
    101 #if CPU(X86_64)
    102 constexpr int32_t archThresholdForJITAfterWarmUp = 610;
    103 constexpr int32_t archThresholdForJITSoon = 89;
    104 constexpr int32_t archThresholdForOptimizeAfterWarmUp = 864;
    105 constexpr int32_t archThresholdForOptimizeAfterLongWarmUp = 1489;
    106 constexpr int32_t archThresholdForOptimizeSoon = 864;
    107 constexpr int32_t archExecutionCounterIncrementForLoop = 2;
    108 constexpr int32_t archExecutionCounterIncrementForEntry = 16;
    109 constexpr int32_t archThresholdForFTLOptimizeAfterWarmUp = 109160;
    110 constexpr int32_t archThresholdForFTLOptimizeSoon = 825;
    111 constexpr int32_t archFtlTierUpCounterIncrementForLoop = 8;
    112 constexpr int32_t archFtlTierUpCounterIncrementForReturn = 27;
    113 constexpr unsigned archFtlOSREntryRetryThreshold = 109;
    114 constexpr double archOptimizationThresholdScalingFactorA = 0.0258587392234135;
    115 constexpr double archOptimizationThresholdScalingFactorB = 1.2428014544978696;
    116 constexpr double archOptimizationThresholdScalingFactorC = 0.0013276440240339;
    117 constexpr double archOptimizationThresholdScalingFactorD = 1.3130654609331458;
    118 constexpr int32_t archEvalThresholdMultiplier = 12;
    119 #else
    120 constexpr int32_t archThresholdForJITAfterWarmUp = 500;
    121 constexpr int32_t archThresholdForJITSoon = 100;
    122 constexpr int32_t archThresholdForOptimizeAfterWarmUp = 1000;
    123 constexpr int32_t archThresholdForOptimizeAfterLongWarmUp = 1000;
    124 constexpr int32_t archThresholdForOptimizeSoon = 1000;
    125 constexpr int32_t archExecutionCounterIncrementForLoop = 1;
    126 constexpr int32_t archExecutionCounterIncrementForEntry = 15;
    127 constexpr int32_t archThresholdForFTLOptimizeAfterWarmUp = 100000;
    128 constexpr int32_t archThresholdForFTLOptimizeSoon = 1000;
    129 constexpr int32_t archFtlTierUpCounterIncrementForLoop = 1;
    130 constexpr int32_t archFtlTierUpCounterIncrementForReturn = 15;
    131 constexpr unsigned archFtlOSREntryRetryThreshold = 100;
    132 constexpr double archOptimizationThresholdScalingFactorA = 0.061504;
    133 constexpr double archOptimizationThresholdScalingFactorB = 1.02406;
    134 constexpr double archOptimizationThresholdScalingFactorC = 0.0;
    135 constexpr double archOptimizationThresholdScalingFactorD = 0.825914;
    136 constexpr int32_t archEvalThresholdMultiplier = 10;
    137 #endif
    138100
    139101typedef OptionRange optionRange;
     
    292254    v(double, jitPolicyScale, 1.0, Normal, "scale JIT thresholds to this specified ratio between 0.0 (compile ASAP) and 1.0 (compile like normal).") \
    293255    v(bool, forceEagerCompilation, false, Normal, nullptr) \
    294     v(int32, thresholdForJITAfterWarmUp, archThresholdForJITAfterWarmUp, Normal, nullptr) \
    295     v(int32, thresholdForJITSoon, archThresholdForJITSoon, Normal, nullptr) \
    296     \
    297     v(int32, thresholdForOptimizeAfterWarmUp, archThresholdForOptimizeAfterWarmUp, Normal, nullptr) \
    298     v(int32, thresholdForOptimizeAfterLongWarmUp, archThresholdForOptimizeAfterLongWarmUp, Normal, nullptr) \
    299     v(int32, thresholdForOptimizeSoon, archThresholdForOptimizeSoon, Normal, nullptr) \
    300     v(int32, executionCounterIncrementForLoop, archExecutionCounterIncrementForLoop, Normal, nullptr) \
    301     v(int32, executionCounterIncrementForEntry, archExecutionCounterIncrementForEntry, Normal, nullptr) \
    302     \
    303     v(int32, thresholdForFTLOptimizeAfterWarmUp, archThresholdForFTLOptimizeAfterWarmUp, Normal, nullptr) \
    304     v(int32, thresholdForFTLOptimizeSoon, archThresholdForFTLOptimizeSoon, Normal, nullptr) \
    305     v(int32, ftlTierUpCounterIncrementForLoop, archFtlTierUpCounterIncrementForLoop, Normal, nullptr) \
    306     v(int32, ftlTierUpCounterIncrementForReturn, archFtlTierUpCounterIncrementForReturn, Normal, nullptr) \
     256    v(int32, thresholdForJITAfterWarmUp, 500, Normal, nullptr) \
     257    v(int32, thresholdForJITSoon, 100, Normal, nullptr) \
     258    \
     259    v(int32, thresholdForOptimizeAfterWarmUp, 1000, Normal, nullptr) \
     260    v(int32, thresholdForOptimizeAfterLongWarmUp, 1000, Normal, nullptr) \
     261    v(int32, thresholdForOptimizeSoon, 1000, Normal, nullptr) \
     262    v(int32, executionCounterIncrementForLoop, 1, Normal, nullptr) \
     263    v(int32, executionCounterIncrementForEntry, 15, Normal, nullptr) \
     264    \
     265    v(int32, thresholdForFTLOptimizeAfterWarmUp, 100000, Normal, nullptr) \
     266    v(int32, thresholdForFTLOptimizeSoon, 1000, Normal, nullptr) \
     267    v(int32, ftlTierUpCounterIncrementForLoop, 1, Normal, nullptr) \
     268    v(int32, ftlTierUpCounterIncrementForReturn, 15, Normal, nullptr) \
    307269    v(unsigned, ftlOSREntryFailureCountForReoptimization, 15, Normal, nullptr) \
    308     v(unsigned, ftlOSREntryRetryThreshold, archFtlOSREntryRetryThreshold, Normal, nullptr) \
    309     \
    310     v(double, optimizationThresholdScalingFactorA, archOptimizationThresholdScalingFactorA, Normal, nullptr) \
    311     v(double, optimizationThresholdScalingFactorB, archOptimizationThresholdScalingFactorB, Normal, nullptr) \
    312     v(double, optimizationThresholdScalingFactorC, archOptimizationThresholdScalingFactorC, Normal, nullptr) \
    313     v(double, optimizationThresholdScalingFactorD, archOptimizationThresholdScalingFactorD, Normal, nullptr) \
    314     \
    315     v(int32, evalThresholdMultiplier, archEvalThresholdMultiplier, Normal, nullptr) \
     270    v(unsigned, ftlOSREntryRetryThreshold, 100, Normal, nullptr) \
     271    \
     272    v(int32, evalThresholdMultiplier, 10, Normal, nullptr) \
    316273    v(unsigned, maximumEvalCacheableSourceLength, 256, Normal, nullptr) \
    317274    \
  • trunk/Tools/ChangeLog

    r203325 r203329  
     12016-07-16  Chris Dumez  <cdumez@apple.com>
     2
     3        Unreviewed, rolling out r203318.
     4
     5        Regressed most JS Benchmarks on MacBook Air by ~2% (7% on
     6        SunSpider)
     7
     8        Reverted changeset:
     9
     10        "[JSC] Change some parameters based on a random search"
     11        https://bugs.webkit.org/show_bug.cgi?id=158514
     12        http://trac.webkit.org/changeset/203318
     13
    1142016-07-16  Chris Dumez  <cdumez@apple.com>
    215
  • trunk/Tools/Scripts/run-jsc-stress-tests

    r203318 r203329  
    430430BASE_OPTIONS = ["--useFTLJIT=false", "--useFunctionDotArguments=true", "--maxPerThreadStackUsage=1572864"]
    431431EAGER_OPTIONS = ["--thresholdForJITAfterWarmUp=10", "--thresholdForJITSoon=10", "--thresholdForOptimizeAfterWarmUp=20", "--thresholdForOptimizeAfterLongWarmUp=20", "--thresholdForOptimizeSoon=20", "--thresholdForFTLOptimizeAfterWarmUp=20", "--thresholdForFTLOptimizeSoon=20", "--maximumEvalCacheableSourceLength=150000"]
    432 NO_CJIT_OPTIONS = ["--useConcurrentJIT=false", "--thresholdForJITAfterWarmUp=100", "--thresholdForJITSoon=100", "--thresholdForOptimizeAfterWarmUp=1000", "--thresholdForOptimizeAfterLongWarmUp=1000", "--thresholdForOptimizeSoon=1000", "--executionCounterIncrementForLoop=1", "--executionCounterIncrementForEntry=15", "--thresholdForFTLOptimizeAfterWarmUp=100000", "--thresholdForFTLOptimizeSoon=1000", "--ftlTierUpCounterIncrementForLoop=1", "--ftlTierUpCounterIncrementForReturn=15", "--evalThresholdMultiplier=10", "--optimizationThresholdScalingFactorA=0.061504", "--optimizationThresholdScalingFactorB=1.02406", "--optimizationThresholdScalingFactorC=0.0", "--optimizationThresholdScalingFactorD=0.825914"]
     432NO_CJIT_OPTIONS = ["--useConcurrentJIT=false", "--thresholdForJITAfterWarmUp=100"]
    433433FTL_OPTIONS = ["--useFTLJIT=true"]
    434434
Note: See TracChangeset for help on using the changeset viewer.