review request (M): 7121140: Allocation paths require explicit memory synchronization operations for RMO systems

[Tom Rodriguez]

On Dec 15, 2011, at 6:45 AM, Roland Westrelin wrote:

>> I wouldn't expect this to affect performance since it's not emitting an extra instructions but we are emitting some extra barriers which tangles of the memory graph a bit.  Could you do some quick performance to make sure nothing's changed?
> 
> Below are the results. 

Could you do a longer set of specjvm98 to confirm the jess difference?  Looking at the data, I don't really think it's real but another, longer set of runs should tell us that.  Try 20.

tom

> 
> Roland.
> 
> ============================================================================
> tt1
>  Benchmark         Samples        Mean     Stdev
>  specjbb2005            10   196409.50      0.01
>    peak                 10   219067.61      0.02
>    peak_warehouse       10        4.00      0.00
>    last                 10   196409.49      0.01
>    interval_average     10    11553.50      0.01
>    first                10    35924.81      0.01
>    overall_average      10   167704.10      0.01
>    last_warehouse       10        8.00      0.00
> ============================================================================
> tt2
>  Benchmark         Samples        Mean     Stdev   %Diff    P   Significant
>  specjbb2005            10   196016.36      0.01   -0.20 0.737            *
>    peak                 10   216797.36      0.02   -1.04 0.262            *
>    peak_warehouse       10        4.00      0.00   -0.00 0.000            *
>    last                 10   196016.35      0.01   -0.20 0.737            *
>    interval_average     10    11530.40      0.01   -0.20 0.737            *
>    first                10    35558.66      0.00   -1.02 0.000          Yes
>    overall_average      10   167822.07      0.01    0.07 0.852            *
>    last_warehouse       10        8.00      0.00   -0.00 0.000            *
> ============================================================================
> 
> ============================================================================
> t1
>  Benchmark         Samples        Mean     Stdev
>  specjvm98              10      667.76      0.00
>    javac                10      392.03      0.01
>    db                   10      438.88      0.00
>    jess                 10      700.73      0.00
>    jack                 10      594.09      0.01
>    compress             10      536.31      0.00
>    mtrt                 10     1808.20      0.02
>    mpegaudio            10      852.60      0.01
> ============================================================================
> t2
>  Benchmark         Samples        Mean     Stdev   %Diff    P   Significant
>  specjvm98              10      666.48      0.00   -0.19 0.183            *
>    javac                10      391.46      0.01   -0.15 0.827            *
>    db                   10      439.52      0.00    0.15 0.501            *
>    jess                 10      701.00      0.01    0.04 0.870            *
>    jack                 10      589.77      0.00   -0.73 0.006          Yes
>    compress             10      537.17      0.01    0.16 0.534            *
>    mtrt                 10     1788.32      0.02   -1.10 0.229            *
>    mpegaudio            10      855.14      0.01    0.30 0.468            *
> ============================================================================
>