RFR (L): 7153771: array bound check elimination for c1
Vladimir Kozlov
vladimir.kozlov at oracle.com
Thu Feb 21 15:44:22 PST 2013
Roland
The research paper said about 55% improvement for scimark.LU. You got
18.7%. There is big effect of cpu cache size on scimark results. Could
you rerun scimark with small matrix size? :
benchscripts/scimark/run
< scimark_args="-large"
---
> scimark_args=""
Thanks,
Vladimir
On 2/20/13 8:56 AM, Roland Westrelin wrote:
> I forgot the perf results.
>
> Roland.
>
> ============================================================================
> tserver-base-client: reference_server
> Benchmark Samples Mean Stdev
> jetstream 10 71.18 0.01
> Copy 10 923.50 0.03
> Parse 10 55.50 0.01
> Read 10 24.00 0.03
> Write 10 317.10 0.02
> scimark 10 307.82 0.00
> Sparse 10 219.86 0.01
> LU 10 535.98 0.00
> SOR 10 622.70 0.00
> FFT 10 32.32 0.00
> Monte 10 128.23 0.00
> specjbb2000 10 309060.88 0.01
> First_Warehouse 10 45224.53 0.00
> Last_Warehouse 10 309060.89 0.01
> specjbb2005 10 129684.88 0.01
> peak 10 137062.81 0.01
> peak_warehouse 10 4.00 0.00
> last 10 129684.88 0.01
> interval_average 10 7628.70 0.01
> first 10 19314.00 0.00
> overall_average 10 108614.46 0.01
> last_warehouse 10 8.00 0.00
> specjvm98 10 477.99 0.00
> javac 10 319.46 0.00
> db 10 344.64 0.00
> jess 10 381.30 0.00
> jack 10 379.54 0.01
> compress 10 505.53 0.00
> mtrt 10 1018.25 0.02
> mpegaudio 10 695.23 0.00
> volano25 10 171270.09 0.01
> connections 10 400.00 0.00
> time 10 4.67 0.01
> --------------------------------------------------------------------------
> Weighted Geomean 12848.65
> ============================================================================
> tserver-abce-client: reference_server
> Benchmark Samples Mean Stdev %Diff P Significant
> jetstream 10 73.59 0.01 3.38 0.000 Yes
> Copy 10 908.20 0.04 1.66 0.324 *
> Parse 10 53.50 0.01 3.60 0.000 Yes
> Read 10 22.20 0.02 7.50 0.000 Yes
> Write 10 316.60 0.02 0.16 0.834 *
> scimark 10 343.38 0.00 11.56 0.000 Yes
> Sparse 10 231.70 0.01 5.39 0.000 Yes
> LU 10 636.11 0.00 18.68 0.000 Yes
> SOR 10 685.85 0.00 10.14 0.000 Yes
> FFT 10 32.43 0.00 0.32 0.000 Yes
> Monte 10 130.84 0.00 2.03 0.000 Yes
> specjbb2000 10 313336.28 0.01 1.38 0.013 *
> First_Warehouse 10 45396.47 0.00 0.38 0.060 *
> Last_Warehouse 10 313336.27 0.01 1.38 0.013 *
> specjbb2005 10 135469.84 0.01 4.46 0.000 Yes
> peak 10 143285.94 0.00 4.54 0.000 Yes
> peak_warehouse 10 4.00 0.00 -0.00 0.000 *
> last 10 135469.84 0.01 4.46 0.000 Yes
> interval_average 10 7968.80 0.01 4.46 0.000 Yes
> first 10 20434.39 0.00 5.80 0.000 Yes
> overall_average 10 113769.13 0.00 4.75 0.000 Yes
> last_warehouse 10 8.00 0.00 -0.00 0.000 *
> specjvm98 10 496.53 0.00 3.88 0.000 Yes
> javac 10 319.00 0.00 -0.14 0.442 *
> db 10 346.65 0.00 0.58 0.000 Yes
> jess 10 397.70 0.00 4.30 0.000 Yes
> jack 10 386.73 0.01 1.90 0.000 Yes
> compress 10 531.25 0.00 5.09 0.000 Yes
> mtrt 10 1017.30 0.02 -0.09 0.924 *
> mpegaudio 10 809.66 0.00 16.46 0.000 Yes
> volano25 10 170535.09 0.01 -0.43 0.338 *
> connections 10 400.00 0.00 0.00 0.000 *
> time 10 4.69 0.01 -0.43 0.342 *
> --------------------------------------------------------------------------
> Weighted Geomean 13340.24 3.83
> ============================================================================
>
>
More information about the hotspot-compiler-dev
mailing list