Performance

From OpenCV on the Cell

This page describes the comparison of PPE and SPE execution times.

The conditions for measurement were as follows:

• The number of SPE is 6.
• Image size is 640x480 pixels.
• The result is an average of 10 executions.

If you are interested in benchmark program, get and run them according to the following. These programs need Ruby and eRuby:

$ svn co https://cvcell.svn.sourceforge.net/svnroot/cvcell/trunk/benchmark
$ cd image-benchmark
$ make bench-spe bench-ppe bench-spe-pure

and run viewresult.rb

$ ruby viewresult.rb
           cvAbsDiff,16u :               11.437                3.606   3.2
            cvAbsDiff,8u :                8.655                2.329   3.7
           cvAbsDiff,32s :                9.625                7.348   1.3
               cvAdd,16u :                8.255                0.625  13.2
                cvAdd,8u :               10.431                0.484  21.6
               cvAdd,32s :                8.831                0.822  10.7
       cvAddWeighted,16u :               61.750                9.951   6.2
        cvAddWeighted,8u :               15.854                5.098   3.1
       cvAddWeighted,32s :               50.629               15.874   3.2
               cvAnd,16u :                5.850                1.389   4.2

This table is generated by 'gentable.rb' program. 'gentable.rb' is contained in image-benchmark

Target functions		Performance				Comment
		PPE original code [ms]	SPE optimized code [ms]	Ratio 0 100 % fast slow <--- --->	Pure SPE (without PPE-SPE overhead) [ms]
cvAbsDiff	16u	2.356	0.182	7.7 %	0.134
	8u	3.350	0.138	4.1 %	0.090
	32s	3.629	0.282	7.8 %	0.227
cvAbsDiffS	16u	3.075	0.203	6.6 %	0.151
	8u	4.354	0.135	3.1 %	0.086
	32s	2.466	0.302	12.2 %	0.252
cvAdd	16u	2.881	0.185	6.4 %	0.134
	8u	3.556	0.139	3.9 %	0.090
	32s	3.251	0.279	8.6 %	0.226
cvAddS	16u	2.591	0.198	7.6 %	0.149
	8u	4.057	0.135	3.3 %	0.087
	32s	1.905	0.303	15.9 %	0.247
cvAddWeighted	16u	23.663	0.653	2.8 %	0.149
	8u	5.171	0.446	8.6 %	0.088
	32s	19.934	0.990	5.0 %	0.250
cvAnd	16u	1.658	0.180	10.9 %	0.133
	8u	0.828	0.139	16.8 %	0.090
	32s	3.273	0.281	8.6 %	0.225
cvAndS	16u	1.225	0.213	17.4 %	0.148
	8u	0.530	0.159	30.0 %	0.086
	32s	2.092	0.302	14.4 %	0.248
cvCalcArrHist	32x1	6.558	1.004	15.3 %	0.908
	32x32	6.538	1.058	16.2 %	0.910
cvCmp	cmpop0	3.514	0.143	4.1 %	0.091
	cmpop1	3.355	0.139	4.1 %	0.090
cvCmpS	cmpop0	3.165	0.136	4.3 %	0.089
	cmpop1	4.965	0.136	2.7 %	0.089
cvConvertScale	8u32s	3.275	0.545	16.6 %	0.504
	16u8u	10.485	0.431	4.1 %	0.397
	8u16u	3.687	0.406	11.0 %	0.373
	32s8u	12.714	0.690	5.4 %	0.642
cvCvtColor	BGR2GRAY	3.460	0.207	6.0 %	0.154
	BGR2YCrCb	16.488	1.988	12.1 %	1.924
	BGR2HSV	12.978	0.683	5.3 %	0.624
	BGR2Lab	24.602	3.031	12.3 %	2.967
	GRAY2BGR	1.441	0.768	53.3 %	0.703
cvDilate	ksize=3,ch=1,shape=rect	5.344	0.205	3.8 %	0.114
	ksize=7,ch=3,shape=cross	49.044	0.834	1.7 %	0.732
	ksize=7,ch=1,shape=ellipse	39.792	0.462	1.2 %	0.367
	ksize=5,ch=1,shape=ellipse	21.152	0.292	1.4 %	0.204
	ksize=7,ch=3,shape=rect	30.499	0.310	1.0 %	0.220
	ksize=9,ch=3,shape=cross	60.694	0.995	1.6 %	0.893
	ksize=9,ch=1,shape=cross	20.240	0.433	2.1 %	0.338
	ksize=9,ch=3,shape=ellipse	200.637	1.463	0.7 %	1.353
	ksize=7,ch=3,shape=ellipse	119.366	1.088	0.9 %	0.982
	ksize=9,ch=3,shape=rect	37.507	0.361	1.0 %	0.267
	ksize=9,ch=1,shape=rect	12.625	0.256	2.0 %	0.166
	ksize=7,ch=1,shape=rect	10.147	0.234	2.3 %	0.147
	ksize=5,ch=1,shape=rect	7.663	0.213	2.8 %	0.128
	ksize=7,ch=1,shape=cross	16.526	0.376	2.3 %	0.285
	ksize=5,ch=3,shape=ellipse	63.040	0.601	1.0 %	0.494
	ksize=3,ch=3,shape=ellipse	21.067	0.341	1.6 %	0.247
	ksize=3,ch=1,shape=ellipse	7.089	0.208	2.9 %	0.122
	ksize=5,ch=3,shape=cross	35.018	0.522	1.5 %	0.424
	ksize=5,ch=1,shape=cross	11.712	0.268	2.3 %	0.181
	ksize=9,ch=1,shape=ellipse	66.979	0.593	0.9 %	0.491
	ksize=5,ch=3,shape=rect	23.243	0.297	1.3 %	0.179
	ksize=3,ch=3,shape=cross	21.103	0.344	1.6 %	0.247
	ksize=3,ch=1,shape=cross	7.066	0.210	3.0 %	0.122
	ksize=3,ch=3,shape=rect	16.293	0.243	1.5 %	0.150
cvDiv	16u	2.732	0.211	7.7 %	0.160
	8u	2.280	0.218	9.6 %	0.168
	32s	2.512	0.285	11.3 %	0.232
cvDotProduct	32f	3.470	0.247	7.1 %	0.212
	64f	6.069	0.727	12.0 %	0.686
cvEqualizeHist	8u	8.382	0.832	9.9 %	0.327
cvErode	ksize=3,ch=1,shape=rect	5.078	0.200	3.9 %	0.114
	ksize=7,ch=3,shape=cross	49.053	0.863	1.8 %	0.736
	ksize=7,ch=1,shape=ellipse	39.764	0.468	1.2 %	0.367
	ksize=5,ch=1,shape=ellipse	21.087	0.289	1.4 %	0.202
	ksize=7,ch=3,shape=rect	27.510	0.312	1.1 %	0.220
	ksize=9,ch=3,shape=cross	60.625	1.005	1.7 %	0.903
	ksize=9,ch=1,shape=cross	20.207	0.427	2.1 %	0.338
	ksize=9,ch=3,shape=ellipse	200.581	1.464	0.7 %	1.358
	ksize=7,ch=3,shape=ellipse	119.097	1.085	0.9 %	0.983
	ksize=9,ch=3,shape=rect	33.323	0.363	1.1 %	0.267
	ksize=9,ch=1,shape=rect	11.237	0.279	2.5 %	0.166
	ksize=7,ch=1,shape=rect	9.145	0.250	2.7 %	0.147
	ksize=5,ch=1,shape=rect	7.007	0.236	3.4 %	0.129
	ksize=7,ch=1,shape=cross	16.420	0.405	2.5 %	0.285
	ksize=5,ch=3,shape=ellipse	63.085	0.588	0.9 %	0.489
	ksize=3,ch=3,shape=ellipse	21.049	0.335	1.6 %	0.245
	ksize=3,ch=1,shape=ellipse	7.080	0.207	2.9 %	0.121
	ksize=5,ch=3,shape=cross	35.051	0.518	1.5 %	0.416
	ksize=5,ch=1,shape=cross	11.722	0.267	2.3 %	0.178
	ksize=9,ch=1,shape=ellipse	66.994	0.590	0.9 %	0.490
	ksize=5,ch=3,shape=rect	21.385	0.267	1.2 %	0.180
	ksize=3,ch=3,shape=cross	21.047	0.334	1.6 %	0.245
	ksize=3,ch=1,shape=cross	7.062	0.210	3.0 %	0.121
	ksize=3,ch=3,shape=rect	15.534	0.241	1.6 %	0.150
cvFilter2D	kernelsize=3	8.782	0.356	4.1 %	0.305
	kernelsize=5	8.817	0.367	4.2 %	0.313
	kernelsize=7	8.793	0.508	5.8 %	0.450
	kernelsize=9	7.902	4.518	57.2 %	4.455
cvFindStereoCorrespondence	8u	361.165	16.753	4.6 %	13.929
cvGEMM	ch=1	415.827	46.363	11.1 %	45.233
	ch=2	908.749	83.554	9.2 %	81.557
cvInRange	8u	10.129	0.518	5.1 %	0.157
cvInRangeS	8u	7.732	0.442	5.7 %	0.131
cvLUT	8u_8u	2.725	0.378	13.9 %	0.327
	8u_16u	1.152	0.398	34.5 %	0.347
	8u_32s	1.158	0.483	41.7 %	0.435
cvMahalanobis	32f	8.146	1.610	19.8 %	1.553
cvMax	16u	2.357	0.181	7.7 %	0.134
	8u	4.062	0.140	3.4 %	0.091
	32s	3.649	0.297	8.1 %	0.226
cvMaxS	16u	1.618	0.200	12.4 %	0.150
	8u	2.752	0.137	5.0 %	0.087
	32s	2.424	0.304	12.5 %	0.251
cvMin	16u	2.206	0.181	8.2 %	0.134
	8u	3.856	0.140	3.6 %	0.091
	32s	3.485	0.276	7.9 %	0.228
cvMinS	16u	1.614	0.202	12.5 %	0.150
	8u	2.752	0.136	4.9 %	0.089
	32s	2.398	0.301	12.6 %	0.250
cvMorphologyEx	ksize=9,ch=1,op=4	27.432	0.907	3.3 %	0.094
	ksize=7,ch=3,op=1	57.827	1.106	1.9 %	0.598
	ksize=5,ch=3,op=2	55.826	0.787	1.4 %	0.184
	ksize=5,ch=1,op=1	14.634	0.630	4.3 %	0.328
	ksize=3,ch=3,op=4	42.809	0.962	2.2 %	0.185
	ksize=3,ch=1,op=2	14.081	0.540	3.8 %	0.093
	ksize=7,ch=3,op=2	68.986	0.862	1.2 %	0.185
	ksize=7,ch=1,op=0	19.336	0.722	3.7 %	0.335
	ksize=5,ch=1,op=2	18.350	0.569	3.1 %	0.094
	ksize=7,ch=3,op=3	68.924	1.386	2.0 %	0.186
	ksize=7,ch=1,op=1	19.199	0.687	3.6 %	0.354
	ksize=5,ch=3,op=3	55.666	1.197	2.2 %	0.186
	ksize=5,ch=1,op=3	18.142	0.765	4.2 %	0.094
	ksize=3,ch=1,op=3	13.814	0.732	5.3 %	0.093
	ksize=7,ch=3,op=4	68.888	1.417	2.1 %	0.187
	ksize=7,ch=1,op=2	22.906	0.612	2.7 %	0.094
	ksize=5,ch=3,op=4	55.655	1.165	2.1 %	0.185
	ksize=5,ch=1,op=4	18.188	0.777	4.3 %	0.093
	ksize=3,ch=1,op=4	13.813	0.710	5.1 %	0.093
	ksize=7,ch=1,op=3	22.744	0.861	3.8 %	0.094
	ksize=9,ch=3,op=0	70.726	1.349	1.9 %	0.734
	ksize=7,ch=1,op=4	22.857	0.875	3.8 %	0.094
	ksize=9,ch=3,op=1	70.735	1.358	1.9 %	0.764
	ksize=9,ch=3,op=2	81.795	0.959	1.2 %	0.184
	ksize=9,ch=1,op=0	23.946	0.744	3.1 %	0.393
	ksize=3,ch=3,op=0	31.805	0.712	2.2 %	0.364
	ksize=9,ch=3,op=3	81.757	1.619	2.0 %	0.188
	ksize=9,ch=1,op=1	23.845	0.750	3.1 %	0.417
	ksize=3,ch=3,op=1	31.768	0.718	2.3 %	0.398
	ksize=9,ch=3,op=4	81.793	1.630	2.0 %	0.185
	ksize=9,ch=1,op=2	27.533	0.758	2.8 %	0.093
	ksize=7,ch=3,op=0	57.935	1.125	1.9 %	0.700
	ksize=5,ch=3,op=0	44.528	0.917	2.1 %	0.490
	ksize=5,ch=1,op=0	14.663	0.630	4.3 %	0.295
	ksize=3,ch=3,op=2	42.998	0.705	1.6 %	0.185
	ksize=3,ch=1,op=0	10.379	0.588	5.7 %	0.286
	ksize=9,ch=1,op=3	27.610	0.922	3.3 %	0.094
	ksize=5,ch=3,op=1	44.480	0.945	2.1 %	0.472
	ksize=3,ch=3,op=3	42.813	0.995	2.3 %	0.185
	ksize=3,ch=1,op=1	10.359	0.553	5.3 %	0.274
cvMul	16u	4.254	0.184	4.3 %	0.136
	8u	5.006	0.145	2.9 %	0.095
	32s	4.304	0.281	6.5 %	0.229
cvNot	16u	0.974	0.198	20.3 %	0.146
	8u	0.409	0.137	33.5 %	0.085
	32s	1.932	0.300	15.5 %	0.246
cvOr	16u	1.650	0.181	11.0 %	0.135
	8u	0.817	0.148	18.1 %	0.090
	32s	3.273	0.283	8.6 %	0.226
cvOrS	16u	1.196	0.199	16.6 %	0.148
	8u	0.507	0.135	26.6 %	0.086
	32s	2.128	0.303	14.2 %	0.248
cvPerspectiveTransform	3d	29.865	5.162	17.3 %	5.082
	2d	21.359	2.440	11.4 %	2.367
cvRandArr	normal-16u	206.601	182.178	88.2 %	182.119	This is running on a single SPE. The slow result is due to sequential operation. We need an algorithm that is suited to SIMD.
	normal-8u	204.119	181.861	89.1 %	181.805
	normal-32s	180.506	180.444	100.0 %	180.372
cvScaleAdd	ch=1	3.757	0.375	10.0 %	0.323
	ch=2	9.431	0.662	7.0 %	0.590
cvScaleAdd	ch=1	3.767	0.374	9.9 %	0.322
	ch=2	9.447	0.662	7.0 %	0.592
cvSub	16u	3.260	0.183	5.6 %	0.135
	8u	3.593	0.143	4.0 %	0.091
	32s	3.275	0.283	8.6 %	0.227
cvSubRS	16u	2.583	0.200	7.7 %	0.148
	8u	3.227	0.137	4.2 %	0.088
	32s	1.920	0.301	15.7 %	0.247
cvSubS	16u	2.589	0.199	7.7 %	0.148
	8u	4.078	0.139	3.4 %	0.088
	32s	1.911	0.303	15.9 %	0.248
cvSum	8u3c	1.719	0.279	16.2 %	0.226
	8u1c	0.399	0.167	41.9 %	0.122
	32s3c	9.112	0.457	5.0 %	0.395
	32s1c	2.954	0.247	8.4 %	0.198
cvTranspose	64f	3.792	0.575	15.2 %	0.517
	32s	2.054	0.335	16.3 %	0.285
cvXor	16u	1.658	0.183	11.0 %	0.133
	8u	0.832	0.138	16.6 %	0.091
	32s	3.298	0.281	8.5 %	0.225
cvXorS	16u	1.213	0.200	16.5 %	0.148
	8u	0.474	0.140	29.5 %	0.087
	32s	2.164	0.374	17.3 %	0.248
cvIntegral	8u32s,mode=0	1.687	0.869	51.5 %	0.278
	8u32s,mode=1	3.579	2.054	57.4 %	0.811
	8u32s,mode=2	5.503	2.330	42.3 %	0.753
cvHaarDetectObjects	lena,flag=0	1454.043	147.656	10.2 %	141.958
	lena,flag=1	1499.969	1502.872	100.2 %	0.285
	lena,flag=2	2690.345	88.363	3.3 %	83.755
cvPyrDown	8u	2.347	0.155	6.6 %	0.108
cvCornerHarris	ksize=3	46.130	1.398	3.0 %	1.335
	ksize=5	45.843	1.540	3.4 %	1.477
cvCornerMinEigenVal	ksize=3	67.329	1.640	2.4 %	1.568
	ksize=5	67.155	1.771	2.6 %	1.710
cvGoodFeaturesToTrack	harris	55.266	4.109	7.4 %	0.262
	mineigen	79.428	5.132	6.5 %	0.263
cvMinMaxLoc	int	2.591	0.114	4.4 %	0.062
	float	3.002	0.114	3.8 %	0.062
cvThreshold	uchar	2.825	0.093	3.3 %	0.044
	float	3.408	0.173	5.1 %	0.124

Contents 1 An example of poor performance 2 Hameescalia said ... 3 fitzgibbon said ... 4 brooksonea said ... 5 reedballa said ... 6 rydgesegal said ... 7 ocelfaover said ... 8 shanikavar said ... 9 edrahuth said ... 10 hadenholde said ... 11 pitneysutt said ... 12 wattsonand said ... 13 renayfunkh said ... 14 orrickcott said ... 15 ediblylig said ... 16 payday loans said ...

An example of poor performance

The following function has been ommitted as a candidate for optimization. The reasons are described below.

• Generally, when small data is processed, the overhead to processing time ratio increases.
• Typically, an argument of this function contains small amount of data.

This table shows the execution time of PPE is faster than that of PPE-SPEs. "Pure SPE" means that an execution time excluding DMA and PPE-SPEs communication time.

Target functions		Performance
		PPE original code [ms]	SPE optimized code [ms]	Pure SPE (without PPE-SPE overhead) [ms]
cvCompareHist	32x1	0.001	0.043	0.002
	32x32	0.004	0.073	0.021

---

Hameescalia said ...

fitzgibbon said ...

brooksonea said ...

reedballa said ...

rydgesegal said ...

ocelfaover said ...

shanikavar said ...

edrahuth said ...

hadenholde said ...

pitneysutt said ...

wattsonand said ...

renayfunkh said ...

orrickcott said ...

ediblylig said ...

payday loans said ...