(Public copy) Ruy benchmarks on Pixel4, January 2020

	A	B	C	D	E
1	Note: Eigen's multithreading relies on OpenMP, not available in production on mobile platforms. Eigen is what TensorFlow Lite used before ruy. EigenTensor and ruy's multithreading use only c++11 threads. OpenBLAS' multithreading was not easily available to us so was left out in these benchmarks.
2
3
4
5
6
7
8
9
10
11
12
13	XXXXXXXXXXX Float cubic
14
15
16
17
18
19
20
21
22	### cpumask=0f numthreads=1
23
24
25	size	ruy	Eigen	EigenTensor	OpenBLAS
26	16	2.679	1.049	1.14	1.203
27	24	2.688	1.346	1.473	1.512
28	32	4.587	2.099	2.418	2.3
29	48	5.774	2.789	3.123	3.044
30	64	6.17	3.201	3.381	3.568
31	96	7.037	3.821	3.604	4.401
32	128	7.474	3.925	3.999	4.816
33	192	7.602	4.533	4.801	5.376
34	256	6.914	4.36	4.357	5.117
35	384	8.124	5.038	4.928	5.802
36	512	8.313	5.535	5.671	6.108
37	768	9.071	5.307	5.308	6.473
38	1024	9.113	5.342	5.348	6.739
39	1536	9.44	5.509	5.557	6.987
40	2048	9.381	5.41	5.359	7.099
41	3072	7.298	5.766	5.767	7.309
42	4096	8.803	5.649	5.706	7.283
43
44
45
46	### cpumask=0f numthreads=4
47
48
49	size	ruy	Eigen	EigenTensor
50	16	2.585	1.028	1.099
51	24	2.775	1.344	1.46
52	32	4.742	2.084	2.38
53	48	10.24	3.5	3.104
54	64	14.8	5.502	3.433
55	96	20.38	7.749	4.84
56	128	25.03	7.587	7.489
57	192	28.36	10.74	8.525
58	256	28.01	11.75	4.251
59	384	31.86	14.69	6.857
60	512	32.83	16.27	10.75
61	768	36.56	18.04	16.46
62	1024	35.84	19.08	18.32
63	1536	37.23	20.38	19.84
64	2048	36.7	20.94	19.86
65	3072	34.84	21.9	21.47
66	4096	37.47	22.18	21.43
67
68
69
70	### cpumask=0f numthreads=1 BENCHMARK_ONLY_MATMUL=1
71
72
73	size	ruy	Eigen	EigenTensor	OpenBLAS
74	16	2.612	1.893	1.653	2.162
75	24	2.826	2.037	1.872	2.304
76	32	4.549	3.388	3.168	3.665
77	48	5.724	4.149	3.933	4.569
78	64	6.299	4.496	4.074	5.071
79	96	7.016	5.174	4.205	5.846
80	128	7.352	5.022	4.613	6.251
81	192	7.454	5.377	5.281	6.25
82	256	7.02	4.877	4.59	5.692
83	384	8.079	5.466	5.211	6.319
84	512	8.307	5.655	5.477	6.513
85	768	9.227	5.411	5.273	6.754
86	1024	9.122	5.495	5.358	6.88
87	1536	9.347	5.614	5.632	7.074
88	2048	9.448	5.52	5.909	7.188
89	3072	7.243	5.802	5.776	7.411
90	4096	7.273	5.739	5.635	7.381
91
92
93
94	### cpumask=0f numthreads=4 BENCHMARK_ONLY_MATMUL=1
95
96
97	size	ruy	Eigen	EigenTensor
98	16	2.68	1.892	1.63
99	24	2.891	2.022	1.915
100	32	4.75	3.363	3.151