Performance data was obtained on the J721S2 EVM. EVM warm cycle obtained by profiling the kernel's compute code execution after a cold run of the same code. Please refer to the kernel's documentation for more information about the parameters in the tables shown below.

MATHLIB kernels

This section contains tables that depict the expected performance numbers for the MATHLIB kernels.

Kernel Name	Vector Length	EVM Warm Cycles – Single Precision	Cycles per Sample
MATHLIB_acos	512	969	1.8926
	1024	1748	1.707
	2048	3348	1.6348
MATHLIB_acosh	512	1244	2.4297
	1024	2235	2.1826
	2048	4174	2.0381
MATHLIB_asin	512	779	1.5215
	1024	1366	1.3340
	2048	2582	1.2607
MATHLIB_asinh	512	1369	2.6738
	1024	2457	2.3994
	2048	4633	2.2622
MATHLIB_atan	512	727	1.4199
	1024	1282	1.2520
	2048	2434	1.1885
MATHLIB_atanh	512	1435	2.8027
	1024	2429	2.3721
	2048	4605	2.2485
MATHLIB_atan2	512	1021	1.9941
	1024	1915	1.8701
	2048	3709	1.8110
MATHLIB_cos	512	484	0.9453
	1024	783	0.7646
	2048	1423	0.6948
MATHLIB_cosh	512	1501	2.9316
	1024	2520	2.4609
	2048	4824	2.3555
MATHLIB_div	512	200	0.3906
	1024	243	0.2373
	2048	369	0.1802
MATHLIB_exp	512	764	1.4922
	1024	1353	1.3213
	2048	2569	1.2544
MATHLIB_exp2	512	820	1.6016
	1024	1492	1.4570
	2048	2836	1.3848
MATHLIB_exp10	512	802	1.5664
	1024	1410	1.3770
	2048	2626	1.2822
MATHLIB_log	512	792	1.5469
	1024	1411	1.3779
	2048	2729	1.3325
MATHLIB_log2	512	923	1.8027
	1024	1723	1.6826
	2048	3323	1.6226
MATHLIB_log10	512	923	1.8027
	1024	1723	1.6826
	2048	3323	1.6226
MATHLIB_pow	512	2668	5.2109
	1024	4896	4.7813
	2048	9433	4.6060
MATHLIB_rsqrt	512	313	0.6113
	1024	505	0.4932
	2048	889	0.4341
MATHLIB_sin	512	420	0.8203
	1024	656	0.6406
	2048	1168	0.5703
MATHLIB_sinh	512	1387	2.7090
	1024	2278	2.2246
	2048	4326	2.1123
MATHLIB_sqrt	512	254	0.4961
	1024	362	0.3535
	2048	618	0.3018
MATHLIB_tan	512	1322	2.5820
	1024	2453	2.3955
	2048	4757	2.3228
MATHLIB_tanh	512	1313	2.5645
	1024	2218	2.1660
	2048	4201	2.0513

MATHLIB double precision kernels

This section contains tables that depict the expected performance numbers for the MATHLIB kernels for double precision implementation.

Kernel Name	Vector Length	EVM Warm Cycles – Double Precision	Cycles per Sample	Scalar Implementation Cycles	Cycles Per Sample
MATHLIB_atan2	512	4184	8.2	120835	236
	1024	7960	7.8	243183	237.5
	2048	15512	7.6	483317	236
MATHLIB_cos	512	1464	2.9	41888	81.8
	1024	2333	2.3	83776	81.8
	2048	4509	2.2	167552	81.8
MATHLIB_log10	512	2332	4.6	76720	149.8
	1024	4316	4.2	153436	149.8
	2048	8330	4.1	306406	149.6
MATHLIB_pow	512	4856	9.5	132897	259.6
	1024	8992	8.8	265845	259.6
	2048	17440	8.5	531627	259.6
MATHLIB_sin	512	1186	2.3	39840	77.8
	1024	2210	2.2	79680	77.8
	2048	4322	2.1	159360	77.8
MATHLIB_sqrt	512	976	1.9	37281	72.8
	1024	1808	1.8	74560	72.8
	2048	3472	1.7	149120	72.8