strlen

Variants

Name	Description
stock	MI C version
SSE2	pcmpeqb and pmovmskb
SSE4.2	pcmpestri and pcpmestrm
AVX	128-bit vpcmpeqb and vpmovmskb
ERMS	repne scasb for machines with ERMS

Note: clang was too smart and optimized plain strlen calls away, so I had to create a copy of the C version called strlen_mi() to fool it.

Machines Tested

CPU	Speed (GHz)	Notes
AMD FX-8120	3.11	1 x 8 zoo.freebsd.org
AMD Opteron 6328	3.20	2 x 8 Supermicro H8DG6/H8DGi
Intel Xeon X5365	3.00	2 x 4 Supermicro X7DBU
Intel Xeon X5482	3.20	2 x 4 Supermicro X7DWN+
Intel Xeon X5675	3.07	Westmere 2 x 6 Supermicro X8DTU
Intel Core i5-2520M	2.50	Sandy Bridge 1 x 4 Thinkpad X220 (4286)
Intel Core i5-2500K	3.30	Sandy Bridge 1 x 4 MSI Z77A-G45 (MS-7752)
Intel Xeon E5-2680	2.70	Romley 2 x 8 Supermicro X9DRW
Intel Xeon E5-2667 v2	3.30	Romley V2 2 x 8 Supermicro X9DRW (supports ERMS)

Test Cases

Name	Description
page	aligned string one page - 1 long
short	aligned string 14 characters long
short2	aligned string 32 characters long
short3	aligned string 48 characters long
offset	4 byte offset string 126 characters long
offset2	7 byte offset string 95 characters long

Results

The numbers are the min value in the distribution where the values are a TSC delta across a single invocation of the test.

Bold indicates the lowest time among the given variations in a Test and CPU combination. Green text is used for times faster than the stock implementation, and red text is used for times slower than the stock implementation.

CPU	Test / Variant
	page						short						short2						short3						offset						offset2
	stock	SSE2	SSSE4.2	AVX	ERMS		stock	SSE2	SSSE4.2	AVX	ERMS		stock	SSE2	SSSE4.2	AVX	ERMS		stock	SSE2	SSSE4.2	AVX	ERMS		stock	SSE2	SSSE4.2	AVX	ERMS		stock	SSE2	SSSE4.2	AVX	ERMS
AMD FX-8120	3899	2017	7245	2006	33064		139	81	95	82	318		168	82	157	81	430		195	86	183	95	551		316	150	354	155	1184		241	137	307	137	938
AMD Opteron 6328	1510	821	2932	585	13543		78	78	80	73	143		90	70	86	73	202		93	73	97	79	253		126	96	166	83	511		114	83	143	79	412
Intel Xeon X5365	1386	864	--	--	16506		81	72	--	--	180		90	72	--	--	252		90	81	--	--	315		144	108	--	--	630		135	108	--	--	504
Intel Xeon X5482	1608	808	--	--	16464		48	40	--	--	136		48	40	--	--	140		56	40	--	--	264		80	80	--	--	592		72	72	--	--	464
Intel Xeon X5675	1592	848	2100	--	8252		60	24	24	--	92		32	46	32	--	124		78	46	83	--	156		76	64	104	--	316		64	56	88	--	256
Intel Core i5-2520M	3525	1950	6463	536	25812		100	75	75	18	300		112	75	87	18	412		112	75	112	21	512		212	100	262	21	1012		175	100	212	24	825
Intel Core i5-2500K	1002	552	1893	573	7350		21	18	18	18	87		33	18	18	18	117		36	21	27	21	147		57	24	69	21	297		45	21	54	24	225
Intel Xeon E5-2680	1496	632	2048	644	8260		36	24	24	28	96		40	24	28	24	132		52	24	32	24	164		80	40	84	24	324		68	36	64	24	264
Intel Xeon E5-2667 v2	1296	632	2076	648	8260		24	24	24	24	100		36	24	28	24	132		48	24	28	28	164		80	44	84	24	324		64	28	64	24	264

Conclusions

• The SSE2 version is generally faster than the stock version.
• The SSE4.2 version is generally slower than the SSE2 version.
• The AVX version often outperforms the SSE2 version.
• It seems that ERMS does not accelerate repne scasb.