i.MX Processors中的主题 Re: NEON memory copy is slower

NEON memory copy is slower

ismailmohammed — Wed, 28 Sep 2016 10:28:18 GMT

Hello,

As per the below article http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.faqs/ka13544.html, I tried NEON memory copy with preload on IMX6 Solo and Quad modules.

For 16 MB data copy, the memory copy performance is as below:

IMXDQ Module

memcpy_libc : 435 MB/s

memcpy_neon: 343 MB/s

IMX Solo Module

memcpy_libc : 472 MB/s

memcpy_neon: 385 MB/s

Why neon memcpy performance is slower than libc memcpy?

Thanks

Ismail

Re: NEON memory copy is slower

igorpadykov — Thu, 29 Sep 2016 00:07:11 GMT

Hi Ismail

neon is not used for memory copy in official linux releases,

could you try with bare metal test.

Best regards
igor
-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

Re: NEON memory copy is slower

Yuri — Thu, 29 Sep 2016 01:20:51 GMT

Hello,

You may try NEON based memcpy(), using source and destination

memory addresses, aligned with DRAM burst size.

Regards,

Yuri.

Re: NEON memory copy is slower

ismailmohammed — Thu, 29 Sep 2016 04:11:40 GMT

Hi Igorpadykov,

I am using WinCE OS.

Hi Yuri,

Where can I find NEON based memcpy function?

Thanks and Regards,

Ismail

Re: NEON memory copy is slower

igorpadykov — Thu, 29 Sep 2016 05:12:13 GMT

Hi Ismail

I am afraid NEON based memcpy function is not used

in WinCE too, so one will have to develop it himself.

Best regards
igor