The memset has addtional time cost after turn on the CONFIG_NEWLIB_LIBC (NCS1.9.1 nrf53)

[David Duan]

Hi Devzone

currently we found the time cost for memset has increased after the project the use NEWLIB_LIBC

inside the project , we use funcation k_cycle_get_32 to calulate the time cost for function memset & memcpy

if we enable the CONFIG_NEWLIB_LIBC the time cost:

pic1

if we disable the CONFIG_NEWLIB_LIBC the time cost:

pic2

How to reproduce:

memset_issue.7z

above project could directly run in NCS 1.9.1.

If you set CONFIG_NEWLIB_LIBC=n in prj.conf, then you will pic2 result in RTT log 

If you set CONFIG_NEWLIB_LIBC=y in prj.conf, then you will pic1 result in RTT log

Question:

The memset & memcpy has high frequtly used in customer's project , and they need NEWLIB_LIBC to supprot their algorithm

Is it possilbe to reduce the time cost for func memset when the project use the NEWLIB_LIBC?

BR

[Simon]

Hello David,

I'll just let you know that I will take a look at this tomorrow/Friday

Best regards,

Simon

[Jørgen Kvalvaag]

By default NEWLIB_LIBC_NANO is used. Which is very focused on size and not speed.

You can try using the "standard" NEWLIB_LIBC by adding CONFIG_NEWLIB_LIBC_NANO=n to your prj.conf, it contains larger, but more optimized implementations.