How to avoid the time-consuming coming from memcpy from linux user space to kernel space?


My use_case is like this.I transfer yuv data from linux host to c6678 through PCIE and

then c6678 do H264 encode.In order to achieve that goal,I decide to do like below steps:


First I transfer yuv data from user space to kernel space.On this step,I use memcpy.


Transfer yuv data from linux kernel space to c6678 DDR through PCIE .On this step.

I use EDMA as PCIE user guide said.

But after my profile,I find step1 using memcpy is the bottleneck.I want to cut the time-wasting

on memcpy from user space to kernel space.So,How to avoid that time consuming ?


Best Regards!


Target chip : TCI6638k2k

Target EVM : XTCIEVMK2X   Rev1.1

CCS  ver    :  CCS

MCSDK version: 



1 Reply