TDA4VM: the performance of memcpy in qnx

Dear customer,

   Which version are you using of SDP? You tested base of SDP7.0?  Thanks.

Linjun

Yes, it is SDP7.0

Hi,

Regarding the "memcpy with neon optimization" portion of the question, please be aware of the below FAQ, for checking if neon is enabled.

https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1075898/faq-tda4vm-neon-support-in-psdk-qnx 

Regards,

kb

Hi:

 Our configuration as follows:

/home/root#pidin syspage=cpuinfo
Header size=0x00000108, Total Size=0x00000e38, #Cpu=2, Type=257
Section:cpuinfo offset:0x00000340 size:0x00000040 elsize:0x00000020
0) cpu:411fd080 flg:c0000c7a spd:2000 hwid:80000000 cache i/d:0/1 name:111
history:fffffffffffffff7
1) cpu:411fd080 flg:c0000c7a spd:2000 hwid:80000001 cache i/d:0/1 name:111
history:ffffffffffffffee

The "flg:c0000c7a"  indicates that NEON (0x40) functionality is viewed as enabled by H/W & QNX.

Regards,

kb

When comparing functionality between QNX / Linux care must be taken to ensure that the setups under test are the same.  For example having L3 cache enabled, may make a difference for one HLOS vs, another, as will the tool set used to build the binaries, and the compile flags that were used during building.

In general, expectation is that performance the memcpy() for QNX and Linux should be comparable as you have seen without NEON.   If you feel the issue has been isolated to NEON, please check the compile flags being used for QNX to ensure that the memcpy() under test has been built for NEON support.

Recommend contacting QNX for further details on memcpy(), implementation. 

Regards,

kb