PROCESSOR-SDK-J721E: tivxMemAlloc function have problem in J721S2 SDK

Hello,

Which core are you allocating this memory from on the J721S2?  Is it on the A72?

And from the list below, which memory region are you allocating from?

https://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-jacinto7/latest/exports/docs/tiovx/docs/user_guide/group__group__tivx__mem.html#ga453747d90eef05b464e633ea644ac0ed

Regards,

Lucas

tivxMemAlloc(asize*long_size, TIVX_MEM_EXTERNAL);

Hi Lucas Weaver,

This is the function i used. i apply ddr from TIVX_MEM_EXTERNAL region in A72 core.

Please help to check this problem. 

Thank you very much!

Hi xie jc,

Interesting, few questions,

1, when you are using calloc or malloc, is it allocated on A72? 

2, where do you do memcpy? Is it on A72/Linux? 

Regards,

Brijesh

Hi Brijesh Jadav,

1, when you are using calloc or malloc, is it allocated on A72? 

Yes, i use two way to get ddr. one is use calloc function, the other is tivxMemAlloc. They are all run on LINUX A72 core.

2, where do you do memcpy? Is it on A72/Linux?

I write a test demo. They are on A72 core linux system. they other code is the same, only ddr is get from different ways.

This is the code as below.

Hi xie jc,

Lets look at only tivxMemAlloc function, since it is a part of the SDK and it allocates physically continuous memory. 

The one of the reason for such drastic change in performance could be cache. If the buffer is mapped as uncached, then there could be performance impact, so can you please check if this is the cache?

Lucas Weaver, can you please help here? Are we mapping EXTERNAL region as cached on A72? 

Regards,

Brijesh

Hi Lucas Weaver & Brijesh Jadav 

Thanks for your reply.

I test it again after modify as last advise. Now the write and read rate of tivxMemAlloc is improved than default config.

But the rate is less than calloc function. This is the report as below.

from EVM board, the tivxMemAlloc function is more efficient than calloc function. Is there anything need to update?

Hi xie jc,

Not sure how calloc works, does it also allocate physically contiguous cached buffer in the application space? May be calloc is also initializing the memory with 0x0.. 

Regards,

Brijesh

Hi Brijesh Jadav ,

calloc can apply contiguous buffer in application space. it does memory copy in my test.

so it is not affect test data when memory is initialized or not.

compare with EVM board, the tivxMemAlloc function is more efficient than calloc function.

but in ECO board, it is almost 80% of alloc function.

Is there anything need to update?

They are at least the same level of reading and writing rate.

Hi xie jc,

I dont see there is any difference between TDA4 and TDA4-eco, so calloc should also give same performance.. Are you sure you are running same application? 

Also could you help us understand what is your goal? Where are you using calloc and where tivxMemAlloc function? 

Regards,

Brijesh