PROCESSOR-SDK-AM62A: AM6254ATCGHAALW , phyCORE-AM62Ax

Part Number: PROCESSOR-SDK-AM62A
Other Parts Discussed in Thread: FFTLIB

Hello,

I'm currently investigating the possibility of changing the C7x DSP core L2 SRAM to cache on the AM62Ax processor. Do you know if this is possible and if there are any examples available in CCS demonstrating how to achieve this?

I've found some suggestions on E2E forums that for products with a similar structure and featuring the C7x, it might be possible to make this change by increasing the cache size and configuring MMU page tables. Would this be the correct strategy for the AM62Ax?

Within the DSP examples (located under the /source folder), there are some APIs (e.g., FFTLIB_TEST_c7xSetL2CC) for setting L2 cache registers. However, it seems that these APIs haven't been utilized in the AM62Ax SDK, and there are TODO comments in the code (e.g., TI_cache.c):

/* TODO: Configure DSP L1 and L2 cache registers, and call TI_meminit() */
/*       Check c6x implementation                                       */
int32_t TI_cache_init (void) { return (0); }

So, it appears that converting L2 into cache on the AM62Ax is technically feasible, although it hasn't been implemented in the SDK yet. If my assumption is correct, is there an expected timeline for when we might see examples demonstrating how to achieve this?

I'm currently assisting a client working with the PHYTEC phyCORE-AM62Ax development kit using MCU-PLUS-SDK-AM62AX v09.01.00.39, and they are interested in exploring this possibility to enhance the performance of the DSP. Having an idea of the expected timeline would be greatly beneficial for our planning purposes.

  • Hello Madelyn,

    Thanks for reaching out to Texas Instruments E2E support forum.

    I have taken your inputs and working on it. Please allow some time to revert back.

    Regards,

    Tushar

  • Hi Madelyn,

    It  is our pleasure to help. 

    AM62A uses the C7504 variation of the DSP which does not support configuring L2 memory as cache. This feature is supported on other variations of the DSP such as C7100. 

    The FFTLIB you referred is scaled to support both C7504 and C71x. For this reason, you might see references to cache setup but this is for only C71x devices. 


    I would like to highlight, as you may already know, that the Streaming Engines available on C7504 efficiently move data from the L2 memory to the core DSP computing units which improve whole kernel performance.  

    Best regards,

    Qutaiba