• State TI Thinks Resolved
  • Locked Locked
  • Replies 2 replies
  • Answers 1 answer
  • Subscribers 99 subscribers
  • Views 536 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

SK-TDA4VM: Usage of C7x intrinsic in MMA

MadhuBala E
Intellectual 275 points
Part Number: SK-TDA4VM
Other Parts Discussed in Thread: FFTLIB, TDA4VM

Hi 
I am using SDK version "ti-processor-sdk-rtos-j721e-evm-09_00_01_01" which has the MMA library functions implemented for FFTLIB_FFT_highRadixDecompositions under the file location 
ti-processor-sdk-rtos-j721e-evm-09_00_01_01\fftlib_02_06_01_00\ti\fftlib\src\fft_c7xmma\FFTLIB_FFT_highRadixDecompositions_ixX_cxX_oxX. Here I could see how mma config is performed and utilized the intrinsic functions like __HWAOPEN() and __HWALDB() to open the configured mma MACRO and loading the B matrix MMA memory .

My question here is we have many vector C7x intrinsic functions/operations like __permute() , __vmatmpysp_vww() etc which perform SIMD operations/vector operations , can we implement these type of intrinsic functions which will perform matrix operations like __vmatmpysp_vww() in MMA since we have only HWA based intrinsic. If so is there any documentation or code sample available to check how we can configure MMA MACROS based on our requirement?

Thanks

Madhu

  • 0
    TI__Intellectual 2440 points

    Hi Madhu,

    Processor SDK also include a library called "MMALIB" which provides functions to accelerate computation on the MMA.  Of the ones you mentioned, matrix-matrix multiply is supported in MMALIB for 8, 16, 32-bit integers on TDA4VM.

    Regards,

    Will

  • 0
    TI__Genius 10170 points

    Hi Madhu,

    As Will has mentioned, if you are looking for a matrix-multiplication function that utilizes the MMA, we recommend using the function provided by MMALIB to do so which is packaged in the PSDK RTOS package. 

    With our current support strategy for MMA, we are not supporting the development of custom functions or utilization of the MMA outside of the context of TI provided interfaces, namely TIDL, FFTLIB, and MMALIB. We will support questions regarding using these SDK-provided libraries. 

    Best,

    Asha