Optimizing speex resampler for 24kec with DSP ASE

This topic contains 1 reply, has 2 voices, and was last updated by  keng 3 years, 3 months ago.

Viewing 2 posts - 1 through 2 (of 2 total)
  • Author
    Posts
  • #31754

    manu_ag
    Member

    Hi,

    I am using the Android provided prebuilt MIPS compiler for compiling the speex audio resampler. I see the performance is very bad even after using the -mdspr2 and -O3 compiler flags. Some googling revealed that only the MIPS SDE compiler supports the optimized instructions. Now the only way forward for me is to re-write the core function of the re-sampler in assembly.

    I would like to know is there any other alternate compiler/approach that can be used in order to optimize the code.

    Thanks in advance,
    manu

    #38866

    keng
    Member

    Hi Manu:

    Need clarification from you. Are you referring library in external/speex/libspeex? If so, couple comments

    1) I believe this library does offer either fixed point or floating point implementation. Floating implementation is default option. For 24KEc, please use fixed point implementation. We do have cores that do support FPU but just happen to be 24KEc does not have it.

    2) 24KEc only support DSP ASE R1 so -mdspr2 flag should not be used. -mdspr2 should only be used for our higher end core that support DSP ASE R2 extension.

    3) Another quick option for you is using -funroll-loops compiler option. By default, this should be disabled even with -O3. This library seem to have some nested loop. Unroll the loop might help.

    Keng

Viewing 2 posts - 1 through 2 (of 2 total)
You must be logged in to reply to this topic.