Hi, I am evaluating the opus codec for use in an embedded design. I have successfully ported opus on a TI C5502 (fixed poin dspt) but need to use a more powerful dsp processor for our application. With the C5502 it takes 4.5ms to encode 2 (mono) channels with a 5ms frame @80kbits per channel. I need to reduce the computing time to reduce latency and leave some time for the DSP to perform some other tasks.The ADSP-21489 with it's floating point architecture seems like a good fit. It also runs faster @ 450Mhz instead of 300Mhz for the C5502. My question is : can I expect faster computing when opus is compiled in float mode rather than in fixed mode ? Thanks for your input. Robert -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.xiph.org/pipermail/opus/attachments/20130716/523490eb/attachment.htm
Robert Radrua wrote:> My question is : can I expect faster computing when opus is compiled in > float mode > rather than in fixed mode ?This depends a lot on the FPU available. With a good FPU, it should be true. For example, on OMAP 4460 (Cortex A9), float is _slightly_ faster than fixed. On OMAP 3430 (Cortex A8), float is _much_ slower, even though both have a hardware FPU.
Reasonably Related Threads
- how to improve the voice quality (run speex on ADSP-BF533)
- [PATCH 6/6] rpmsg: add a device ID to also bind to the ADSP device
- [RFC 12/12] rpmsg: add a device ID to also bind to the ADSP device
- [RFC 12/12] rpmsg: add a device ID to also bind to the ADSP device
- Porting Speex on C5509A and CELP Algorithm Documentation