similar to: Opus's x86 Optimization

Hello guys I have profiled opus code(cpu sampling) and identified functions that are expensive for our use case (mostly x86 silk mode). Those functions could be optimized for x86 using intrinsics. Looking at source tree and there is already optimization work going on ,especially Cisco folks and Jonathan have already contributed for x86. We would like to have some optimization included in future

I know the focus for Opus is low delay, but I've been watching its development with interest because of the potential for audiobook/podcast use, where latency is practically irrelevant. I hear the upcoming USAC codec will give good results for this niche (though listening test results don't seem to be available to the public yet), but I also hear it'll be extremely patent

This is observed on a live call between webRTC browser client and another legacy client. Our server is there in between and transcoding from opus to another codec and this is observed while decoding the opus. Anyway, I'll try to capture/dump the packets in the server before feeding to the opus_decode and share with you. But this will not have the first 8 bytes (length+enc range) to directly

On 3 March 2015 at 21:59, Jonathan Lennox <jonathan at vidyo.com> wrote: > Viswenath, > > My patch should be against the tip, but it?s the very recent tip, > including some changes this past Friday (27 Feb). I mentioned in the IRC > room a problem I discovered in creating my patch, and then later improved > the fix Tim had made for the problem. Where do you get conflicts

Hi Jean-Marc, Could you please update if you got a chance to look into. As I mentioned, I don't see the same issue in 1.1.1, but I don't see any difference in 1.1.1 other than optimization based on the architecture. This optimization could have fixed some stack overflow issue in some specific cases? Thanks Suresh On 13 April 2015 at 12:39, Suresh Thiriveedi <sthiriveedi at

Hi Jean-Marc, Thanks for your response. Please find the details as below. *Backtrace we got for this crash:* #0 0x0000000000800c54 in opus_decode_frame (st=0x38906b8f99d09c5, data=0xf0aa10b4ef1008ae <Address 0xf0aa10b4ef1008ae out of bounds>, len=-188613428, pcm=0x6e80016085efd57, frame_size=44037315, decode_fec=58716895) at src/opus_decoder.c:384 #1 0x00000000008009c0 in

Viswenath, My patch should be against the tip, but it?s the very recent tip, including some changes this past Friday (27 Feb). I mentioned in the IRC room a problem I discovered in creating my patch, and then later improved the fix Tim had made for the problem. Where do you get conflicts merging it to tip? In terms of merging, you posted your patch before I posted mine, so probably I should be

I just tried decoding with v1.1: ./opus_demo -d 48000 2 opus_encoded_crash.opus out.pcm and I see no issue (including with valgrind). Does the same command-line create problems for you? What compile flags did you use? fixed-point or float, any assembly, ...? Could be assembly here, or even a compiler bug wouldn't be unheard of. Cheers, Jean-Marc On 20/04/15 07:27 AM, Suresh Thiriveedi

Hello Jonathan, I am unable to apply your patch cleanly on tip. Timothy/opus-dev, This patch has some conflicts with my ARM patch that does fft optimizations http://lists.xiph.org/pipermail/opus/2015-March/002904.html http://lists.xiph.org/pipermail/opus/2015-March/002905.html One of us probably has to rebase depending on which patch goes into opus first. Regards, Vish On 1 March 2015 at

> I don't believe this is an actual error. If it's truly possible for > these areas to overlap (I don't think it is), then something much more > serious than using memmove instead of memcpy needs to be done about it. In the C# version of this code, these two copy regions are stored in separate arrays entirely. I agree that there should be no normal way to have the memcpy

On 3 March 2015 at 22:17, Jonathan Lennox <jonathan at vidyo.com> wrote: > > On Mar 3, 2015, at 11:08 PM, Viswanath Puttagunta > <viswanath.puttagunta at linaro.org> wrote: > > > > On 3 March 2015 at 21:59, Jonathan Lennox <jonathan at vidyo.com> wrote: >> >> Viswenath, >> >> My patch should be against the tip, but it?s the very recent

To be decodable by opus_demo, you'll have to add the 8-byte "header". Just put in the length of the packet followed by "0" for the encoder range (0 means "not present"). That being said, from previous experience, the most likely cause of the crash is a bug in your software causing a corruption in Opus. So it's safe to assume that if you can't reproduce

Please provide the input file that produces this with opus_demo. On 16/04/15 03:24 AM, Suresh Thiriveedi wrote: > Hi Jean-Marc, > > Could you please update if you got a chance to look into. As I > mentioned, I don't see the same issue in 1.1.1, but I don't see any > difference in 1.1.1 other than optimization based on the architecture. > This optimization could have

Hi Dmitry, So it's not explicitly in your report, but it looks like the crash is due to a divide-by-zero at: min_coef = silk_DIV32_16( silk_int16_MAX, silk_RSHIFT( psSilk_VAD->counter, 4 ) + 1 ); which happens because counter is -16 (which means (-16 >> 4) + 1 == 0). Now, this could be caused by an integer wrap-around, but it should only happen after encoding around

Hi, We are able to reproduce the issue with the 1.1 opus_demo (sample file). We captured the frames in our server just before the opus_decode and fed the file to opus_demo (1.1) and it is crashing. Same file is tested with 1.1.1 and it is fine. So this is in line with our server testing observation and I think here we can conclude that the 1.1 library is crashing while handling a specific mode

The patch prevents KEIL MDK compile warnings, like:   warning:  #1035-d: single-precision operand implicitly converted to double-precision Actually ARM Cortex-M4F has only a *single precision* (float) FPU. It's suit for all platforms. See the comment at the begin of patch file. Sincerely Forrest Zhang -------------- next part -------------- Specify the floating point constant with single

Hello Jonathan, Just FYI, I started doing review of your patch and will get back to you in few days. After review, I would like to rebase your patch (as necessary) myself and do some testing.. and re-submit. Regards, Vish On 4 March 2015 at 09:00, Viswanath Puttagunta <viswanath.puttagunta at linaro.org> wrote: > > On 3 March 2015 at 22:17, Jonathan Lennox <jonathan at

Still can't reproduce. What OS and compiler version? Jean-Marc On 21/04/15 02:48 AM, Suresh Thiriveedi wrote: > Hi, > > There is no change in the compiler flags. I'm using as it is from the > original code. No change in the Makefile and I believe it is using the > floating point only by default. > > We are using 8k samples and mono so the commands is as follows.

On Mar 3, 2015, at 11:08 PM, Viswanath Puttagunta <viswanath.puttagunta at linaro.org<mailto:viswanath.puttagunta at linaro.org>> wrote: On 3 March 2015 at 21:59, Jonathan Lennox <jonathan at vidyo.com<mailto:jonathan at vidyo.com>> wrote: Viswenath, My patch should be against the tip, but it?s the very recent tip, including some changes this past Friday (27 Feb). I

We have found that it is possible to achieve a 30 to 50% reduction in MHz requirement for implementation of OPUS on Cortex M4 compared to the public version (v1.3 beta/v1.2.1). For the CELT configuration you mention (complexity 0, 16kHz, mono, 20ms) we are measuring a 4ms encode time and a 3ms decode time for that platform (32kbit/s). An important issue that I haven't seen much discussion