Okay, Not exactly the answer I was looking for. This sounds like a big change. I don't mind re-writing the LSP quantizer but re-training code books and breaking compatibility is not what I want to do. I am working on an optimization for an ARM cortex-A8. It is desirable to process things in 4 element blocks. Is there a simpler approach you could recommend? Vasant Shridhar ________________________________________ From: Jean-Marc Valin [jmvalin at jmvalin.ca] Sent: Monday, April 11, 2011 5:29 PM To: Shridhar, Vasant Cc: speex-dev at xiph.org Subject: Re: [Speex-dev] lpcSize On 11-04-11 11:05 AM, Shridhar, Vasant wrote:> I would like to make lpcSize a multiple of 4. In the current library for > narrow band mode lpcSize is currently set to 10. I would like to > increase this 12 for optimization reasons. Is it a simple matter of just > changing lpcSize in the structures in mode.c or are there other > implications to doing this?Nothing major except you'd need to rewrite the LSP quantizer, retrain half the codebooks, change the bit-allocation and break compatibility with the format. Cheers, Jean-Marc
On 11-04-11 06:48 PM, Shridhar, Vasant wrote:> Okay, > > Not exactly the answer I was looking for. This sounds like a big > change. I don't mind re-writing the LSP quantizer but re-training > code books and breaking compatibility is not what I want to do. I am > working on an optimization for an ARM cortex-A8. It is desirable to > process things in 4 element blocks. Is there a simpler approach you > could recommend?Well, nothing prevents you from simply padding with zeros in the functions you're trying to optimize. Jean-Marc> Vasant Shridhar ________________________________________ From: > Jean-Marc Valin [jmvalin at jmvalin.ca] Sent: Monday, April 11, 2011 > 5:29 PM To: Shridhar, Vasant Cc: speex-dev at xiph.org Subject: Re: > [Speex-dev] lpcSize > > On 11-04-11 11:05 AM, Shridhar, Vasant wrote: >> I would like to make lpcSize a multiple of 4. In the current >> library for narrow band mode lpcSize is currently set to 10. I >> would like to increase this 12 for optimization reasons. Is it a >> simple matter of just changing lpcSize in the structures in mode.c >> or are there other implications to doing this? > > Nothing major except you'd need to rewrite the LSP quantizer, > retrain half the codebooks, change the bit-allocation and break > compatibility with the format. > > Cheers, > > Jean-Marc > >
I thought about this. However this would mean I would have to copy the data over to a temp array which can be expensive on this part. The size of the filters are small which tends to remove the benefit for this part. I will look into this a bit more. How are the code books trained? Vas ________________________________________ From: Jean-Marc Valin [jmvalin at jmvalin.ca] Sent: Monday, April 11, 2011 7:18 PM To: Shridhar, Vasant Cc: speex-dev at xiph.org Subject: Re: [Speex-dev] lpcSize On 11-04-11 06:48 PM, Shridhar, Vasant wrote:> Okay, > > Not exactly the answer I was looking for. This sounds like a big > change. I don't mind re-writing the LSP quantizer but re-training > code books and breaking compatibility is not what I want to do. I am > working on an optimization for an ARM cortex-A8. It is desirable to > process things in 4 element blocks. Is there a simpler approach you > could recommend?Well, nothing prevents you from simply padding with zeros in the functions you're trying to optimize. Jean-Marc> Vasant Shridhar ________________________________________ From: > Jean-Marc Valin [jmvalin at jmvalin.ca] Sent: Monday, April 11, 2011 > 5:29 PM To: Shridhar, Vasant Cc: speex-dev at xiph.org Subject: Re: > [Speex-dev] lpcSize > > On 11-04-11 11:05 AM, Shridhar, Vasant wrote: >> I would like to make lpcSize a multiple of 4. In the current >> library for narrow band mode lpcSize is currently set to 10. I >> would like to increase this 12 for optimization reasons. Is it a >> simple matter of just changing lpcSize in the structures in mode.c >> or are there other implications to doing this? > > Nothing major except you'd need to rewrite the LSP quantizer, > retrain half the codebooks, change the bit-allocation and break > compatibility with the format. > > Cheers, > > Jean-Marc > >