Re: [Audacity-devel] VC2012 and wxWidgets30

[Vaughan J. <va...@au...>]

On 2/19/2014 3:17 PM, James Crook wrote:
> Hi Andrew,
>
> Can you give a status update on the shift to vs2012 and wx3.0?
> Are your changes ready for prime time?
>
>
> My understanding is:
> - The switch from wx2.8 to wx3.0 is controlled by a #define.  So we
> could take the code right now, and stay on 2.8 and vs2008 with no issues.
> - wx3.0 opens a route to supporting retina display on Mac.  Not
> practical with wx2.8.
> - There are other advantages to shifting to wx3.0 and vc2012. (please
> spell them out).
> - The combination wx3.0 + vs2012 works.
> - The combination wx2.8 + vs2012 works.
> - The combination wx2.8 + vs2008 works.
> - The combination wx3.0 + vs2008 doesn't work.  Please say a bit more
> about the Unicode problem you encountered here.  I don't think we
> necessarily would need that combination, but it would be good to
> understand whether it is 'our problem' or wxWidgets'.
>
> We do want to be able to support retina display.
> At some point we want to move to vs2012.
> At some point we want to move to wx3.0.
>
> As (it sounds like) you have done the work,

+1.


>now could be a very good
> time to switch both, and we just need to get agreement that that is a
> good step to make.

Very much -1. There's already a great deal of new stuff in 2.0.6-alpha. 
Even with the possible safe encapsulation via #define (and we've seen it 
fail before), I'm opposed to adding substantial changes until we release 
2.0.6.

And I think we should release 2.0.6 soon. That requires some discussions 
on team@, not here, though.

- Vaughan


>   From the sounds of things I would be keen at the very least to get the
> code that works in both wx2.8 and wx3.0 into svn, and if no one has
> objections, then take the next step and switch to both wx3.0 and vs2012.
>
> --James.
>
>
>
>
> On 20/01/2014 04:50, Andrew Hallendorff wrote:
>> On 1/19/2014 7:20 PM, Vaughan Johnson wrote:
>>> On 1/18/2014 5:43 PM, Andrew Hallendorff wrote:
>>>> I don't mind being the only one who can run on 2012 wx3. wx2.8 is not
>>>> supported in 2010+.
>>> Really? Leland and Martyn got it working with VS2012, afaik.
>>>
>>>
>> I couldn't get it to work. When I compiled with the
>>
>>        #if !defined __WXWINCE__ && !defined NEED_PBT_H
>>        #include <pbt.h>
>>        #endif
>>
>> removed from widgets. It would run but you couldn't load a file. It
>> would puke out an
>>
>>        Unhandled exception at 0x59CD73B3 (msvcr110d.dll) in Audacity.exe:
>> An invalid parameter was passed to a function that considers invalid
>> parameters fatal.
>>
>> Basically it seemed to be trying to load  a file with a unicode string
>> and puking.
>>
>>
>>
>> So time for a funny story.
>>
>> I did some work to put AVX into the EQ code just to see how it would
>> run. (vc2008 does not do AVX btw) In doing so I doubled the buffers when
>> AVX is present as it basically uses 8x the data the same way sse uses 4x
>> the data. This does not prevent sse from running in the same buffers. So
>> I put a little function to half the size of the buffer in the sse
>> routine if it detects AVX being active. I put in.
>>
>>        int
>> subBufferSize=mBufferCount<8?(mSubBufferSize>>1):mSubBufferSize; // half
>> the buffers if avx is active
>>
>> but it should of been.
>>
>>        int
>> subBufferSize=mBufferCount==8?(mSubBufferSize>>1):mSubBufferSize; //
>> half the buffers if avx is active
>>
>> Because of this the 4x routines copied half the data in and out when
>> they executed. Thus doing twice the work and ignoring half the processed
>> calculations.
>>
>> So after this change, I went back to testing. I started getting not so
>> great results for sse. So instead of looking for the problem I went to
>> tweaking VS to see why this was the case. After this didn't work I wrote
>> a threaded single operation math path version and got results almost
>> equal to the sse code. After reading somewhere on the net about VS2008
>> not handling sse well, I said to myself the next logical step is to test
>> on vc2012. After spending half a day not getting vc2012 to work with
>> wx28, I looked at the wxWidgets site and read the wx28 was not
>> compatible with vc2012. So then I found the posts about getting wx3 to
>> work and followed them. I didn't get very far.
>>
>> Edgar's post in the forums was a great start but there was no final
>> resolution. After making a few of the changes necessary I figured out
>> why. Although many of the changes were trivial, some were kind of
>> advanced requiring new virtual functions and such. I decided if I was
>> going to do all the changes I might as well make every change reversible
>> to the wx28 base regardless of how trivial they were. Believe me it was
>> hard to leave some of them in. This is good because they are marked and
>> can be validated later. I ironically spent about 3hrs trying to chase
>> down why FileDialog would link error for some odd reason. It turns out
>> that the project of FileDialog has its own path to widgets and because
>> of this it was still pointing to the wx28 directory. Yes I decided to do
>> this work in such a way that both 2.8 and 3.0 can exist on the same
>> machine.
>>
>> A real strange error was http://trac.wxwidgets.org/ticket/11482 in
>> XMLTagHandler.cpp on a wx void yeald()=0 virtual function.
>>
>> So I finally finish debug, it seems to run fine. Thinking I'm done I
>> move on to release and it dies on all these files. So after chasing a
>> few rabbits, it turns out that some files were inheriting their headers
>> from other files and some files didn't get the functions they needed.
>> Tedious but manageable, I finally get release to run.
>>
>> I get to testing and one thing you begin to notice in 2012 is debug is
>> kind of slow. I don't know if it's doing extra checks during runtime but
>> it makes comparing things hard. I run the same tests in release and I
>> get basically the same results as vc2008. So now I know the error is in
>> the routines. Luckily it actually didn't take me long to figure out the
>> subBufferSize error above.
>>
>> So all the above is a byproduct of a misplaced inequality.
>>
>> Andrew Hallendorff
>
>
> ------------------------------------------------------------------------------
> Managing the Performance of Cloud-Based Applications
> Take advantage of what the Cloud has to offer - Avoid Common Pitfalls.
> Read the Whitepaper.
> http://pubads.g.doubleclick.net/gampad/clk?id=121054471&iu=/4140/ostg.clktrk
> _______________________________________________
> audacity-devel mailing list
> aud...@li...
> https://lists.sourceforge.net/lists/listinfo/audacity-devel
>