[Webkit-unassigned] [Bug 75522] Enable IPP for FFTFrame

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=75522





--- Comment #11 from Raymond Toy <rtoy at chromium.org>  2012-01-06 09:02:31 PST ---
(In reply to comment #10)
> (In reply to comment #9)
> > 
> > I tried the convolution layout test and failed with the result:
> > *******
> > Tests ConvolverNode processing a mono channel with mono impulse response.
> > 
> > On success, you will see a series of "PASS" messages, followed by "TEST COMPLETE".
> > 
> > PASS Initial latency of convolver is silent.
> > FAIL Triangular portion of convolution is not incorrect.  Max deviation = 2.0867834820236123e-7
> > FAIL First part of tail of convolution is not sufficiently small: 0.00244140625
> > PASS Rendered signal after tail of convolution is silent.
> > FAIL Test signal was not correctly convolved.
> > PASS successfullyParsed is true
> > 
> > TEST COMPLETE
> > ******
> > 
> > I found the threshold values are experimental, the "Max deviation" and "small" I got are only larger than the values in test script a little. I wonder to know whether such values are tolerable?
> 
> I believe this is well within an acceptable tolerance, and the performance improvement is significant.  It would be great to go with the "fast" option.
> 
> Ray, can you sanity check this and refine the layout test by re-defining the "Max deviation" and "not sufficiently small" in terms of decibels. I can see that "Max deviation" in terms of dB deviation is very slight.  The "not sufficiently small" can be compared to the "peak" value of the triangle to see how far down "in the noise" we are.

I believe the values returned are sufficiently close.  I will modify the current test to compute a "dB" value instead.  I will leave the constants as is.  When the new FFT is landed, the thresholds should be adjusted then so we have a record of what the new FFT did.  (I also need to test this on OSX in case the OSX FFT produces different results.)

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.