[Csnd] pvsblur and reinit = Segmentation fault
| Date | 2011-04-21 01:31 |
| From | peiman khosravi |
| Subject | [Csnd] pvsblur and reinit = Segmentation fault |
Hello,
I have been debugging a csd for hours and have finally found the
culprit. It seems that "pvsblur" causes a segmentation fault when FFT
overlap size is increased to fft/8 during performance, using the
"reinit" opcode:
ktrig changed gkfftsize,gkol, gkwindmult, gkwindtype
if ktrig = 1 then
reinit pass
endif
pass:
ifftsize = i(gkfftsize)
iol = i(gkol)
iNumBins = ifftsize/2 + 1
ihop = ifftsize/iol
iwindow = ifftsize*i(gkwindmult)
iwindtype = i(gkwindtype)
For full csd please see below. Commenting the pvsblur line fixes
everything.
Thanks
Peiman
|
| Date | 2011-04-21 09:16 |
| From | Victor Lazzarini |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
I think I know why. I will check and see if I can fix it. Victor On 21 Apr 2011, at 01:31, peiman khosravi wrote: > Hello, > > I have been debugging a csd for hours and have finally found the > culprit. It seems that "pvsblur" causes a segmentation fault when > FFT overlap size is increased to fft/8 during performance, using the > "reinit" opcode: > > ktrig changed gkfftsize,gkol, gkwindmult, gkwindtype > if ktrig = 1 then > reinit pass > endif > > pass: > > ifftsize = i(gkfftsize) > iol = i(gkol) > iNumBins = ifftsize/2 + 1 > ihop = ifftsize/iol > iwindow = ifftsize*i(gkwindmult) > iwindtype = i(gkwindtype) > > > For full csd please see below. Commenting the pvsblur line fixes > everything. > > Thanks > > Peiman > > > |
| Date | 2011-04-21 09:39 |
| From | Victor Lazzarini |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
As I suspected; fixed in GIT. Thanks for the report. On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: > I think I know why. I will check and see if I can fix it. > > Victor > On 21 Apr 2011, at 01:31, peiman khosravi wrote: > >> Hello, >> >> I have been debugging a csd for hours and have finally found the >> culprit. It seems that "pvsblur" causes a segmentation fault when >> FFT overlap size is increased to fft/8 during performance, using >> the "reinit" opcode: >> >> ktrig changed gkfftsize,gkol, gkwindmult, gkwindtype >> if ktrig = 1 then >> reinit pass >> endif >> >> pass: >> >> ifftsize = i(gkfftsize) >> iol = i(gkol) >> iNumBins = ifftsize/2 + 1 >> ihop = ifftsize/iol >> iwindow = ifftsize*i(gkwindmult) >> iwindtype = i(gkwindtype) >> >> >> For full csd please see below. Commenting the pvsblur line fixes >> everything. >> >> Thanks >> >> Peiman >> >> >> |
| Date | 2011-04-21 11:15 |
| From | peiman khosravi |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
Thanks very much Victor. Best, Peiman On 21/04/2011 09:39, Victor Lazzarini wrote: > As I suspected; fixed in GIT. Thanks for the report. > On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: > >> I think I know why. I will check and see if I can fix it. >> >> Victor >> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >> >>> Hello, >>> >>> I have been debugging a csd for hours and have finally found the >>> culprit. It seems that "pvsblur" causes a segmentation fault when >>> FFT overlap size is increased to fft/8 during performance, using the >>> "reinit" opcode: >>> >>> ktrig changed gkfftsize,gkol, gkwindmult, gkwindtype >>> if ktrig = 1 then >>> reinit pass >>> endif >>> >>> pass: >>> >>> ifftsize = i(gkfftsize) >>> iol = i(gkol) >>> iNumBins = ifftsize/2 + 1 >>> ihop = ifftsize/iol >>> iwindow = ifftsize*i(gkwindmult) >>> iwindtype = i(gkwindtype) >>> >>> >>> For full csd please see below. Commenting the pvsblur line fixes >>> everything. >>> >>> Thanks >>> >>> Peiman >>> >>> >>> |
| Date | 2011-05-12 11:41 |
| From | peiman khosravi |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
Hi Victor, I am just wondering if there is a temporary workaround to this? Thanks Peiman On 21/04/2011 09:39, Victor Lazzarini wrote: > As I suspected; fixed in GIT. Thanks for the report. > On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: > >> I think I know why. I will check and see if I can fix it. >> >> Victor >> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >> >>> Hello, >>> >>> I have been debugging a csd for hours and have finally found the >>> culprit. It seems that "pvsblur" causes a segmentation fault when >>> FFT overlap size is increased to fft/8 during performance, using the >>> "reinit" opcode: >>> >>> ktrig changed gkfftsize,gkol, gkwindmult, gkwindtype >>> if ktrig = 1 then >>> reinit pass >>> endif >>> >>> pass: >>> >>> ifftsize = i(gkfftsize) >>> iol = i(gkol) >>> iNumBins = ifftsize/2 + 1 >>> ihop = ifftsize/iol >>> iwindow = ifftsize*i(gkwindmult) >>> iwindtype = i(gkwindtype) >>> >>> >>> For full csd please see below. Commenting the pvsblur line fixes >>> everything. >>> >>> Thanks >>> >>> Peiman >>> >>> >>> |
| Date | 2011-05-12 11:44 |
| From | peiman khosravi |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
Sorry I think I found one. Just zero the averaging time before changing fftzise. Best Peiman On 12/05/2011 11:41, peiman khosravi wrote: > > Hi Victor, > > I am just wondering if there is a temporary workaround to this? > > Thanks > > Peiman > > On 21/04/2011 09:39, Victor Lazzarini wrote: >> As I suspected; fixed in GIT. Thanks for the report. >> On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: >> >>> I think I know why. I will check and see if I can fix it. >>> >>> Victor >>> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >>> >>>> Hello, >>>> >>>> I have been debugging a csd for hours and have finally found the >>>> culprit. It seems that "pvsblur" causes a segmentation fault when >>>> FFT overlap size is increased to fft/8 during performance, using >>>> the "reinit" opcode: >>>> >>>> ktrig changed gkfftsize,gkol, gkwindmult, gkwindtype >>>> if ktrig = 1 then >>>> reinit pass >>>> endif >>>> >>>> pass: >>>> >>>> ifftsize = i(gkfftsize) >>>> iol = i(gkol) >>>> iNumBins = ifftsize/2 + 1 >>>> ihop = ifftsize/iol >>>> iwindow = ifftsize*i(gkwindmult) >>>> iwindtype = i(gkwindtype) >>>> >>>> >>>> For full csd please see below. Commenting the pvsblur line fixes >>>> everything. >>>> >>>> Thanks >>>> >>>> Peiman >>>> >>>> >>>> |
| Date | 2011-05-12 12:20 |
| From | peiman khosravi |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
Spoke too soon. That doesn't work! On 12/05/2011 11:44, peiman khosravi wrote: > Sorry I think I found one. Just zero the averaging time before > changing fftzise. > > Best > > Peiman > > On 12/05/2011 11:41, peiman khosravi wrote: >> >> Hi Victor, >> >> I am just wondering if there is a temporary workaround to this? >> >> Thanks >> >> Peiman >> >> On 21/04/2011 09:39, Victor Lazzarini wrote: >>> As I suspected; fixed in GIT. Thanks for the report. >>> On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: >>> >>>> I think I know why. I will check and see if I can fix it. >>>> >>>> Victor >>>> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >>>> >>>>> Hello, >>>>> >>>>> I have been debugging a csd for hours and have finally found the >>>>> culprit. It seems that "pvsblur" causes a segmentation fault when >>>>> FFT overlap size is increased to fft/8 during performance, using >>>>> the "reinit" opcode: >>>>> >>>>> ktrig changed gkfftsize,gkol, gkwindmult, gkwindtype >>>>> if ktrig = 1 then >>>>> reinit pass >>>>> endif >>>>> >>>>> pass: >>>>> >>>>> ifftsize = i(gkfftsize) >>>>> iol = i(gkol) >>>>> iNumBins = ifftsize/2 + 1 >>>>> ihop = ifftsize/iol >>>>> iwindow = ifftsize*i(gkwindmult) >>>>> iwindtype = i(gkwindtype) >>>>> >>>>> >>>>> For full csd please see below. Commenting the pvsblur line fixes >>>>> everything. >>>>> >>>>> Thanks >>>>> >>>>> Peiman >>>>> >>>>> >>>>> |
| Date | 2011-05-12 13:08 |
| From | Victor Lazzarini |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
The problem is caused by trying to re-use a buffer that is not big enough, so if your fftsize is going to double, you can try to allocate a max average size that is twice what you need. You might also reduce it to 1/2 when you double the fftsize, eg. fftsize = 1024 max blur time = 1.0 fftsize = 2048 max blur time = 0.5 etc. Victor On 12 May 2011, at 12:20, peiman khosravi wrote: > Spoke too soon. That doesn't work! > > On 12/05/2011 11:44, peiman khosravi wrote: >> Sorry I think I found one. Just zero the averaging time before >> changing fftzise. >> >> Best >> >> Peiman >> >> On 12/05/2011 11:41, peiman khosravi wrote: >>> >>> Hi Victor, >>> >>> I am just wondering if there is a temporary workaround to this? >>> >>> Thanks >>> >>> Peiman >>> >>> On 21/04/2011 09:39, Victor Lazzarini wrote: >>>> As I suspected; fixed in GIT. Thanks for the report. >>>> On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: >>>> >>>>> I think I know why. I will check and see if I can fix it. >>>>> >>>>> Victor >>>>> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >>>>> >>>>>> Hello, >>>>>> >>>>>> I have been debugging a csd for hours and have finally found >>>>>> the culprit. It seems that "pvsblur" causes a segmentation >>>>>> fault when FFT overlap size is increased to fft/8 during >>>>>> performance, using the "reinit" opcode: >>>>>> >>>>>> ktrig changed gkfftsize,gkol, gkwindmult, >>>>>> gkwindtype >>>>>> if ktrig = 1 then >>>>>> reinit pass >>>>>> endif >>>>>> >>>>>> pass: >>>>>> >>>>>> ifftsize = i(gkfftsize) >>>>>> iol = i(gkol) >>>>>> iNumBins = ifftsize/2 + 1 >>>>>> ihop = ifftsize/iol >>>>>> iwindow = ifftsize*i(gkwindmult) >>>>>> iwindtype = i(gkwindtype) >>>>>> >>>>>> >>>>>> For full csd please see below. Commenting the pvsblur line >>>>>> fixes everything. >>>>>> >>>>>> Thanks >>>>>> >>>>>> Peiman >>>>>> >>>>>> >>>>>> |
| Date | 2011-05-12 13:37 |
| From | peiman khosravi |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
Thanks, And where does overlap size come in? It seems that even when I don't change fftsize changing overlap can cause a seg fault. Peiman On 12/05/2011 13:08, Victor Lazzarini wrote: > The problem is caused by trying to re-use a buffer that is not big > enough, so if your fftsize is going to double, > you can try to allocate a max average size that is twice what you > need. You might also reduce it to 1/2 when > you double the fftsize, eg. > > > fftsize = 1024 max blur time = 1.0 > fftsize = 2048 max blur time = 0.5 > > etc. > > > Victor > > On 12 May 2011, at 12:20, peiman khosravi wrote: > >> Spoke too soon. That doesn't work! >> >> On 12/05/2011 11:44, peiman khosravi wrote: >>> Sorry I think I found one. Just zero the averaging time before >>> changing fftzise. >>> >>> Best >>> >>> Peiman >>> >>> On 12/05/2011 11:41, peiman khosravi wrote: >>>> >>>> Hi Victor, >>>> >>>> I am just wondering if there is a temporary workaround to this? >>>> >>>> Thanks >>>> >>>> Peiman >>>> >>>> On 21/04/2011 09:39, Victor Lazzarini wrote: >>>>> As I suspected; fixed in GIT. Thanks for the report. >>>>> On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: >>>>> >>>>>> I think I know why. I will check and see if I can fix it. >>>>>> >>>>>> Victor >>>>>> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >>>>>> >>>>>>> Hello, >>>>>>> >>>>>>> I have been debugging a csd for hours and have finally found the >>>>>>> culprit. It seems that "pvsblur" causes a segmentation fault >>>>>>> when FFT overlap size is increased to fft/8 during performance, >>>>>>> using the "reinit" opcode: >>>>>>> >>>>>>> ktrig changed gkfftsize,gkol, gkwindmult, gkwindtype >>>>>>> if ktrig = 1 then >>>>>>> reinit pass >>>>>>> endif >>>>>>> >>>>>>> pass: >>>>>>> >>>>>>> ifftsize = i(gkfftsize) >>>>>>> iol = i(gkol) >>>>>>> iNumBins = ifftsize/2 + 1 >>>>>>> ihop = ifftsize/iol >>>>>>> iwindow = ifftsize*i(gkwindmult) >>>>>>> iwindtype = i(gkwindtype) >>>>>>> >>>>>>> >>>>>>> For full csd please see below. Commenting the pvsblur line fixes >>>>>>> everything. >>>>>>> >>>>>>> Thanks >>>>>>> >>>>>>> Peiman >>>>>>> >>>>>>> >>>>>>> |
| Date | 2011-05-12 13:58 |
| From | Victor Lazzarini |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
The overlap is the same thing. Double the number of overlaps will need double the buffer space. So overlap = 256 max blur time = 1.0 overlap =128 max blur time = 0.5 On 12 May 2011, at 13:37, peiman khosravi wrote: > Thanks, > > And where does overlap size come in? It seems that even when I don't > change fftsize changing overlap can cause a seg fault. > > Peiman > > On 12/05/2011 13:08, Victor Lazzarini wrote: >> The problem is caused by trying to re-use a buffer that is not big >> enough, so if your fftsize is going to double, >> you can try to allocate a max average size that is twice what you >> need. You might also reduce it to 1/2 when >> you double the fftsize, eg. >> >> >> fftsize = 1024 max blur time = 1.0 >> fftsize = 2048 max blur time = 0.5 >> >> etc. >> >> >> Victor >> >> On 12 May 2011, at 12:20, peiman khosravi wrote: >> >>> Spoke too soon. That doesn't work! >>> >>> On 12/05/2011 11:44, peiman khosravi wrote: >>>> Sorry I think I found one. Just zero the averaging time before >>>> changing fftzise. >>>> >>>> Best >>>> >>>> Peiman >>>> >>>> On 12/05/2011 11:41, peiman khosravi wrote: >>>>> >>>>> Hi Victor, >>>>> >>>>> I am just wondering if there is a temporary workaround to this? >>>>> >>>>> Thanks >>>>> >>>>> Peiman >>>>> >>>>> On 21/04/2011 09:39, Victor Lazzarini wrote: >>>>>> As I suspected; fixed in GIT. Thanks for the report. >>>>>> On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: >>>>>> >>>>>>> I think I know why. I will check and see if I can fix it. >>>>>>> >>>>>>> Victor >>>>>>> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >>>>>>> >>>>>>>> Hello, >>>>>>>> >>>>>>>> I have been debugging a csd for hours and have finally found >>>>>>>> the culprit. It seems that "pvsblur" causes a segmentation >>>>>>>> fault when FFT overlap size is increased to fft/8 during >>>>>>>> performance, using the "reinit" opcode: >>>>>>>> >>>>>>>> ktrig changed gkfftsize,gkol, gkwindmult, >>>>>>>> gkwindtype >>>>>>>> if ktrig = 1 then >>>>>>>> reinit pass >>>>>>>> endif >>>>>>>> >>>>>>>> pass: >>>>>>>> >>>>>>>> ifftsize = i(gkfftsize) >>>>>>>> iol = i(gkol) >>>>>>>> iNumBins = ifftsize/2 + 1 >>>>>>>> ihop = ifftsize/iol >>>>>>>> iwindow = ifftsize*i(gkwindmult) >>>>>>>> iwindtype = i(gkwindtype) >>>>>>>> >>>>>>>> >>>>>>>> For full csd please see below. Commenting the pvsblur line >>>>>>>> fixes everything. >>>>>>>> >>>>>>>> Thanks >>>>>>>> >>>>>>>> Peiman >>>>>>>> >>>>>>>> >>>>>>>> |
| Date | 2011-05-12 14:10 |
| From | peiman khosravi |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
So does this mean that increasing fftsize once and then increasing overlap the buffer must be doubled twice? Thanks Peiman On 12/05/2011 13:58, Victor Lazzarini wrote: > The overlap is the same thing. Double the number of overlaps will need > double the buffer space. > So > > overlap = 256 max blur time = 1.0 > overlap =128 max blur time = 0.5 > > > On 12 May 2011, at 13:37, peiman khosravi wrote: > >> Thanks, >> >> And where does overlap size come in? It seems that even when I don't >> change fftsize changing overlap can cause a seg fault. >> >> Peiman >> >> On 12/05/2011 13:08, Victor Lazzarini wrote: >>> The problem is caused by trying to re-use a buffer that is not big >>> enough, so if your fftsize is going to double, >>> you can try to allocate a max average size that is twice what you >>> need. You might also reduce it to 1/2 when >>> you double the fftsize, eg. >>> >>> >>> fftsize = 1024 max blur time = 1.0 >>> fftsize = 2048 max blur time = 0.5 >>> >>> etc. >>> >>> >>> Victor >>> >>> On 12 May 2011, at 12:20, peiman khosravi wrote: >>> >>>> Spoke too soon. That doesn't work! >>>> >>>> On 12/05/2011 11:44, peiman khosravi wrote: >>>>> Sorry I think I found one. Just zero the averaging time before >>>>> changing fftzise. >>>>> >>>>> Best >>>>> >>>>> Peiman >>>>> >>>>> On 12/05/2011 11:41, peiman khosravi wrote: >>>>>> >>>>>> Hi Victor, >>>>>> >>>>>> I am just wondering if there is a temporary workaround to this? >>>>>> >>>>>> Thanks >>>>>> >>>>>> Peiman >>>>>> >>>>>> On 21/04/2011 09:39, Victor Lazzarini wrote: >>>>>>> As I suspected; fixed in GIT. Thanks for the report. >>>>>>> On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: >>>>>>> >>>>>>>> I think I know why. I will check and see if I can fix it. >>>>>>>> >>>>>>>> Victor >>>>>>>> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >>>>>>>> >>>>>>>>> Hello, >>>>>>>>> >>>>>>>>> I have been debugging a csd for hours and have finally found >>>>>>>>> the culprit. It seems that "pvsblur" causes a segmentation >>>>>>>>> fault when FFT overlap size is increased to fft/8 during >>>>>>>>> performance, using the "reinit" opcode: >>>>>>>>> >>>>>>>>> ktrig changed gkfftsize,gkol, gkwindmult, >>>>>>>>> gkwindtype >>>>>>>>> if ktrig = 1 then >>>>>>>>> reinit pass >>>>>>>>> endif >>>>>>>>> >>>>>>>>> pass: >>>>>>>>> >>>>>>>>> ifftsize = i(gkfftsize) >>>>>>>>> iol = i(gkol) >>>>>>>>> iNumBins = ifftsize/2 + 1 >>>>>>>>> ihop = ifftsize/iol >>>>>>>>> iwindow = ifftsize*i(gkwindmult) >>>>>>>>> iwindtype = i(gkwindtype) >>>>>>>>> >>>>>>>>> >>>>>>>>> For full csd please see below. Commenting the pvsblur line >>>>>>>>> fixes everything. >>>>>>>>> >>>>>>>>> Thanks >>>>>>>>> >>>>>>>>> Peiman >>>>>>>>> >>>>>>>>> >>>>>>>>> |
| Date | 2011-05-12 15:50 |
| From | Victor Lazzarini |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
yes, I guess so. Start with the largest buffer you need, then you probably will not have any issues. On 12 May 2011, at 14:10, peiman khosravi wrote: > So does this mean that increasing fftsize once and then increasing > overlap the buffer must be doubled twice? > > Thanks > > Peiman > > On 12/05/2011 13:58, Victor Lazzarini wrote: >> The overlap is the same thing. Double the number of overlaps will >> need double the buffer space. >> So >> >> overlap = 256 max blur time = 1.0 >> overlap =128 max blur time = 0.5 >> >> >> On 12 May 2011, at 13:37, peiman khosravi wrote: >> >>> Thanks, >>> >>> And where does overlap size come in? It seems that even when I >>> don't change fftsize changing overlap can cause a seg fault. >>> >>> Peiman >>> >>> On 12/05/2011 13:08, Victor Lazzarini wrote: >>>> The problem is caused by trying to re-use a buffer that is not >>>> big enough, so if your fftsize is going to double, >>>> you can try to allocate a max average size that is twice what you >>>> need. You might also reduce it to 1/2 when >>>> you double the fftsize, eg. >>>> >>>> >>>> fftsize = 1024 max blur time = 1.0 >>>> fftsize = 2048 max blur time = 0.5 >>>> >>>> etc. >>>> >>>> >>>> Victor >>>> >>>> On 12 May 2011, at 12:20, peiman khosravi wrote: >>>> >>>>> Spoke too soon. That doesn't work! >>>>> >>>>> On 12/05/2011 11:44, peiman khosravi wrote: >>>>>> Sorry I think I found one. Just zero the averaging time before >>>>>> changing fftzise. >>>>>> >>>>>> Best >>>>>> >>>>>> Peiman >>>>>> >>>>>> On 12/05/2011 11:41, peiman khosravi wrote: >>>>>>> >>>>>>> Hi Victor, >>>>>>> >>>>>>> I am just wondering if there is a temporary workaround to this? >>>>>>> >>>>>>> Thanks >>>>>>> >>>>>>> Peiman >>>>>>> >>>>>>> On 21/04/2011 09:39, Victor Lazzarini wrote: >>>>>>>> As I suspected; fixed in GIT. Thanks for the report. >>>>>>>> On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: >>>>>>>> >>>>>>>>> I think I know why. I will check and see if I can fix it. >>>>>>>>> >>>>>>>>> Victor >>>>>>>>> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >>>>>>>>> >>>>>>>>>> Hello, >>>>>>>>>> >>>>>>>>>> I have been debugging a csd for hours and have finally >>>>>>>>>> found the culprit. It seems that "pvsblur" causes a >>>>>>>>>> segmentation fault when FFT overlap size is increased to >>>>>>>>>> fft/8 during performance, using the "reinit" opcode: >>>>>>>>>> >>>>>>>>>> ktrig changed gkfftsize,gkol, gkwindmult, >>>>>>>>>> gkwindtype >>>>>>>>>> if ktrig = 1 then >>>>>>>>>> reinit pass >>>>>>>>>> endif >>>>>>>>>> >>>>>>>>>> pass: >>>>>>>>>> >>>>>>>>>> ifftsize = i(gkfftsize) >>>>>>>>>> iol = i(gkol) >>>>>>>>>> iNumBins = ifftsize/2 + 1 >>>>>>>>>> ihop = ifftsize/iol >>>>>>>>>> iwindow = ifftsize*i(gkwindmult) >>>>>>>>>> iwindtype = i(gkwindtype) >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> For full csd please see below. Commenting the pvsblur line >>>>>>>>>> fixes everything. >>>>>>>>>> >>>>>>>>>> Thanks >>>>>>>>>> >>>>>>>>>> Peiman >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> |
| Date | 2011-05-12 16:00 |
| From | peiman khosravi |
| Subject | Re: [Csnd] pvsblur and reinit = Segmentation fault |
Thanks, I am going to try that. It's just that I use pvsblur to average my noise-prints. Best, Peiman On 12/05/2011 15:50, Victor Lazzarini wrote: > yes, I guess so. Start with the largest buffer you need, then you > probably will not have any issues. > On 12 May 2011, at 14:10, peiman khosravi wrote: > >> So does this mean that increasing fftsize once and then increasing >> overlap the buffer must be doubled twice? >> >> Thanks >> >> Peiman >> >> On 12/05/2011 13:58, Victor Lazzarini wrote: >>> The overlap is the same thing. Double the number of overlaps will >>> need double the buffer space. >>> So >>> >>> overlap = 256 max blur time = 1.0 >>> overlap =128 max blur time = 0.5 >>> >>> >>> On 12 May 2011, at 13:37, peiman khosravi wrote: >>> >>>> Thanks, >>>> >>>> And where does overlap size come in? It seems that even when I >>>> don't change fftsize changing overlap can cause a seg fault. >>>> >>>> Peiman >>>> >>>> On 12/05/2011 13:08, Victor Lazzarini wrote: >>>>> The problem is caused by trying to re-use a buffer that is not big >>>>> enough, so if your fftsize is going to double, >>>>> you can try to allocate a max average size that is twice what you >>>>> need. You might also reduce it to 1/2 when >>>>> you double the fftsize, eg. >>>>> >>>>> >>>>> fftsize = 1024 max blur time = 1.0 >>>>> fftsize = 2048 max blur time = 0.5 >>>>> >>>>> etc. >>>>> >>>>> >>>>> Victor >>>>> >>>>> On 12 May 2011, at 12:20, peiman khosravi wrote: >>>>> >>>>>> Spoke too soon. That doesn't work! >>>>>> >>>>>> On 12/05/2011 11:44, peiman khosravi wrote: >>>>>>> Sorry I think I found one. Just zero the averaging time before >>>>>>> changing fftzise. >>>>>>> >>>>>>> Best >>>>>>> >>>>>>> Peiman >>>>>>> >>>>>>> On 12/05/2011 11:41, peiman khosravi wrote: >>>>>>>> >>>>>>>> Hi Victor, >>>>>>>> >>>>>>>> I am just wondering if there is a temporary workaround to this? >>>>>>>> >>>>>>>> Thanks >>>>>>>> >>>>>>>> Peiman >>>>>>>> >>>>>>>> On 21/04/2011 09:39, Victor Lazzarini wrote: >>>>>>>>> As I suspected; fixed in GIT. Thanks for the report. >>>>>>>>> On 21 Apr 2011, at 09:16, Victor Lazzarini wrote: >>>>>>>>> >>>>>>>>>> I think I know why. I will check and see if I can fix it. >>>>>>>>>> >>>>>>>>>> Victor >>>>>>>>>> On 21 Apr 2011, at 01:31, peiman khosravi wrote: >>>>>>>>>> >>>>>>>>>>> Hello, >>>>>>>>>>> >>>>>>>>>>> I have been debugging a csd for hours and have finally found >>>>>>>>>>> the culprit. It seems that "pvsblur" causes a segmentation >>>>>>>>>>> fault when FFT overlap size is increased to fft/8 during >>>>>>>>>>> performance, using the "reinit" opcode: >>>>>>>>>>> >>>>>>>>>>> ktrig changed gkfftsize,gkol, gkwindmult, >>>>>>>>>>> gkwindtype >>>>>>>>>>> if ktrig = 1 then >>>>>>>>>>> reinit pass >>>>>>>>>>> endif >>>>>>>>>>> >>>>>>>>>>> pass: >>>>>>>>>>> >>>>>>>>>>> ifftsize = i(gkfftsize) >>>>>>>>>>> iol = i(gkol) >>>>>>>>>>> iNumBins = ifftsize/2 + 1 >>>>>>>>>>> ihop = ifftsize/iol >>>>>>>>>>> iwindow = ifftsize*i(gkwindmult) >>>>>>>>>>> iwindtype = i(gkwindtype) >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> For full csd please see below. Commenting the pvsblur line >>>>>>>>>>> fixes everything. >>>>>>>>>>> >>>>>>>>>>> Thanks >>>>>>>>>>> >>>>>>>>>>> Peiman >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> |