Execution of cufftExecC2C() failed

[o-smirnov]

Running this (on jake -- K10 --- /home/oms/RP-3C147):

$ time mt-imager conf=mtimager.properties image.nx=1024 image.ny=1024 channel=-1
2014-04-15 13:25:06 SEVERE  MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)  Leap second table TAI_UTC seems out-of-date. 
2014-04-15 13:25:06 SEVERE  MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)+ Until table is updated (see aips++ manager) times and coordinates
2014-04-15 13:25:06 SEVERE  MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)+ derived from UTC could be wrong by 1s or more.
terminate called after throwing an instance of 'GAFW::GPU::StandardOperators::cuFFTException'
  what():  
Description: Execution of cufftExecC2C() failed
Type: cuFFT Exception
cuFFT Error: CUFFT_EXEC_FAILED
Object Name: Operator::IFFT2D
Object ObjectName: Finalizer-ChoosenChannel-8.UVIfft
File Name: /home/oms/src/mt-imager/code/gafw/Implimentation/StandardOperators/IFFT2D.cpp
Line Number: 108

Aborted (core dumped)

real    0m27.105s
user    0m42.348s
sys 0m18.428s

Though occasionally the same command segfaults earlier on:

$ time mt-imager conf=mtimager.properties image.nx=1024 image.ny=1024 channel=-1
2014-04-15 13:33:59 SEVERE  MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)  Leap second table TAI_UTC seems out-of-date. 
2014-04-15 13:33:59 SEVERE  MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)+ Until table is updated (see aips++ manager) times and coordinates
2014-04-15 13:33:59 SEVERE  MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)+ derived from UTC could be wrong by 1s or more.
Segmentation fault (core dumped)

[o-smirnov]

See also issue #3 !

[ddmusc]

Are you able to generate same error over gpu cards?. The cufft exception given is thrown because the cufft did not execute...by the way is a k10 a Kepler card?

Regards Daniel Muscat

On 15 Apr 2014, at 15:36, oms-ratt notifications@github.com wrote:

Running this (on jake -- K10 --- /home/oms/RP-3C147):

$ time mt-imager conf=mtimager.properties image.nx=1024 image.ny=1024 channel=-1 2014-04-15 13:25:06 SEVERE MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657) Leap second table TAI_UTC seems out-of-date. 2014-04-15 13:25:06 SEVERE MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)+ Until table is updated (see aips++ manager) times and coordinates 2014-04-15 13:25:06 SEVERE MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)+ derived from UTC could be wrong by 1s or more. terminate called after throwing an instance of 'GAFW::GPU::StandardOperators::cuFFTException' what(): Description: Execution of cufftExecC2C() failed Type: cuFFT Exception cuFFT Error: CUFFT_EXEC_FAILED Object Name: Operator::IFFT2D Object ObjectName: Finalizer-ChoosenChannel-8.UVIfft File Name: /home/oms/src/mt-imager/code/gafw/Implimentation/StandardOperators/IFFT2D.cpp Line Number: 108

Aborted (core dumped)

real 0m27.105s user 0m42.348s sys 0m18.428s

Though occasionally the same command segfaults earlier on:

$ time mt-imager conf=mtimager.properties image.nx=1024 image.ny=1024 channel=-1 2014-04-15 13:33:59 SEVERE MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657) Leap second table TAI_UTC seems out-of-date. 2014-04-15 13:33:59 SEVERE MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)+ Until table is updated (see aips++ manager) times and coordinates 2014-04-15 13:33:59 SEVERE MeasTable::dUTC(Double) (file /build/buildd/casacore-1.5.0+svn21411/measures/Measures/MeasTable.cc, line 6657)+ derived from UTC could be wrong by 1s or more. Segmentation fault (core dumped)

— Reply to this email directly or view it on GitHubhttps://github.com/ska-sa/mt-imager/issues/4 .

[o-smirnov]

Works for channel=0 though, it's only the multiple-channel mode that fails.

Yes, the K10 is a Kepler.

[o-smirnov]

P.S. And yes, this particular error seems to be unique to the K10.