args: deepcl_unittests.exe --gtest_filter=-SLOW*
Note: Google Test filter = -SLOW*
[==========] Running 159 tests from 29 test cases.
[----------] Global test environment set-up.
[----------] 7 tests from testClBlas
[ RUN      ] testClBlas.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
clblas teardown
[       OK ] testClBlas.basic (430 ms)
[ RUN      ] testClBlas.transA
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
1 2 9
3 7 5
initializing clblas
clblas teardown
[       OK ] testClBlas.transA (90 ms)
[ RUN      ] testClBlas.transB
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
3
-1
initializing clblas
clblas teardown
[       OK ] testClBlas.transB (100 ms)
[ RUN      ] testClBlas.colMajor
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
clblas teardown
[       OK ] testClBlas.colMajor (80 ms)
[ RUN      ] testClBlas.colMajor2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
clblas teardown
[       OK ] testClBlas.colMajor2 (90 ms)
[ RUN      ] testClBlas.colMajorTransA
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
clblas teardown
[       OK ] testClBlas.colMajorTransA (100 ms)
[ RUN      ] testClBlas.colMajorTransB
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
clblas teardown
[       OK ] testClBlas.colMajorTransB (90 ms)
[----------] 7 tests from testClBlas (980 ms total)

[----------] 1 test from testDeepCL
[ RUN      ] testDeepCL.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
expected number of output: 4
clblas teardown
[       OK ] testDeepCL.basic (430 ms)
[----------] 1 test from testDeepCL (430 ms total)

[----------] 23 tests from testupdateweights
[ RUN      ] testupdateweights.conv1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=2 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=5 numFilters=2 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:SquareLossLayer{}

layer 0:InputLayer{ outputPlanes=2 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=5 numFilters=2 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:SquareLossLayer{}

batchSize: 4
inputtotalsize=200 outputTotalSize=72
layer ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=5 numFilters=2 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
weightsize=36 biassize=0
statefultimer v0.7
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=2 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=5 numFilters=2 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:SquareLossLayer{}
Parameters overview: (skipping 2 layers with 0 params)
layer 1: params=36	100.0%
TOTAL  : params=36
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
idx=8 predicted losschange=0.000111445 actual=0.000112534
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
idx=13 predicted losschange=-0.000886715 actual=-0.000884056
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
idx=0 predicted losschange=0.000210491 actual=0.000212669
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 440ms
idx=22 predicted losschange=-0.000164224 actual=0.000212669
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 440ms
   forward layer selected kernel 1
idx=22 predicted losschange=-0.000164224 actual=-0.000163078
idx=35 predicted losschange=-0.000391028 actual=-0.000391006
idx=26 predicted losschange=2.23142e-05 actual=2.57492e-05
idx=27 predicted losschange=9.38328e-05 actual=9.44138e-05
idx=27 predicted losschange=9.38328e-05 actual=9.44138e-05
idx=10 predicted losschange=0.00186697 actual=0.00187111
clblas teardown
[       OK ] testupdateweights.conv1 (1380 ms)
[ RUN      ] testupdateweights.conv1z
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=2 outputSize=3 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=3 numFilters=2 filterSize=3 outputSize=3 padZeros=1 biased=0 skip=0} }
layer 2:SquareLossLayer{}

layer 0:InputLayer{ outputPlanes=2 outputSize=3 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=3 numFilters=2 filterSize=3 outputSize=3 padZeros=1 biased=0 skip=0} }
layer 2:SquareLossLayer{}

batchSize: 4
inputtotalsize=72 outputTotalSize=72
layer ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=3 numFilters=2 filterSize=3 outputSize=3 padZeros=1 biased=0 skip=0} }
weightsize=36 biassize=0
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=2 outputSize=3 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=3 numFilters=2 filterSize=3 outputSize=3 padZeros=1 biased=0 skip=0} }
layer 2:SquareLossLayer{}
Parameters overview: (skipping 2 layers with 0 params)
layer 1: params=36	100.0%
TOTAL  : params=36
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
idx=8 predicted losschange=0.00039831 actual=0.000397682
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
idx=13 predicted losschange=-0.000426502 actual=-0.000426292
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
idx=0 predicted losschange=0.000143287 actual=0.000144005
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, padzeros must be disabled
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 430ms
idx=22 predicted losschange=-1.7916e-06 actual=0.000144005
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 430ms
   forward layer selected kernel 1
idx=22 predicted losschange=-1.7916e-06 actual=0
idx=35 predicted losschange=-2.82565e-05 actual=-2.76566e-05
idx=26 predicted losschange=3.62191e-05 actual=3.71933e-05
idx=27 predicted losschange=-0.000319862 actual=-0.000317574
idx=27 predicted losschange=-0.000319862 actual=-0.000317574
idx=10 predicted losschange=-0.000883857 actual=-0.000883102
clblas teardown
[       OK ] testupdateweights.conv1z (1390 ms)
[ RUN      ] testupdateweights.numericallytest
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 10ms
layer 0:InputLayer{ outputPlanes=1 outputSize=1 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=1 numFilters=1 filterSize=1 outputSize=1 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=1	100.0%
TOTAL  : params=1
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=1 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=1 numFilters=1 filterSize=1 outputSize=1 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=1	100.0%
TOTAL  : params=1
 loss 0.0367983 loss2 0.0367913 change: 7.01472e-06
sumweightsdiff -0.000264842
 loss change              7.01472e-06
 estimatedLossChangeFromW 7.01413e-06
[       OK ] testupdateweights.numericallytest (850 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=3 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=3 numFilters=1 filterSize=1 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=1	100.0%
TOTAL  : params=1
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=3 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=3 numFilters=1 filterSize=1 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=1	100.0%
TOTAL  : params=1
 loss 1.23358 loss2 1.21612 change: 0.0174606
sumweightsdiff -0.0132709
 loss change              0.0174606
 estimatedLossChangeFromW 0.0176118
[       OK ] testupdateweights.numericallytest_imagesize3 (860 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize5
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=1 filterSize=1 outputSize=5 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=1	100.0%
TOTAL  : params=1
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=1 filterSize=1 outputSize=5 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=1	100.0%
TOTAL  : params=1
 loss 4.12958 loss2 4.11952 change: 0.0100665
sumweightsdiff -0.0101708
 loss change              0.0100665
 estimatedLossChangeFromW 0.0103444
[       OK ] testupdateweights.numericallytest_imagesize5 (890 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize9
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=9 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=9 numFilters=1 filterSize=1 outputSize=9 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=1	100.0%
TOTAL  : params=1
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=9 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=9 numFilters=1 filterSize=1 outputSize=9 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=1	100.0%
TOTAL  : params=1
 loss 13.4341 loss2 13.4339 change: 0.000207901
sumweightsdiff 0.00153953
 loss change              0.000207901
 estimatedLossChangeFromW 0.000237015
[       OK ] testupdateweights.numericallytest_imagesize9 (890 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize9_filtersize9
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=9 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=9 numFilters=1 filterSize=9 outputSize=1 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=81	100.0%
TOTAL  : params=81
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=9 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=9 numFilters=1 filterSize=9 outputSize=1 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=81	100.0%
TOTAL  : params=81
 loss 0.135896 loss2 0.0848782 change: 0.0510182
sumweightsdiff -0.0322406
 loss change              0.0510182
 estimatedLossChangeFromW 0.0555841
[       OK ] testupdateweights.numericallytest_imagesize9_filtersize9 (930 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize9_filtersize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=9 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=9 numFilters=1 filterSize=3 outputSize=7 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=9	100.0%
TOTAL  : params=9
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=9 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=9 numFilters=1 filterSize=3 outputSize=7 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=9	100.0%
TOTAL  : params=9
 loss 7.70633 loss2 7.41581 change: 0.290529
sumweightsdiff -0.0898812
 loss change              0.290529
 estimatedLossChangeFromW 0.316231
[       OK ] testupdateweights.numericallytest_imagesize9_filtersize3 (940 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize3_filtersize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=3 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=3 numFilters=1 filterSize=3 outputSize=1 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=9	100.0%
TOTAL  : params=9
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=3 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=3 numFilters=1 filterSize=3 outputSize=1 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=9	100.0%
TOTAL  : params=9
 loss 0.0719101 loss2 0.0694461 change: 0.00246406
sumweightsdiff -0.0110647
 loss change              0.00246406
 estimatedLossChangeFromW 0.00248372
[       OK ] testupdateweights.numericallytest_imagesize3_filtersize3 (880 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize5_filtersize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=9	100.0%
TOTAL  : params=9
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 10ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=9	100.0%
TOTAL  : params=9
 loss 1.20022 loss2 1.17241 change: 0.0278131
sumweightsdiff -0.0203888
 loss change              0.0278131
 estimatedLossChangeFromW 0.0280929
[       OK ] testupdateweights.numericallytest_imagesize5_filtersize3 (910 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize5_filtersize3_batchsize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=9	100.0%
TOTAL  : params=9
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=9	100.0%
TOTAL  : params=9
 loss 4.97142 loss2 4.78768 change: 0.183744
sumweightsdiff -0.056004
 loss change              0.183744
 estimatedLossChangeFromW 0.193264
[       OK ] testupdateweights.numericallytest_imagesize5_filtersize3_batchsize3 (900 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize5_filtersize3_planes3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=3 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=3 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=27	100.0%
TOTAL  : params=27
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=3 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=3 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=27	100.0%
TOTAL  : params=27
 loss 1.08887 loss2 0.9575 change: 0.13137
sumweightsdiff -0.00764531
 loss change              0.13137
 estimatedLossChangeFromW 0.134379
[       OK ] testupdateweights.numericallytest_imagesize5_filtersize3_planes3 (940 ms)
[ RUN      ] testupdateweights.numericallytest_imagesize5_filtersize3_planes3_batchsize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 10ms
layer 0:InputLayer{ outputPlanes=3 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=3 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=27	100.0%
TOTAL  : params=27
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 0:InputLayer{ outputPlanes=3 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=3 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:ActivationLayer{ TANH }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=27	100.0%
TOTAL  : params=27
 loss 4.76631 loss2 4.18154 change: 0.584769
sumweightsdiff 0.029606
 loss change              0.584769
 estimatedLossChangeFromW 0.620442
[       OK ] testupdateweights.numericallytest_imagesize5_filtersize3_planes3_batchsize3 (940 ms)
[ RUN      ] testupdateweights.backprop_weights_2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=1 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
mismatch for i 0
[       OK ] testupdateweights.backprop_weights_2 (110 ms)
[ RUN      ] testupdateweights.backprop_weights_2_upstreamimagesize2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=2 -D gInputSizeSquared=4 -D gNumFilters=1 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=2 -D gOutputSizeSquared=4 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=2 -DgInputStripeOuterNumRows=2 -DgInputStripeInnerSize=4 -DgInputStripeOuterSize=4 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=2 -DgOutputStripeSize=4
mismatch for i 0
[       OK ] testupdateweights.backprop_weights_2_upstreamimagesize2 (110 ms)
[ RUN      ] testupdateweights.backprop_weights_2_upstreamimagesize3_filtersize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=1 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
mismatch for i 0
mismatch for i 1
mismatch for i 2
mismatch for i 3
mismatch for i 4
mismatch for i 5
mismatch for i 6
mismatch for i 7
mismatch for i 8
[       OK ] testupdateweights.backprop_weights_2_upstreamimagesize3_filtersize3 (110 ms)
[ RUN      ] testupdateweights.backprop_weights_2_upstreamimagesize4_filtersize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=4 -D gInputSizeSquared=16 -D gNumFilters=1 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=2 -D gOutputSizeSquared=4 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=4 -DgInputStripeOuterNumRows=8 -DgInputStripeInnerSize=16 -DgInputStripeOuterSize=32 -DgInputStripeMarginSize=8 -DgOutputStripeNumRows=2 -DgOutputStripeSize=4
mismatch for i 0
mismatch for i 8
[       OK ] testupdateweights.backprop_weights_2_upstreamimagesize4_filtersize3 (120 ms)
[ RUN      ] testupdateweights.backprop_weights_2_upstreamimagesize5_filtersize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=1 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=3 -D gOutputSizeSquared=9 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=3 -DgOutputStripeSize=9
mismatch for i 0
mismatch for i 4
mismatch for i 8
[       OK ] testupdateweights.backprop_weights_2_upstreamimagesize5_filtersize3 (120 ms)
[ RUN      ] testupdateweights.backprop_weights_2_upstreamimagesize3_filtersize1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=1 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=3 -D gOutputSizeSquared=9 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=3 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=9 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=3 -DgOutputStripeSize=9
mismatch for i 0
[       OK ] testupdateweights.backprop_weights_2_upstreamimagesize3_filtersize1 (110 ms)
[ RUN      ] testupdateweights.backprop_weights_2_upstreamimagesize16_filtersize1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=16 -D gInputSizeSquared=256 -D gNumFilters=1 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=16 -D gOutputSizeSquared=256 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=16 -DgInputStripeOuterNumRows=16 -DgInputStripeInnerSize=256 -DgInputStripeOuterSize=256 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=16 -DgOutputStripeSize=256
mismatch for i 0
[       OK ] testupdateweights.backprop_weights_2_upstreamimagesize16_filtersize1 (240 ms)
[ RUN      ] testupdateweights.backprop_weights_2_upstreamimagesize17_filtersize1
LayerDimensions{ inputPlanes=1 inputSize=17 numFilters=1 filterSize=1 outputSize=17 padZeros=0 biased=0 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=17 -D gInputSizeSquared=289 -D gNumFilters=1 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=17 -D gOutputSizeSquared=289 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=17 -DgInputStripeOuterNumRows=17 -DgInputStripeInnerSize=289 -DgInputStripeOuterSize=289 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=17 -DgOutputStripeSize=289
mismatch for i 0
[       OK ] testupdateweights.backprop_weights_2_upstreamimagesize17_filtersize1 (390 ms)
[ RUN      ] testupdateweights.backprop_weights_2_upstreamimagesize17_filtersize1_moredata
expectedresult: -958.715
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=17 -D gInputSizeSquared=289 -D gNumFilters=1 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=17 -D gOutputSizeSquared=289 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=17 -DgInputStripeOuterNumRows=17 -DgInputStripeInnerSize=289 -DgInputStripeOuterSize=289 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=17 -DgOutputStripeSize=289
mismatch for i 0
[       OK ] testupdateweights.backprop_weights_2_upstreamimagesize17_filtersize1_moredata (380 ms)
[ RUN      ] testupdateweights.backprop_instance3_smaller2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
numweights: 36
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=96 -D gInputSizeSquared=9216 -D gNumFilters=1 -D gFilterSize=6 -D gHalfFilterSize=3 -D gFilterSizeSquared=36 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=91 -D gOutputSizeSquared=8281 -D gPadZeros=0 -D gMargin=0 -D gEven=1 -D gSkip=0 -DgNumStripes=16 -DgInputStripeMarginRows=5 -DgInputStripeInnerNumRows=6 -DgInputStripeOuterNumRows=16 -DgInputStripeInnerSize=576 -DgInputStripeOuterSize=1536 -DgInputStripeMarginSize=480 -DgOutputStripeNumRows=6 -DgOutputStripeSize=546
138 0 0 0 0 0
132 0 0 0 0 0
138 0 0 0 0 0
138 0 0 0 0 0
138 0 0 0 0 0
132 0 0 0 0 0

138 0 0 0 0 0
132 0 0 0 0 0
138 0 0 0 0 0
138 0 0 0 0 0
138 0 0 0 0 0
132 0 0 0 0 0

......
......
......
......
......
......

0=0 0 0 0 0 0 0 0
1=0 0 0 0 0 0 0 0
2=0 0 0 0 0 0 0 0
3=0 0 0 0 0 0 0 0
4=0 0 0 0 0 0 0 0
5=0 0 0 0 0 0 0 0
6=0 0 0 0 0 0 0 0
7=0 0 0 0 0 0 0 0
8=0 0 0 0 0 0 0 0
9=0 0 0 0 0 0 0 0
10=0 0 0 0 0 0 0 0
11=0 0 0 0 0 0 0 0

0=0 0 0 0 0 0 0 0
1=0 0 0 0 0 0 0 0
2=0 0 0 0 0 0 0 0
3=0 0 0 0 0 0 0 0
4=0 0 0 0 0 0 0 0
5=0 0 0 0 0 0 0 0
6=0 0 0 0 0 0 0 0
7=0 0 0 0 0 0 0 0
8=0 0 0 0 0 0 0 0
9=0 0 0 0 0 0 0 0
10=0 0 0 0 0 0 0 0
11=0 0 0 0 0 0 0 0
12=0 0 0 0 0 0 0 0
13=0 0 0 0 0 0 0 0
14=0 0 0 0 0 0 0 0
15=0 0 0 0 0 0 0 0
16=0 0 0 0 0 0 0 0
17=0 0 0 0 0 0 0 0
18=0 0 0 0 0 0 0 0
19=0 0 0 0 0 0 0 0
[       OK ] testupdateweights.backprop_instance3_smaller2 (800 ms)
[----------] 23 tests from testupdateweights (15190 ms total)

[----------] 17 tests from testforward
[ RUN      ] testforward.imagesize2_nopadzeros
expected number of output: 4
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testforward.imagesize2_nopadzeros (430 ms)
[ RUN      ] testforward.imagesize2_padzeros
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
 checking result[0]=0 expecting: 0
 checking result[1]=0 expecting: 0
 checking result[2]=0 expecting: 0
 checking result[3]=0.2 expecting: 0.2
 checking result[4]=-0.13 expecting: -0.13
 checking result[5]=-0.15 expecting: -0.15
 checking result[6]=0 expecting: 0
 checking result[7]=0 expecting: 0
 checking result[8]=0 expecting: 0
 checking result[9]=0 expecting: 0
 checking result[10]=0 expecting: 0
 checking result[11]=0 expecting: 0
 checking result[12]=-0.55 expecting: -0.55
 checking result[13]=0.02 expecting: 0.02
 checking result[14]=0.21 expecting: 0.21
 checking result[27]=-14.3 expecting: -14.3
 checking result[28]=-9.6 expecting: -9.6
 checking result[29]=11.9 expecting: 11.9
 checking result[35]=0.46 expecting: 0.46
[       OK ] testforward.imagesize2_padzeros (170 ms)
[ RUN      ] testforward.imagesize3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
test1 ok
[       OK ] testforward.imagesize3 (180 ms)
[ RUN      ] testforward.test2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testforward.test2 (160 ms)
[ RUN      ] testforward.test3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testforward.test3 (170 ms)
[ RUN      ] testforward.compare_0_1_biased_nopad
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
dump enabled=0
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
clblas teardown
[       OK ] testforward.compare_0_1_biased_nopad (340 ms)
[ RUN      ] testforward.compare_0_1_biased_pad
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
dump enabled=0
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0}
clblas teardown
[       OK ] testforward.compare_0_1_biased_pad (360 ms)
[ RUN      ] testforward.compare_1_n_biased_nopad
instance: 2
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
dump enabled=0
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
clblas teardown
instance: 3
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
dump enabled=0
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
clblas teardown
instance: 4
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
dump enabled=0
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
clblas teardown
instance: 6
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=15 padZeros=0 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
clblas teardown
unknown file: error: C++ exception with description "memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size" thrown in the test body.
[  FAILED  ] testforward.compare_1_n_biased_nopad (2230 ms)
[ RUN      ] testforward.compare_1_n_biased_pad
instance: 2
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
clblas teardown
unknown file: error: C++ exception with description "cannot use forward2, since outputimagesize * outputimagesize > maxworkgroupsize" thrown in the test body.
[  FAILED  ] testforward.compare_1_n_biased_pad (310 ms)
[ RUN      ] testforward.compare_1_5_biased_nopad
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=19 outputSize=1 padZeros=0 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLB4E.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
dump enabled=0
LayerDimensions{ inputPlanes=8 inputSize=19 numFilters=8 filterSize=19 outputSize=1 padZeros=0 biased=1 skip=0}
clblas teardown
[       OK ] testforward.compare_1_5_biased_nopad (820 ms)
[ RUN      ] testforward.compare_1_4_fcscenario
LayerDimensions{ inputPlanes=10 inputSize=24 numFilters=10 filterSize=24 outputSize=1 padZeros=0 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
dump enabled=0
LayerDimensions{ inputPlanes=10 inputSize=24 numFilters=10 filterSize=24 outputSize=1 padZeros=0 biased=1 skip=0}
clblas teardown
[       OK ] testforward.compare_1_4_fcscenario (1200 ms)
[ RUN      ] testforward.compare_break1_0_1
LayerDimensions{ inputPlanes=1 inputSize=33 numFilters=1 filterSize=1 outputSize=33 padZeros=0 biased=0 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 1
dump enabled=0
batch 0 batchsize 1
dump enabled=0
LayerDimensions{ inputPlanes=1 inputSize=33 numFilters=1 filterSize=1 outputSize=33 padZeros=0 biased=0 skip=0}
clblas teardown
[       OK ] testforward.compare_break1_0_1 (170 ms)
[ RUN      ] testforward.compare_break1_0_4
LayerDimensions{ inputPlanes=1 inputSize=33 numFilters=1 filterSize=1 outputSize=33 padZeros=0 biased=0 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
batch 0 batchsize 1
dump enabled=0
batch 0 batchsize 1
dump enabled=0
LayerDimensions{ inputPlanes=1 inputSize=33 numFilters=1 filterSize=1 outputSize=33 padZeros=0 biased=0 skip=0}
clblas teardown
[       OK ] testforward.compare_break1_0_4 (160 ms)
[ RUN      ] testforward.comparespecific_break2
LayerDimensions{ inputPlanes=64 inputSize=19 numFilters=64 filterSize=19 outputSize=1 padZeros=0 biased=0 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL14AC.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


batch 0 batchsize 4
dump enabled=0
batch 0 batchsize 4
dump enabled=0
LayerDimensions{ inputPlanes=64 inputSize=19 numFilters=64 filterSize=19 outputSize=1 padZeros=0 biased=0 skip=0}
clblas teardown
[       OK ] testforward.comparespecific_break2 (850 ms)
[ RUN      ] testforward.softmax
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
output[0]=0.0320586
output[1]=0.0871443
output[2]=0.643914
output[3]=0.236883
loss 0.44019
loss 3.44019
loss 2.44019
loss 1.44019
[       OK ] testforward.softmax (20 ms)
[ RUN      ] testforward.softmax_byplane
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
output[0]=0.0320586
output[1]=0.0871443
output[2]=0.643914
output[3]=0.236883
loss 0.44019
loss 3.44019
loss 2.44019
loss 1.44019
[       OK ] testforward.softmax_byplane (10 ms)
[ RUN      ] testforward.crash_from_jm
 -D gNumInputPlanes=32 -D gInputPlanes=32 -D gInputSize=28 -D gInputSizeSquared=784 -D gNumFilters=20 -D gFilterSize=28 -D gHalfFilterSize=14 -D gFilterSizeSquared=784 -D gNumOutputPlanes=20 -D gOutputPlanes=20 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=1 -D gSkip=0
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
dump enabled=0
[       OK ] testforward.crash_from_jm (410 ms)
[----------] 17 tests from testforward (7990 ms total)

[----------] 2 tests from testfilehelper
[ RUN      ] testfilehelper.testfilehelper
[       OK ] testfilehelper.testfilehelper (10 ms)
[ RUN      ] testfilehelper.testreadchunk
[       OK ] testfilehelper.testreadchunk (0 ms)
[----------] 2 tests from testfilehelper (10 ms total)

[----------] 12 tests from testsimpleconvolvenet
[ RUN      ] testsimpleconvolvenet.imagesize1_planes2_filters2_unbiased_tanh
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 0.141046
 accuracy: 2/2 100%
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 290ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL1C51.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 10ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 290ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 200ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 10ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 200ms
   forward layer selected kernel 1
loss, E, 0.0733092
 accuracy: 2/2 100%
loss, E, 0.0426809
 accuracy: 2/2 100%
loss, E, 0.0262452
 accuracy: 2/2 100%
loss, E, 0.0164245
 accuracy: 2/2 100%
loss, E, 0.0107573
 accuracy: 2/2 100%
accuracy: 2/2
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize1_planes2_filters2_unbiased_tanh (1920 ms)
[ RUN      ] testsimpleconvolvenet.imagesize1_planes2_filters2_tanh
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 0.964924
 accuracy: 1/2 50%
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 420ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL247A.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 420ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 200ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 200ms
   forward layer selected kernel 1
loss, E, 0.00570461
 accuracy: 2/2 100%
loss, E, 1.34828e-05
 accuracy: 2/2 100%
loss, E, 3.62078e-08
 accuracy: 2/2 100%
accuracy: 2/2
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize1_planes2_filters2_tanh (2050 ms)
[ RUN      ] testsimpleconvolvenet.imagesize3_n4_filtersize3_tanh
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 10ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 1.13283
 accuracy: 3/4 75%
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=2 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 490ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL2D1F.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 490ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 260ms
loss, E, 0.00996342
 accuracy: 4/4 100%
   forward kernel 0: cannot be used
   forward kernel 1 time: 10ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 260ms
   forward layer selected kernel 2
loss, E, 4.70669e-05
 accuracy: 4/4 100%
loss, E, 4.0975e-07
 accuracy: 4/4 100%
accuracy: 4/4
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize3_n4_filtersize3_tanh (2250 ms)
[ RUN      ] testsimpleconvolvenet.imagesize1_2planes_filtersize1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 0.751601
 accuracy: 2/2 100%
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 430ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL3507.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 430ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 210ms
loss, E, 0.195916
 accuracy: 2/2 100%
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 210ms
   forward layer selected kernel 1
loss, E, 0.0679117
 accuracy: 2/2 100%
loss, E, 0.023677
 accuracy: 2/2 100%
loss, E, 0.00825563
 accuracy: 2/2 100%
loss, E, 0.00287856
 accuracy: 2/2 100%
loss, E, 0.00100369
 accuracy: 2/2 100%
loss, E, 0.000349964
 accuracy: 2/2 100%
 accuracy: 2/2 100%
accuracy: 2/2
loss, E, 0.000150648
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize1_2planes_filtersize1 (1950 ms)
[ RUN      ] testsimpleconvolvenet.imagesize3_n4_filtersize3_relu
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 1.48951
 accuracy: 2/4 50%
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=2 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 500ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL3DDA.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 500ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 250ms
loss, E, 1.12957
 accuracy: 2/4 50%
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 250ms
   forward layer selected kernel 1
loss, E, 0.070782
 accuracy: 4/4 100%
loss, E, 0.003026
 accuracy: 4/4 100%
loss, E, 0.00021158
 accuracy: 4/4 100%
loss, E, 1.96858e-05
 accuracy: 4/4 100%
loss, E, 2.03002e-06
 accuracy: 4/4 100%
loss, E, 2.15572e-07
 accuracy: 4/4 100%
loss, E, 2.3083e-08
 accuracy: 4/4 100%
loss, E, 2.48239e-09
 accuracy: 4/4 100%
loss, E, 4.14442e-10
 accuracy: 4/4 100%
accuracy: 4/4
loss, E, 4.14442e-10
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize3_n4_filtersize3_relu (2330 ms)
[ RUN      ] testsimpleconvolvenet.imagesize3_n4_filtersize3_linear
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 10ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 0.50604
 accuracy: 4/4 100%
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=2 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 510ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL466E.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 510ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 260ms
loss, E, 0.0565529
 accuracy: 4/4 100%
   forward kernel 0: cannot be used
   forward kernel 1 time: 10ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 260ms
   forward layer selected kernel 2
loss, E, 0.00777245
 accuracy: 4/4 100%
loss, E, 0.00106831
 accuracy: 4/4 100%
loss, E, 0.000218376
 accuracy: 4/4 100%
accuracy: 4/4
loss, E, 0.000218376
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize3_n4_filtersize3_linear (2110 ms)
[ RUN      ] testsimpleconvolvenet.imagesize1_n2_2layers_unbiased
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
epoch 0 loss, E, 0.0559531
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
epoch 1 loss, E, 0.0254554
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 223ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=2 -D gInputPlanes=2 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
epoch 2 loss, E, 0.0172943
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 223ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 430ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 70ms
epoch 3 loss, E, 0.0138013
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL52A9.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL52F8.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 430ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 70ms
   calcGradWeights layer selected kernel 1
epoch 4 loss, E, 0.0115848
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 200ms
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 80ms
epoch 5 loss, E, 0.00987036
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 200ms
   forward layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 80ms
   forward layer selected kernel 1
epoch 6 loss, E, 0.00844797
epoch 7 loss, E, 0.00724182
epoch 8 loss, E, 0.00621212
epoch 9 loss, E, 0.00533106
epoch 10 loss, E, 0.00457645
epoch 11 loss, E, 0.00392979
epoch 12 loss, E, 0.00337539
epoch 13 loss, E, 0.00289992
epoch 14 loss, E, 0.002492
epoch 15 loss, E, 0.00214191
epoch 16 loss, E, 0.00184138
epoch 17 loss, E, 0.00158331
epoch 18 loss, E, 0.00136164
epoch 19 loss, E, 0.0011712
epoch 20 loss, E, 0.00100754
epoch 21 loss, E, 0.000866877
epoch 22 loss, E, 0.000745946
epoch 23 loss, E, 0.000641966
epoch 24 loss, E, 0.000552543
epoch 25 loss, E, 0.000475625
epoch 26 loss, E, 0.000409454
epoch 27 loss, E, 0.000352522
epoch 28 loss, E, 0.000303531
epoch 29 loss, E, 0.00026137
epoch 30 loss, E, 0.000225082
epoch 31 loss, E, 0.000193845
epoch 32 loss, E, 0.000166954
epoch 33 loss, E, 0.000143801
epoch 34 loss, E, 0.000123866
epoch 35 loss, E, 0.000106699
epoch 36 loss, E, 9.19176e-05
epoch 37 loss, E, 7.91864e-05
epoch 38 loss, E, 6.82211e-05
epoch 39 loss, E, 5.87767e-05
layer 0:InputLayer{ outputPlanes=1 outputSize=1 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=1 numFilters=2 filterSize=1 outputSize=1 padZeros=0 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=1 numFilters=2 filterSize=1 outputSize=1 padZeros=0 biased=1 skip=0} }
layer 4:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=4	40.0%
layer 3: params=6	60.0%
TOTAL  : params=10
loss, E, 5.87767e-05
 accuracy: 2/2 100%
accuracy: 2/2
loss, E, 5.87767e-05
loss, E, 5.87767e-05
layer 0:InputLayer{ outputPlanes=1 outputSize=1 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=1 numFilters=2 filterSize=1 outputSize=1 padZeros=0 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=1 numFilters=2 filterSize=1 outputSize=1 padZeros=0 biased=1 skip=0} }
layer 4:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=4	40.0%
layer 3: params=6	60.0%
TOTAL  : params=10
float weights1[] = {-0.303866f, -1.59823f};
float weights3[] = {0.426358f, -0.719592f, -0.420361f, 0.719566f};
float bias1[] = {-0.324465f, 0.60279f};
float bias3[] = {0.506862f, -0.506837f};
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize1_n2_2layers_unbiased (3450 ms)
[ RUN      ] testsimpleconvolvenet.imagesize1_n2_2layers_biased
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 1.19067
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 217ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=2 -D gInputPlanes=2 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 16ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 217ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 421ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 78ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL6040.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL607F.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 421ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 16ms
   calcGradWeights kernel 4 time: 78ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 218ms
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 78ms
loss, E, 0.0667568
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 218ms
   forward layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 78ms
   forward layer selected kernel 1
loss, E, 0.00923595
loss, E, 0.00112611
loss, E, 0.0001174
loss, E, 1.15642e-05
dump enabled=0
loss, E, 1.78564e-06
 accuracy: 2/2 100%
accuracy: 2/2
loss, E, 1.78564e-06
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize1_n2_2layers_biased (3385 ms)
[ RUN      ] testsimpleconvolvenet.imagesize_5_4_2layers_filtersize_2_4_biased_n3
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 1.6207
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 249ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=3 -D gInputPlanes=3 -D gInputSize=4 -D gInputSizeSquared=16 -D gNumFilters=3 -D gFilterSize=4 -D gHalfFilterSize=2 -D gFilterSizeSquared=16 -D gNumOutputPlanes=3 -D gOutputPlanes=3 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=1 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=3 -DgInputStripeInnerNumRows=4 -DgInputStripeOuterNumRows=10 -DgInputStripeInnerSize=16 -DgInputStripeOuterSize=40 -DgInputStripeMarginSize=12 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=3 -D gFilterSize=2 -D gHalfFilterSize=1 -D gFilterSizeSquared=4 -D gNumOutputPlanes=3 -D gOutputPlanes=3 -D gOutputSize=4 -D gOutputSizeSquared=16 -D gPadZeros=0 -D gMargin=0 -D gEven=1 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=1 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=35 -DgInputStripeMarginSize=5 -DgOutputStripeNumRows=4 -DgOutputStripeSize=16
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 249ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 468ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 437ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 234ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL7234.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 468ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 437ms
   calcGradWeights layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 234ms
   forward layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 234ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 234ms
   forward layer selected kernel 1
loss, E, 0.000427028
loss, E, 8.40991e-08
loss, E, 3.03482e-11
loss, E, 2.59792e-13
loss, E, 1.15907e-13
loss, E, 8.03801e-14
loss, E, 7.14984e-14
loss, E, 6.08402e-14
loss, E, 6.9722e-14
loss, E, 6.9722e-14
 accuracy: 3/3 100%
accuracy: 3/3
loss, E, 6.9722e-14
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize_5_4_2layers_filtersize_2_4_biased_n3 (6365 ms)
[ RUN      ] testsimpleconvolvenet.imagesize_5_4_2layers_filtersize_2_4_biased_n6
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 16ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 3.64011
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 265ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=3 -D gInputPlanes=3 -D gInputSize=4 -D gInputSizeSquared=16 -D gNumFilters=3 -D gFilterSize=4 -D gHalfFilterSize=2 -D gFilterSizeSquared=16 -D gNumOutputPlanes=3 -D gOutputPlanes=3 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=1 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=3 -DgInputStripeInnerNumRows=4 -DgInputStripeOuterNumRows=10 -DgInputStripeInnerSize=16 -DgInputStripeOuterSize=40 -DgInputStripeMarginSize=12 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=3 -D gFilterSize=2 -D gHalfFilterSize=1 -D gFilterSizeSquared=4 -D gNumOutputPlanes=3 -D gOutputPlanes=3 -D gOutputSize=4 -D gOutputSizeSquared=16 -D gPadZeros=0 -D gMargin=0 -D gEven=1 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=1 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=35 -DgInputStripeMarginSize=5 -DgOutputStripeNumRows=4 -DgOutputStripeSize=16
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 16ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 265ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 452ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 452ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 234ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL8B45.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 452ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 452ms
   calcGradWeights layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 16ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 16ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 234ms
   forward layer selected kernel 2
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 234ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 234ms
   forward layer selected kernel 1
loss, E, 4.07297e-10
loss, E, 2.30926e-14
loss, E, 3.9968e-15
loss, E, 3.9968e-15
loss, E, 1.55431e-14
 accuracy: 6/6 100%
accuracy: 6/6
loss, E, 1.55431e-14
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize_5_4_2layers_filtersize_2_4_biased_n6 (5398 ms)
[ RUN      ] testsimpleconvolvenet.imagesize_5_3_2layers_filtersize_3_3_biased_n6
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 4.00796
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 280ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=3 -D gInputPlanes=3 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=3 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=3 -D gOutputPlanes=3 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 16ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=3 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=3 -D gOutputPlanes=3 -D gOutputSize=3 -D gOutputSizeSquared=9 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=3 -DgOutputStripeSize=9
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 280ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 499ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 640ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 437ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLA243.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 16ms
   calcGradWeights kernel 4 time: 499ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 640ms
   calcGradWeights layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 437ms
   forward layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 249ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 249ms
   forward layer selected kernel 1
loss, E, 1.87712e-08
loss, E, 5.01821e-14
loss, E, 8.88178e-15
 accuracy: 6/6 100%
accuracy: 6/6
loss, E, 8.88178e-15
clblas teardown
[       OK ] testsimpleconvolvenet.imagesize_5_3_2layers_filtersize_3_3_biased_n6 (5304 ms)
[ RUN      ] testsimpleconvolvenet.imagesize_5_3_2layers_filtersize_3_3_biased_n18
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
loss, E, 17.9931
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 15ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 16ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 281ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=3 -D gInputPlanes=3 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=3 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=3 -D gOutputPlanes=3 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=3 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=3 -D gOutputPlanes=3 -D gOutputSize=3 -D gOutputSizeSquared=9 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=3 -DgOutputStripeSize=9
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 281ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 484ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 655ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 437ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLB730.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 16ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 484ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 655ms
   calcGradWeights layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 437ms
   forward layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 249ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 15ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 249ms
   forward layer selected kernel 1
loss, E, 2.93736
loss, E, 2.74045
loss, E, 2.72813
loss, E, 2.72734
loss, E, 2.72728
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
loss, E, 2.72727
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=3 filterSize=3 outputSize=3 padZeros=0 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:ConvolutionalLayer{ LayerDimensions{ inputPlanes=3 inputSize=3 numFilters=3 filterSize=3 outputSize=1 padZeros=0 biased=1 skip=0} }
layer 4:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 1: params=30	26.3%
layer 3: params=84	73.7%
TOTAL  : params=114
loss, E, 2.72727
 accuracy: 13/18 72.2222%
accuracy: 13/18
C:\Users\pz\Documents\ml\DeepCL\test\testsimpleconvolvenet.cpp(1055): error: Value of: N
  Actual: 18
Expected: numCorrect
Which is: 13
loss, E, 2.72727
C:\Users\pz\Documents\ml\DeepCL\test\testsimpleconvolvenet.cpp(1059): error: Expected: (0.1f) >= (loss), actual: 0.1 vs 2.72727
clblas teardown
[  FAILED  ] testsimpleconvolvenet.imagesize_5_3_2layers_filtersize_3_3_biased_n18 (11310 ms)
[----------] 12 tests from testsimpleconvolvenet (47822 ms total)

[----------] 3 tests from testlogicaloperators
[ RUN      ] testlogicaloperators.Convolve_1layer_biased_And
And
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
Loss L 2.13088
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=2 -D gInputPlanes=2 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 421ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLD91E.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 421ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 203ms
Loss L 0.679527
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 203ms
   forward layer selected kernel 1
Loss L 0.398398
Loss L 0.301735
accuracy: 4/4
loss, E, 0.27227
clblas teardown
[       OK ] testlogicaloperators.Convolve_1layer_biased_And (1919 ms)
[ RUN      ] testlogicaloperators.Convolve_1layerbiased_Or
Or, convolve
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 15ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
Loss L 5.59056
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=2 -D gInputPlanes=2 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 422ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLE0B6.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 422ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 203ms
Loss L 1.22162
   forward kernel 0: cannot be used
   forward kernel 1 time: 15ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 203ms
   forward layer selected kernel 2
Loss L 0.583397
Loss L 0.366216
 accuracy: 4/4 100%
loss, E, 0.300027
clblas teardown
[       OK ] testlogicaloperators.Convolve_1layerbiased_Or (1934 ms)
[ RUN      ] testlogicaloperators.Convolve_2layers_relu_Xor
Xor, convolve
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
hand-setting weights...
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
Loss L 0.152638
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 15ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 219ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=2 -D gInputPlanes=2 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=2 -D gInputPlanes=2 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=2 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=2 -D gOutputPlanes=2 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 219ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 421ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 78ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLECF3.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLED42.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 421ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 78ms
   calcGradWeights layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 187ms
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 78ms
Loss L 0.00640068
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 187ms
   forward layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 15ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 78ms
   forward layer selected kernel 1
Loss L 0.00139435
Loss L 0.000383307
Loss L 0.000117079
Loss L 4.63626e-05
Loss L 1.8873e-05
Loss L 7.15534e-06
Loss L 2.83958e-06
Loss L 1.12727e-06
Loss L 4.44109e-07
Loss L 1.72233e-07
Loss L 6.82345e-08
Loss L 2.76343e-08
Loss L 1.04286e-08
Loss L 4.13357e-09
Loss L 1.67201e-09
Loss L 6.29148e-10
Loss L 2.4837e-10
Loss L 1.00833e-10
Loss L 3.80673e-11
Loss L 1.5131e-11
Loss L 5.84421e-12
Loss L 2.16893e-12
Loss L 9.52127e-13
Loss L 3.58824e-13
Loss L 1.56319e-13
Loss L 9.9476e-14
Loss L 9.9476e-14
Loss L 9.9476e-14
Loss L 9.9476e-14
Loss L 9.23706e-14
Loss L 9.23706e-14
Loss L 9.41469e-14
Loss L 8.70415e-14
Loss L 9.41469e-14
Loss L 8.52651e-14
Loss L 8.52651e-14
Loss L 8.52651e-14
Loss L 8.52651e-14
layer 0:InputLayer{ outputPlanes=2 outputSize=1 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=1 numFilters=2 filterSize=1 outputSize=1 padZeros=0 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=1 numFilters=2 filterSize=1 outputSize=1 padZeros=0 biased=1 skip=0} }
layer 4:ActivationLayer{ RELU }
layer 5:SquareLossLayer{}
Parameters overview: (skipping 4 layers with 0 params)
layer 1: params=6	50.0%
layer 3: params=6	50.0%
TOTAL  : params=12
 accuracy: 4/4 100%
loss, E, 8.52651e-14
clblas teardown
[       OK ] testlogicaloperators.Convolve_2layers_relu_Xor (3916 ms)
[----------] 3 tests from testlogicaloperators (7769 ms total)

[----------] 12 tests from testbackward
[ RUN      ] testbackward.squareloss
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
layer 0:InputLayer{ outputPlanes=3 outputSize=5 }
layer 1:ForceBackpropLayer{ outputPlanes=3 outputSize=5 }
layer 2:SquareLossLayer{}

inputtotalsize=2400 outputTotalSize=2400
layer 0:InputLayer{ outputPlanes=3 outputSize=5 }
layer 1:ForceBackpropLayer{ outputPlanes=3 outputSize=5 }
layer 2:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
TOTAL  : params=0
idx=44 predicted losschange=-0.000912508 actual=-0.000976563
idx=2245 predicted losschange=0.00785823 actual=0.00805664
idx=648 predicted losschange=0.00965759 actual=0.00976563
idx=586 predicted losschange=0.0136895 actual=0.0136719
idx=730 predicted losschange=0.00117897 actual=0.00146484
idx=611 predicted losschange=0.00152302 actual=0.00195313
idx=1130 predicted losschange=0.0159167 actual=0.0161133
idx=15 predicted losschange=0.0434798 actual=0.0439453
idx=1923 predicted losschange=-0.00790002 actual=-0.0078125
idx=670 predicted losschange=0.0335141 actual=0.0336914
[       OK ] testbackward.squareloss (15 ms)
[ RUN      ] testbackward.crossentropyloss
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
layer 0:InputLayer{ outputPlanes=3 outputSize=5 }
layer 1:ForceBackpropLayer{ outputPlanes=3 outputSize=5 }
layer 2:Layer{}

inputtotalsize=300 outputTotalSize=300
layer 0:InputLayer{ outputPlanes=3 outputSize=5 }
layer 1:ForceBackpropLayer{ outputPlanes=3 outputSize=5 }
layer 2:Layer{}
Parameters overview: (skipping 3 layers with 0 params)
TOTAL  : params=0
idx=44 predicted losschange=0.000274935 actual=0.000274658
idx=145 predicted losschange=-0.000885784 actual=-0.00088501
idx=48 predicted losschange=-0.000859834 actual=-0.000854492
idx=286 predicted losschange=0.00713042 actual=0.00717163
idx=130 predicted losschange=-0.000264829 actual=-0.000244141
idx=11 predicted losschange=-1.98163e-05 actual=0
idx=230 predicted losschange=-0.000594819 actual=-0.000610352
idx=15 predicted losschange=-0.0006499 actual=-0.000640869
idx=123 predicted losschange=-0.000846121 actual=-0.000823975
idx=70 predicted losschange=0.000790196 actual=0.000793457
[       OK ] testbackward.crossentropyloss (16 ms)
[ RUN      ] testbackward.softmaxloss
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:SoftMaxLayer{ perPlane=0 numPlanes=5 imageSize=1 }

inputtotalsize=10 outputTotalSize=10
layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:SoftMaxLayer{ perPlane=0 numPlanes=5 imageSize=1 }
Parameters overview: (skipping 3 layers with 0 params)
TOTAL  : params=0
idx=4 predicted losschange=0.000113075 actual=0.00011301
idx=5 predicted losschange=0.000145627 actual=0.000145674
idx=8 predicted losschange=3.16699e-05 actual=3.19481e-05
idx=6 predicted losschange=4.89271e-06 actual=5.24521e-06
idx=0 predicted losschange=2.29469e-05 actual=2.28882e-05
idx=1 predicted losschange=-8.26119e-05 actual=-8.27312e-05
idx=0 predicted losschange=2.29469e-05 actual=2.28882e-05
idx=5 predicted losschange=0.000145627 actual=0.000145674
idx=3 predicted losschange=-5.50179e-05 actual=-5.50747e-05
idx=0 predicted losschange=2.29469e-05 actual=2.28882e-05
[       OK ] testbackward.softmaxloss (0 ms)
[ RUN      ] testbackward.squareloss2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:SquareLossLayer{}

layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:SquareLossLayer{}

batchSize: 32
inputtotalsize=160 outputTotalSize=160
layer SquareLossLayer{}
layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
TOTAL  : params=0
idx=44 predicted losschange=0.000126406 actual=0.000125885
idx=5 predicted losschange=0.00461891 actual=0.00464439
idx=8 predicted losschange=0.000356787 actual=0.000356674
idx=106 predicted losschange=0.00716324 actual=0.00719643
idx=90 predicted losschange=0.000474759 actual=0.000480652
idx=131 predicted losschange=0.000979017 actual=0.000984192
idx=10 predicted losschange=0.000660134 actual=0.000663757
idx=15 predicted losschange=0.00961313 actual=0.00965118
idx=3 predicted losschange=0.00264732 actual=0.00267029
idx=30 predicted losschange=0.00865312 actual=0.00868607
[       OK ] testbackward.squareloss2 (31 ms)
[ RUN      ] testbackward.crossentropy2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:Layer{}

layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:Layer{}

batchSize: 2
inputtotalsize=10 outputTotalSize=10
layer Layer{}
layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:Layer{}
Parameters overview: (skipping 3 layers with 0 params)
TOTAL  : params=0
idx=4 predicted losschange=0.00258649 actual=-nan(ind)
idx=5 predicted losschange=0.0227095 actual=-nan(ind)
idx=8 predicted losschange=-0.00202714 actual=-nan(ind)
idx=6 predicted losschange=-0.000846508 actual=-nan(ind)
idx=0 predicted losschange=-0.000424821 actual=-nan(ind)
idx=1 predicted losschange=-0.00171216 actual=-nan(ind)
idx=0 predicted losschange=-0.000424821 actual=-nan(ind)
idx=5 predicted losschange=0.0227095 actual=-nan(ind)
idx=3 predicted losschange=0.0123444 actual=-nan(ind)
idx=0 predicted losschange=-0.000424821 actual=-nan(ind)
[       OK ] testbackward.crossentropy2 (31 ms)
[ RUN      ] testbackward.softmax2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:SoftMaxLayer{ perPlane=0 numPlanes=5 imageSize=1 }

layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:SoftMaxLayer{ perPlane=0 numPlanes=5 imageSize=1 }

batchSize: 2
inputtotalsize=10 outputTotalSize=10
layer SoftMaxLayer{ perPlane=0 numPlanes=5 imageSize=1 }
layer 0:InputLayer{ outputPlanes=5 outputSize=1 }
layer 1:ForceBackpropLayer{ outputPlanes=5 outputSize=1 }
layer 2:SoftMaxLayer{ perPlane=0 numPlanes=5 imageSize=1 }
Parameters overview: (skipping 3 layers with 0 params)
TOTAL  : params=0
idx=4 predicted losschange=0.00035729 actual=0.000357628
idx=5 predicted losschange=0.0015055 actual=0.00151086
idx=8 predicted losschange=-5.63632e-05 actual=-5.65052e-05
idx=6 predicted losschange=-1.48864e-05 actual=-1.4782e-05
idx=0 predicted losschange=1.96542e-05 actual=1.97887e-05
idx=1 predicted losschange=-0.000287167 actual=-0.000287056
idx=0 predicted losschange=1.96542e-05 actual=1.97887e-05
idx=5 predicted losschange=0.0015055 actual=0.00151086
idx=3 predicted losschange=-0.000152824 actual=-0.00014782
idx=0 predicted losschange=1.96542e-05 actual=1.97887e-05
[       OK ] testbackward.softmax2 (63 ms)
[ RUN      ] testbackward.conv1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=2 outputSize=4 }
layer 1:ForceBackpropLayer{ outputPlanes=2 outputSize=4 }
layer 2:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=4 numFilters=2 filterSize=3 outputSize=2 padZeros=0 biased=0 skip=0} }
layer 3:SquareLossLayer{}

layer 0:InputLayer{ outputPlanes=2 outputSize=4 }
layer 1:ForceBackpropLayer{ outputPlanes=2 outputSize=4 }
layer 2:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=4 numFilters=2 filterSize=3 outputSize=2 padZeros=0 biased=0 skip=0} }
layer 3:SquareLossLayer{}

batchSize: 4
inputtotalsize=128 outputTotalSize=32
layer ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=4 numFilters=2 filterSize=3 outputSize=2 padZeros=0 biased=0 skip=0} }
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=2 outputSize=4 }
layer 1:ForceBackpropLayer{ outputPlanes=2 outputSize=4 }
layer 2:ConvolutionalLayer{ LayerDimensions{ inputPlanes=2 inputSize=4 numFilters=2 filterSize=3 outputSize=2 padZeros=0 biased=0 skip=0} }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 2: params=36	100.0%
TOTAL  : params=36
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 16ms
idx=44 predicted losschange=0.000198655 actual=0.000199318
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
idx=37 predicted losschange=-0.00664573 actual=-0.00663185
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
idx=40 predicted losschange=0.00305358 actual=0.00306416
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 266ms
idx=106 predicted losschange=0.000651619 actual=0.00306416
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 16ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 266ms
   forward layer selected kernel 1
idx=122 predicted losschange=0.0040653 actual=0.00407314
idx=99 predicted losschange=-0.000240484 actual=-0.00024128
idx=10 predicted losschange=0.00158175 actual=0.00158405
idx=47 predicted losschange=0.00140132 actual=0.00140285
idx=67 predicted losschange=-0.00154732 actual=-0.00154686
idx=126 predicted losschange=-0.000393638 actual=-0.000391006
clblas teardown
[       OK ] testbackward.conv1 (1294 ms)
[ RUN      ] testbackward.fc1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=2 outputSize=4 }
layer 1:ForceBackpropLayer{ outputPlanes=2 outputSize=4 }
layer 2:FullyConnectedLayer{ numPlanes=4 imageSize=1 }
layer 3:SquareLossLayer{}

layer 0:InputLayer{ outputPlanes=2 outputSize=4 }
layer 1:ForceBackpropLayer{ outputPlanes=2 outputSize=4 }
layer 2:FullyConnectedLayer{ numPlanes=4 imageSize=1 }
layer 3:SquareLossLayer{}

batchSize: 4
inputtotalsize=128 outputTotalSize=16
layer FullyConnectedLayer{ numPlanes=4 imageSize=1 }
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
layer 0:InputLayer{ outputPlanes=2 outputSize=4 }
layer 1:ForceBackpropLayer{ outputPlanes=2 outputSize=4 }
layer 2:FullyConnectedLayer{ numPlanes=4 imageSize=1 }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 3 layers with 0 params)
layer 2: params=128	100.0%
TOTAL  : params=128
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
idx=44 predicted losschange=-2.78137e-06 actual=-2.86102e-06
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
idx=37 predicted losschange=-0.000552869 actual=-0.000545502
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
idx=40 predicted losschange=0.00245549 actual=0.00246334
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLFB2C.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
idx=106 predicted losschange=0.00259146 actual=0.00259662
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 250ms
idx=122 predicted losschange=0.000431057 actual=0.00259662
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 250ms
   forward layer selected kernel 1
idx=99 predicted losschange=-0.00116097 actual=-0.00116014
idx=10 predicted losschange=-0.000360866 actual=-0.00036025
idx=47 predicted losschange=0.000165997 actual=0.000166655
idx=67 predicted losschange=-0.000468417 actual=-0.000465631
idx=126 predicted losschange=3.95745e-05 actual=4.1008e-05
clblas teardown
[       OK ] testbackward.fc1 (1389 ms)
[ RUN      ] testbackward.act1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
layer 0:InputLayer{ outputPlanes=1 outputSize=2 }
layer 1:ForceBackpropLayer{ outputPlanes=1 outputSize=2 }
layer 2:ActivationLayer{ RELU }
layer 3:SquareLossLayer{}

layer 0:InputLayer{ outputPlanes=1 outputSize=2 }
layer 1:ForceBackpropLayer{ outputPlanes=1 outputSize=2 }
layer 2:ActivationLayer{ RELU }
layer 3:SquareLossLayer{}

batchSize: 1
inputtotalsize=4 outputTotalSize=4
layer ActivationLayer{ RELU }
layer 0:InputLayer{ outputPlanes=1 outputSize=2 }
layer 1:ForceBackpropLayer{ outputPlanes=1 outputSize=2 }
layer 2:ActivationLayer{ RELU }
layer 3:SquareLossLayer{}
Parameters overview: (skipping 4 layers with 0 params)
TOTAL  : params=0
idx=0 predicted losschange=-0.000880961 actual=-0.00088048
idx=1 predicted losschange=-0.00151209 actual=-0.00151044
idx=0 predicted losschange=-0.000880961 actual=-0.00088048
idx=2 predicted losschange=-0.00245153 actual=-0.0024423
idx=2 predicted losschange=-0.00245153 actual=-0.0024423
idx=3 predicted losschange=-0.00214455 actual=-0.00212085
idx=2 predicted losschange=-0.00245153 actual=-0.0024423
idx=3 predicted losschange=-0.00214455 actual=-0.00212085
idx=3 predicted losschange=-0.00214455 actual=-0.00212085
idx=2 predicted losschange=-0.00245153 actual=-0.0024423
[       OK ] testbackward.act1 (140 ms)
[ RUN      ] testbackward.checknumerically
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
 loss 0.0986296 loss2 0.0984814 change: 0.000148199
sumweightsdiff 0.0038507
 loss change              0.000148199
 estimatedLossChangeFromW 0.000148279
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL3D3.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 15ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL422.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 156ms
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 78ms
 loss 0.0984814 loss2 0.0983336 change: 0.000147872
sumweightsdiff 0.00384641
 loss change              0.000147872
 estimatedLossChangeFromW 0.000147948
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 15ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 156ms
   forward layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 78ms
   forward layer selected kernel 1
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 140ms
calcGradWeights try kernel 3
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=1 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=1 -D gInputSizeSquared=1 -D gNumFilters=1 -D gFilterSize=1 -D gHalfFilterSize=0 -D gFilterSizeSquared=1 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=0 -DgInputStripeInnerNumRows=1 -DgInputStripeOuterNumRows=1 -DgInputStripeInnerSize=1 -DgInputStripeOuterSize=1 -DgInputStripeMarginSize=0 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
 loss 0.0983336 loss2 0.098186 change: 0.000147544
sumweightsdiff 0.00384223
 loss change              0.000147544
 estimatedLossChangeFromW 0.000147628
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 140ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 218ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 78ms
 loss 0.098186 loss2 0.0980388 change: 0.000147216
sumweightsdiff 0.00383794
 loss change              0.000147216
 estimatedLossChangeFromW 0.000147298
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 218ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 78ms
   calcGradWeights layer selected kernel 1
 loss 0.0980388 loss2 0.0978919 change: 0.000146888
sumweightsdiff 0.00383377
 loss change              0.000146888
 estimatedLossChangeFromW 0.000146978
clblas teardown
[       OK ] testbackward.checknumerically (3027 ms)
[ RUN      ] testbackward.checknumerically_imagesize5_filter3_relu
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
 loss 630.466 loss2 608.021 change: 22.4443
sumweightsdiff -0.035685
 loss change              22.4443
 estimatedLossChangeFromW 22.6629
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 281ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 94ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 281ms
   forward layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 94ms
   forward layer selected kernel 1
 loss 608.021 loss2 586.349 change: 21.672
sumweightsdiff -0.0350289
 loss change              21.672
 estimatedLossChangeFromW 21.7974
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 406ms
calcGradWeights try kernel 3
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=1 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=5 -D gOutputSizeSquared=25 -D gPadZeros=1 -D gMargin=1 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=5 -DgOutputStripeSize=25
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=1 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=1 -D gOutputPlanes=1 -D gOutputSize=5 -D gOutputSizeSquared=25 -D gPadZeros=1 -D gMargin=1 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=5 -DgOutputStripeSize=25
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
 loss 586.349 loss2 565.324 change: 21.025
sumweightsdiff -0.0345262
 loss change              21.025
 estimatedLossChangeFromW 21.2378
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 406ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 359ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 109ms
 loss 565.324 loss2 545.133 change: 20.1916
sumweightsdiff -0.0338754
 loss change              20.1916
 estimatedLossChangeFromW 20.3956
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 359ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 109ms
   calcGradWeights layer selected kernel 1
 loss 545.133 loss2 525.742 change: 19.3912
sumweightsdiff -0.0332378
 loss change              19.3912
 estimatedLossChangeFromW 19.5872
 loss 525.742 loss2 507.119 change: 18.6229
sumweightsdiff -0.0326132
 loss change              18.6229
 estimatedLossChangeFromW 18.8111
 loss 507.119 loss2 489.233 change: 17.8853
sumweightsdiff -0.032001
 loss change              17.8853
 estimatedLossChangeFromW 18.066
 loss 489.233 loss2 472.056 change: 17.1772
sumweightsdiff -0.0314012
 loss change              17.1772
 estimatedLossChangeFromW 17.3506
 loss 472.056 loss2 455.559 change: 16.4975
sumweightsdiff -0.0308135
 loss change              16.4975
 estimatedLossChangeFromW 16.6639
 loss 455.559 loss2 439.714 change: 15.8447
sumweightsdiff -0.0302379
 loss change              15.8447
 estimatedLossChangeFromW 16.0046
 loss 439.714 loss2 424.416 change: 15.2976
sumweightsdiff -0.0296733
 loss change              15.2976
 estimatedLossChangeFromW 15.3717
 loss 424.416 loss2 409.545 change: 14.871
sumweightsdiff -0.0299227
 loss change              14.871
 estimatedLossChangeFromW 15.0234
 loss 409.545 loss2 395.271 change: 14.274
sumweightsdiff -0.0293575
 loss change              14.274
 estimatedLossChangeFromW 14.4202
 loss 395.271 loss2 381.57 change: 13.7013
sumweightsdiff -0.0288033
 loss change              13.7013
 estimatedLossChangeFromW 13.8415
 loss 381.57 loss2 368.418 change: 13.1519
sumweightsdiff -0.0282608
 loss change              13.1519
 estimatedLossChangeFromW 13.2864
 loss 368.418 loss2 355.794 change: 12.6248
sumweightsdiff -0.0277294
 loss change              12.6248
 estimatedLossChangeFromW 12.7538
 loss 355.794 loss2 343.675 change: 12.119
sumweightsdiff -0.027209
 loss change              12.119
 estimatedLossChangeFromW 12.2429
 loss 343.675 loss2 332.041 change: 11.634
sumweightsdiff -0.0266991
 loss change              11.634
 estimatedLossChangeFromW 11.7526
 loss 332.041 loss2 320.872 change: 11.1684
sumweightsdiff -0.0261997
 loss change              11.1684
 estimatedLossChangeFromW 11.2823
 loss 320.872 loss2 310.15 change: 10.7218
sumweightsdiff -0.0257105
 loss change              10.7218
 estimatedLossChangeFromW 10.8312
clblas teardown
[       OK ] testbackward.checknumerically_imagesize5_filter3_relu (3915 ms)
[ RUN      ] testbackward.compare_1_n_kgsgo_32c5
 -D BIASED -D gNumInputPlanes=32 -D gInputPlanes=32 -D gInputSize=19 -D gInputSizeSquared=361 -D gNumFilters=32 -D gFilterSize=5 -D gHalfFilterSize=2 -D gFilterSizeSquared=25 -D gNumOutputPlanes=32 -D gOutputPlanes=32 -D gOutputSize=19 -D gOutputSizeSquared=361 -D gPadZeros=1 -D gMargin=2 -D gEven=0 -D gSkip=0
batchsize=8 LayerDimensions{ inputPlanes=32 inputSize=19 numFilters=32 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
LayerDimensions{ inputPlanes=32 inputSize=19 numFilters=32 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0}
output[0]=-0.0308112 -0.0308112 SAME  || -0.129603  || -0.048413  || 0.07916  || -0.118675  || 0.0416933  || 0.100887  || -0.106013
output[1]=-0.0574008 -0.0574008 SAME  || 0.099984  || 0.0155394  || 0.00411644  || 0.131031  || -0.0107744  || 0.121347  || 0.0437087
output[2]=-0.0227139 -0.0227139 SAME  || -0.0115189  || -0.190989  || -0.0445787  || -0.013341  || -0.04953  || -0.109186  || 0.104814
output[3]=-0.0805896 -0.0805896 SAME  || 0.0216207  || -0.128649  || -0.0159031  || 0.0534839  || 0.0301581  || 0.104269  || -0.0841106
output[4]=-0.0723994 -0.0723994 SAME  || -0.0164838  || -0.00649171  || -0.042007  || 0.147102  || -0.0702085  || -0.0120931  || 0.0597854
output[5]=0.130336 0.130336 SAME  || -0.0816751  || -0.272227  || 0.0707071  || 0.133967  || 0.0323092  || 0.124248  || -0.0138626
output[6]=-0.00415662 -0.00415662 SAME  || -0.0920411  || 0.0352436  || 0.0541946  || 0.00491123  || -0.0805987  || 0.0834764  || 0.0631893
output[7]=-0.0915931 -0.0915931 SAME  || -0.0358497  || 0.0445722  || -0.0472172  || 0.0778742  || -0.0550363  || -0.179262  || -0.0812755
output[8]=0.0556533 0.0556533 SAME  || -0.0684331  || -0.0243033  || -0.0822076  || -0.0104788  || -0.043145  || -0.0481164  || 0.0538944
output[9]=-0.0725742 -0.0725742 SAME  || 0.0486592  || -0.0286811  || -0.0249626  || 0.0394469  || -0.144496  || 0.0909432  || -0.0152857
output[10]=-0.0153476 -0.0153476 SAME  || -0.0677297  || -0.140709  || -0.0161164  || 0.131645  || 0.0545684  || -0.0210541  || 0.0611338
output[11]=-0.0212713 -0.0212713 SAME  || 0.100494  || 0.2122  || -0.0812487  || 0.0532493  || -0.0183774  || -0.0937923  || -0.069912
output[12]=0.0389741 0.0389741 SAME  || 0.0809882  || 0.0370538  || 0.0241565  || -0.0582968  || 0.0437625  || 0.139931  || -0.065007
output[13]=0.0349705 0.0349705 SAME  || -0.0251775  || -0.0759114  || 0.0945214  || 0.00389841  || -0.0377205  || 0.17624  || -0.114476
output[14]=0.0366689 0.0366689 SAME  || -0.0348694  || -0.0581568  || 0.0376178  || -0.0298947  || -0.0299259  || -0.0913825  || -0.0745193
output[15]=0.0186965 0.0186965 SAME  || 0.0281147  || 0.00937999  || 0.108983  || -0.0505074  || -0.0573388  || 0.067382  || 0.0387854
output[16]=0.0658136 0.0658136 SAME  || -0.0412163  || -0.128719  || 0.150029  || 0.0555238  || -0.0203267  || -0.0795422  || -0.123847
output[17]=0.0705919 0.0705919 SAME  || 0.147334  || 0.151016  || -0.0122364  || 0.0360484  || -0.0609187  || 0.0166715  || -0.141399
output[18]=-0.0508929 -0.0508929 SAME  || 0.0131358  || -0.0101773  || -0.120741  || -0.00821514  || 0.00894922  || -0.117651  || 0.0631629
output[19]=-0.0110406 -0.0110406 SAME  || 0.189081  || 0.0665268  || 0.0622702  || 0.151629  || -0.0172241  || -0.0215623  || 0.0457666
clblas teardown
batchsize=8 LayerDimensions{ inputPlanes=32 inputSize=19 numFilters=32 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0}
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
clblas teardown
unknown file: error: C++ exception with description "
kernel source:
1: // Copyright Hugh Perkins 2014, 2015 hughperkins at gmail
2: //
3: // This Source Code Form is subject to the terms of the Mozilla Public License,
4: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
5: // obtain one at http://mozilla.org/MPL/2.0/.
6:
7: void copyLocal(local float *target, global float const *source, int N) {
8:     int numLoops = (N + get_local_size(0) - 1) / get_local_size(0);
9:     for (int loop = 0; loop < numLoops; loop++) {
10:         int offset = loop * get_local_size(0) + get_local_id(0);
11:         if (offset < N) {
12:             target[offset] = source[offset];
13:         }
14:     }
15: }
16:
17: // as calcGradInput, but with local cache
18: // convolve weights with gradOutput to produce gradInput
19: // workgroupid: [n][inputPlane]
20: // localid: [upstreamrow][upstreamcol]
21: // per-thread aggregation: [outPlane][filterRow][filterCol]
22: // need to store locally:
23: // - _gradOutputPlane. size = outputSizeSquared
24: // - _filterPlane. size = filtersizesquared
25: // note: currently doesnt use bias as input.  thats probably an error?
26: // inputs: gradOutput :convolve: filters => gradInput
27: //
28: // global:
29: // gradOutput: [n][outPlane][outRow][outCol] 128 * 32 * 19 * 19 * 4
30: // weights: [filterId][upstreamplane][filterRow][filterCol] 32 * 32 * 5 * 5 * 4
31: // per workgroup:
32: // gradOutput: [outPlane][outRow][outCol] 32 * 19 * 19 * 4 = 46KB
33: // weights: [filterId][filterRow][filterCol] 32 * 5 * 5 * 4 = 3.2KB
34: // gradOutputforupstream: [n][upstreamPlane][upstreamRow][upstreamCol]
35: void kernel calcGradInputCached(
36:         const int batchSize,
37:         global const float *gradOutputGlobal,
38:         global const float *filtersGlobal,
39:         global float *gradInput,
40:         local float *_gradOutputPlane,
41:         local float *_filterPlane) {
42:
43:     #define globalId get_global_id(0)
44:     #define localId get_local_id(0)
45:     #define workgroupId get_group_id(0)
46:     #define workgroupSize get_local_size(0)
47:
48:     const int n = workgroupId / gInputPlanes;
49:     const int upstreamPlane = workgroupId % gInputPlanes;
50:
51:     const int upstreamRow = localId / gInputSize;
52:     const int upstreamCol = localId % gInputSize;
53:
54:     float sumWeightTimesOutError = 0;
55:     for (int outPlane = 0; outPlane < gNumFilters; outPlane++) {
56:         barrier(CLK_LOCAL_MEM_FENCE);
57:         copyLocal(_filterPlane, filtersGlobal + (outPlane * gInputPlanes + upstreamPlane) * gFilterSizeSquared, gFilterSizeSquared);
58:         copyLocal(_gradOutputPlane, gradOutputGlobal + (n * gNumFilters + outPlane) * gOutputSizeSquared, gOutputSizeSquared);
59:         barrier(CLK_LOCAL_MEM_FENCE);
60:         for (int filterRow = 0; filterRow < gFilterSize; filterRow++) {
61:             int outRow = upstreamRow + gMargin - filterRow;
62:             for (int filterCol = 0; filterCol < gFilterSize; filterCol++) {
63:                 int outCol = upstreamCol + gMargin - filterCol;
64:                 if (outCol >= 0 && outCol < gOutputSize && outRow >= 0 && outRow < gOutputSize) {
65:                     float thisWeightTimesError =
66:                         _gradOutputPlane[outRow * gOutputSize + outCol] *
67:                         _filterPlane[filterRow * gFilterSize + filterCol];
68:                     sumWeightTimesOutError += thisWeightTimesError;
69:                 }
70:             }
71:         }
72:     }
73:     const int upstreamImageGlobalOffset = (n * gInputPlanes + upstreamPlane) * gInputSizeSquared;
74:     if (localId < gInputSizeSquared) {
75:         gradInput[upstreamImageGlobalOffset + localId] = sumWeightTimesOutError;
76:     }
77: }
78:
79:

Something went wrong, code -55" thrown in the test body.
[  FAILED  ] testbackward.compare_1_n_kgsgo_32c5 (843 ms)
[----------] 12 tests from testbackward (10764 ms total)

[----------] 6 tests from testsinglebatch
[ RUN      ] testsinglebatch.imagesize5_filtersize3_batchsize2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=5 filterSize=3 outputSize=3 padZeros=0 biased=1 skip=0} }
layer 2:ActivationLayer{ LINEAR }
layer 3:FullyConnectedLayer{ numPlanes=5 imageSize=1 }
layer 4:ActivationLayer{ TANH }
layer 5:SquareLossLayer{}
Parameters overview: (skipping 4 layers with 0 params)
layer 1: params=50	17.9%
layer 3: params=230	82.1%
TOTAL  : params=280
weightsTotalSize=280
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 16ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 15ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 702ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL2857.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 421ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=5 -D gInputPlanes=5 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=5 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=5 -D gOutputPlanes=5 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=5 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=5 -D gOutputPlanes=5 -D gOutputSize=3 -D gOutputSizeSquared=9 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=3 -DgOutputStripeSize=9
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 702ms
   forward layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 312ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 15ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 312ms
   forward layer selected kernel 1
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 421ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 843ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 1092ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 16ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 843ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 1092ms
   calcGradWeights layer selected kernel 1
batch time 5694 ms
dump enabled=0
clblas teardown
[       OK ] testsinglebatch.imagesize5_filtersize3_batchsize2 (6130 ms)
[ RUN      ] testsinglebatch.imagesize5_filtersize3_batchsize2_10filters
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=10 filterSize=3 outputSize=3 padZeros=0 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:FullyConnectedLayer{ numPlanes=10 imageSize=1 }
layer 4:ActivationLayer{ TANH }
layer 5:SquareLossLayer{}
Parameters overview: (skipping 4 layers with 0 params)
layer 1: params=100	9.9%
layer 3: params=910	90.1%
TOTAL  : params=1010
weightsTotalSize=1010
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 702ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL410C.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 374ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=10 -D gInputPlanes=10 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=10 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=10 -D gOutputPlanes=10 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=10 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=10 -D gOutputPlanes=10 -D gOutputSize=3 -D gOutputSizeSquared=9 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=3 -DgOutputStripeSize=9
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 702ms
   forward layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 312ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 312ms
   forward layer selected kernel 1
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 374ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 827ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 1076ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 827ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 1076ms
   calcGradWeights layer selected kernel 1
batch time 6037 ms
dump enabled=0
clblas teardown
[       OK ] testsinglebatch.imagesize5_filtersize3_batchsize2_10filters (6490 ms)
[ RUN      ] testsinglebatch.imagesize28
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=1 outputSize=28 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=28 numFilters=10 filterSize=3 outputSize=26 padZeros=0 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:FullyConnectedLayer{ numPlanes=10 imageSize=1 }
layer 4:ActivationLayer{ TANH }
layer 5:SquareLossLayer{}
Parameters overview: (skipping 4 layers with 0 params)
layer 1: params=100	0.1%
layer 3: params=67610	99.9%
TOTAL  : params=67710
weightsTotalSize=67710
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 2
ForwardAuto: kernel 2: this instance cant be used: cannot use forward2, since outputimagesize * outputimagesize > maxworkgroupsize
   ... not valid
forward try kernel 3
ForwardAuto: kernel 3: this instance cant be used: cannot use forward3, since outputimagesize * outputimagesize > maxworkgroupsize
   ... not valid
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 562ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 this instance cant be used:
kernel source:
1: // Copyright Hugh Perkins 2014, 2015 hughperkins at gmail
2: //
3: // This Source Code Form is subject to the terms of the Mozilla Public License,
4: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
5: // obtain one at http://mozilla.org/MPL/2.0/.
6:
7: void copyLocal(local float *target, global float const *source, int N) {
8:     int numLoops = (N + get_local_size(0) - 1) / get_local_size(0);
9:     for (int loop = 0; loop < numLoops; loop++) {
10:         int offset = loop * get_local_size(0) + get_local_id(0);
11:         if (offset < N) {
12:             target[offset] = source[offset];
13:         }
14:     }
15: }
16:
17: // as calcGradInput, but with local cache
18: // convolve weights with gradOutput to produce gradInput
19: // workgroupid: [n][inputPlane]
20: // localid: [upstreamrow][upstreamcol]
21: // per-thread aggregation: [outPlane][filterRow][filterCol]
22: // need to store locally:
23: // - _gradOutputPlane. size = outputSizeSquared
24: // - _filterPlane. size = filtersizesquared
25: // note: currently doesnt use bias as input.  thats probably an error?
26: // inputs: gradOutput :convolve: filters => gradInput
27: //
28: // global:
29: // gradOutput: [n][outPlane][outRow][outCol] 128 * 32 * 19 * 19 * 4
30: // weights: [filterId][upstreamplane][filterRow][filterCol] 32 * 32 * 5 * 5 * 4
31: // per workgroup:
32: // gradOutput: [outPlane][outRow][outCol] 32 * 19 * 19 * 4 = 46KB
33: // weights: [filterId][filterRow][filterCol] 32 * 5 * 5 * 4 = 3.2KB
34: // gradOutputforupstream: [n][upstreamPlane][upstreamRow][upstreamCol]
35: void kernel calcGradInputCached(
36:         const int batchSize,
37:         global const float *gradOutputGlobal,
38:         global const float *filtersGlobal,
39:         global float *gradInput,
40:         local float *_gradOutputPlane,
41:         local float *_filterPlane) {
42:
43:     #define globalId get_global_id(0)
44:     #define localId get_local_id(0)
45:     #define workgroupId get_group_id(0)
46:     #define workgroupSize get_local_size(0)
47:
48:     const int n = workgroupId / gInputPlanes;
49:     const int upstreamPlane = workgroupId % gInputPlanes;
50:
51:     const int upstreamRow = localId / gInputSize;
52:     const int upstreamCol = localId % gInputSize;
53:
54:     float sumWeightTimesOutError = 0;
55:     for (int outPlane = 0; outPlane < gNumFilters; outPlane++) {
56:         barrier(CLK_LOCAL_MEM_FENCE);
57:         copyLocal(_filterPlane, filtersGlobal + (outPlane * gInputPlanes + upstreamPlane) * gFilterSizeSquared, gFilterSizeSquared);
58:         copyLocal(_gradOutputPlane, gradOutputGlobal + (n * gNumFilters + outPlane) * gOutputSizeSquared, gOutputSizeSquared);
59:         barrier(CLK_LOCAL_MEM_FENCE);
60:         for (int filterRow = 0; filterRow < gFilterSize; filterRow++) {
61:             int outRow = upstreamRow + gMargin - filterRow;
62:             for (int filterCol = 0; filterCol < gFilterSize; filterCol++) {
63:                 int outCol = upstreamCol + gMargin - filterCol;
64:                 if (outCol >= 0 && outCol < gOutputSize && outRow >= 0 && outRow < gOutputSize) {
65:                     float thisWeightTimesError =
66:                         _gradOutputPlane[outRow * gOutputSize + outCol] *
67:                         _filterPlane[filterRow * gFilterSize + filterCol];
68:                     sumWeightTimesOutError += thisWeightTimesError;
69:                 }
70:             }
71:         }
72:     }
73:     const int upstreamImageGlobalOffset = (n * gInputPlanes + upstreamPlane) * gInputSizeSquared;
74:     if (localId < gInputSizeSquared) {
75:         gradInput[upstreamImageGlobalOffset + localId] = sumWeightTimesOutError;
76:     }
77: }
78:
79:

Something went wrong, code -55
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 889ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 this instance cant be used:
kernel source:
1: // Copyright Hugh Perkins 2014,2015 hughperkins at gmail
2: //
3: // This Source Code Form is subject to the terms of the Mozilla Public License,
4: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
5: // obtain one at http://mozilla.org/MPL/2.0/.
6:
7: // expected defines:
8: // BIASED (or not)
9:
10: // including cl/copyLocal.cl:
11: // Copyright Hugh Perkins 2015 hughperkins at gmail
12: //
13: // This Source Code Form is subject to the terms of the Mozilla Public License,
14: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
15: // obtain one at http://mozilla.org/MPL/2.0/.
16:
17: void copyLocal(local float *target, global float const *source, int N) {
18:     int numLoops = (N + get_local_size(0) - 1) / get_local_size(0);
19:     for (int loop = 0; loop < numLoops; loop++) {
20:         int offset = loop * get_local_size(0) + get_local_id(0);
21:         if (offset < N) {
22:             target[offset] = source[offset];
23:         }
24:     }
25: }
26:
27: void copyGlobal(global float *target, local float const *source, int N) {
28:     int numLoops = (N + get_local_size(0) - 1) / get_local_size(0);
29:     for (int loop = 0; loop < numLoops; loop++) {
30:         int offset = loop * get_local_size(0) + get_local_id(0);
31:         if (offset < N) {
32:             target[offset] = source[offset];
33:         }
34:     }
35: }
36:
37:
38: // including cl/ids.cl:
39: // Copyright Hugh Perkins 2015 hughperkins at gmail
40: //
41: // This Source Code Form is subject to the terms of the Mozilla Public License,
42: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
43: // obtain one at http://mozilla.org/MPL/2.0/.
44:
45: #define globalId (get_global_id(0))
46: #define localId (get_local_id(0)  )
47: #define workgroupId (get_group_id(0))
48: #define workgroupSize (get_local_size(0))
49:
50:
51:
52:
53: // workgroupId: [outputPlane][inputPlane]
54: // localId: [filterRow][filterCol]
55: // per-thread iteration: [n][outputRow][outputCol]
56: // local: errorimage: outputSize * outputSize
57: //        imageimage: inputSize * inputSize
58: void kernel backprop_floats_withscratch_dobias(
59:         const float learningRateMultiplier, const int batchSize,
60:          global const float *gradOutput, global const float *images,
61:         global float *gradWeights,
62:         #ifdef BIASED
63:              global float *gradBiasWeights,
64:         #endif
65:         local float *_errorImage, local float *_imageImage
66:  ) {
67:     const int filterRow = localId / gFilterSize;
68:     const int filterCol = localId % gFilterSize;
69:
70:     #define outPlane (workgroupId / gInputPlanes)
71:     #define upstreamPlane (workgroupId % gInputPlanes)
72:
73:     // gradWeights:     [outPlane][upstreamPlane][filterRow][filterCol]
74:     //       aggregate over:  [outRow][outCol][n]
75:     float thiswchange = 0;
76: #ifdef BIASED
77:     float thisbiaschange = 0;
78: #endif
79:     for (int n = 0; n < batchSize; n++) {
80:         barrier(CLK_LOCAL_MEM_FENCE);
81:         copyLocal(_imageImage, images + (n * gInputPlanes + upstreamPlane) * gInputSizeSquared, gInputSizeSquared);
82:         copyLocal(_errorImage, gradOutput + (n * gNumFilters + outPlane) * gOutputSizeSquared, gOutputSizeSquared);
83:         barrier(CLK_LOCAL_MEM_FENCE);
84:         if (localId < gFilterSizeSquared) {
85:             for (int outRow = 0; outRow < gOutputSize; outRow++) {
86:                 int upstreamRow = outRow - gMargin + filterRow;
87:                 for (int outCol = 0; outCol < gOutputSize; outCol++) {
88:                     const int upstreamCol = outCol - gMargin + filterCol;
89:                     #define proceed (upstreamRow >= 0 && upstreamCol >= 0 && upstreamRow < gInputSize && upstreamCol < gInputSize)
90:                     if (proceed) {
91:                         // these defines reduce register pressure, compared to const
92:                         // giving a 40% speedup on nvidia :-)
93:                         #define resultIndex (outRow * gOutputSize + outCol)
94:                         #define error (_errorImage[resultIndex])
95:                         //const float error = _errorImage[resultIndex];
96:                         #define upstreamDataIndex (upstreamRow * gInputSize + upstreamCol)
97:                         #define upstreamResult (_imageImage[upstreamDataIndex])
98:                         thiswchange += upstreamResult * error;
99:     #ifdef BIASED
100:                         thisbiaschange += error;
101:     #endif
102:                     }
103:                 }
104:             }
105:         }
106:     }
107:     if (localId < gFilterSizeSquared) {
108:         gradWeights[ workgroupId * gFilterSizeSquared + localId ] = learningRateMultiplier * thiswchange;
109:     }
110: #ifdef BIASED
111:     #define writeBias (upstreamPlane == 0 && filterRow == gMargin && filterCol == gMargin)
112:     if (writeBias) {
113:         gradBiasWeights[outPlane] = learningRateMultiplier * thisbiaschange;
114:     }
115: #endif
116:     // gradWeights:     [outPlane][upstreamPlane][filterRow][filterCol]
117:     //       aggregate over:  [outRow][outCol][n]
118: }
119:
120:

Something went wrong, code -55
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=10 -D gInputPlanes=10 -D gInputSize=26 -D gInputSizeSquared=676 -D gNumFilters=10 -D gFilterSize=26 -D gHalfFilterSize=13 -D gFilterSizeSquared=676 -D gNumOutputPlanes=10 -D gOutputPlanes=10 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=1 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=25 -DgInputStripeInnerNumRows=26 -DgInputStripeOuterNumRows=76 -DgInputStripeInnerSize=676 -DgInputStripeOuterSize=1976 -DgInputStripeMarginSize=650 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 this instance cant be used:
kernel source:
1: // Copyright Hugh Perkins 2014,2015 hughperkins at gmail
2: //
3: // This Source Code Form is subject to the terms of the Mozilla Public License,
4: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
5: // obtain one at http://mozilla.org/MPL/2.0/.
6:
7: // expected defines:
8: // BIASED (or not)
9:
10: // workgroupId: [outputPlane][inputPlane]
11: // localId: [filterRow][filterCol]
12: // per-thread iteration: [n][outputRow][outputCol]
13: // local: errorimage: outputSize * outputSize
14: //        imageimage: inputSize * inputSize
15: // specific characteristic: load one stripe of each image at a time,
16: // so we dont run out of memory
17: // number of stripes set in: gNumStripes
18: // note that whilst we can stripe the gradOutput simply,
19: // we actually need to add a half-filter widthed additional few rows
20: // onto the images stripe, otherwise we will be missing data
21: //   we will call the size of the non-overlapping image stripes: gInputStripeInnerSize
22: //      the outersize, including the two margins is: gInputStripeOuterSize
23: //      of course, the first and last stripes will be missing a bit off the top/bottom, where the
24: //      corresponding outer margin would be
25: void kernel backprop_floats_withscratch_dobias_striped(
26:         const float learningRateMultiplier, const int batchSize,
27:          global const float *gradOutput, global const float *images,
28:         global float *gradWeights,
29:         #ifdef BIASED
30:              global float *gradBiasWeights,
31:         #endif
32:         local float *_errorStripe, local float *_imageStripe
33:  ) {
34:     // gHalfFilterSize
35:     // gInputSize
36:     //
37:     // gInputStripeMarginRows => basically equal to gHalfFilterSize
38:     // gInputStripeInnerNumRows = gInputSize / gNumStripes
39:     // gInputStripeOuterNumRows = gInputStripeInnerNumRows + 2 * gHalfFilterSize  (note: one row less than
40:     //                                                         if we just added gFilterSize)
41:     // gInputStripeInnerSize = gInputStripeInnerNumRows * gInputSize
42:     // gInputStripeOuterSize = gInputStripeOuterNumRows * gInputSize
43:     // gInputStripeMarginSize = gInputStripeMarginRows * gInputSize
44:     //
45:     // gOutputStripeNumRows
46:     // gOutputStripeSize
47:
48:     const int globalId = get_global_id(0);
49:     const int localId = get_local_id(0);
50:     const int workgroupId = get_group_id(0);
51:     const int workgroupSize = get_local_size(0);
52:
53:     const int filterRow = localId / gFilterSize;
54:     const int filterCol = localId % gFilterSize;
55:
56:     const int outPlane = workgroupId / gInputPlanes;
57:     const int upstreamPlane = workgroupId % gInputPlanes;
58:
59:     // gradWeights:     [outPlane][upstreamPlane][filterRow][filterCol]
60:     //       aggregate over:  [outRow][outCol][n]
61:     float thiswchange = 0;
62: #ifdef BIASED
63:     float thisbiaschange = 0;
64: #endif
65:     const int numLoopsForImageStripe = (gInputStripeOuterSize + workgroupSize - 1) / workgroupSize;
66:     const int numLoopsForErrorStripe = (gOutputSizeSquared + workgroupSize - 1) / workgroupSize;
67:     for (int n = 0; n < batchSize; n++) {
68:         const int imageImageGlobalOffset = (n * gInputPlanes + upstreamPlane) * gInputSizeSquared;
69:         const int imageImageGlobalOffsetAfter = imageImageGlobalOffset + gInputSizeSquared;
70:         const int errorImageGlobalOffset = (n * gNumFilters + outPlane) * gOutputSizeSquared;
71:         const int errorImageGlobalOffsetAfter = errorImageGlobalOffset + gOutputSizeSquared;
72:         for (int stripe = 0; stripe < gNumStripes; stripe++) {
73:             const int imageStripeInnerOffset = imageImageGlobalOffset + stripe * gInputStripeInnerSize;
74:             const int imageStripeOuterOffset = imageStripeInnerOffset - gInputStripeMarginSize;
75:             // need to fetch the image, but it's bigger than us, so will need to loop...
76:             barrier(CLK_LOCAL_MEM_FENCE);
77:             for (int i = 0; i < numLoopsForImageStripe; i++) {
78:                 int thisOffset = i * workgroupSize + localId;
79:                 int thisGlobalImagesOffset = imageStripeOuterOffset + thisOffset;
80:                 bool process = thisOffset < gInputStripeOuterSize
81:                     && thisGlobalImagesOffset >= imageImageGlobalOffset
82:                     && thisGlobalImagesOffset < imageImageGlobalOffsetAfter;
83:                 if (process) {
84:                     _imageStripe[thisOffset] = images[ thisGlobalImagesOffset ];
85:                 }
86:             }
87:             int errorStripeOffset = errorImageGlobalOffset + stripe * gOutputStripeSize;
88:             for (int i = 0; i < numLoopsForErrorStripe; i++) {
89:                 int thisOffset = i * workgroupSize + localId;
90:                 int globalErrorsOffset = errorStripeOffset + thisOffset;
91:                 bool process = thisOffset < gOutputStripeSize
92:                     && globalErrorsOffset < errorImageGlobalOffsetAfter;
93:                 if (process) {
94:                     _errorStripe[thisOffset ] = gradOutput[globalErrorsOffset];
95:                 }
96:             }
97:             const int stripeOutRowStart = stripe * gOutputStripeNumRows;
98:             const int stripeOutRowEndExcl = stripeOutRowStart + gOutputStripeNumRows;
99:             barrier(CLK_LOCAL_MEM_FENCE);
100: //            if (localId == 13) {
101: //                for (int i = 0; i < 12; i++) {
102: //                    gradWeights[100 + stripe * 12 + i ] = _errorStripe[i * gOutputSize];
103: //                }
104: //                for (int i = 0; i < 20; i++) {
105: //                    gradWeights[200 + stripe * 20 + i ] = _imageStripe[i * gInputSize];
106: //                }
107: //            }
108:             if (localId < gFilterSizeSquared) {
109:                 for (int outRow = stripeOutRowStart; outRow < stripeOutRowEndExcl; outRow++) {
110:                     int upstreamRow = outRow - gMargin + filterRow;
111:                     for (int outCol = 0; outCol < gOutputSize; outCol++) {
112:                         int upstreamCol = outCol - gMargin + filterCol;
113:                         bool proceed =
114:                             upstreamRow >= 0 && upstreamCol >= 0
115:                             && upstreamRow < gInputSize && upstreamCol < gInputSize
116:                             && outRow < gOutputSize;
117:                         if (proceed) {
118:                             int resultIndex = outRow * gOutputSize + outCol;
119:                             float error = _errorStripe[resultIndex - stripe * gOutputStripeSize];
120:                             int upstreamDataIndex = upstreamRow * gInputSize + upstreamCol;
121:                             float upstreamResult = _imageStripe[upstreamDataIndex +  gInputStripeMarginSize
122:                                         - stripe * gInputStripeInnerSize ];
123:                             thiswchange += upstreamResult * error;
124:         #ifdef BIASED
125:                             thisbiaschange += error;
126:         #endif
127:                         }
128:                     }
129:                 }
130:             }
131:         }
132:     }
133:     if (localId < gFilterSizeSquared) {
134:         gradWeights[ workgroupId * gFilterSizeSquared + localId ] = learningRateMultiplier * thiswchange;
135: //        weightChanges[ workgroupId * gFilterSizeSquared + localId ] = workgroupId;
136:     }
137: #ifdef BIASED
138:     bool writeBias = upstreamPlane == 0 && filterRow == gMargin && filterCol == gMargin;
139:     if (writeBias) {
140:         gradBiasWeights[outPlane] = learningRateMultiplier * thisbiaschange;
141:     }
142: #endif
143:     // gradWeights:     [outPlane][upstreamPlane][filterRow][filterCol]
144:     //       aggregate over:  [outRow][outCol][n]
145: }
146:
147:

Something went wrong, code -55
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 1341ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2: cannot be used
   forward kernel 3: cannot be used
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 562ms
   forward layer selected kernel 1
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCL7D02.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2: cannot be used
   backward kernel 3 time: 889ms
   backward layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2: cannot be used
   calcGradWeights kernel 3: cannot be used
   calcGradWeights kernel 4 time: 1341ms
   calcGradWeights layer selected kernel 1
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=28 -D gInputSizeSquared=784 -D gNumFilters=10 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=10 -D gOutputPlanes=10 -D gOutputSize=26 -D gOutputSizeSquared=676 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=28 -DgInputStripeOuterNumRows=32 -DgInputStripeInnerSize=784 -DgInputStripeOuterSize=896 -DgInputStripeMarginSize=56 -DgOutputStripeNumRows=26 -DgOutputStripeSize=676
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 827ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 827ms
   forward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 889ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 889ms
   calcGradWeights layer selected kernel 1
batch time 15116 ms
dump enabled=0
clblas teardown
[       OK ] testsinglebatch.imagesize28 (15553 ms)
[ RUN      ] testsinglebatch.imagesize28_filtersize5
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=1 outputSize=28 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=28 numFilters=10 filterSize=5 outputSize=24 padZeros=0 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:FullyConnectedLayer{ numPlanes=10 imageSize=1 }
layer 4:ActivationLayer{ TANH }
layer 5:SquareLossLayer{}
Parameters overview: (skipping 4 layers with 0 params)
layer 1: params=260	0.4%
layer 3: params=57610	99.6%
TOTAL  : params=57870
weightsTotalSize=57870
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 15ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
forward try kernel 2
ForwardAuto: kernel 2: this instance cant be used: cannot use forward2, since outputimagesize * outputimagesize > maxworkgroupsize
   ... not valid
forward try kernel 3
ForwardAuto: kernel 3: this instance cant be used: cannot use forward3, since outputimagesize * outputimagesize > maxworkgroupsize
   ... not valid
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 562ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 this instance cant be used:
kernel source:
1: // Copyright Hugh Perkins 2014, 2015 hughperkins at gmail
2: //
3: // This Source Code Form is subject to the terms of the Mozilla Public License,
4: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
5: // obtain one at http://mozilla.org/MPL/2.0/.
6:
7: void copyLocal(local float *target, global float const *source, int N) {
8:     int numLoops = (N + get_local_size(0) - 1) / get_local_size(0);
9:     for (int loop = 0; loop < numLoops; loop++) {
10:         int offset = loop * get_local_size(0) + get_local_id(0);
11:         if (offset < N) {
12:             target[offset] = source[offset];
13:         }
14:     }
15: }
16:
17: // as calcGradInput, but with local cache
18: // convolve weights with gradOutput to produce gradInput
19: // workgroupid: [n][inputPlane]
20: // localid: [upstreamrow][upstreamcol]
21: // per-thread aggregation: [outPlane][filterRow][filterCol]
22: // need to store locally:
23: // - _gradOutputPlane. size = outputSizeSquared
24: // - _filterPlane. size = filtersizesquared
25: // note: currently doesnt use bias as input.  thats probably an error?
26: // inputs: gradOutput :convolve: filters => gradInput
27: //
28: // global:
29: // gradOutput: [n][outPlane][outRow][outCol] 128 * 32 * 19 * 19 * 4
30: // weights: [filterId][upstreamplane][filterRow][filterCol] 32 * 32 * 5 * 5 * 4
31: // per workgroup:
32: // gradOutput: [outPlane][outRow][outCol] 32 * 19 * 19 * 4 = 46KB
33: // weights: [filterId][filterRow][filterCol] 32 * 5 * 5 * 4 = 3.2KB
34: // gradOutputforupstream: [n][upstreamPlane][upstreamRow][upstreamCol]
35: void kernel calcGradInputCached(
36:         const int batchSize,
37:         global const float *gradOutputGlobal,
38:         global const float *filtersGlobal,
39:         global float *gradInput,
40:         local float *_gradOutputPlane,
41:         local float *_filterPlane) {
42:
43:     #define globalId get_global_id(0)
44:     #define localId get_local_id(0)
45:     #define workgroupId get_group_id(0)
46:     #define workgroupSize get_local_size(0)
47:
48:     const int n = workgroupId / gInputPlanes;
49:     const int upstreamPlane = workgroupId % gInputPlanes;
50:
51:     const int upstreamRow = localId / gInputSize;
52:     const int upstreamCol = localId % gInputSize;
53:
54:     float sumWeightTimesOutError = 0;
55:     for (int outPlane = 0; outPlane < gNumFilters; outPlane++) {
56:         barrier(CLK_LOCAL_MEM_FENCE);
57:         copyLocal(_filterPlane, filtersGlobal + (outPlane * gInputPlanes + upstreamPlane) * gFilterSizeSquared, gFilterSizeSquared);
58:         copyLocal(_gradOutputPlane, gradOutputGlobal + (n * gNumFilters + outPlane) * gOutputSizeSquared, gOutputSizeSquared);
59:         barrier(CLK_LOCAL_MEM_FENCE);
60:         for (int filterRow = 0; filterRow < gFilterSize; filterRow++) {
61:             int outRow = upstreamRow + gMargin - filterRow;
62:             for (int filterCol = 0; filterCol < gFilterSize; filterCol++) {
63:                 int outCol = upstreamCol + gMargin - filterCol;
64:                 if (outCol >= 0 && outCol < gOutputSize && outRow >= 0 && outRow < gOutputSize) {
65:                     float thisWeightTimesError =
66:                         _gradOutputPlane[outRow * gOutputSize + outCol] *
67:                         _filterPlane[filterRow * gFilterSize + filterCol];
68:                     sumWeightTimesOutError += thisWeightTimesError;
69:                 }
70:             }
71:         }
72:     }
73:     const int upstreamImageGlobalOffset = (n * gInputPlanes + upstreamPlane) * gInputSizeSquared;
74:     if (localId < gInputSizeSquared) {
75:         gradInput[upstreamImageGlobalOffset + localId] = sumWeightTimesOutError;
76:     }
77: }
78:
79:

Something went wrong, code -55
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 562ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 this instance cant be used:
kernel source:
1: // Copyright Hugh Perkins 2014,2015 hughperkins at gmail
2: //
3: // This Source Code Form is subject to the terms of the Mozilla Public License,
4: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
5: // obtain one at http://mozilla.org/MPL/2.0/.
6:
7: // expected defines:
8: // BIASED (or not)
9:
10: // including cl/copyLocal.cl:
11: // Copyright Hugh Perkins 2015 hughperkins at gmail
12: //
13: // This Source Code Form is subject to the terms of the Mozilla Public License,
14: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
15: // obtain one at http://mozilla.org/MPL/2.0/.
16:
17: void copyLocal(local float *target, global float const *source, int N) {
18:     int numLoops = (N + get_local_size(0) - 1) / get_local_size(0);
19:     for (int loop = 0; loop < numLoops; loop++) {
20:         int offset = loop * get_local_size(0) + get_local_id(0);
21:         if (offset < N) {
22:             target[offset] = source[offset];
23:         }
24:     }
25: }
26:
27: void copyGlobal(global float *target, local float const *source, int N) {
28:     int numLoops = (N + get_local_size(0) - 1) / get_local_size(0);
29:     for (int loop = 0; loop < numLoops; loop++) {
30:         int offset = loop * get_local_size(0) + get_local_id(0);
31:         if (offset < N) {
32:             target[offset] = source[offset];
33:         }
34:     }
35: }
36:
37:
38: // including cl/ids.cl:
39: // Copyright Hugh Perkins 2015 hughperkins at gmail
40: //
41: // This Source Code Form is subject to the terms of the Mozilla Public License,
42: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
43: // obtain one at http://mozilla.org/MPL/2.0/.
44:
45: #define globalId (get_global_id(0))
46: #define localId (get_local_id(0)  )
47: #define workgroupId (get_group_id(0))
48: #define workgroupSize (get_local_size(0))
49:
50:
51:
52:
53: // workgroupId: [outputPlane][inputPlane]
54: // localId: [filterRow][filterCol]
55: // per-thread iteration: [n][outputRow][outputCol]
56: // local: errorimage: outputSize * outputSize
57: //        imageimage: inputSize * inputSize
58: void kernel backprop_floats_withscratch_dobias(
59:         const float learningRateMultiplier, const int batchSize,
60:          global const float *gradOutput, global const float *images,
61:         global float *gradWeights,
62:         #ifdef BIASED
63:              global float *gradBiasWeights,
64:         #endif
65:         local float *_errorImage, local float *_imageImage
66:  ) {
67:     const int filterRow = localId / gFilterSize;
68:     const int filterCol = localId % gFilterSize;
69:
70:     #define outPlane (workgroupId / gInputPlanes)
71:     #define upstreamPlane (workgroupId % gInputPlanes)
72:
73:     // gradWeights:     [outPlane][upstreamPlane][filterRow][filterCol]
74:     //       aggregate over:  [outRow][outCol][n]
75:     float thiswchange = 0;
76: #ifdef BIASED
77:     float thisbiaschange = 0;
78: #endif
79:     for (int n = 0; n < batchSize; n++) {
80:         barrier(CLK_LOCAL_MEM_FENCE);
81:         copyLocal(_imageImage, images + (n * gInputPlanes + upstreamPlane) * gInputSizeSquared, gInputSizeSquared);
82:         copyLocal(_errorImage, gradOutput + (n * gNumFilters + outPlane) * gOutputSizeSquared, gOutputSizeSquared);
83:         barrier(CLK_LOCAL_MEM_FENCE);
84:         if (localId < gFilterSizeSquared) {
85:             for (int outRow = 0; outRow < gOutputSize; outRow++) {
86:                 int upstreamRow = outRow - gMargin + filterRow;
87:                 for (int outCol = 0; outCol < gOutputSize; outCol++) {
88:                     const int upstreamCol = outCol - gMargin + filterCol;
89:                     #define proceed (upstreamRow >= 0 && upstreamCol >= 0 && upstreamRow < gInputSize && upstreamCol < gInputSize)
90:                     if (proceed) {
91:                         // these defines reduce register pressure, compared to const
92:                         // giving a 40% speedup on nvidia :-)
93:                         #define resultIndex (outRow * gOutputSize + outCol)
94:                         #define error (_errorImage[resultIndex])
95:                         //const float error = _errorImage[resultIndex];
96:                         #define upstreamDataIndex (upstreamRow * gInputSize + upstreamCol)
97:                         #define upstreamResult (_imageImage[upstreamDataIndex])
98:                         thiswchange += upstreamResult * error;
99:     #ifdef BIASED
100:                         thisbiaschange += error;
101:     #endif
102:                     }
103:                 }
104:             }
105:         }
106:     }
107:     if (localId < gFilterSizeSquared) {
108:         gradWeights[ workgroupId * gFilterSizeSquared + localId ] = learningRateMultiplier * thiswchange;
109:     }
110: #ifdef BIASED
111:     #define writeBias (upstreamPlane == 0 && filterRow == gMargin && filterCol == gMargin)
112:     if (writeBias) {
113:         gradBiasWeights[outPlane] = learningRateMultiplier * thisbiaschange;
114:     }
115: #endif
116:     // gradWeights:     [outPlane][upstreamPlane][filterRow][filterCol]
117:     //       aggregate over:  [outRow][outCol][n]
118: }
119:
120:

Something went wrong, code -55
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=10 -D gInputPlanes=10 -D gInputSize=24 -D gInputSizeSquared=576 -D gNumFilters=10 -D gFilterSize=24 -D gHalfFilterSize=12 -D gFilterSizeSquared=576 -D gNumOutputPlanes=10 -D gOutputPlanes=10 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=1 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=23 -DgInputStripeInnerNumRows=24 -DgInputStripeOuterNumRows=70 -DgInputStripeInnerSize=576 -DgInputStripeOuterSize=1680 -DgInputStripeMarginSize=552 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 this instance cant be used:
kernel source:
1: // Copyright Hugh Perkins 2014,2015 hughperkins at gmail
2: //
3: // This Source Code Form is subject to the terms of the Mozilla Public License,
4: // v. 2.0. If a copy of the MPL was not distributed with this file, You can
5: // obtain one at http://mozilla.org/MPL/2.0/.
6:
7: // expected defines:
8: // BIASED (or not)
9:
10: // workgroupId: [outputPlane][inputPlane]
11: // localId: [filterRow][filterCol]
12: // per-thread iteration: [n][outputRow][outputCol]
13: // local: errorimage: outputSize * outputSize
14: //        imageimage: inputSize * inputSize
15: // specific characteristic: load one stripe of each image at a time,
16: // so we dont run out of memory
17: // number of stripes set in: gNumStripes
18: // note that whilst we can stripe the gradOutput simply,
19: // we actually need to add a half-filter widthed additional few rows
20: // onto the images stripe, otherwise we will be missing data
21: //   we will call the size of the non-overlapping image stripes: gInputStripeInnerSize
22: //      the outersize, including the two margins is: gInputStripeOuterSize
23: //      of course, the first and last stripes will be missing a bit off the top/bottom, where the
24: //      corresponding outer margin would be
25: void kernel backprop_floats_withscratch_dobias_striped(
26:         const float learningRateMultiplier, const int batchSize,
27:          global const float *gradOutput, global const float *images,
28:         global float *gradWeights,
29:         #ifdef BIASED
30:              global float *gradBiasWeights,
31:         #endif
32:         local float *_errorStripe, local float *_imageStripe
33:  ) {
34:     // gHalfFilterSize
35:     // gInputSize
36:     //
37:     // gInputStripeMarginRows => basically equal to gHalfFilterSize
38:     // gInputStripeInnerNumRows = gInputSize / gNumStripes
39:     // gInputStripeOuterNumRows = gInputStripeInnerNumRows + 2 * gHalfFilterSize  (note: one row less than
40:     //                                                         if we just added gFilterSize)
41:     // gInputStripeInnerSize = gInputStripeInnerNumRows * gInputSize
42:     // gInputStripeOuterSize = gInputStripeOuterNumRows * gInputSize
43:     // gInputStripeMarginSize = gInputStripeMarginRows * gInputSize
44:     //
45:     // gOutputStripeNumRows
46:     // gOutputStripeSize
47:
48:     const int globalId = get_global_id(0);
49:     const int localId = get_local_id(0);
50:     const int workgroupId = get_group_id(0);
51:     const int workgroupSize = get_local_size(0);
52:
53:     const int filterRow = localId / gFilterSize;
54:     const int filterCol = localId % gFilterSize;
55:
56:     const int outPlane = workgroupId / gInputPlanes;
57:     const int upstreamPlane = workgroupId % gInputPlanes;
58:
59:     // gradWeights:     [outPlane][upstreamPlane][filterRow][filterCol]
60:     //       aggregate over:  [outRow][outCol][n]
61:     float thiswchange = 0;
62: #ifdef BIASED
63:     float thisbiaschange = 0;
64: #endif
65:     const int numLoopsForImageStripe = (gInputStripeOuterSize + workgroupSize - 1) / workgroupSize;
66:     const int numLoopsForErrorStripe = (gOutputSizeSquared + workgroupSize - 1) / workgroupSize;
67:     for (int n = 0; n < batchSize; n++) {
68:         const int imageImageGlobalOffset = (n * gInputPlanes + upstreamPlane) * gInputSizeSquared;
69:         const int imageImageGlobalOffsetAfter = imageImageGlobalOffset + gInputSizeSquared;
70:         const int errorImageGlobalOffset = (n * gNumFilters + outPlane) * gOutputSizeSquared;
71:         const int errorImageGlobalOffsetAfter = errorImageGlobalOffset + gOutputSizeSquared;
72:         for (int stripe = 0; stripe < gNumStripes; stripe++) {
73:             const int imageStripeInnerOffset = imageImageGlobalOffset + stripe * gInputStripeInnerSize;
74:             const int imageStripeOuterOffset = imageStripeInnerOffset - gInputStripeMarginSize;
75:             // need to fetch the image, but it's bigger than us, so will need to loop...
76:             barrier(CLK_LOCAL_MEM_FENCE);
77:             for (int i = 0; i < numLoopsForImageStripe; i++) {
78:                 int thisOffset = i * workgroupSize + localId;
79:                 int thisGlobalImagesOffset = imageStripeOuterOffset + thisOffset;
80:                 bool process = thisOffset < gInputStripeOuterSize
81:                     && thisGlobalImagesOffset >= imageImageGlobalOffset
82:                     && thisGlobalImagesOffset < imageImageGlobalOffsetAfter;
83:                 if (process) {
84:                     _imageStripe[thisOffset] = images[ thisGlobalImagesOffset ];
85:                 }
86:             }
87:             int errorStripeOffset = errorImageGlobalOffset + stripe * gOutputStripeSize;
88:             for (int i = 0; i < numLoopsForErrorStripe; i++) {
89:                 int thisOffset = i * workgroupSize + localId;
90:                 int globalErrorsOffset = errorStripeOffset + thisOffset;
91:                 bool process = thisOffset < gOutputStripeSize
92:                     && globalErrorsOffset < errorImageGlobalOffsetAfter;
93:                 if (process) {
94:                     _errorStripe[thisOffset ] = gradOutput[globalErrorsOffset];
95:                 }
96:             }
97:             const int stripeOutRowStart = stripe * gOutputStripeNumRows;
98:             const int stripeOutRowEndExcl = stripeOutRowStart + gOutputStripeNumRows;
99:             barrier(CLK_LOCAL_MEM_FENCE);
100: //            if (localId == 13) {
101: //                for (int i = 0; i < 12; i++) {
102: //                    gradWeights[100 + stripe * 12 + i ] = _errorStripe[i * gOutputSize];
103: //                }
104: //                for (int i = 0; i < 20; i++) {
105: //                    gradWeights[200 + stripe * 20 + i ] = _imageStripe[i * gInputSize];
106: //                }
107: //            }
108:             if (localId < gFilterSizeSquared) {
109:                 for (int outRow = stripeOutRowStart; outRow < stripeOutRowEndExcl; outRow++) {
110:                     int upstreamRow = outRow - gMargin + filterRow;
111:                     for (int outCol = 0; outCol < gOutputSize; outCol++) {
112:                         int upstreamCol = outCol - gMargin + filterCol;
113:                         bool proceed =
114:                             upstreamRow >= 0 && upstreamCol >= 0
115:                             && upstreamRow < gInputSize && upstreamCol < gInputSize
116:                             && outRow < gOutputSize;
117:                         if (proceed) {
118:                             int resultIndex = outRow * gOutputSize + outCol;
119:                             float error = _errorStripe[resultIndex - stripe * gOutputStripeSize];
120:                             int upstreamDataIndex = upstreamRow * gInputSize + upstreamCol;
121:                             float upstreamResult = _imageStripe[upstreamDataIndex +  gInputStripeMarginSize
122:                                         - stripe * gInputStripeInnerSize ];
123:                             thiswchange += upstreamResult * error;
124:         #ifdef BIASED
125:                             thisbiaschange += error;
126:         #endif
127:                         }
128:                     }
129:                 }
130:             }
131:         }
132:     }
133:     if (localId < gFilterSizeSquared) {
134:         gradWeights[ workgroupId * gFilterSizeSquared + localId ] = learningRateMultiplier * thiswchange;
135: //        weightChanges[ workgroupId * gFilterSizeSquared + localId ] = workgroupId;
136:     }
137: #ifdef BIASED
138:     bool writeBias = upstreamPlane == 0 && filterRow == gMargin && filterCol == gMargin;
139:     if (writeBias) {
140:         gradBiasWeights[outPlane] = learningRateMultiplier * thisbiaschange;
141:     }
142: #endif
143:     // gradWeights:     [outPlane][upstreamPlane][filterRow][filterCol]
144:     //       aggregate over:  [outRow][outCol][n]
145: }
146:
147:

Something went wrong, code -55
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 1014ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 15ms
   forward kernel 2: cannot be used
   forward kernel 3: cannot be used
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 562ms
   forward layer selected kernel 4
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLB324.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2: cannot be used
   backward kernel 3 time: 562ms
   backward layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2: cannot be used
   calcGradWeights kernel 3: cannot be used
   calcGradWeights kernel 4 time: 1014ms
   calcGradWeights layer selected kernel 1
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=28 -D gInputSizeSquared=784 -D gNumFilters=10 -D gFilterSize=5 -D gHalfFilterSize=2 -D gFilterSizeSquared=25 -D gNumOutputPlanes=10 -D gOutputPlanes=10 -D gOutputSize=24 -D gOutputSizeSquared=576 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=4 -DgInputStripeInnerNumRows=28 -DgInputStripeOuterNumRows=36 -DgInputStripeInnerSize=784 -DgInputStripeOuterSize=1008 -DgInputStripeMarginSize=112 -DgOutputStripeNumRows=24 -DgOutputStripeSize=576
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 499ms
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 499ms
   forward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 702ms
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 702ms
   calcGradWeights layer selected kernel 1
batch time 12714 ms
dump enabled=0
clblas teardown
[       OK ] testsinglebatch.imagesize28_filtersize5 (13151 ms)
[ RUN      ] testsinglebatch.imagesize5_filtersize3_batchsize2_softmax
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=5 filterSize=3 outputSize=5 padZeros=1 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:ConvolutionalLayer{ LayerDimensions{ inputPlanes=5 inputSize=5 numFilters=5 filterSize=3 outputSize=5 padZeros=1 biased=1 skip=0} }
layer 4:ActivationLayer{ RELU }
layer 5:FullyConnectedLayer{ numPlanes=5 imageSize=1 }
layer 6:SoftMaxLayer{ perPlane=0 numPlanes=5 imageSize=1 }
Parameters overview: (skipping 4 layers with 0 params)
layer 1: params=50	5.5%
layer 3: params=230	25.3%
layer 5: params=630	69.2%
TOTAL  : params=910
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 16ms
layer 1 offset: 0
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 1
    from w: 0
    actual: -3.3898
layer 2 offset: 50
layer 3 offset: 50
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 16ms
layer 3
    from w: 0
    actual: -3.3898
layer 4 offset: 280
layer 5 offset: 280
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 687ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 109ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLCC8F.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
layer 5
    from w: 0
    actual: -3.38981
layer 6 offset: 910
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 687ms
   forward layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 109ms
   forward layer selected kernel 1
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 328ms
full thisloss: 3.3898
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 16ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 328ms
   forward layer selected kernel 1
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 436ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=5 -D gInputPlanes=5 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=5 -D gFilterSize=5 -D gHalfFilterSize=2 -D gFilterSizeSquared=25 -D gNumOutputPlanes=5 -D gOutputPlanes=5 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=4 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=13 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=65 -DgInputStripeMarginSize=20 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 1108ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=5 -D gInputPlanes=5 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=5 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=5 -D gOutputPlanes=5 -D gOutputSize=5 -D gOutputSizeSquared=25 -D gPadZeros=1 -D gMargin=1 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=5 -DgOutputStripeSize=25
   ... seems valid
BackpropWeightsAuto: kernel 3 15ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=5 -D gInputSizeSquared=25 -D gNumFilters=5 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=5 -D gOutputPlanes=5 -D gOutputSize=5 -D gOutputSizeSquared=25 -D gPadZeros=1 -D gMargin=1 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=5 -DgInputStripeOuterNumRows=9 -DgInputStripeInnerSize=25 -DgInputStripeOuterSize=45 -DgInputStripeMarginSize=10 -DgOutputStripeNumRows=5 -DgOutputStripeSize=25
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 436ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 858ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 1108ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 1076ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 109ms
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 858ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 15ms
   calcGradWeights kernel 4 time: 1076ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 16ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 109ms
   calcGradWeights layer selected kernel 2
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 3
    from w: 0
    actual: 0
layer 4 offset: 280
layer 5 offset: 280
layer 5
    from w: 0
    actual: 0
layer 6 offset: 910
full thisloss: 3.3898
batch time 8252 ms
dump enabled=0
clblas teardown
[       OK ] testsinglebatch.imagesize5_filtersize3_batchsize2_softmax (8705 ms)
[ RUN      ] testsinglebatch.imagesize4_filtersize3_batchsize2_pooling
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
initializing clblas
layer 0:InputLayer{ outputPlanes=1 outputSize=12 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=12 numFilters=5 filterSize=3 outputSize=12 padZeros=1 biased=1 skip=0} }
layer 2:ActivationLayer{ RELU }
layer 3:PoolingLayer{ inputPlanes=5 inputSize=12 poolingSize=2 }
layer 4:ConvolutionalLayer{ LayerDimensions{ inputPlanes=5 inputSize=6 numFilters=5 filterSize=3 outputSize=6 padZeros=1 biased=1 skip=0} }
layer 5:ActivationLayer{ RELU }
layer 6:PoolingLayer{ inputPlanes=5 inputSize=6 poolingSize=2 }
layer 7:FullyConnectedLayer{ numPlanes=5 imageSize=1 }
layer 8:SoftMaxLayer{ perPlane=0 numPlanes=5 imageSize=1 }
Parameters overview: (skipping 6 layers with 0 params)
layer 1: params=50	9.8%
layer 4: params=230	45.1%
layer 7: params=230	45.1%
TOTAL  : params=510
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 16ms
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
forward try kernel 2
   ... seems valid
ForwardAuto: kernel 2 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
backward try kernel 0
  ... not plausibly optimal, skipping
backward try kernel 1
   ... seems valid
BackwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
layer 1 offset: 0
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
forward try kernel 3
   ... seems valid
ForwardAuto: kernel 3 0ms
layer 1
    from w: 0
    actual: -3.3063
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
forward try kernel 4
   ... seems valid
ForwardAuto: kernel 4 0ms
layer 4
    from w: 0
    actual: -3.3063
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 561ms
forward try kernel 5
ForwardAuto: kernel 5: this instance cant be used: For ForwardFc, filtersize and inputimagesize must be identical
   ... not valid
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 562ms
forward try kernel 5
cl/forward_fc_wgperrow.cl build log:
"C:\Users\pz\AppData\Local\Temp\OCLF1F7.tmp.cl", line 75: warning: variable
          "loopsPerExample" was declared but never referenced
      const int loopsPerExample = (gInputSize + workgroupSize - 1) / workgroupSize;
                ^


   ... seems valid
ForwardAuto: kernel 5 0ms
layer 7
    from w: 0
    actual: -3.3063
layer 8 offset: 510
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 561ms
   forward layer selected kernel 1
   forward kernel 0: cannot be used
   forward kernel 1 time: 16ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5: cannot be used
   forward kernel 6: cannot be used
   forward kernel 7 time: 562ms
   forward layer selected kernel 2
forward try kernel 6
   ... seems valid
ForwardAuto: kernel 6 this instance cant be used: memallocsize too small to use this kernel on this device.  Need: 0MB, but only have: -1984MB max alloc size
forward try kernel 7
   ... seems valid
ForwardAuto: kernel 7 328ms
full thisloss: 3.3063
   forward kernel 0: cannot be used
   forward kernel 1 time: 0ms
   forward kernel 2 time: 0ms
   forward kernel 3 time: 0ms
   forward kernel 4 time: 0ms
   forward kernel 5 time: 0ms
   forward kernel 6: cannot be used
   forward kernel 7 time: 328ms
   forward layer selected kernel 1
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
backward try kernel 2
   ... seems valid
BackwardAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
calcGradWeights try kernel 2
   ... seems valid
BackpropWeightsAuto: kernel 2 0ms
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 421ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=5 -D gInputPlanes=5 -D gInputSize=3 -D gInputSizeSquared=9 -D gNumFilters=5 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=5 -D gOutputPlanes=5 -D gOutputSize=1 -D gOutputSizeSquared=1 -D gPadZeros=0 -D gMargin=0 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=3 -DgInputStripeOuterNumRows=7 -DgInputStripeInnerSize=9 -DgInputStripeOuterSize=21 -DgInputStripeMarginSize=6 -DgOutputStripeNumRows=1 -DgOutputStripeSize=1
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
backward try kernel 3
   ... seems valid
BackwardAuto: kernel 3 920ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=5 -D gInputPlanes=5 -D gInputSize=6 -D gInputSizeSquared=36 -D gNumFilters=5 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=5 -D gOutputPlanes=5 -D gOutputSize=6 -D gOutputSizeSquared=36 -D gPadZeros=1 -D gMargin=1 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=6 -DgInputStripeOuterNumRows=10 -DgInputStripeInnerSize=36 -DgInputStripeOuterSize=60 -DgInputStripeMarginSize=12 -DgOutputStripeNumRows=6 -DgOutputStripeSize=36
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
calcGradWeights try kernel 3
options:  -D BIASED -D gNumInputPlanes=1 -D gInputPlanes=1 -D gInputSize=12 -D gInputSizeSquared=144 -D gNumFilters=5 -D gFilterSize=3 -D gHalfFilterSize=1 -D gFilterSizeSquared=9 -D gNumOutputPlanes=5 -D gOutputPlanes=5 -D gOutputSize=12 -D gOutputSizeSquared=144 -D gPadZeros=1 -D gMargin=1 -D gEven=0 -D gSkip=0 -DgNumStripes=1 -DgInputStripeMarginRows=2 -DgInputStripeInnerNumRows=12 -DgInputStripeOuterNumRows=16 -DgInputStripeInnerSize=144 -DgInputStripeOuterSize=192 -DgInputStripeMarginSize=24 -DgOutputStripeNumRows=12 -DgOutputStripeSize=144
   ... seems valid
BackpropWeightsAuto: kernel 3 0ms
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 421ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 842ms
   backward kernel 0: cannot be used
   backward kernel 1 time: 0ms
   backward kernel 2 time: 0ms
   backward kernel 3 time: 920ms
   backward layer selected kernel 1
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 889ms
calcGradWeights try kernel 4
   ... seems valid
BackpropWeightsAuto: kernel 4 546ms
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 842ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 889ms
   calcGradWeights layer selected kernel 1
   calcGradWeights kernel 0: cannot be used
   calcGradWeights kernel 1 time: 0ms
   calcGradWeights kernel 2 time: 0ms
   calcGradWeights kernel 3 time: 0ms
   calcGradWeights kernel 4 time: 546ms
   calcGradWeights layer selected kernel 1
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
layer 1 offset: 0
layer 1
    from w: 0
    actual: 0
layer 2 offset: 50
layer 3 offset: 50
layer 4 offset: 50
layer 4
    from w: 0
    actual: 0
layer 5 offset: 280
layer 6 offset: 280
layer 7 offset: 280
layer 7
    from w: 0
    actual: 0
layer 8 offset: 510
full thisloss: 3.3063
batch time 8954 ms
dump enabled=0
clblas teardown
[       OK ] testsinglebatch.imagesize4_filtersize3_batchsize2_pooling (9734 ms)
[----------] 6 tests from testsinglebatch (59763 ms total)

[----------] 9 tests from testpoolingforward
[ RUN      ] testpoolingforward.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.basic (94 ms)
[ RUN      ] testpoolingforward.basic_2plane_batchsize2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.basic_2plane_batchsize2 (78 ms)
[ RUN      ] testpoolingforward.fromwrappers
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.fromwrappers (93 ms)
[ RUN      ] testpoolingforward.comparespecific_0_1_pooling2
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.comparespecific_0_1_pooling2 (94 ms)
[ RUN      ] testpoolingforward.comparespecific_0_1_pooling3
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.comparespecific_0_1_pooling3 (94 ms)
[ RUN      ] testpoolingforward.comparespecific_0_1_pooling2_pz
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.comparespecific_0_1_pooling2_pz (93 ms)
[ RUN      ] testpoolingforward.comparespecific_0_1_pooling3_pz
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.comparespecific_0_1_pooling3_pz (109 ms)
[ RUN      ] testpoolingforward.comparespecific_0_1_pooling3_small
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.comparespecific_0_1_pooling3_small (78 ms)
[ RUN      ] testpoolingforward.comparespecific_0_1_pooling3_small2
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingforward.comparespecific_0_1_pooling3_small2 (78 ms)
[----------] 9 tests from testpoolingforward (811 ms total)

[----------] 2 tests from testpoolingbackward
[ RUN      ] testpoolingbackward.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingbackward.basic (16 ms)
[ RUN      ] testpoolingbackward.basic_2plane_batchsize2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testpoolingbackward.basic_2plane_batchsize2 (16 ms)
[----------] 2 tests from testpoolingbackward (32 ms total)

[----------] 1 test from testNorbLoader
[ RUN      ] testNorbLoader.load1000
unknown file: error: C++ exception with description "failed to open file: ..\data\norb\training-shuffled-dat.mat" thrown in the test body.
[  FAILED  ] testNorbLoader.load1000 (0 ms)
[----------] 1 test from testNorbLoader (0 ms total)

[----------] 7 tests from teststringhelper
[ RUN      ] teststringhelper.split
[       OK ] teststringhelper.split (0 ms)
[ RUN      ] teststringhelper.split2
[       OK ] teststringhelper.split2 (0 ms)
[ RUN      ] teststringhelper.split3
[       OK ] teststringhelper.split3 (0 ms)
[ RUN      ] teststringhelper.tolower
[       OK ] teststringhelper.tolower (0 ms)
[ RUN      ] teststringhelper.replace
[       OK ] teststringhelper.replace (0 ms)
[ RUN      ] teststringhelper.replaceglobal
[       OK ] teststringhelper.replaceglobal (0 ms)
[ RUN      ] teststringhelper.strcpy_safe
[       OK ] teststringhelper.strcpy_safe (0 ms)
[----------] 7 tests from teststringhelper (0 ms total)

[----------] 1 test from testGtestGlobals
[ RUN      ] testGtestGlobals.basic
There are 1 parameters:
   argv[0]=deepcl_unittests.exe
[       OK ] testGtestGlobals.basic (0 ms)
[----------] 1 test from testGtestGlobals (0 ms total)

[----------] 1 test from testMemset
[ RUN      ] testMemset.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
myArray[0]=99
myArray[1]=99
myArray[2]=99
myArray[3]=99
myArray[4]=99
myArray[5]=99
myArray[6]=99
myArray[7]=99
myArray[8]=99
myArray[9]=99
[       OK ] testMemset.basic (78 ms)
[----------] 1 test from testMemset (78 ms total)

[----------] 2 tests from testCopyBuffer
[ RUN      ] testCopyBuffer.floats
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
3
4
5
6
7
8
9
10
11
12
3
4
5
6
7
8
9
10
11
12
 3
 4
 5
 6
 7
 8
 9
 10
 11
 12
[       OK ] testCopyBuffer.floats (187 ms)
[ RUN      ] testCopyBuffer.nits
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
3
4
5
6
7
8
9
10
11
12
3
4
5
6
7
8
9
10
11
12
 3
 4
 5
 6
 7
 8
 9
 10
 11
 12
[       OK ] testCopyBuffer.nits (171 ms)
[----------] 2 tests from testCopyBuffer (358 ms total)

[----------] 2 tests from testCopyBlock
[ RUN      ] testCopyBlock.testPos
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
in[0]=3076
in[1]=8
in[2]=14
res[0]=3
res[1]=4
res[2]=8206
res[3]=8
res[4]=14
[       OK ] testCopyBlock.testPos (110 ms)
[ RUN      ] testCopyBlock.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
2 3 4
6 7 8

0 0 0 0

5 6 7
9 10 11

0 0 0 0

[       OK ] testCopyBlock.basic (93 ms)
[----------] 2 tests from testCopyBlock (203 ms total)

[----------] 1 test from testCopyLocal
[ RUN      ] testCopyLocal.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
0 0 0 0
1 2 3 4
5 6 7 8
9 10 11 12

0 0 0 0
[       OK ] testCopyLocal.basic (78 ms)
[----------] 1 test from testCopyLocal (78 ms total)

[----------] 8 tests from testNetdefToNet
[ RUN      ] testNetdefToNet.empty
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testNetdefToNet.empty (16 ms)
[ RUN      ] testNetdefToNet.onefc
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testNetdefToNet.onefc (171 ms)
[ RUN      ] testNetdefToNet.onefclinear
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testNetdefToNet.onefclinear (156 ms)
[ RUN      ] testNetdefToNet.150n_10n
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testNetdefToNet.150n_10n (172 ms)
[ RUN      ] testNetdefToNet.3xfclinear
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
nnString: [3]
repeatNum 3
remainderString [150n]
inner [150n]
multiplied string: 150n-150n-150n
layer 0:InputLayer{ outputPlanes=1 outputSize=19 }
layer 1:FullyConnectedLayer{ numPlanes=150 imageSize=1 }
layer 2:FullyConnectedLayer{ numPlanes=150 imageSize=1 }
layer 3:FullyConnectedLayer{ numPlanes=150 imageSize=1 }
layer 4:SoftMaxLayer{ perPlane=0 numPlanes=150 imageSize=1 }
Parameters overview: (skipping 2 layers with 0 params)
layer 1: params=54300	54.5%
layer 2: params=22650	22.7%
layer 3: params=22650	22.7%
TOTAL  : params=99600
[       OK ] testNetdefToNet.3xfclinear (156 ms)
[ RUN      ] testNetdefToNet.mp2_3x32c5z_10n
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
prefix: [mp2]
nnString: [3]
repeatNum 3
remainderString [32c5z-10n ]
postfix [10n ]
inner [32c5z]
multiplied string: mp2-32c5z-32c5z-32c5z-10n
layer 0:InputLayer{ outputPlanes=1 outputSize=19 }
layer 1:PoolingLayer{ inputPlanes=1 inputSize=19 poolingSize=2 }
layer 2:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=9 numFilters=32 filterSize=5 outputSize=9 padZeros=1 biased=1 skip=0} }
layer 3:ConvolutionalLayer{ LayerDimensions{ inputPlanes=32 inputSize=9 numFilters=32 filterSize=5 outputSize=9 padZeros=1 biased=1 skip=0} }
layer 4:ConvolutionalLayer{ LayerDimensions{ inputPlanes=32 inputSize=9 numFilters=32 filterSize=5 outputSize=9 padZeros=1 biased=1 skip=0} }
layer 5:FullyConnectedLayer{ numPlanes=10 imageSize=1 }
layer 6:SoftMaxLayer{ perPlane=0 numPlanes=10 imageSize=1 }
Parameters overview: (skipping 3 layers with 0 params)
layer 2: params=832	1.1%
layer 3: params=25632	32.9%
layer 4: params=25632	32.9%
layer 5: params=25930	33.2%
TOTAL  : params=78026
[       OK ] testNetdefToNet.mp2_3x32c5z_10n (343 ms)
[ RUN      ] testNetdefToNet.3x32c5zmp2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
nnString: [3]
repeatNum 3
remainderString [(32c5z-mp2)-10n]
inner [32c5z-mp2]
newRemainder [-10n]
postfix [10n]
multiplied string: 32c5z-mp2-32c5z-mp2-32c5z-mp2-10n
layer 0:InputLayer{ outputPlanes=1 outputSize=128 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=128 numFilters=32 filterSize=5 outputSize=128 padZeros=1 biased=1 skip=0} }
layer 2:PoolingLayer{ inputPlanes=32 inputSize=128 poolingSize=2 }
layer 3:ConvolutionalLayer{ LayerDimensions{ inputPlanes=32 inputSize=64 numFilters=32 filterSize=5 outputSize=64 padZeros=1 biased=1 skip=0} }
layer 4:PoolingLayer{ inputPlanes=32 inputSize=64 poolingSize=2 }
layer 5:ConvolutionalLayer{ LayerDimensions{ inputPlanes=32 inputSize=32 numFilters=32 filterSize=5 outputSize=32 padZeros=1 biased=1 skip=0} }
layer 6:PoolingLayer{ inputPlanes=32 inputSize=32 poolingSize=2 }
layer 7:FullyConnectedLayer{ numPlanes=10 imageSize=1 }
layer 8:SoftMaxLayer{ perPlane=0 numPlanes=10 imageSize=1 }
Parameters overview: (skipping 5 layers with 0 params)
layer 1: params=832	0.6%
layer 3: params=25632	19.1%
layer 5: params=25632	19.1%
layer 7: params=81930	61.1%
TOTAL  : params=134026
[       OK ] testNetdefToNet.3x32c5zmp2 (702 ms)
[ RUN      ] testNetdefToNet.2x32c7_3x32c5z
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
nnString: [2]
repeatNum 2
remainderString [32c7z-3*32c5z-10n]
postfix [3*32c5z-10n]
inner [32c7z]
nnString: [3]
repeatNum 3
remainderString [32c5z-10n]
postfix [10n]
inner [32c5z]
multiplied string: 32c5z-32c5z-32c5z-10n
multiplied string: 32c7z-32c7z-32c5z-32c5z-32c5z-10n
layer 0:InputLayer{ outputPlanes=1 outputSize=19 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=19 numFilters=32 filterSize=7 outputSize=19 padZeros=1 biased=1 skip=0} }
layer 2:ConvolutionalLayer{ LayerDimensions{ inputPlanes=32 inputSize=19 numFilters=32 filterSize=7 outputSize=19 padZeros=1 biased=1 skip=0} }
layer 3:ConvolutionalLayer{ LayerDimensions{ inputPlanes=32 inputSize=19 numFilters=32 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0} }
layer 4:ConvolutionalLayer{ LayerDimensions{ inputPlanes=32 inputSize=19 numFilters=32 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0} }
layer 5:ConvolutionalLayer{ LayerDimensions{ inputPlanes=32 inputSize=19 numFilters=32 filterSize=5 outputSize=19 padZeros=1 biased=1 skip=0} }
layer 6:FullyConnectedLayer{ numPlanes=10 imageSize=1 }
layer 7:SoftMaxLayer{ perPlane=0 numPlanes=10 imageSize=1 }
Parameters overview: (skipping 2 layers with 0 params)
layer 1: params=1600	0.7%
layer 2: params=50208	20.6%
layer 3: params=25632	10.5%
layer 4: params=25632	10.5%
layer 5: params=25632	10.5%
layer 6: params=115530	47.3%
TOTAL  : params=244234
[       OK ] testNetdefToNet.2x32c7_3x32c5z (172 ms)
[----------] 8 tests from testNetdefToNet (1888 ms total)

[----------] 10 tests from testactivationforward
[ RUN      ] testactivationforward.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.basic (15 ms)
[ RUN      ] testactivationforward.basic_2plane_batchsize2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.basic_2plane_batchsize2 (16 ms)
[ RUN      ] testactivationforward.fromwrappers
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.fromwrappers (78 ms)
[ RUN      ] testactivationforward.comparespecific_0_1_activation2
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.comparespecific_0_1_activation2 (78 ms)
[ RUN      ] testactivationforward.comparespecific_0_1_activation3
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.comparespecific_0_1_activation3 (94 ms)
[ RUN      ] testactivationforward.comparespecific_0_1_activation2_pz
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.comparespecific_0_1_activation2_pz (78 ms)
[ RUN      ] testactivationforward.comparespecific_0_1_activation3_pz
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.comparespecific_0_1_activation3_pz (78 ms)
[ RUN      ] testactivationforward.comparespecific_0_1_activation3_small
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.comparespecific_0_1_activation3_small (78 ms)
[ RUN      ] testactivationforward.comparespecific_0_1_activation3_small2
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.comparespecific_0_1_activation3_small2 (78 ms)
[ RUN      ] testactivationforward.comparespecific_0_1_activation3_small2_tanh
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testactivationforward.comparespecific_0_1_activation3_small2_tanh (109 ms)
[----------] 10 tests from testactivationforward (702 ms total)

[----------] 2 tests from testactivationbackward
[ RUN      ] testactivationbackward.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
 gradInput=3
 gradInput=0
 gradInput=-2.7
 gradInput=2
 gradInput=-0
 gradInput=2.1
 gradInput=0
 gradInput=-1.1
 gradInput=0
[       OK ] testactivationbackward.basic (0 ms)
[ RUN      ] testactivationbackward.basic_2plane_batchsize2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
 gradInput=3
 gradInput=0
 gradInput=0
 gradInput=9
[       OK ] testactivationbackward.basic_2plane_batchsize2 (15 ms)
[----------] 2 tests from testactivationbackward (15 ms total)

[----------] 1 test from testRandomSingleton
[ RUN      ] testRandomSingleton.testMockRandom
0.549356
0.634521
0.5968
0.863601
0.982891
0.637683
0.248837
0.351605
0.225401
0.220224
[       OK ] testRandomSingleton.testMockRandom (0 ms)
[----------] 1 test from testRandomSingleton (0 ms total)

[----------] 10 tests from testdropoutforward
[ RUN      ] testdropoutforward.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.basic (16 ms)
[ RUN      ] testdropoutforward.basic_2plane_batchsize2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.basic_2plane_batchsize2 (16 ms)
[ RUN      ] testdropoutforward.fromwrappers
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.fromwrappers (15 ms)
[ RUN      ] testdropoutforward.comparespecific_0_1_dropout2
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.comparespecific_0_1_dropout2 (78 ms)
[ RUN      ] testdropoutforward.comparespecific_0_1_dropout3
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.comparespecific_0_1_dropout3 (78 ms)
[ RUN      ] testdropoutforward.comparespecific_0_1_dropout2_pz
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.comparespecific_0_1_dropout2_pz (94 ms)
[ RUN      ] testdropoutforward.comparespecific_0_1_dropout3_pz
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.comparespecific_0_1_dropout3_pz (78 ms)
[ RUN      ] testdropoutforward.comparespecific_0_1_dropout3_small
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.comparespecific_0_1_dropout3_small (93 ms)
[ RUN      ] testdropoutforward.comparespecific_0_1_dropout3_small2
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.comparespecific_0_1_dropout3_small2 (78 ms)
[ RUN      ] testdropoutforward.comparespecific_0_1_dropout3_small2_tanh
instance0: 0
instance1: 1
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutforward.comparespecific_0_1_dropout3_small2_tanh (78 ms)
[----------] 10 tests from testdropoutforward (624 ms total)

[----------] 3 tests from testdropoutbackward
[ RUN      ] testdropoutbackward.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutbackward.basic (94 ms)
[ RUN      ] testdropoutbackward.basic_2plane_batchsize2
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutbackward.basic_2plane_batchsize2 (78 ms)
[ RUN      ] testdropoutbackward.compare_args
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testdropoutbackward.compare_args (78 ms)
[----------] 3 tests from testdropoutbackward (250 ms total)

[----------] 1 test from testsgd
[ RUN      ] testsgd.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
layer 0:InputLayer{ outputPlanes=1 outputSize=5 }
layer 1:ConvolutionalLayer{ LayerDimensions{ inputPlanes=1 inputSize=5 numFilters=1 filterSize=3 outputSize=3 padZeros=0 biased=0 skip=0} }
layer 2:SquareLossLayer{}

inputtotalsize=50 outputTotalSize=18
forward try kernel 0
  ... not plausibly optimal, skipping
forward try kernel 1
   ... seems valid
ForwardAuto: kernel 1 0ms
calcGradWeights try kernel 0
  ... not plausibly optimal, skipping
calcGradWeights try kernel 1
   ... seems valid
BackpropWeightsAuto: kernel 1 0ms
[       OK ] testsgd.basic (577 ms)
[----------] 1 test from testsgd (577 ms total)

[----------] 9 tests from testCLMathWrapper
[ RUN      ] testCLMathWrapper.assign
a[0]=4
a[1]=2.1
a[2]=5
a[3]=3
a[4]=9.2
[       OK ] testCLMathWrapper.assign (78 ms)
[ RUN      ] testCLMathWrapper.assignScalar
a[0]=3.4
a[1]=3.4
a[2]=3.4
a[3]=3.4
a[4]=3.4
[       OK ] testCLMathWrapper.assignScalar (78 ms)
[ RUN      ] testCLMathWrapper.addinplace
a[0]=5
a[1]=5.1
a[2]=14
a[3]=15.5
a[4]=11.7
[       OK ] testCLMathWrapper.addinplace (78 ms)
[ RUN      ] testCLMathWrapper.multiplyinplace
a[0]=1.5
a[1]=4.5
a[2]=13.5
a[3]=18.75
a[4]=3.75
[       OK ] testCLMathWrapper.multiplyinplace (78 ms)
[ RUN      ] testCLMathWrapper.addscalar
a[0]=2.5
a[1]=4.5
a[2]=10.5
a[3]=14
a[4]=4
[       OK ] testCLMathWrapper.addscalar (78 ms)
[ RUN      ] testCLMathWrapper.sqrt
a[0]=1
a[1]=1.73205
a[2]=3
a[3]=3.53553
a[4]=1.58114
[       OK ] testCLMathWrapper.sqrt (78 ms)
[ RUN      ] testCLMathWrapper.squared
a[0]=1
a[1]=9
a[2]=81
a[3]=156.25
a[4]=6.25
[       OK ] testCLMathWrapper.squared (78 ms)
[ RUN      ] testCLMathWrapper.inverse
a[0]=1
a[1]=0.333333
a[2]=0.111111
a[3]=0.08
a[4]=0.4
[       OK ] testCLMathWrapper.inverse (78 ms)
[ RUN      ] testCLMathWrapper.perelementmult
a[0]=4
a[1]=6.3
a[2]=45
a[3]=37.5
a[4]=23
[       OK ] testCLMathWrapper.perelementmult (78 ms)
[----------] 9 tests from testCLMathWrapper (702 ms total)

[----------] 1 test from testreducesegments
[ RUN      ] testreducesegments.basic
Using Advanced Micro Devices, Inc. , OpenCL platform: AMD Accelerated Parallel Processing
Using OpenCL device: Tahiti
[       OK ] testreducesegments.basic (78 ms)
[----------] 1 test from testreducesegments (78 ms total)

[----------] 4 tests from testGpuOp
[ RUN      ] testGpuOp.addinplace
a[0]=5
a[1]=5.1
a[2]=14
a[3]=15.5
a[4]=11.7
[       OK ] testGpuOp.addinplace (78 ms)
[ RUN      ] testGpuOp.addoutofplace
a[0]=1
a[1]=3
a[2]=9
a[3]=12.5
a[4]=2.5
c[0]=5
c[1]=5.1
c[2]=14
c[3]=15.5
c[4]=11.7
[       OK ] testGpuOp.addoutofplace (94 ms)
[ RUN      ] testGpuOp.inverse
a[0]=1
a[1]=0.333333
a[2]=0.111111
a[3]=0.08
a[4]=0.4
[       OK ] testGpuOp.inverse (78 ms)
[ RUN      ] testGpuOp.addscalarinplace
a[0]=5.2
a[1]=7.2
a[2]=13.2
a[3]=16.7
a[4]=6.7
[       OK ] testGpuOp.addscalarinplace (78 ms)
[----------] 4 tests from testGpuOp (328 ms total)

[----------] 1 test from testjpeghelper
[ RUN      ] testjpeghelper.writeread