add_tensor4d throws error

[JZZQuant]

include("cifar10.jl") WARNING: Method definition info(Any...) in module Base at util.jl:334 overwritten in module Logging at C:\Users\sragh.julia\v0.4\Logging\src\Logging.jl:61. WARNING: Method definition warn(Any...) in module Base at util.jl:364 overwritten in module Logging at C:\Users\sragh.julia\v0.4\Logging\src\Logging.jl:61. Configuring Mocha...

• CUDA enabled [DEV=0](MOCHA_USE_CUDA environment variable detected)
• Native Ext disabled by default Mocha configured, continue loading module... DefaultBackend = Mocha.GPUBackend 02-Jan 22:24:06:INFO:root:Initializing CuDNN backend... 02-Jan 22:24:08:INFO:root:CuDNN backend initialized! 02-Jan 22:24:09:INFO:root:Constructing net CIFAR10-train on Mocha.GPUBackend... 02-Jan 22:24:09:INFO:root:Topological sorting 11 layers... 02-Jan 22:24:09:INFO:root:Setup layers... 02-Jan 22:24:13:INFO:root:Network constructed! 02-Jan 22:24:13:INFO:root:Constructing net CIFAR10-test on Mocha.GPUBackend... 02-Jan 22:24:13:INFO:root:Topological sorting 11 layers... 02-Jan 22:24:13:INFO:root:Setup layers... 02-Jan 22:24:13:DEBUG:root:ConvolutionLayer(conv1): sharing filters and bias 02-Jan 22:24:13:DEBUG:root:ConvolutionLayer(conv2): sharing filters and bias 02-Jan 22:24:13:DEBUG:root:ConvolutionLayer(conv3): sharing filters and bias 02-Jan 22:24:13:DEBUG:root:InnerProductLayer(ip1): sharing weights and bias 02-Jan 22:24:14:INFO:root:Network constructed! 02-Jan 22:24:16:DEBUG:root:#DEBUG Checking network topology for back-propagation 02-Jan 22:24:16:DEBUG:root:Init network CIFAR10-train 02-Jan 22:24:16:DEBUG:root:Init parameter filter for layer conv1 02-Jan 22:24:16:DEBUG:root:Init parameter bias for layer conv1 02-Jan 22:24:16:DEBUG:root:Init parameter filter for layer conv2 02-Jan 22:24:16:DEBUG:root:Init parameter bias for layer conv2 02-Jan 22:24:16:DEBUG:root:Init parameter filter for layer conv3 02-Jan 22:24:16:DEBUG:root:Init parameter bias for layer conv3 02-Jan 22:24:16:DEBUG:root:Init parameter weight for layer ip1 02-Jan 22:24:16:DEBUG:root:Init parameter bias for layer ip1 WARNING: convert(::Type{Ptr}, ::Int64) methods should be converted to be methods of unsafe_convert in depwarn at deprecated.jl:73 [inlined code] from deprecated.jl:418 in unsafe_convert at no file:0 [inlined code] from C:\Users\sragh.julia\v0.4\Mocha\src\cuda/cudnn.jl:51 in convolution_forward at C:\Users\sragh.julia\v0.4\Mocha\src\cuda/cudnn.jl:323 in forward at C:\Users\sragh.julia\v0.4\Mocha\src\cuda\layers/convolution.jl:83 in forward at C:\Users\sragh.julia\v0.4\Mocha\src\net.jl:148 in init_solve at C:\Users\sragh.julia\v0.4\Mocha\src\solvers.jl:189 in solve at C:\Users\sragh.julia\v0.4\Mocha\src\solvers.jl:234 in include at boot.jl:261 in include_from_node1 at loading.jl:304 while loading C:\Users\sragh.julia\v0.4\Mocha\examples\cifar10\cifar10.jl, in expression starting on line 62 ERROR: LoadError: Not supported [inlined code] from C:\Users\sragh.julia\v0.4\Mocha\src\cuda/cudnn.jl:53 in add_tensor4d at C:\Users\sragh.julia\v0.4\Mocha\src\cuda/cudnn.jl:172 in forward at C:\Users\sragh.julia\v0.4\Mocha\src\cuda\layers/convolution.jl:89 in forward at C:\Users\sragh.julia\v0.4\Mocha\src\net.jl:148 in init_solve at C:\Users\sragh.julia\v0.4\Mocha\src\solvers.jl:189 in solve at C:\Users\sragh.julia\v0.4\Mocha\src\solvers.jl:234 in include at boot.jl:261 in include_from_node1 at loading.jl:304 while loading C:\Users\sragh.julia\v0.4\Mocha\examples\cifar10\cifar10.jl, in expression starting on line 62

@assert CUDNN_ADD_IMAGE <= mode <= CUDNN_ADD_FULL_TENSOR line throws an error

my configuration : cudnn64_4.dll and cuda 7.5 on windows 10 (could reproduce both on release build and also the bleeding edge)

[pluskid]

It seems cudnn v4 introduced some new algorithms that were not presented before. I have not gotten a chance to test cudnn v4 yet. But you could try to comment out that assertion line and see if it works.

[pluskid]

This should be fixed now in a recent commit for fixing cuDNN v4 API changing.