I think I solved this issue. It popped up when I updated to CUDA11. I
had to patch up
three functions dnn_fwd.c, dnn_gi.c, dnn_gw.c. I added the config
option gcc.cxxflags=-DCUDAVERSION=11, then added checks in the
code. It's can use cleaning up. If someone wants, they can take this fix
and check it in to the repo.
cheers, Paul
In dnn_fwd.c, I put in this check:
#if CUDAVERSION > 10
// PMB
#define CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT 2
int ret_alg_cnt;
cudnnConvolutionFwdAlgoPerf_t ta;
err = cudnnGetConvolutionForwardAlgorithm_v7(
params->handle, APPLY_SPECIFIC(input), APPLY_SPECIFIC(kerns),
desc, APPLY_SPECIFIC(output),
CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT, &ret_alg_cnt,
&ta);
algo=ta.algo;
#else
err = cudnnGetConvolutionForwardAlgorithm(
params->handle, APPLY_SPECIFIC(input), APPLY_SPECIFIC(kerns),
desc, APPLY_SPECIFIC(output),
CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT, maxfree, &algo);
#endif
In dnn_gi.c, I put in this check:
#if CUDAVERSION > 10
int ret_alg_cnt;
cudnnConvolutionBwdDataAlgoPerf_t at;
#define CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT 2
err = cudnnGetConvolutionBackwardDataAlgorithm_v7(
params->handle, APPLY_SPECIFIC(kerns), APPLY_SPECIFIC(output),
desc, APPLY_SPECIFIC(input),
CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT,&ret_alg_cnt, &at);
algo=at.algo;
#else
err = cudnnGetConvolutionBackwardDataAlgorithm(
params->handle, APPLY_SPECIFIC(kerns), APPLY_SPECIFIC(output),
desc, APPLY_SPECIFIC(input),
CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT, maxfree,
&algo);
#endif
In dnn_gw.c, I put in this check:
#if CUDAVERSION > 10
// PMB
#define CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT 2
cudnnConvolutionBwdFilterAlgoPerf_t ta;
int alg_cnt;
err = cudnnGetConvolutionBackwardFilterAlgorithm_v7(
params->handle, APPLY_SPECIFIC(input), APPLY_SPECIFIC(output),
desc, APPLY_SPECIFIC(kerns),
CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT,
&alg_cnt, &ta);
algo=ta.algo;
#else
err = cudnnGetConvolutionBackwardFilterAlgorithm(
params->handle, APPLY_SPECIFIC(input), APPLY_SPECIFIC(output),
desc, APPLY_SPECIFIC(kerns),
CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT,
maxfree, &algo);
#endif
On Thursday, September 1, 2022 at 8:37:48 AM UTC+2 [email protected] wrote:
>
>
>
>
> *My installed packages are:CUDA 11.7, cudnn 8.5, theano 1.0.5When I run
> the convolution network in the test file of Michael Nielsen's tutorial on
> Deep Learning
> <https://github.com/MichalDanielDobrzanski/DeepLearningPython>:*
> import network3
> from network3 import Network, ConvPoolLayer, FullyConnectedLayer,
> SoftmaxLayer
> training_data, validation_data, test_data = network3.load_data_shared()
> mini_batch_size = 10
> net = Network([
> ConvPoolLayer(image_shape=(mini_batch_size, 1, 28,
> 28),filter_shape=(20, 1, 5, 5), poolsize=(2,
> 2)),FullyConnectedLayer(n_in=20*12*12, n_out=100),SoftmaxLayer(n_in=100,
> n_out=10)], mini_batch_size)
> net.SGD(training_data, 60, mini_batch_size, 0.1,
> validation_data, test_data)
>
> * Problem occurred during compilation with the command line below:*
> dnn_fwd.c:326:60: error: invalid conversion from 'size_t {aka long
> unsigned int}' to 'int*' [-fpermissive]
> dnn_fwd.c:326:60: error: cannot convert 'cudnnConvolutionFwdAlgo_t*'
> to 'cudnnConvolutionFwdAlgoPerf_t* {aka
> cudnnConvolutionFwdAlgoPerfStruct*}' for argument '8' to 'cudnnStatus_t
> cudnnGetConvolutionForwardAlgorithm_v7(cudnnHandle_t,
> cudnnTensorDescriptor_t, cudnnFilterDescriptor_t,
> cudnnConvolutionDescriptor_t, cudnnTensorDescriptor_t, int, int*,
> cudnnConvolutionFwdAlgoPerf_t*)'
>
> *I understand that incompatibility of packages should be the reason,
> however I was wondering if there is any way to resolve this by modify the
> module or the theano's config file without downgrading the packages?*
>
>
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/theano-users/4221f407-2af1-4c46-86bf-6384c993c69bn%40googlegroups.com.
