Installation errors with CUDNN v5

[iNeil77]

I have cuda v7.5 and cudnn v5 installed and upon having pre-installed the necessary dependencies through apt-get and pip, I proceeded to make all in my caffe master directory with my makefile config having USE_CUDNN := 1 set, which led to the following error. Can anybody tell me where am I going wrong?
Thanks in advance!

neilpaul77@NeilRig77:~/Documents/Git/caffe$ make all
CXX src/caffe/util/im2col.cpp
In file included from ./include/caffe/util/device_alternate.hpp:40:0,
                 from ./include/caffe/common.hpp:19,
                 from ./include/caffe/util/math_functions.hpp:9,
                 from src/caffe/util/im2col.cpp:4:
./include/caffe/util/cudnn.hpp: In function ‘void caffe::cudnn::createPoolingDesc(cudnnPoolingStruct**, caffe::PoolingParameter_PoolMethod, cudnnPoolingMode_t*, int, int, int, int, int, int)’:
./include/caffe/util/cudnn.hpp:127:41: error: too few arguments to function ‘cudnnStatus_t cudnnSetPooling2dDescriptor(cudnnPoolingDescriptor_t, cudnnPoolingMode_t, cudnnNanPropagation_t, int, int, int, int, int, int)’
         pad_h, pad_w, stride_h, stride_w));
                                         ^
./include/caffe/util/cudnn.hpp:15:28: note: in definition of macro ‘CUDNN_CHECK’
     cudnnStatus_t status = condition; \
                            ^
In file included from ./include/caffe/util/cudnn.hpp:5:0,
                 from ./include/caffe/util/device_alternate.hpp:40,
                 from ./include/caffe/common.hpp:19,
                 from ./include/caffe/util/math_functions.hpp:9,
                 from src/caffe/util/im2col.cpp:4:
/usr/local/cuda/include/cudnn.h:799:27: note: declared here
 cudnnStatus_t CUDNNWINAPI cudnnSetPooling2dDescriptor(
                           ^
Makefile:572: recipe for target '.build_release/src/caffe/util/im2col.o' failed
make: *** [.build_release/src/caffe/util/im2col.o] Error 1

[seanbell]

I'm pretty sure that cuDNN v5 isn't supported yet. I think caffe is still on v3. cuDNN v4 is the current supported version.

See #3919

[naibaf7]

@iNeil77
cuDNN V4/V5 is supported (unofficially though) on https://github.com/BVLC/caffe/tree/opencl
Otherwise as @seanbell pointed out, you have to go to V3 or wait for #3919 to be merged (or make a fork and merge that PR in yourself).

[flx42]

cuDNN v4 works fine with the master branch, since December. I don't know why people keep saying it doesn't :)
It's just that there is no cuDNN BN layer, but you will benefit from all the performance improvements from v4.

[samchorlton]

iNeil77 did you manage to fix this in the end? If so how as I have the exact same problem? (I have tried version 3,4 and 5

[iNeil77]

@samchorlton No I needed it setup quickly, so I just proceeded to install with v3

[flx42]

@samchorlton @iNeil77 again, v4 works fine too. If that's not the case you have a problem with your setup.

[iNeil77]

Didnt try v4, just stuck to what everyone says works

[shelhamer]

To reiterate, cuDNN v4 works perfectly fine as of #3439 in Dec. 2015 as @flx42 noted.

[atticcas]

Since I am using bleeding edge Theano, they force me to upgrade to v5, any way to make caffe to work with v5?

[flx42]

@atticcas you will need this PR: #3919

[atticcas]

MANY THANKS!
wonder why this hasn't been pulled to master yet.
Lets see what happens after I pull those three commits.

[ghost]

I met the same error

[youngwanLEE]

I also met this same error. What should I do? eventually, I have to re-install ubuntu14.04 and start at the beginning?? :(

[ghost]

it now support cudnnv4 not v5

On Mon, May 2, 2016 at 12:36 PM youngwanLEE [email protected]
wrote:

I also this same error. What should I do? eventually, I have to re-install
ubuntu14.04 and start at the beginning?? :(

—
You are receiving this because you commented.
Reply to this email directly or view it on GitHub
#3969 (comment)

[ghost]

use cudnn v4

On Mon, May 2, 2016 at 12:36 PM youngwanLEE [email protected]
wrote:

I also this same error. What should I do? eventually, I have to re-install
ubuntu14.04 and start at the beginning?? :(

—
You are receiving this because you commented.
Reply to this email directly or view it on GitHub
#3969 (comment)

[Clayton-Davis]

just using V4 worked for me

[flx42]

@Clayton-Davis v5 is supported now actually

[ghost]

@flx42 Really? I will have a try now. Thank you for the inform.

[oztc]

@RuiLiu1217 v5 can work ?

[oztc]

@flx42 when I run v5, it reports the error:
-- Performing Test BLAS_F2C_DOUBLE_WORKS - Failed

-- Performing Test BLAS_F2C_FLOAT_WORKS

after I change to v4, make runtest can go through properly.

[ekcheng]

Can someone point me to the commit(s) that made cuDNN v5 work?

[lukeyeager]

@ekcheng: #4159

[vguptai]

@ekcheng

You can try this fork and let me know if it suffices.

https://github.com/vgupta-ai/py-faster-rcnn