cunn/SpatialFullConvolution.cu(80): error: identifier "THCudaBlas_gemm" is undefined

75 views
Skip to first unread message

Raghuram Vadapalli

unread,
Oct 22, 2016, 4:33:36 PM10/22/16
to torch7
Hi,
I am trying to run LSTM with attention written for SNLI data i.e https://github.com/cheng6076/SNLI-attention . I am required to checkout to the mentioned commits in nn and cunn as said in README of the mentioned link (
a4b42a74cb10c52f7916e31e28ebb604fc32fbf1
and 
3f5a8ba2bd4e6babf112d6369c98f37be86d2391
respectively
). nn compiled perfectly but I am getting this error - 
cunn/SpatialFullConvolution.cu(80): error: identifier "THCudaBlas_gemm" is undefined
while installing cunn.

soumith

unread,
Oct 24, 2016, 11:11:20 AM10/24/16
to torch7 on behalf of Raghuram Vadapalli
You prob also need to get cutorch around that commit date

--
You received this message because you are subscribed to the Google Groups "torch7" group.
To unsubscribe from this group and stop receiving emails from it, send an email to torch7+unsubscribe@googlegroups.com.
To post to this group, send email to tor...@googlegroups.com.
Visit this group at https://groups.google.com/group/torch7.
For more options, visit https://groups.google.com/d/optout.

Raghuram Vadapalli

unread,
Oct 30, 2016, 1:48:20 AM10/30/16
to torch7
Can you please tell the exact commit id. Saves me a lot of time. Thanks :)


On Monday, October 24, 2016 at 8:41:20 PM UTC+5:30, smth chntla wrote:
You prob also need to get cutorch around that commit date
On Sat, Oct 22, 2016 at 4:33 PM, Raghuram Vadapalli via torch7 <torch7+APn2wQc5ybJv0jr2MyxdTiZhcSp0jTpmDsEyqgTykPSfPqNX5QQ9Tox4r@googlegroups.com> wrote:
Hi,
I am trying to run LSTM with attention written for SNLI data i.e https://github.com/cheng6076/SNLI-attention . I am required to checkout to the mentioned commits in nn and cunn as said in README of the mentioned link (
a4b42a74cb10c52f7916e31e28ebb604fc32fbf1
and 
3f5a8ba2bd4e6babf112d6369c98f37be86d2391
respectively
). nn compiled perfectly but I am getting this error - 
cunn/SpatialFullConvolution.cu(80): error: identifier "THCudaBlas_gemm" is undefined
while installing cunn.

--
You received this message because you are subscribed to the Google Groups "torch7" group.
To unsubscribe from this group and stop receiving emails from it, send an email to torch7+un...@googlegroups.com.

Raghuram Vadapalli

unread,
Oct 30, 2016, 2:04:54 AM10/30/16
to torch7
As you said, I checked out to commit e778c19a4 which is on Dec 19 (before the mentioned cunn commit on Dec25). But cutorch doesn't compile at that commit. I get error error: identifier "THInf" is undefined.

soumith

unread,
Oct 30, 2016, 10:30:45 PM10/30/16
to torch7 on behalf of Raghuram Vadapalli
Well, in that case cutorch depends on a torch commit as well.
Here's my recommendation.
Erase your torch, and do this:


cd ~/torch
git reset --hard 99081d95148631c581891d74e626d09078895a93 # commit hash from Dec 19th
./install.sh

On Sun, Oct 30, 2016 at 2:04 AM, Raghuram Vadapalli via torch7 <torch7+APn2wQc5ybJv0jr2MyxdTiZhc...@googlegroups.com> wrote:
As you said, I checked out to commit e778c19a4 which is on Dec 19 (before the mentioned cunn commit on Dec25). But cutorch doesn't compile at that commit. I get error error: identifier "THInf" is undefined.

--
You received this message because you are subscribed to the Google Groups "torch7" group.
To unsubscribe from this group and stop receiving emails from it, send an email to torch7+unsubscribe@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages