connorgoggins opened a new issue #17493: im2col fails in GPU context regardless 
of input
URL: https://github.com/apache/incubator-mxnet/issues/17493
 
 
   ## Description
   Calling the im2col operator within a GPU context results in a dimension 
mismatch error - even if the shapes used are entirely valid. I encountered this 
error while testing the operator on CPU and GPU machines, and found that while 
it passed on CPU, it failed on GPU with the exact same input in every case.
   
   ### Error Message
   ```mxnet.base.MXNetError: Traceback (most recent call last):
     File "include/mxnet/./tuple.h", line 558
   MXNetError: Check failed: dim == ndim() (3 vs. 4) : dimension do not match 
target dimension 3 vs 4```
   
   ### Steps to reproduce
   Open a Python shell and run the following:
   
   1. `import mxnet as mx`
   2. `run_performance_test(mx.nd.im2col, inputs=[{'data':(32,3,256,256), 
'kernel':(3,)}], ctx=mx.gpu())`
   
   Feel free to replace data/kernel with any other valid shape - the error will 
remain the same. Note that if `ctx` is switched to `mx.cpu()` the error 
disappears.
   
   ## What have you tried to solve it?
   
   1. Tried adjusting the shape of `data`, shape of `kernel`, and all other 
parameters. The error occurred regardless of what shapes I used.
   2. Tried running the op without `run_backwards` enabled, but the error still 
occurred.
   
   ## Environment
   Using Ubuntu 16.04, DL AMI v26
   
   ✔ CUDA, ✔ CUDNN, ✖ NCCL, ✖ CUDA_RTC, ✖ TENSORRT, ✔ CPU_SSE, ✔ CPU_SSE2, ✔ 
CPU_SSE3, ✔ CPU_SSE4_1, ✔ CPU_SSE4_2, ✖ CPU_SSE4A, ✔ CPU_AVX, ✖ CPU_AVX2, ✔ 
OPENMP, ✖ SSE, ✔ F16C, ✖ JEMALLOC, ✔ BLAS_OPEN, ✖ BLAS_ATLAS, ✖ BLAS_MKL, ✖ 
BLAS_APPLE, ✔ LAPACK, ✖ MKLDNN, ✖ OPENCV, ✖ CAFFE, ✖ PROFILER, ✖ DIST_KVSTORE, 
✖ CXX14, ✔ INT64_TENSOR_SIZE, ✖ SIGNAL_HANDLER, ✖ DEBUG, ✖ TVM_OP
   

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to