[GitHub] anirudh2290 closed pull request #13433: Fix deconvolution / PR 13421

[GitBox <gi...@apache.org>]

anirudh2290 closed pull request #13433: Fix deconvolution  / PR 13421
URL: https://github.com/apache/incubator-mxnet/pull/13433
 
 
   

This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:

As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):

diff --git a/src/operator/nn/mkldnn/mkldnn_deconvolution.cc b/src/operator/nn/mkldnn/mkldnn_deconvolution.cc
index a6d6b24235c..4da48fa3f83 100644
--- a/src/operator/nn/mkldnn/mkldnn_deconvolution.cc
+++ b/src/operator/nn/mkldnn/mkldnn_deconvolution.cc
@@ -262,12 +262,14 @@ void MKLDNNDeconvForward::SetDataHandle(const DeconvolutionParam& param,
     // For inference, we want to reorder the weight array so we don't need to
     // reorder data every time.
     if (weight.IsDefaultData()) {
+      weight_mem = GetWeights(weight, fwd_pd.weights_primitive_desc(), param.num_group);
       // We also need to modify the layout on the original weight array. The
       // data conversion happens after the weight array is used.
       const_cast<NDArray&>(weight).MKLDNNDataReorderAsync(fwd_pd.weights_primitive_desc());
+    } else {
+      weight_mem = weight.GetMKLDNNData();
+      CHECK(weight_mem->get_primitive_desc() == fwd_pd.weights_primitive_desc());
     }
-    weight_mem = weight.GetMKLDNNData();
-    CHECK(weight_mem->get_primitive_desc() == fwd_pd.weights_primitive_desc());
   }
   auto out_mem = CreateMKLDNNMem(out_data[deconv::kOut],
       fwd_pd.diff_src_primitive_desc(), req[deconv::kOut]);
diff --git a/tests/python/gpu/test_operator_gpu.py b/tests/python/gpu/test_operator_gpu.py
index 8054937a84c..7a7c6f69dd7 100644
--- a/tests/python/gpu/test_operator_gpu.py
+++ b/tests/python/gpu/test_operator_gpu.py
@@ -683,11 +683,11 @@ def test_deconvolution_options():
     check_consistency_NxM([sym, sym_no_cudnn], ctx_list)
 
     # 2D deconvolution
-    ctx_list = [{'ctx': mx.gpu(0), 'deconv_data': (2, 2, 10, 10), 'type_dict': {'deconv_data': np.float64}},
-                {'ctx': mx.gpu(0), 'deconv_data': (2, 2, 10, 10), 'type_dict': {'deconv_data': np.float32}},
-                {'ctx': mx.gpu(0), 'deconv_data': (2, 2, 10, 10), 'type_dict': {'deconv_data': np.float16}},
-                {'ctx': mx.cpu(0), 'deconv_data': (2, 2, 10, 10), 'type_dict': {'deconv_data': np.float64}},
-                {'ctx': mx.cpu(0), 'deconv_data': (2, 2, 10, 10), 'type_dict': {'deconv_data': np.float32}}]
+    ctx_list = [{'ctx': mx.gpu(0), 'deconv_data': (2, 8, 10, 10), 'type_dict': {'deconv_data': np.float64}},
+                {'ctx': mx.gpu(0), 'deconv_data': (2, 8, 10, 10), 'type_dict': {'deconv_data': np.float32}},
+                {'ctx': mx.gpu(0), 'deconv_data': (2, 8, 10, 10), 'type_dict': {'deconv_data': np.float16}},
+                {'ctx': mx.cpu(0), 'deconv_data': (2, 8, 10, 10), 'type_dict': {'deconv_data': np.float64}},
+                {'ctx': mx.cpu(0), 'deconv_data': (2, 8, 10, 10), 'type_dict': {'deconv_data': np.float32}}]
     # Pad > 0
     sym = mx.sym.Deconvolution(num_filter=2, kernel=(3,3), pad=(1,1), name='deconv')
     sym_no_cudnn = mx.sym.Deconvolution(num_filter=2, kernel=(3,3), pad=(1,1), cudnn_off=True, name='deconv')


 

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


With regards,
Apache Git Services