Merge pull request #23560 from WanliZhong:eltwise_cuda_bug

asmorkalov · web-flow · commit 59ca444b26cb · 2023-05-16T14:16:37.000+03:00
DNN/CUDA: Solve the bug of same shape broadcast with CUDA
diff --git a/modules/dnn/src/cuda/eltwise_ops.cu b/modules/dnn/src/cuda/eltwise_ops.cu
@@ -150,7 +150,7 @@ void eltwise_op(const Stream& stream, TensorSpan<T> output, TensorView<T> x, Ten
          */
         for (int r = 0; r < output.rank(); r++)
         {
-            while (x.get_axis_size(r) == 1 && y.get_axis_size(r) == 1) {
+            while (x.rank() > r && y.rank() > r && x.get_axis_size(r) == 1 && y.get_axis_size(r) == 1) {
                 CV_Assert(output.get_axis_size(r) == 1);
 
                 x.squeeze(r);
@@ -183,6 +183,9 @@ void eltwise_op(const Stream& stream, TensorSpan<T> output, TensorView<T> x, Ten
                     auto new_size = inShape1[i] * inShape1[j];
                     inShape1[i] = new_size;
                     inShape2[i] = new_size;
+                    // outShape should be changed after merged
+                    auto output_new_size = outShape[i] * outShape[j];
+                    outShape[i] = output_new_size;
 
                     /* delete axis `j` */
                     inShape1.erase(std::begin(inShape1) + j);
diff --git a/modules/dnn/src/layers/nary_eltwise_layers.cpp b/modules/dnn/src/layers/nary_eltwise_layers.cpp
@@ -51,8 +51,6 @@ class NaryEltwiseLayerImpl CV_FINAL : public NaryEltwiseLayer
         WHERE,
     } op;
 
-    // If the eltwise implementation is modified, you need to force enable the 'Layer_Test_Eltwise_bcast'
-    // test in the 'test_layers.cpp' file to make sure it all passes
     NaryEltwiseLayerImpl(const LayerParams& params)
     {
         setParamsFrom(params);
diff --git a/modules/dnn/test/test_layers.cpp b/modules/dnn/test/test_layers.cpp
@@ -1949,6 +1949,9 @@ struct Layer_Test_Eltwise_bcast : testing::TestWithParam<tuple<string, int, tupl
         int backend = get<0>(backend_target);
         int target = get<1>(backend_target);
 
+        if (backend == DNN_BACKEND_CUDA && dim > 4)
+            applyTestTag(CV_TEST_TAG_LONG);
+
         vector<vector<int>> dim_shape_list;
         get_all_arr(dim_shape_list, dim);
         replace(dim_shape_list, 1, 3);
@@ -2028,7 +2031,7 @@ struct Layer_Test_Eltwise_bcast : testing::TestWithParam<tuple<string, int, tupl
     }
 };
 
-TEST_P(Layer_Test_Eltwise_bcast, DISABLED_brute_force)
+TEST_P(Layer_Test_Eltwise_bcast, brute_force)
 {
     test_bcast();
 }

Original file line number	Diff line number	Diff line change
`@@ -51,8 +51,6 @@ class NaryEltwiseLayerImpl CV_FINAL : public NaryEltwiseLayer`
`51`	`51`	`WHERE,`
`52`	`52`	`} op;`
`53`	`53`
`54`		`- // If the eltwise implementation is modified, you need to force enable the 'Layer_Test_Eltwise_bcast'`
`55`		`- // test in the 'test_layers.cpp' file to make sure it all passes`
`56`	`54`	`NaryEltwiseLayerImpl(const LayerParams& params)`
`57`	`55`	`{`
`58`	`56`	`setParamsFrom(params);`
Original file line number	Diff line number	Diff line change
`@@ -1949,6 +1949,9 @@ struct Layer_Test_Eltwise_bcast : testing::TestWithParam<tuple<string, int, tupl`
`1949`	`1949`	`int backend = get<0>(backend_target);`
`1950`	`1950`	`int target = get<1>(backend_target);`
`1951`	`1951`
	`1952`	`+ if (backend == DNN_BACKEND_CUDA && dim > 4)`
	`1953`	`+ applyTestTag(CV_TEST_TAG_LONG);`
	`1954`	`+`
`1952`	`1955`	`vector<vector<int>> dim_shape_list;`
`1953`	`1956`	`get_all_arr(dim_shape_list, dim);`
`1954`	`1957`	`replace(dim_shape_list, 1, 3);`
`@@ -2028,7 +2031,7 @@ struct Layer_Test_Eltwise_bcast : testing::TestWithParam<tuple<string, int, tupl`
`2028`	`2031`	`}`
`2029`	`2032`	`};`
`2030`	`2033`
`2031`		`-TEST_P(Layer_Test_Eltwise_bcast, DISABLED_brute_force)`
	`2034`	`+TEST_P(Layer_Test_Eltwise_bcast, brute_force)`
`2032`	`2035`	`{`
`2033`	`2036`	`test_bcast();`
`2034`	`2037`	`}`