Merge pull request #21884 from rogday:cuda_cleanup

Fix CUDA compilation issues and adjust thresholds.

* Fix CUDA compilation issues and adjust thresholds.

* add conformance tests to denylist

modules/dnn/src/cuda/activations.cu

@@ -260,7 +260,7 @@ void shrink(const Stream& stream, Span<T> output, View<T> input, T bias, T lambd
 
 template <class T>
 void reciprocal(const Stream& stream, Span<T> output, View<T> input) {
-    generic_op<T, SignFunctor<T>>(stream, output, input);
+    generic_op<T, ReciprocalFunctor<T>>(stream, output, input);
 }
 
 template <class T>

modules/dnn/src/cuda/functors.hpp

@@ -732,7 +732,8 @@ struct SignFunctor {
         CUDA4DNN_HOST_DEVICE Params() {}
     };
 
-    CUDA4DNN_DEVICE SignFunctor() : SignFunctor(Params{}) { }
+    CUDA4DNN_DEVICE SignFunctor() { }
+    CUDA4DNN_DEVICE SignFunctor(const Params& params) { }
 
     CUDA4DNN_DEVICE T operator()(T value) {
         return value > T(0) ? T(1) : (value < T(0) ? T(-1) : T(0));
@@ -747,7 +748,7 @@ struct ShrinkFunctor {
         T bias, lambd;
     };
 
-    CUDA4DNN_DEVICE ShrinkFunctor() : bias(0), lambd(0.5) { }
+    CUDA4DNN_DEVICE ShrinkFunctor() : ShrinkFunctor(Params{}) { }
     CUDA4DNN_DEVICE ShrinkFunctor(const Params& params) : bias{params.bias}, lambd{params.lambd} { }
 
     CUDA4DNN_DEVICE T operator()(T value) {
@@ -763,10 +764,11 @@ struct ReciprocalFunctor {
         CUDA4DNN_HOST_DEVICE Params() {}
     };
 
-    CUDA4DNN_DEVICE ReciprocalFunctor() : ReciprocalFunctor(Params{}) { }
+    CUDA4DNN_DEVICE ReciprocalFunctor() { }
+    CUDA4DNN_DEVICE ReciprocalFunctor(const Params& params) { }
 
     CUDA4DNN_DEVICE T operator()(T value) {
-        return T(1.0f)/value;
+        return T(1.f)/value;
     }
 };
 

modules/dnn/src/layers/convolution_layer.cpp

@@ -2080,6 +2080,7 @@ public:
     {
         auto context = reinterpret_cast<csl::CSLContext*>(context_);
 
+        // TODO: extract bias from inputs and pass it
         CV_Assert(inputs.size() == 1 || inputs.size() == 2);
         auto input_wrapper = inputs[0].dynamicCast<CUDABackendWrapper>();
         auto input_shape = input_wrapper->getShape();

modules/dnn/src/layers/elementwise_layers.cpp

@@ -2282,7 +2282,7 @@ struct SignFunctor : public BaseDefaultFunctor<SignFunctor>
 
     inline float calculate(float x) const
     {
-        return x > 0 ? 1 : (x < 0 ? -1 : 0);
+        return x > 0.f ? 1.f : (x < 0.f ? -1.f : 0.f);
     }
 
 #ifdef HAVE_CUDA
@@ -2315,13 +2315,13 @@ struct ShrinkFunctor : public BaseDefaultFunctor<ShrinkFunctor>
 
     inline float calculate(float x) const
     {
-        return x > lambd ? x - bias : (x < -lambd ? x + bias : 0);
+        return x > lambd ? x - bias : (x < -lambd ? x + bias : 0.f);
     }
 
 #ifdef HAVE_CUDA
     Ptr<BackendNode> initCUDA(int target, csl::Stream stream)
     {
-        return make_cuda_node<cuda4dnn::ShrinkOp>(target, stream);
+        return make_cuda_node<cuda4dnn::ShrinkOp>(target, stream, bias, lambd);
     }
 #endif
 
@@ -2343,7 +2343,7 @@ struct ReciprocalFunctor : public BaseDefaultFunctor<ReciprocalFunctor>
 
     inline float calculate(float x) const
     {
-        return 1.0/x;
+        return 1.f/x;
     }
 
 #ifdef HAVE_CUDA

modules/dnn/src/opencl/activations.cl

@@ -320,7 +320,7 @@ __kernel void SignForward(const int n, __global T* in, __global T* out)
 {
     int index = get_global_id(0);
     if(index < n)
-        out[index] = in[index] > 0.f ? 1.0f : (in[index] < 0.f) ? -1.0f : 0.0f);
+        out[index] = in[index] > 0.f ? 1.0f : ((in[index] < 0.f) ? -1.0f : 0.0f);
 }
 
 __kernel void ReciprocalForward(const int n, __global T* in, __global T* out)

modules/dnn/test/test_backends.cpp

@@ -516,7 +516,7 @@ TEST_P(DNNTestNetwork, DenseNet_121)
     else if (target == DNN_TARGET_CUDA_FP16)
     {
         l1 = 0.008;
-        lInf = 0.05;
+        lInf = 0.06;
     }
     processNet("dnn/DenseNet_121.caffemodel", "dnn/DenseNet_121.prototxt", Size(224, 224), "", "", l1, lInf);
     if (target != DNN_TARGET_MYRIAD || getInferenceEngineVPUType() != CV_DNN_INFERENCE_ENGINE_VPU_TYPE_MYRIAD_X)

modules/dnn/test/test_misc.cpp

@@ -844,8 +844,9 @@ TEST_P(Test_two_inputs, basic)
     Mat ref;
     addWeighted(firstInp, kScale, secondInp, kScaleInv, 0, ref, CV_32F);
 
-    double l1 = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 0.06 : 1e-6;
-    double lInf = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 0.3 : 1e-5;
+    double l1 = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD || targetId == DNN_TARGET_CUDA_FP16) ? 0.06 : 1e-6;
+    double lInf = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD || targetId == DNN_TARGET_CUDA_FP16) ? 0.3 : 1e-5;
+
     normAssert(out, ref, "", l1, lInf);
 
     if (cvtest::debugLevel > 0 || HasFailure())

modules/dnn/test/test_model.cpp

@@ -512,7 +512,7 @@ TEST_P(Test_Model, DetectionMobilenetSSD)
     }
     else if (target == DNN_TARGET_CUDA_FP16)
     {
-        scoreDiff = 0.002;
+        scoreDiff = 0.0021;
         iouDiff = 1e-2;
     }
     float confThreshold = FLT_MIN;
@@ -661,7 +661,8 @@ TEST_P(Test_Model, Segmentation)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH, CV_TEST_TAG_DNN_SKIP_IE_VERSION);
 #endif
 
-    if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
+    if ((backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
+        || (backend == DNN_BACKEND_CUDA && target == DNN_TARGET_CUDA_FP16))
     {
         norm = 2.0f;  // l1 = 0.01 lInf = 2
     }

modules/dnn/test/test_onnx_conformance.cpp

@@ -954,7 +954,7 @@ public:
 
         if (target == DNN_TARGET_CUDA_FP16 || target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
         {
-            default_l1 = 4e-3;
+            default_l1 = 7e-3;
             default_lInf = 2e-2;
         }
         else

modules/dnn/test/test_onnx_conformance_layer_filter__cuda_denylist.inl.hpp

@@ -66,6 +66,15 @@
 "test_maxunpool_export_with_output_shape",
 "test_mul_bcast",
 "test_mul_uint8",
+"test_reduce_prod_default_axes_keepdims_example", // FP16 only
+"test_reduce_prod_default_axes_keepdims_random", // FP16 only
+"test_reduce_prod_do_not_keepdims_random", // FP16 only
+"test_reduce_prod_keepdims_random", // FP16 only
+"test_reduce_prod_negative_axes_keepdims_random", // FP16 only
+"test_reduce_sum_square_default_axes_keepdims_random", // FP16 only
+"test_reduce_sum_square_do_not_keepdims_random", // FP16 only
+"test_reduce_sum_square_keepdims_random", // FP16 only
+"test_reduce_sum_square_negative_axes_keepdims_random", // FP16 only
 "test_softmax_default_axis",
 "test_softmax_large_number",  // FP16 only
 "test_softmax_large_number_expanded",  // FP16 only

modules/dnn/test/test_onnx_importer.cpp

@@ -169,16 +169,17 @@ TEST_P(Test_ONNX_layers, Convolution_variable_weight_bias)
          backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019) && target == DNN_TARGET_MYRIAD)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_MYRIAD, CV_TEST_TAG_DNN_SKIP_IE_NN_BUILDER, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
 
-    if (backend == DNN_BACKEND_CUDA)
-        applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA); // not supported
-    if (backend == DNN_BACKEND_VKCOM)
-        applyTestTag(CV_TEST_TAG_DNN_SKIP_VULKAN); // not supported
-
     if (backend == DNN_BACKEND_INFERENCE_ENGINE_NGRAPH && target == DNN_TARGET_CPU &&
         getInferenceEngineCPUType() == CV_DNN_INFERENCE_ENGINE_CPU_TYPE_ARM_COMPUTE)
         applyTestTag(CV_TEST_TAG_DNN_SKIP_IE_ARM_CPU, CV_TEST_TAG_DNN_SKIP_IE_NGRAPH);
 #endif
 
+    if (backend == DNN_BACKEND_CUDA)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA); // supports only <= 2 inputs
+
+    if (backend == DNN_BACKEND_VKCOM)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_VULKAN); // not supported
+
     String basename = "conv_variable_wb";
     Net net = readNetFromONNX(_tf("models/" + basename + ".onnx"));
     ASSERT_FALSE(net.empty());
@@ -464,11 +465,15 @@ TEST_P(Test_ONNX_layers, Scale)
 
 TEST_P(Test_ONNX_layers, Scale_broadcast)
 {
+    if (backend == DNN_BACKEND_CUDA)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA); // doesn't support broadcasting
     testONNXModels("scale_broadcast", npy, 0, 0, false, true, 3);
 }
 
 TEST_P(Test_ONNX_layers, Scale_broadcast_mid)
 {
+    if (backend == DNN_BACKEND_CUDA)
+        applyTestTag(CV_TEST_TAG_DNN_SKIP_CUDA); // doesn't support broadcasting
     testONNXModels("scale_broadcast_mid", npy, 0, 0, false, true, 2);
 }
 
@@ -2131,7 +2136,7 @@ TEST_P(Test_ONNX_nets, Emotion_ferplus)
     double lInf = default_lInf;
 
     // Output values are in range [-2.011, 2.111]
-    if (backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16)
+    if ((backend == DNN_BACKEND_OPENCV && target == DNN_TARGET_OPENCL_FP16) || (target == DNN_TARGET_CUDA_FP16))
         l1 = 0.007;
     else if (backend == DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 && target == DNN_TARGET_OPENCL_FP16)
     {