Merge pull request opencv#17295 from dkurt:dnn_fusion_ftz

Author: alalek

modules/dnn/src/layers/convolution_layer.cpp

@@ -1230,6 +1230,13 @@ class ConvolutionLayerImpl CV_FINAL : public BaseConvolutionLayerImpl
         CV_TRACE_FUNCTION();
         CV_TRACE_ARG_VALUE(name, "name", name.c_str());
 
+#if CV_TRY_SSE
+        uint32_t ftzMode = _MM_GET_FLUSH_ZERO_MODE();
+        uint32_t dazMode = _MM_GET_DENORMALS_ZERO_MODE();
+        _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON);
+        _MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON);
+#endif
+
         CV_OCL_RUN(IS_DNN_OPENCL_TARGET(preferableTarget),
                    forward_ocl(inputs_arr, outputs_arr, internals_arr))
 
@@ -1312,6 +1319,10 @@ class ConvolutionLayerImpl CV_FINAL : public BaseConvolutionLayerImpl
             ParallelConv::run(inputs[0], outputs[0], weightsMat, biasvec, reluslope,
                             kernel_size, strides, pads_begin, pads_end, dilations, activ.get(), ngroups, nstripes);
         }
+#if CV_TRY_SSE
+        _MM_SET_FLUSH_ZERO_MODE(ftzMode);
+        _MM_SET_DENORMALS_ZERO_MODE(dazMode);
+#endif
     }
 
     virtual int64 getFLOPS(const std::vector<MatShape> &inputs,