From 4a34c4e3b08de5eeee6bc1fe9ff0f32f8b890a73 Mon Sep 17 00:00:00 2001
From: Deven Desai <deven.desai.amd@gmail.com>
Date: Wed, 5 Feb 2020 21:22:38 +0000
Subject: [PATCH] working around a known gcc/hcc interface kernel args passing
 bug

---
 tensorflow/core/kernels/relu_op.cc        |  4 +---
 tensorflow/core/kernels/relu_op.h         |  4 ++--
 tensorflow/core/kernels/relu_op_functor.h | 18 ++++++++++++++----
 3 files changed, 17 insertions(+), 9 deletions(-)

diff --git a/tensorflow/core/kernels/relu_op.cc b/tensorflow/core/kernels/relu_op.cc
index 75f6649e983..f5d23ca9d25 100644
--- a/tensorflow/core/kernels/relu_op.cc
+++ b/tensorflow/core/kernels/relu_op.cc
@@ -105,9 +105,7 @@ namespace functor {
   extern template struct Relu6Grad<GPUDevice, T>;                              \
                                                                                \
   template <>                                                                  \
-  void LeakyRelu<GPUDevice, T>::operator()(                                    \
-      const GPUDevice& d, typename TTypes<T>::ConstTensor features, T alpha,   \
-      typename TTypes<T>::Tensor activations);                                 \
+  void LeakyRelu<GPUDevice, T>::operator()(LeakyReluArgs args);                \
   extern template struct LeakyRelu<GPUDevice, T>;                              \
                                                                                \
   template <>                                                                  \
diff --git a/tensorflow/core/kernels/relu_op.h b/tensorflow/core/kernels/relu_op.h
index a4638c70c2c..2ef38a62e40 100644
--- a/tensorflow/core/kernels/relu_op.h
+++ b/tensorflow/core/kernels/relu_op.h
@@ -143,8 +143,8 @@ class LeakyReluOp : public UnaryElementWiseOp<T, LeakyReluOp<Device, T>> {
 
   void Operate(OpKernelContext* context, const Tensor& input, Tensor* output) {
     functor::LeakyRelu<Device, T> functor;
-    functor(context->eigen_device<Device>(), input.flat<T>(), alpha_,
-            output->flat<T>());
+    functor({context->eigen_device<Device>(), input.flat<T>(), alpha_,
+             output->flat<T>()});
   }
 
  private:
diff --git a/tensorflow/core/kernels/relu_op_functor.h b/tensorflow/core/kernels/relu_op_functor.h
index 9337282b3d3..913d5f7ced0 100644
--- a/tensorflow/core/kernels/relu_op_functor.h
+++ b/tensorflow/core/kernels/relu_op_functor.h
@@ -98,11 +98,21 @@ struct LeakyRelu {
   //
   // features: any shape.
   // activations: same shape as "features".
-  void operator()(const Device& d, typename TTypes<T>::ConstTensor features,
-                  T alpha, typename TTypes<T>::Tensor activations) {
+
+  // Need to bundle the args (to the LeakyRelu functor) within a struct
+  // Not doing so leads to Eigen kernel args not getting populated
+  // corretly for Eigen::half type (when building on the ROCM platform)
+  struct LeakyReluArgs {
+    const Device& d;
+    typename TTypes<T>::ConstTensor features;
+    T alpha;
+    typename TTypes<T>::Tensor activations;
+  };
+  void operator()(LeakyReluArgs args) {
     // Note that alpha might be > 1 or < 0, so we don't use cwiseMax here.
-    activations.device(d) =
-        (features > static_cast<T>(0)).select(features, features * alpha);
+    args.activations.device(args.d) =
+        (args.features > static_cast<T>(0))
+            .select(args.features, args.features * args.alpha);
   }
 };