COMPMID-3575: Mixed preicision in NEInstanceNormalizationLayerKernel

In order to fix the issue caused by the limited precision of FP16.
mixed precision (float accumulator) is introduced to
NEInstanceNormalizationLayerKernel. Since the reference kernel
is doing the mixed precision, currently mixed preicision computation
is default when it is called from NEInstanceNormalizationLayer.

- Make NEInstanceNormalizationLayerKernel use kernel descriptor
  to enable mixed precision computation
- NEInstanceNormalizationLayer is modified to use the descriptor

Change-Id: I7766622d715df054e303f9b441380b15b51f02b2
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3589
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 12 insertions, 14 deletions

diff --git a/arm_compute/core/NEON/kernels/NEInstanceNormalizationLayerKernel.h b/arm_compute/core/NEON/kernels/NEInstanceNormalizationLayerKernel.h
index 43f6ec3f13..a5bd453ac7 100644
--- a/arm_compute/core/NEON/kernels/NEInstanceNormalizationLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEInstanceNormalizationLayerKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 Arm Limited.
+ * Copyright (c) 2019-2020 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,6 +29,7 @@
 namespace arm_compute
 {
 class ITensor;
+struct InstanceNormalizationLayerKernelInfo;
 
 /** Interface for performing an instance normalization */
 class NEInstanceNormalizationLayerKernel : public INEKernel
@@ -52,26 +53,22 @@ public:
     ~NEInstanceNormalizationLayerKernel() = default;
     /** Set the input and output tensors.
      *
-     * @param[in, out] input   Source tensor. Data types supported: F16/F32. Data layout supported: NCHW
-     *                         In case of @p output tensor = nullptr this tensor will store the result of the normalization.
-     * @param[out]     output  Destination tensor. Data types and data layouts supported: same as @p input.
-     * @param[in]      gamma   (Optional) The scale scalar value applied to the normalized tensor. Defaults to 1.0
-     * @param[in]      beta    (Optional) The offset scalar value applied to the normalized tensor. Defaults to 0.0
-     * @param[in]      epsilon (Optional) Lower bound value for the normalization. Defaults to 1e-12
+     * @param[in, out] input  Source tensor. Data types supported: F16/F32. Data layout supported: NCHW
+     *                        In case of @p output tensor = nullptr this tensor will store the result of the normalization.
+     * @param[out]     output Destination tensor. Data types and data layouts supported: same as @p input.
+     * @param[in]      info   Kernel meta-data descriptor
      */
-    void configure(ITensor *input, ITensor *output, float gamma = 1.0f, float beta = 0.0f, float epsilon = 1e-12f);
+    void configure(ITensor *input, ITensor *output, const InstanceNormalizationLayerKernelInfo &info);
 
     /** Static function to check if given info will lead to a valid configuration of @ref NEInstanceNormalizationLayer.
      *
-     * @param[in] input   Source tensor info. Data types supported: F16/F32. Data layout supported: NCHW
-     * @param[in] output  Destination tensor info. Data types and data layouts supported: same as @p input.
-     * @param[in] gamma   (Optional) The scale scalar value applied to the normalized tensor. Defaults to 1.0
-     * @param[in] beta    (Optional) The offset scalar value applied to the normalized tensor. Defaults to 0.0
-     * @param[in] epsilon (Optional) Lower bound value for the normalization. Defaults to 1e-12
+     * @param[in] input  Source tensor info. Data types supported: F16/F32. Data layout supported: NCHW
+     * @param[in] output Destination tensor info. Data types and data layouts supported: same as @p input.
+     * @param[in] info   Kernel meta-data descriptor
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, float gamma = 1.0f, float beta = 0.0f, float epsilon = 1e-12f);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const InstanceNormalizationLayerKernelInfo &info);
 
     // Inherited methods overridden:
     void run(const Window &window, const ThreadInfo &info) override;
@@ -93,6 +90,7 @@ private:
     float                  _gamma;
     float                  _beta;
     float                  _epsilon;
+    bool                   _use_mixed_precision{ true };
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_NEINSTANCENORMALIZATIONLAYERKERNEL_H */