From 4406fd6cc4abded564d3791324e1f48bdfd34273 Mon Sep 17 00:00:00 2001
From: Frank Lei <frank.lei@arm.com>
Date: Thu, 1 Feb 2018 14:47:14 +0800
Subject: APPBROWSER-391: Fix GLES COMPUTE alignment issues

APPBROWSER-402: Performance optimization for squeezenet/xray model

Change-Id: If31b186b99a6d6087164019fe94d3ac9279e3204
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/119526
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
---
 .../core/GLES_COMPUTE/kernels/GCDepthConcatenateLayerKernel.h      | 3 ++-
 arm_compute/core/GLES_COMPUTE/kernels/GCScaleKernel.h              | 7 ++++---
 arm_compute/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.h        | 1 +
 3 files changed, 7 insertions(+), 4 deletions(-)

(limited to 'arm_compute/core/GLES_COMPUTE/kernels')

diff --git a/arm_compute/core/GLES_COMPUTE/kernels/GCDepthConcatenateLayerKernel.h b/arm_compute/core/GLES_COMPUTE/kernels/GCDepthConcatenateLayerKernel.h
index ce220cc564..06a54dd0b3 100644
--- a/arm_compute/core/GLES_COMPUTE/kernels/GCDepthConcatenateLayerKernel.h
+++ b/arm_compute/core/GLES_COMPUTE/kernels/GCDepthConcatenateLayerKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -71,6 +71,7 @@ private:
     IGCTensor       *_output;
     int              _top_bottom;
     int              _left_right;
+    int              _depth_offset;
 };
 }
 #endif /* __ARM_COMPUTE_GCDEPTHCONCATENATEKERNEL_H__ */
diff --git a/arm_compute/core/GLES_COMPUTE/kernels/GCScaleKernel.h b/arm_compute/core/GLES_COMPUTE/kernels/GCScaleKernel.h
index 3f936f6cb9..bdbc50c735 100644
--- a/arm_compute/core/GLES_COMPUTE/kernels/GCScaleKernel.h
+++ b/arm_compute/core/GLES_COMPUTE/kernels/GCScaleKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 ARM Limited.
+ * Copyright (c) 2016-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,7 +24,7 @@
 #ifndef __ARM_COMPUTE_GCSCALEKERNEL_H__
 #define __ARM_COMPUTE_GCSCALEKERNEL_H__
 
-#include "arm_compute/core/GLES_COMPUTE/IGCSimple2DKernel.h"
+#include "arm_compute/core/GLES_COMPUTE/IGCSimple3DKernel.h"
 #include "arm_compute/core/Types.h"
 
 namespace arm_compute
@@ -32,7 +32,7 @@ namespace arm_compute
 class IGCTensor;
 
 /** Interface for the scale kernel */
-class GCScaleKernel : public IGCSimple2DKernel
+class GCScaleKernel : public IGCSimple3DKernel
 {
 public:
     /** Initialise the kernel's inputs, output and interpolation policy
@@ -47,6 +47,7 @@ public:
     void configure(const IGCTensor *input, IGCTensor *output, InterpolationPolicy policy, bool border_undefined, SamplingPolicy sampling_policy = SamplingPolicy::CENTER);
 
     // Inherited methods overridden:
+    void run(const Window &window) override;
     BorderSize border_size() const override;
 };
 } // namespace arm_compute
diff --git a/arm_compute/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.h b/arm_compute/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.h
index 5f108764b4..452caae146 100644
--- a/arm_compute/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.h
+++ b/arm_compute/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.h
@@ -78,6 +78,7 @@ public:
 private:
     IGCTensor    *_input;
     gles::NDRange _lws;
+    int           _left_padding;
 };
 }
 #endif /*__ARM_COMPUTE_GCTENSORSHIFTKERNEL_H__ */
-- 
cgit v1.2.1