Rework OpenCL Depthwise Convolution

- Remove dedicated kernels for NCHW. Now we only use NHWC with permute
- Remove specialized kernels for 3x3 NHWC
- Simplify CLDepthwiseConvolutionLayer.cpp to call just the native
  implementation for both floating-point and quantized data types
- Develop two parametric opencl kernels for depthwise convolution layer NHWC
  (floating-point and quantized)
- Add support to export the weights to cl_image
- Extend test for depthwise convolution on opencl

Resolves COMPMID-4417

Change-Id: I253dd5d959a70783c82e62b1771a5e9f91621cb0
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5806
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>

1 files changed, 7 insertions, 2 deletions

diff --git a/src/core/CL/cl_kernels/activation_quant_helpers.h b/src/core/CL/cl_kernels/activation_quant_helpers.h
index a32e4e94a3..c420578546 100644
--- a/src/core/CL/cl_kernels/activation_quant_helpers.h
+++ b/src/core/CL/cl_kernels/activation_quant_helpers.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -51,7 +51,12 @@ inline TYPE lu_brelu_op(TYPE x)
 // Hard Swish Activation
 inline TYPE hard_swish_op(TYPE x)
 {
-    return  (x * ((min(max((TYPE)(x + (TYPE)3.f), (TYPE)0.f), (TYPE)6.f)) * (TYPE)0.166666667f));
+    return (x * ((min(max((TYPE)(x + (TYPE)3.f), (TYPE)0.f), (TYPE)6.f)) * (TYPE)0.166666667f));
+}
+
+inline TYPE identiy_op(TYPE x)
+{
+    return x;
 }
 
 #define ACTIVATION_OP2(op, x) op##_op(x)