COMPMID-817: Tuner: Port kernels to new design.

Change-Id: Iaabb1153c2abe0400ec79d51a21347debe92d642
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/134062
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

7 files changed, 18 insertions, 4 deletions

diff --git a/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp b/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp
index 676a121a76..c2b24e3c20 100644
--- a/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLDepthwiseConvolutionLayer.cpp
@@ -134,6 +134,7 @@ void CLDepthwiseConvolutionLayer::configure(ICLTensor *input, const ICLTensor *w
     _input_reshaped.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(shape_im2col));
     _im2col_kernel.set_target(gpu_target);
     _im2col_kernel.configure(input, &_input_reshaped, Size2D(weights_w, weights_h), conv_info, append_bias, depth_multiplier);
+    CLScheduler::get().tune_kernel_static(_im2col_kernel);
 
     // Weights reshape configuration
     const TensorShape shape_weights_reshape(patch_size, weights_z);
@@ -149,6 +150,7 @@ void CLDepthwiseConvolutionLayer::configure(ICLTensor *input, const ICLTensor *w
     _v2mm_output.allocator()->init(input->info()->clone()->set_is_resizable(true).reset_padding().set_data_type(v2mm_dt).set_tensor_shape(shape_v2mm_out));
     _v2mm_kernel.set_target(gpu_target);
     _v2mm_kernel.configure(&_input_reshaped, &_weights_reshaped, &_v2mm_output);
+    CLScheduler::get().tune_kernel_static(_v2mm_kernel);
     _output_reshaped.allocator()->init(_v2mm_output.info()->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(output_shape));
     _vector_to_tensor_kernel.configure(&_v2mm_output, (_is_quantized) ? &_output_reshaped : output, conv_w, conv_h);
 
diff --git a/src/runtime/CL/functions/CLFlattenLayer.cpp b/src/runtime/CL/functions/CLFlattenLayer.cpp
index 9f571b2036..f5809a218a 100644
--- a/src/runtime/CL/functions/CLFlattenLayer.cpp
+++ b/src/runtime/CL/functions/CLFlattenLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,6 +25,7 @@
 
 #include "arm_compute/core/CL/kernels/CLIm2ColKernel.h"
 #include "arm_compute/core/Size2D.h"
+#include "arm_compute/runtime/CL/CLScheduler.h"
 #include "support/ToolchainSupport.h"
 
 using namespace arm_compute;
@@ -34,4 +35,5 @@ void CLFlattenLayer::configure(const ICLTensor *input, ICLTensor *output)
     auto k = arm_compute::support::cpp14::make_unique<CLIm2ColKernel>();
     k->configure(input, output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false);
     _kernel = std::move(k);
+    CLScheduler::get().tune_kernel_static(*_kernel);
 }
diff --git a/src/runtime/CL/functions/CLFullyConnectedLayer.cpp b/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
index 44bf28374f..9248bc559b 100644
--- a/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
+++ b/src/runtime/CL/functions/CLFullyConnectedLayer.cpp
@@ -117,6 +117,7 @@ void CLFullyConnectedLayer::configure_conv_fc(const ICLTensor *input, const ICLT
     // Configure im2col kernel
     _memory_group.manage(&_im2col_output);
     _im2col_kernel.configure(input, &_im2col_output, Size2D(1, 1), PadStrideInfo(1, 1, 0, 0), false);
+    CLScheduler::get().tune_kernel_static(_im2col_kernel);
 
     // Configure matrix multiply kernel
     configure_mm(&_im2col_output, weights, output);
diff --git a/src/runtime/CL/functions/CLGEMM.cpp b/src/runtime/CL/functions/CLGEMM.cpp
index 7f37520f10..a0ec66f804 100644
--- a/src/runtime/CL/functions/CLGEMM.cpp
+++ b/src/runtime/CL/functions/CLGEMM.cpp
@@ -143,7 +143,9 @@ void CLGEMM::configure(const ICLTensor *a, const ICLTensor *b, const ICLTensor *
         _transpose_kernel.configure(b, &_tmp_b, mult_transpose1xW_width);
     }
 
+    // Configure and tune matrix multiply kernel
     _mm_kernel.configure(matrix_a, matrix_b, output, alpha, _is_interleaved_transposed, GEMMReshapeInfo(m, n, k, mult_transpose1xW_width, mult_interleave4x4_height));
+    CLScheduler::get().tune_kernel_static(_mm_kernel);
 
     if(_is_interleaved_transposed)
     {
diff --git a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
index 4f87043373..27bed44098 100644
--- a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
@@ -230,10 +230,11 @@ void CLGEMMConvolutionLayer::configure(const ICLTensor *input, const ICLTensor *
     _gemm_output.allocator()->init(info_gemm);
     _memory_group.manage(&_gemm_output);
 
-    // Configure im2col
+    // Configure and tune im2col
     _im2col_kernel.configure(input, &_im2col_output, Size2D(kernel_width, kernel_height), conv_info, append_bias, dilation);
+    CLScheduler::get().tune_kernel_static(_im2col_kernel);
 
-    // Configure GEMM
+    // Configure and tune GEMM
     configure_mm(&_im2col_output, weights, &_gemm_output);
 
     _im2col_output.allocator()->allocate();
@@ -250,8 +251,9 @@ void CLGEMMConvolutionLayer::configure(const ICLTensor *input, const ICLTensor *
         _gemmlowp_output_stage.configure(&_gemm_output, biases, &_tmp_output, output_multiplier, output_shift, output_quant_info.offset);
     }
 
-    // Configure Col2Im
+    // Configure and tune Col2Im
     _col2im_kernel.configure(_is_quantized ? &_tmp_output : &_gemm_output, output, std::make_pair(conv_w, conv_h));
+    CLScheduler::get().tune_kernel_static(_col2im_kernel);
     if(_is_quantized)
     {
         _tmp_output.allocator()->allocate();
diff --git a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
index 986fe00973..31d5cd5a7e 100644
--- a/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
+++ b/src/runtime/CL/functions/CLLocallyConnectedLayer.cpp
@@ -163,6 +163,8 @@ void CLLocallyConnectedLayer::configure(const ICLTensor *input, const ICLTensor
     _weights_reshaped.allocator()->allocate();
     _input_im2col_reshaped.allocator()->allocate();
     _gemm_output.allocator()->allocate();
+
+    CLScheduler::get().tune_kernel_static(_input_im2col_kernel);
 }
 
 void CLLocallyConnectedLayer::run()
diff --git a/src/runtime/CL/functions/CLPoolingLayer.cpp b/src/runtime/CL/functions/CLPoolingLayer.cpp
index 17875a38ad..cbe1ce3b47 100644
--- a/src/runtime/CL/functions/CLPoolingLayer.cpp
+++ b/src/runtime/CL/functions/CLPoolingLayer.cpp
@@ -63,6 +63,9 @@ void CLPoolingLayer::configure(ICLTensor *input, ICLTensor *output, const Poolin
             ARM_COMPUTE_ERROR("Data layout not supported");
     }
     _border_handler.configure(input, _kernel->border_size(), border_mode, pixel_value);
+
+    // Tune kernels
+    CLScheduler::get().tune_kernel_static(*_kernel);
 }
 
 Status CLPoolingLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const PoolingLayerInfo &pool_info)