COMPMID-1145: (API) Introduce prepare() stage (NEON/CL/GLES)

Change-Id: I5b46764f9c3154ec3e3b9c951cc9e6dfbcb81dfb
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/134255
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>

1 files changed, 14 insertions, 21 deletions

diff --git a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
index 27bed44098..82710b6461 100644
--- a/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLGEMMConvolutionLayer.cpp
@@ -91,8 +91,7 @@ void CLConvolutionLayerReshapeWeights::run()
 
 CLGEMMConvolutionLayer::CLGEMMConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager)
     : _memory_group(memory_manager), _reshape_weights(), _im2col_kernel(), _mm_gemm(memory_manager), _mm_gemmlowp(memory_manager), _gemmlowp_output_stage(), _col2im_kernel(), _activationlayer_function(),
-      _original_weights(nullptr), _im2col_output(), _weights_reshaped(), _gemm_output(), _tmp_output(), _is_quantized(false), _is_activationlayer_enabled(false), _is_prepared(false),
-      _retain_internal_weights(false)
+      _original_weights(nullptr), _im2col_output(), _weights_reshaped(), _gemm_output(), _tmp_output(), _is_quantized(false), _is_activationlayer_enabled(false), _is_prepared(false)
 {
 }
 
@@ -166,10 +165,9 @@ void CLGEMMConvolutionLayer::configure(const ICLTensor *input, const ICLTensor *
                                                                 dilation,
                                                                 act_info));
 
-    _is_prepared             = false;
-    _original_weights        = weights;
-    _is_quantized            = is_data_type_quantized_asymmetric(input->info()->data_type());
-    _retain_internal_weights = weights_info.retain_internal_weights();
+    _is_prepared      = weights_info.retain_internal_weights();
+    _original_weights = weights;
+    _is_quantized     = is_data_type_quantized_asymmetric(input->info()->data_type());
 
     const DataType dt = input->info()->data_type();
 
@@ -408,23 +406,18 @@ void CLGEMMConvolutionLayer::prepare()
 {
     if(!_is_prepared)
     {
-        if(!_retain_internal_weights)
-        {
-            // Run weights reshaping and mark as unused
-            ARM_COMPUTE_ERROR_ON(!_original_weights->is_used());
-            _weights_reshaped.allocator()->allocate();
-            _reshape_weights.run();
-            _original_weights->mark_as_unused();
-        }
+        ARM_COMPUTE_ERROR_ON(!_original_weights->is_used());
+
+        // Run weights reshaping and mark original weights tensor as unused
+        _weights_reshaped.allocator()->allocate();
+        _reshape_weights.run();
+        _original_weights->mark_as_unused();
 
-        // Run GEMM prepare
-        if(!_is_quantized)
+        // Prepare GEMM
+        _is_quantized ? _mm_gemmlowp.prepare() : _mm_gemm.prepare();
+        if(!_weights_reshaped.is_used())
         {
-            _mm_gemm.prepare();
-            if(!_weights_reshaped.is_used() && !_retain_internal_weights)
-            {
-                _weights_reshaped.allocator()->free();
-            }
+            _weights_reshaped.allocator()->free();
         }
 
         CLScheduler::get().queue().finish();