commit 8a94e7cec7b09a417a278425e2b56e7af5bf45d9
author Georgios Pinitas <georgios.pinitas@arm.com> Fri Sep 15 19:06:47 2017 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:35:24 2018 +0000
tree e952f39903d4624bbd6445c9cc6c7dbcc1114026
parent 658039bc4e06be34272eccf559a516a6b52f75f5

COMPMID-534: Add MemoryManager support in OpenCL functions

Adds support for:
-CLConvolution
-CLGEMM
-CLGEMMLowp
-CLHOGDescriptor
-CLHOGGradient
-CLHOGMultiDetection
-CLL2Normalize
-CLLocallyConnectedLayer
-CLOpticalFlow
-CLReductionOperation

Change-Id: Ib13354d274ccf32ae933f3fbbad3ac3896cfd3bd
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/87938
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>

src/runtime/CL/functions/CLGEMM.cpp

diff --git a/src/runtime/CL/functions/CLGEMM.cpp b/src/runtime/CL/functions/CLGEMM.cpp
index 9867229..a81d113 100644
--- a/src/runtime/CL/functions/CLGEMM.cpp
+++ b/src/runtime/CL/functions/CLGEMM.cpp
@@ -38,8 +38,8 @@
 
 using namespace arm_compute;
 
-CLGEMM::CLGEMM()
-    : _interleave_kernel(), _transpose_kernel(), _mm_kernel(), _ma_kernel(), _tmp_a(), _tmp_b(), _is_interleaved_transposed(false), _run_addition(false)
+CLGEMM::CLGEMM(std::shared_ptr<IMemoryManager> memory_manager)
+    : _memory_group(std::move(memory_manager)), _interleave_kernel(), _transpose_kernel(), _mm_kernel(), _ma_kernel(), _tmp_a(), _tmp_b(), _is_interleaved_transposed(false), _run_addition(false)
 {
 }
 
@@ -86,6 +86,10 @@
         TensorInfo info_b(shape_tmp_b, 1, b->info()->data_type(), b->info()->fixed_point_position());
         _tmp_b.allocator()->init(info_b);
 
+        // Manage intermediate buffers
+        _memory_group.manage(&_tmp_a);
+        _memory_group.manage(&_tmp_b);
+
         // Configure interleave kernel
         _interleave_kernel.configure(a, &_tmp_a);
 
@@ -115,6 +119,8 @@
 
 void CLGEMM::run()
 {
+    _memory_group.acquire();
+
     if(_is_interleaved_transposed)
     {
         // Run interleave kernel
@@ -132,4 +138,6 @@
     {
         CLScheduler::get().enqueue(_ma_kernel);
     }
+
+    _memory_group.release();
 }