commit 76975a3c36ffbfd2260eb3ed8f31f664c2483a72
author Georgios Pinitas <georgios.pinitas@arm.com> Tue Feb 19 13:42:59 2019 +0000
committer Georgios Pinitas <georgios.pinitas@arm.com> Tue Feb 19 14:56:06 2019 +0000
tree 51edab940788299fb03632cc4c647df033455853
parent c8a149cf9557c90a55ad53fef514b5df26587d86

COMPMID-2006: NEON GEMMLowp assertion failure.

Mark auxilary tensors as resizable when cloning.

Change-Id: I582e6d09a7daadbc43cf02f46a53e51c178daacb
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/726
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

src/runtime/NEON/functions/NEGEMM.cpp

diff --git a/src/runtime/NEON/functions/NEGEMM.cpp b/src/runtime/NEON/functions/NEGEMM.cpp
index 99cddc2..914f088 100644
--- a/src/runtime/NEON/functions/NEGEMM.cpp
+++ b/src/runtime/NEON/functions/NEGEMM.cpp
@@ -91,8 +91,8 @@
             shape_tmp_b.set(0, b->info()->dimension(1) * transpose_w);
             shape_tmp_b.set(1, std::ceil(b->info()->dimension(0) / static_cast<float>(transpose_w)));
 
-            TensorInfo info_a = a->info()->clone()->set_tensor_shape(shape_tmp_a);
-            TensorInfo info_b = b->info()->clone()->set_tensor_shape(shape_tmp_b);
+            TensorInfo info_a = a->info()->clone()->set_tensor_shape(shape_tmp_a).set_is_resizable(true);
+            TensorInfo info_b = b->info()->clone()->set_tensor_shape(shape_tmp_b).set_is_resizable(true);
 
             _tmp_a.allocator()->init(info_a);
             _tmp_b.allocator()->init(info_b);

src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp

diff --git a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
index 98bd95d..5286f11 100644
--- a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
+++ b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp
@@ -97,9 +97,9 @@
         else
         {
             // The interleaved output matrix will have the following shape: [ a_height * 4, ceil(a_width / 4.0f) ]
-            TensorInfo info_a = a->info()->clone()->set_tensor_shape(compute_interleaved_shape(*a->info()));
+            TensorInfo info_a = a->info()->clone()->set_tensor_shape(compute_interleaved_shape(*a->info())).set_is_resizable(true);
             // The transpose1xW output matrix will have the following shape: [ b_height * 16, ceil(b_width / 16.0f) ]
-            TensorInfo info_b = b->info()->clone()->set_tensor_shape(compute_transpose1xW_shape(*b->info()));
+            TensorInfo info_b = b->info()->clone()->set_tensor_shape(compute_transpose1xW_shape(*b->info())).set_is_resizable(true);
             _tmp_a.allocator()->init(info_a);
             _tmp_b.allocator()->init(info_b);
             _memory_group.manage(&_tmp_a);
@@ -241,8 +241,8 @@
             shape_tmp_b.set(0, b->dimension(1) * 16);
             shape_tmp_b.set(1, std::ceil(b->dimension(0) / 16.f));
 
-            TensorInfo info_a = a->clone()->set_tensor_shape(shape_tmp_a);
-            TensorInfo info_b = b->clone()->set_tensor_shape(shape_tmp_b);
+            TensorInfo info_a = a->clone()->set_tensor_shape(shape_tmp_a).set_is_resizable(true);
+            TensorInfo info_b = b->clone()->set_tensor_shape(shape_tmp_b).set_is_resizable(true);
 
             ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMInterleave4x4Kernel::validate(a, &info_a));
             ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMTranspose1xWKernel::validate(b, &info_b));