From 76975a3c36ffbfd2260eb3ed8f31f664c2483a72 Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Tue, 19 Feb 2019 13:42:59 +0000 Subject: COMPMID-2006: NEON GEMMLowp assertion failure. Mark auxilary tensors as resizable when cloning. Change-Id: I582e6d09a7daadbc43cf02f46a53e51c178daacb Signed-off-by: Georgios Pinitas Reviewed-on: https://review.mlplatform.org/726 Reviewed-by: Michalis Spyrou Tested-by: Arm Jenkins --- src/runtime/NEON/functions/NEGEMM.cpp | 4 ++-- src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp | 8 ++++---- 2 files changed, 6 insertions(+), 6 deletions(-) diff --git a/src/runtime/NEON/functions/NEGEMM.cpp b/src/runtime/NEON/functions/NEGEMM.cpp index 99cddc2920..914f088bf5 100644 --- a/src/runtime/NEON/functions/NEGEMM.cpp +++ b/src/runtime/NEON/functions/NEGEMM.cpp @@ -91,8 +91,8 @@ void NEGEMM::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITe shape_tmp_b.set(0, b->info()->dimension(1) * transpose_w); shape_tmp_b.set(1, std::ceil(b->info()->dimension(0) / static_cast(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); diff --git a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp index 98bd95d7df..5286f113a5 100644 --- a/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp +++ b/src/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.cpp @@ -97,9 +97,9 @@ void NEGEMMLowpMatrixMultiplyCore::configure(const ITensor *a, const ITensor *b, 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 @@ Status NEGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITenso 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)); -- cgit v1.2.1