diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2018-06-18 18:13:51 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:53:20 +0000 |
| commit | 17b12307edeaf488cfdf0cc3fa00b8f08293c93e (patch) | |
| tree | d7025d84a03d9897b4811ade16a2a1952bdb09c4 /arm_compute/runtime/NEON | |
| parent | 09b19129b65c5b8d1ca1c3851bab919bb9b7e1a1 (diff) | |
| download | ComputeLibrary-17b12307edeaf488cfdf0cc3fa00b8f08293c93e.tar.gz | |
COMPMID-1293: Handle aligned allocations
Change-Id: I6e642c8cd968240f883c327464519e57e5d0c3e3
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/136088
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'arm_compute/runtime/NEON')
| -rw-r--r-- | arm_compute/runtime/NEON/AssemblyHelper.h | 9 |
1 files changed, 2 insertions, 7 deletions
diff --git a/arm_compute/runtime/NEON/AssemblyHelper.h b/arm_compute/runtime/NEON/AssemblyHelper.h index c4ba1a584e..5801c42684 100644 --- a/arm_compute/runtime/NEON/AssemblyHelper.h +++ b/arm_compute/runtime/NEON/AssemblyHelper.h @@ -89,13 +89,8 @@ public: const auto in1_ptr = reinterpret_cast<const TypeInput *>(_b->buffer()); const int multi_stride_b = _b->info()->strides_in_bytes().z() / sizeof(TypeInput); - // Forcing 128-byte alignment (required by 32-bit kernels) - const unsigned int alignment = 128; - void *raw_ptr = reinterpret_cast<void *>(_pretranspose->buffer()); - size_t space = _pretranspose->info()->total_size(); - void *aligned_ptr = support::cpp11::align(alignment, _gemm_kernel_asm->get_B_pretransposed_array_size(), raw_ptr, space); ARM_COMPUTE_ERROR_ON(_pretranspose == nullptr || _pretranspose->buffer() == nullptr); - _gemm_kernel_asm->pretranspose_B_array(aligned_ptr, in1_ptr, ldb, multi_stride_b); + _gemm_kernel_asm->pretranspose_B_array(_pretranspose->buffer(), in1_ptr, ldb, multi_stride_b); _b->mark_as_unused(); } @@ -169,7 +164,7 @@ using AssemblyKernelGlueS8S32 = AssemblyKernelGlue<int8_t, int32_t>; inline void allocate_workspace(size_t workspace_size, Tensor &workspace, MemoryGroup *memory_group, size_t alignment, unsigned int num_threads) { ARM_COMPUTE_ERROR_ON_MSG(workspace_size == 0, "size cannot be 0"); - workspace.allocator()->init(TensorInfo(TensorShape{ (workspace_size + alignment - 1) * num_threads }, 1, DataType::S8)); + workspace.allocator()->init(TensorInfo(TensorShape{ (workspace_size + alignment) * num_threads }, 1, DataType::S8), alignment); if(memory_group != nullptr) { memory_group->manage(&workspace); |