diff options
author | Narumol Prangnawarat <narumol.prangnawarat@arm.com> | 2020-03-30 16:11:04 +0100 |
---|---|---|
committer | Narumol Prangnawarat <narumol.prangnawarat@arm.com> | 2020-03-31 09:29:40 +0100 |
commit | 250d3927b16abe4d6932cd5dce1184bd7026a2b7 (patch) | |
tree | f73603873c0fbd692fbcbbd242d2a45cef6dc890 /src/backends/neon/workloads/NeonConvertFp32ToBf16Workload.hpp | |
parent | e2062cdf1eb31b87860f9889f0e799e89f0dfa30 (diff) | |
download | armnn-250d3927b16abe4d6932cd5dce1184bd7026a2b7.tar.gz |
IVGCVSW-4633 Add conversion of BF16 support to Neon
* Add NeonConvertBf16ToFp32Workload
* Add NeonConvertFp32ToBf16Workload
* Add BFloat16 type support to NeonConstantWorkload and NeonTensorHandle
* Add ConvertBf16ToFp32Weight when ConvertBf16ToFp32Layer is added
* Unit tests
Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
Change-Id: Id5b44a203add5e0c98c1ca4e2162115741b56644
Diffstat (limited to 'src/backends/neon/workloads/NeonConvertFp32ToBf16Workload.hpp')
-rw-r--r-- | src/backends/neon/workloads/NeonConvertFp32ToBf16Workload.hpp | 26 |
1 files changed, 26 insertions, 0 deletions
diff --git a/src/backends/neon/workloads/NeonConvertFp32ToBf16Workload.hpp b/src/backends/neon/workloads/NeonConvertFp32ToBf16Workload.hpp new file mode 100644 index 0000000000..bc96c16287 --- /dev/null +++ b/src/backends/neon/workloads/NeonConvertFp32ToBf16Workload.hpp @@ -0,0 +1,26 @@ +// +// Copyright © 2020 Arm Ltd. All rights reserved. +// SPDX-License-Identifier: MIT +// + +#pragma once + +#include <backendsCommon/Workload.hpp> +#include <backendsCommon/WorkloadData.hpp> +#include <neon/workloads/NeonWorkloadUtils.hpp> + +namespace armnn +{ + +class NeonConvertFp32ToBf16Workload : public Float32ToBFloat16Workload<ConvertFp32ToBf16QueueDescriptor> +{ +public: + NeonConvertFp32ToBf16Workload(const ConvertFp32ToBf16QueueDescriptor& descriptor, const WorkloadInfo& info); + virtual void Execute() const override; + +private: + using TensorHandlePair = std::pair<const ITensorHandle*, ITensorHandle*>; + std::vector<TensorHandlePair> m_TensorHandlePairs; +}; + +} //namespace armnn |