diff options
author | Matteo Martincigh <matteo.martincigh@arm.com> | 2018-12-18 09:26:39 +0000 |
---|---|---|
committer | Matteo Martincigh <matteo.martincigh@arm.com> | 2019-01-04 17:28:07 +0000 |
commit | 747ef82c88f9afe14a8b80b6b3b34118353e97f2 (patch) | |
tree | a29ac33b84fb96a41103a0a97327189495374cc9 /src/backends/backendsCommon/CpuTensorHandle.hpp | |
parent | 760892724d131c7da4b9baad05cddd49276ad6bb (diff) | |
download | armnn-747ef82c88f9afe14a8b80b6b3b34118353e97f2.tar.gz |
MLCE-77 Depthwise Convolution with depth multiplier > 1 doesn't work
* Unified ArmNN's weight format to [ M, I, H, W ] for the depthwise convolution
* Added conversion utilities to permute/reshape the weights as appropriate
when using CL and Neon backends
* Updated the reference implementation of the convolution
* Updated the relevant unit tests accordingly
!android-nn-driver:459
Change-Id: I07d0818efa9d1ca1e5dad82983aac1fe78eadb18
Diffstat (limited to 'src/backends/backendsCommon/CpuTensorHandle.hpp')
-rw-r--r-- | src/backends/backendsCommon/CpuTensorHandle.hpp | 6 |
1 files changed, 6 insertions, 0 deletions
diff --git a/src/backends/backendsCommon/CpuTensorHandle.hpp b/src/backends/backendsCommon/CpuTensorHandle.hpp index ae13d6c439..b88a0d385b 100644 --- a/src/backends/backendsCommon/CpuTensorHandle.hpp +++ b/src/backends/backendsCommon/CpuTensorHandle.hpp @@ -72,6 +72,9 @@ private: const void* m_Memory; }; +template<> +const void* ConstCpuTensorHandle::GetConstTensor<void>() const; + // Abstract specialization of ConstCpuTensorHandle that allows write access to the same data. class CpuTensorHandle : public ConstCpuTensorHandle { @@ -99,6 +102,9 @@ private: void* m_MutableMemory; }; +template <> +void* CpuTensorHandle::GetTensor<void>() const; + // A CpuTensorHandle that owns the wrapped memory region. class ScopedCpuTensorHandle : public CpuTensorHandle { |