diff options
| author | Narumol Prangnawarat <narumol.prangnawarat@arm.com> | 2022-01-28 17:59:18 +0000 |
|---|---|---|
| committer | Jim Flynn <jim.flynn@arm.com> | 2022-01-31 12:53:51 +0000 |
| commit | e2af6f4322a1e2b8b3c391fb721a6a80c281477f (patch) | |
| tree | b0dd53289e27304a6d724821459cb0f4b6343a39 /include | |
| parent | fd313fef775ed210f8dab84452ea382a0b4164b0 (diff) | |
| download | armnn-e2af6f4322a1e2b8b3c391fb721a6a80c281477f.tar.gz | |
IVGCVSW-6552 Add support of aligned host memory
* Add AllocatedData functions to OutputHandler
* Enable import aligned memory in ImportInputs
* Enable import aligned memory in ImportOutputs
* Allow to import input and output if the memory is aligned
* Implement Reconfigure function on ClConvolution2dWorkload
* End-to-end test on Ref and Cl to ensure that input and output memory
are imported when aligned
Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
Change-Id: I9e5e4c26d1ac2f1d806803ade5f64c6479c51718
Diffstat (limited to 'include')
| -rw-r--r-- | include/armnn/IRuntime.hpp | 15 |
1 files changed, 9 insertions, 6 deletions
diff --git a/include/armnn/IRuntime.hpp b/include/armnn/IRuntime.hpp index d85a3e3724..042271fc2b 100644 --- a/include/armnn/IRuntime.hpp +++ b/include/armnn/IRuntime.hpp @@ -216,18 +216,19 @@ public: TensorInfo GetInputTensorInfo(NetworkId networkId, LayerBindingId layerId) const; TensorInfo GetOutputTensorInfo(NetworkId networkId, LayerBindingId layerId) const; - /// ImportInputs separates the importing and mapping of InputTensors from network execution. /// Allowing for a set of InputTensors to be imported and mapped once, but used in execution many times. /// This function is not thread safe and must not be used while other threads are calling Execute(). - /// Only compatible with AsyncEnabled networks - std::vector<ImportedInputId> ImportInputs(NetworkId networkId, const InputTensors& inputTensors); + /// Only compatible with AsyncEnabled networks and aligned memory import + std::vector<ImportedInputId> ImportInputs(NetworkId networkId, const InputTensors& inputTensors, + MemorySource forceImportMemorySource = MemorySource::Undefined); /// ImportOutputs separates the importing and mapping of OutputTensors from network execution. /// Allowing for a set of OutputTensors to be imported and mapped once, but used in execution many times. /// This function is not thread safe and must not be used while other threads are calling Execute(). - /// Only compatible with AsyncEnabled networks - std::vector<ImportedOutputId> ImportOutputs(NetworkId networkId, const OutputTensors& outputTensors); + /// Only compatible with AsyncEnabled networks and aligned memory import + std::vector<ImportedOutputId> ImportOutputs(NetworkId networkId, const OutputTensors& outputTensors, + MemorySource forceImportMemorySource = MemorySource::Undefined); /// Un-import and delete the imported InputTensor/s /// This function is not thread safe and must not be used while other threads are calling Execute(). @@ -242,7 +243,9 @@ public: /// Evaluates a network using input in inputTensors and outputs filled into outputTensors Status EnqueueWorkload(NetworkId networkId, const InputTensors& inputTensors, - const OutputTensors& outputTensors); + const OutputTensors& outputTensors, + std::vector<ImportedInputId> preImportedInputIds = {}, + std::vector<ImportedOutputId> preImportedOutputIds = {}); /// This is an experimental function. /// Evaluates a network using input in inputTensors and outputs filled into outputTensors. |