aboutsummaryrefslogtreecommitdiff
path: root/src/backends/cl
diff options
context:
space:
mode:
Diffstat (limited to 'src/backends/cl')
-rw-r--r--src/backends/cl/ClLayerSupport.cpp203
-rw-r--r--src/backends/cl/ClLayerSupport.hpp114
-rw-r--r--src/backends/cl/workloads/ClAdditionWorkload.cpp15
-rw-r--r--src/backends/cl/workloads/ClAdditionWorkload.hpp7
-rw-r--r--src/backends/cl/workloads/ClConvertFp16ToFp32Workload.cpp16
-rw-r--r--src/backends/cl/workloads/ClConvertFp16ToFp32Workload.hpp4
-rw-r--r--src/backends/cl/workloads/ClConvertFp32ToFp16Workload.cpp16
-rw-r--r--src/backends/cl/workloads/ClConvertFp32ToFp16Workload.hpp4
-rw-r--r--src/backends/cl/workloads/ClPadWorkload.cpp15
-rw-r--r--src/backends/cl/workloads/ClPadWorkload.hpp7
-rw-r--r--src/backends/cl/workloads/ClSubtractionWorkload.cpp15
-rw-r--r--src/backends/cl/workloads/ClSubtractionWorkload.hpp7
12 files changed, 216 insertions, 207 deletions
diff --git a/src/backends/cl/ClLayerSupport.cpp b/src/backends/cl/ClLayerSupport.cpp
index 434b069092..494b339952 100644
--- a/src/backends/cl/ClLayerSupport.cpp
+++ b/src/backends/cl/ClLayerSupport.cpp
@@ -22,16 +22,16 @@
#include "workloads/ClConvolution2dWorkload.hpp"
#include "workloads/ClDepthwiseConvolutionWorkload.hpp"
#include "workloads/ClDivisionFloatWorkload.hpp"
+#include "workloads/ClFullyConnectedWorkload.hpp"
#include "workloads/ClL2NormalizationFloatWorkload.hpp"
+#include "workloads/ClLstmFloatWorkload.hpp"
#include "workloads/ClMultiplicationWorkload.hpp"
-#include "workloads/ClFullyConnectedWorkload.hpp"
+#include "workloads/ClNormalizationFloatWorkload.hpp"
#include "workloads/ClPadWorkload.hpp"
-#include "workloads/ClPooling2dBaseWorkload.hpp"
#include "workloads/ClPermuteWorkload.hpp"
-#include "workloads/ClNormalizationFloatWorkload.hpp"
+#include "workloads/ClPooling2dBaseWorkload.hpp"
#include "workloads/ClSoftmaxBaseWorkload.hpp"
#include "workloads/ClSubtractionWorkload.hpp"
-#include "workloads/ClLstmFloatWorkload.hpp"
#endif
using namespace boost;
@@ -59,14 +59,14 @@ bool IsMatchingStride(uint32_t actualStride)
return IsMatchingStride<FirstStride>(actualStride) || IsMatchingStride<SecondStride, ValidStrides...>(actualStride);
};
-bool IsClBackendSupported(std::string* reasonIfUnsupported)
+bool IsClBackendSupported(Optional<std::string&> reasonIfUnsupported)
{
#if ARMCOMPUTECL_ENABLED
return true;
#else
- if (reasonIfUnsupported != nullptr)
+ if (reasonIfUnsupported)
{
- *reasonIfUnsupported = "The armnn library has been built without CL support";
+ reasonIfUnsupported.value() = "The armnn library has been built without CL support";
}
return false;
#endif
@@ -80,13 +80,13 @@ bool IsClBackendSupported(std::string* reasonIfUnsupported)
#if ARMCOMPUTECL_ENABLED
template<class FuncType, class... Args>
-inline bool IsWorkloadSupported(FuncType&& func, std::string* reasonIfUnsupported, Args&&... args)
+inline bool IsWorkloadSupported(FuncType&& func, Optional<std::string&> reasonIfUnsupported, Args&&... args)
{
arm_compute::Status aclStatus = func(std::forward<Args>(args)...);
const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK);
if (!supported && reasonIfUnsupported)
{
- *reasonIfUnsupported = aclStatus.error_description();
+ reasonIfUnsupported.value() = aclStatus.error_description();
}
return supported;
}
@@ -101,7 +101,7 @@ inline bool IsWorkloadSupported(FuncType&& func, std::string* reasonIfUnsupporte
} //namespace
template<typename FloatFunc, typename Uint8Func, typename ... Params>
-bool IsSupportedForDataTypeCl(std::string* reasonIfUnsupported,
+bool IsSupportedForDataTypeCl(Optional<std::string&> reasonIfUnsupported,
DataType dataType,
FloatFunc floatFuncPtr,
Uint8Func uint8FuncPtr,
@@ -119,7 +119,7 @@ bool IsSupportedForDataTypeCl(std::string* reasonIfUnsupported,
bool IsActivationSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const ActivationDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClActivationWorkloadValidate,
reasonIfUnsupported,
@@ -131,12 +131,13 @@ bool IsActivationSupportedCl(const TensorInfo& input,
bool IsAdditionSupportedCl(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
- return FORWARD_CL_LAYER_SUPPORT_FUNC(ClAdditionValidate(input0,
- input1,
- output,
- reasonIfUnsupported));
+ FORWARD_WORKLOAD_VALIDATE_FUNC(ClAdditionValidate,
+ reasonIfUnsupported,
+ input0,
+ input1,
+ output);
}
bool IsBatchNormalizationSupportedCl(const TensorInfo& input,
@@ -146,7 +147,7 @@ bool IsBatchNormalizationSupportedCl(const TensorInfo& input,
const TensorInfo& beta,
const TensorInfo& gamma,
const BatchNormalizationDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClBatchNormalizationValidate,
reasonIfUnsupported,
@@ -160,7 +161,7 @@ bool IsBatchNormalizationSupportedCl(const TensorInfo& input,
}
bool IsConstantSupportedCl(const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
return IsSupportedForDataTypeCl(reasonIfUnsupported,
output.GetDataType(),
@@ -201,10 +202,11 @@ bool IsClDirectConvolution2dSupported(const TensorInfo& weightInfo, const Convol
return isSupported;
}
-bool IsDirectConvolution2dParamsSupportedCl(std::string* reasonIfUnsupported,
+bool IsDirectConvolution2dParamsSupportedCl(Optional<std::string&> reasonIfUnsupported,
const Convolution2dDescriptor& parameters,
const TensorInfo& weightInfo)
{
+ ignore_unused(reasonIfUnsupported);
return IsClDirectConvolution2dSupported(weightInfo, parameters);
}
@@ -213,7 +215,7 @@ bool IsConvolution2dSupportedCl(const TensorInfo& input,
const Convolution2dDescriptor& descriptor,
const TensorInfo& weights,
const Optional<TensorInfo>& biases,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClConvolution2dWorkloadValidate,
reasonIfUnsupported,
@@ -229,7 +231,7 @@ bool IsDepthwiseConvolutionSupportedCl(const TensorInfo& input,
const DepthwiseConvolution2dDescriptor& descriptor,
const TensorInfo& weights,
const Optional<TensorInfo>& biases,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClDepthwiseConvolutionWorkloadValidate,
reasonIfUnsupported,
@@ -243,7 +245,7 @@ bool IsDepthwiseConvolutionSupportedCl(const TensorInfo& input,
bool IsDivisionSupportedCl(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClDivisionWorkloadValidate,
reasonIfUnsupported,
@@ -255,12 +257,14 @@ bool IsDivisionSupportedCl(const TensorInfo& input0,
bool IsSubtractionSupportedCl(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
- return FORWARD_CL_LAYER_SUPPORT_FUNC(ClSubtractionValidate(input0,
- input1,
- output,
- reasonIfUnsupported));
+
+ FORWARD_WORKLOAD_VALIDATE_FUNC(ClSubtractionValidate,
+ reasonIfUnsupported,
+ input0,
+ input1,
+ output);
}
bool IsFullyConnectedSupportedCl(const TensorInfo& input,
@@ -268,7 +272,7 @@ bool IsFullyConnectedSupportedCl(const TensorInfo& input,
const TensorInfo& weights,
const TensorInfo& biases,
const FullyConnectedDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClFullyConnectedWorkloadValidate,
reasonIfUnsupported,
@@ -280,7 +284,7 @@ bool IsFullyConnectedSupportedCl(const TensorInfo& input,
}
bool IsInputSupportedCl(const TensorInfo& input,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
return IsSupportedForDataTypeCl(reasonIfUnsupported,
input.GetDataType(),
@@ -291,14 +295,14 @@ bool IsInputSupportedCl(const TensorInfo& input,
bool IsL2NormalizationSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const L2NormalizationDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClL2NormalizationWorkloadValidate, reasonIfUnsupported, input, output, descriptor);
}
bool IsMergerSupportedCl(const std::vector<const TensorInfo*> inputs,
const OriginsDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
ignore_unused(descriptor);
return IsSupportedForDataTypeCl(reasonIfUnsupported,
@@ -310,7 +314,7 @@ bool IsMergerSupportedCl(const std::vector<const TensorInfo*> inputs,
bool IsMultiplicationSupportedCl(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClMultiplicationWorkloadValidate,
reasonIfUnsupported,
@@ -322,13 +326,13 @@ bool IsMultiplicationSupportedCl(const TensorInfo& input0,
bool IsNormalizationSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const NormalizationDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClNormalizationWorkloadValidate, reasonIfUnsupported, input, output, descriptor);
}
bool IsOutputSupportedCl(const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
return IsSupportedForDataTypeCl(reasonIfUnsupported,
output.GetDataType(),
@@ -336,18 +340,10 @@ bool IsOutputSupportedCl(const TensorInfo& output,
&TrueFunc<>);
}
-bool IsPadSupportedCl(const TensorInfo& input,
- const TensorInfo& output,
- const PadDescriptor& descriptor,
- std::string* reasonIfUnsupported)
-{
- return FORWARD_CL_LAYER_SUPPORT_FUNC(ClPadValidate(input, output, descriptor, reasonIfUnsupported));
-}
-
bool IsPermuteSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const PermuteDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
ignore_unused(input);
ignore_unused(output);
@@ -357,13 +353,13 @@ bool IsPermuteSupportedCl(const TensorInfo& input,
bool IsPooling2dSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const Pooling2dDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClPooling2dWorkloadValidate, reasonIfUnsupported, input, output, descriptor);
}
bool IsResizeBilinearSupportedCl(const TensorInfo& input,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
return IsSupportedForDataTypeCl(reasonIfUnsupported,
input.GetDataType(),
@@ -374,7 +370,7 @@ bool IsResizeBilinearSupportedCl(const TensorInfo& input,
bool IsSoftmaxSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const SoftmaxDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
ignore_unused(descriptor);
FORWARD_WORKLOAD_VALIDATE_FUNC(ClSoftmaxWorkloadValidate, reasonIfUnsupported, input, output);
@@ -382,7 +378,7 @@ bool IsSoftmaxSupportedCl(const TensorInfo& input,
bool IsSplitterSupportedCl(const TensorInfo& input,
const ViewsDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
ignore_unused(descriptor);
return IsSupportedForDataTypeCl(reasonIfUnsupported,
@@ -393,23 +389,25 @@ bool IsSplitterSupportedCl(const TensorInfo& input,
bool IsFakeQuantizationSupportedCl(const TensorInfo& input,
const FakeQuantizationDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
ignore_unused(input);
ignore_unused(descriptor);
+ ignore_unused(reasonIfUnsupported);
return false;
}
bool IsReshapeSupportedCl(const TensorInfo& input,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
ignore_unused(input);
+ ignore_unused(reasonIfUnsupported);
return true;
}
bool IsFloorSupportedCl(const TensorInfo& input,
const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
ignore_unused(output);
return IsClBackendSupported(reasonIfUnsupported) &&
@@ -420,59 +418,104 @@ bool IsFloorSupportedCl(const TensorInfo& input,
&FalseFuncU8<>);
}
-bool IsLstmSupportedCl(const TensorInfo& input, const TensorInfo& outputStateIn,
- const TensorInfo& cellStateIn, const TensorInfo& scratchBuffer,
- const TensorInfo& outputStateOut, const TensorInfo& cellStateOut,
- const TensorInfo& output, const LstmDescriptor& descriptor,
- const TensorInfo& inputToForgetWeights, const TensorInfo& inputToCellWeights,
- const TensorInfo& inputToOutputWeights, const TensorInfo& recurrentToForgetWeights,
- const TensorInfo& recurrentToCellWeights, const TensorInfo& recurrentToOutputWeights,
- const TensorInfo& forgetGateBias, const TensorInfo& cellBias,
- const TensorInfo& outputGateBias, const TensorInfo* inputToInputWeights,
- const TensorInfo* recurrentToInputWeights, const TensorInfo* cellToInputWeights,
- const TensorInfo* inputGateBias, const TensorInfo* projectionWeights,
- const TensorInfo* projectionBias, const TensorInfo* cellToForgetWeights,
- const TensorInfo* cellToOutputWeights, std::string* reasonIfUnsupported)
-{
- FORWARD_WORKLOAD_VALIDATE_FUNC(ClLstmFloatWorkloadValidate, reasonIfUnsupported,
- input, outputStateIn, cellStateIn, scratchBuffer, outputStateOut, cellStateOut,
- output, descriptor, inputToForgetWeights, inputToCellWeights,
- inputToOutputWeights, recurrentToForgetWeights,
- recurrentToCellWeights, recurrentToOutputWeights,
- forgetGateBias, cellBias, outputGateBias,
- inputToInputWeights, recurrentToInputWeights,
- cellToInputWeights, inputGateBias, projectionWeights,
- projectionBias, cellToForgetWeights, cellToOutputWeights);
+bool IsLstmSupportedCl(const TensorInfo& input,
+ const TensorInfo& outputStateIn,
+ const TensorInfo& cellStateIn,
+ const TensorInfo& scratchBuffer,
+ const TensorInfo& outputStateOut,
+ const TensorInfo& cellStateOut,
+ const TensorInfo& output,
+ const LstmDescriptor& descriptor,
+ const TensorInfo& inputToForgetWeights,
+ const TensorInfo& inputToCellWeights,
+ const TensorInfo& inputToOutputWeights,
+ const TensorInfo& recurrentToForgetWeights,
+ const TensorInfo& recurrentToCellWeights,
+ const TensorInfo& recurrentToOutputWeights,
+ const TensorInfo& forgetGateBias,
+ const TensorInfo& cellBias,
+ const TensorInfo& outputGateBias,
+ const TensorInfo* inputToInputWeights,
+ const TensorInfo* recurrentToInputWeights,
+ const TensorInfo* cellToInputWeights,
+ const TensorInfo* inputGateBias,
+ const TensorInfo* projectionWeights,
+ const TensorInfo* projectionBias,
+ const TensorInfo* cellToForgetWeights,
+ const TensorInfo* cellToOutputWeights,
+ Optional<std::string&> reasonIfUnsupported)
+{
+ FORWARD_WORKLOAD_VALIDATE_FUNC(ClLstmFloatWorkloadValidate,
+ reasonIfUnsupported,
+ input,
+ outputStateIn,
+ cellStateIn,
+ scratchBuffer,
+ outputStateOut,
+ cellStateOut,
+ output,
+ descriptor,
+ inputToForgetWeights,
+ inputToCellWeights,
+ inputToOutputWeights,
+ recurrentToForgetWeights,
+ recurrentToCellWeights,
+ recurrentToOutputWeights,
+ forgetGateBias,
+ cellBias,
+ outputGateBias,
+ inputToInputWeights,
+ recurrentToInputWeights,
+ cellToInputWeights,
+ inputGateBias,
+ projectionWeights,
+ projectionBias,
+ cellToForgetWeights,
+ cellToOutputWeights);
}
bool IsConvertFp16ToFp32SupportedCl(const TensorInfo& input,
const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClConvertFp16ToFp32WorkloadValidate,
reasonIfUnsupported,
input,
- output,
- reasonIfUnsupported);
+ output);
}
bool IsConvertFp32ToFp16SupportedCl(const TensorInfo& input,
const TensorInfo& output,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
FORWARD_WORKLOAD_VALIDATE_FUNC(ClConvertFp32ToFp16WorkloadValidate,
reasonIfUnsupported,
input,
- output,
- reasonIfUnsupported);
+ output);
}
bool IsMeanSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const MeanDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+ Optional<std::string&> reasonIfUnsupported)
{
+ ignore_unused(input);
+ ignore_unused(output);
+ ignore_unused(descriptor);
+ ignore_unused(reasonIfUnsupported);
return false;
}
+bool IsPadSupportedCl(const TensorInfo& input,
+ const TensorInfo& output,
+ const PadDescriptor& descriptor,
+ Optional<std::string&> reasonIfUnsupported)
+{
+ FORWARD_WORKLOAD_VALIDATE_FUNC(ClPadValidate,
+ reasonIfUnsupported,
+ input,
+ output,
+ descriptor);
+}
+
}
diff --git a/src/backends/cl/ClLayerSupport.hpp b/src/backends/cl/ClLayerSupport.hpp
index 314ac4c73f..5cd756ba1b 100644
--- a/src/backends/cl/ClLayerSupport.hpp
+++ b/src/backends/cl/ClLayerSupport.hpp
@@ -5,6 +5,7 @@
#pragma once
#include <armnn/DescriptorsFwd.hpp>
+#include <armnn/Optional.hpp>
#include <armnn/Types.hpp>
#include <armnn/Tensor.hpp>
#include <armnn/ArmNN.hpp>
@@ -18,19 +19,19 @@ class ClLayerSupport : public ILayerSupport
};
bool IsClDirectConvolution2dSupported(const TensorInfo& weightInfo, const Convolution2dDescriptor& desc);
-bool IsClDepthwiseConvolution2dDescParamsSupported(std::string* reasonIfUnsupported,
+bool IsClDepthwiseConvolution2dDescParamsSupported(Optional<std::string&> reasonIfUnsupported,
const DepthwiseConvolution2dDescriptor& parameters,
const TensorInfo& weights);
bool IsActivationSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const ActivationDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsAdditionSupportedCl(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsBatchNormalizationSupportedCl(const TensorInfo& input,
const TensorInfo& output,
@@ -39,130 +40,143 @@ bool IsBatchNormalizationSupportedCl(const TensorInfo& input,
const TensorInfo& beta,
const TensorInfo& gamma,
const BatchNormalizationDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsConstantSupportedCl(const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsConvolution2dSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const Convolution2dDescriptor& descriptor,
const TensorInfo& weights,
const Optional<TensorInfo>& biases,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsDepthwiseConvolutionSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const DepthwiseConvolution2dDescriptor& descriptor,
const TensorInfo& weights,
const Optional<TensorInfo>& biases,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsDivisionSupportedCl(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsSubtractionSupportedCl(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsFullyConnectedSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const TensorInfo& weights,
const TensorInfo& biases,
const FullyConnectedDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsInputSupportedCl(const TensorInfo& input,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsL2NormalizationSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const L2NormalizationDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
-
-bool IsLstmSupportedCl(const TensorInfo& input, const TensorInfo& outputStateIn,
- const TensorInfo& cellStateIn, const TensorInfo& scratchBuffer,
- const TensorInfo& outputStateOut, const TensorInfo& cellStateOut,
- const TensorInfo& output, const LstmDescriptor& descriptor,
- const TensorInfo& inputToForgetWeights, const TensorInfo& inputToCellWeights,
- const TensorInfo& inputToOutputWeights, const TensorInfo& recurrentToForgetWeights,
- const TensorInfo& recurrentToCellWeights, const TensorInfo& recurrentToOutputWeights,
- const TensorInfo& forgetGateBias, const TensorInfo& cellBias,
- const TensorInfo& outputGateBias, const TensorInfo* inputToInputWeights,
- const TensorInfo* recurrentToInputWeights, const TensorInfo* cellToInputWeights,
- const TensorInfo* inputGateBias, const TensorInfo* projectionWeights,
- const TensorInfo* projectionBias, const TensorInfo* cellToForgetWeights,
- const TensorInfo* cellToOutputWeights, std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
+
+bool IsLstmSupportedCl(const TensorInfo& input,
+ const TensorInfo& outputStateIn,
+ const TensorInfo& cellStateIn,
+ const TensorInfo& scratchBuffer,
+ const TensorInfo& outputStateOut,
+ const TensorInfo& cellStateOut,
+ const TensorInfo& output,
+ const LstmDescriptor& descriptor,
+ const TensorInfo& inputToForgetWeights,
+ const TensorInfo& inputToCellWeights,
+ const TensorInfo& inputToOutputWeights,
+ const TensorInfo& recurrentToForgetWeights,
+ const TensorInfo& recurrentToCellWeights,
+ const TensorInfo& recurrentToOutputWeights,
+ const TensorInfo& forgetGateBias,
+ const TensorInfo& cellBias,
+ const TensorInfo& outputGateBias,
+ const TensorInfo* inputToInputWeights,
+ const TensorInfo* recurrentToInputWeights,
+ const TensorInfo* cellToInputWeights,
+ const TensorInfo* inputGateBias,
+ const TensorInfo* projectionWeights,
+ const TensorInfo* projectionBias,
+ const TensorInfo* cellToForgetWeights,
+ const TensorInfo* cellToOutputWeights,
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsMergerSupportedCl(const std::vector<const TensorInfo*> inputs,
const OriginsDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsMultiplicationSupportedCl(const TensorInfo& input0,
const TensorInfo& input1,
const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsNormalizationSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const NormalizationDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsOutputSupportedCl(const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
-
-bool IsPadSupportedCl(const TensorInfo& input,
- const TensorInfo& output,
- const PadDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsPermuteSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const PermuteDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsPooling2dSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const Pooling2dDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsResizeBilinearSupportedCl(const TensorInfo& input,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsSoftmaxSupportedCl(const TensorInfo& input,
const TensorInfo& output,
const SoftmaxDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsSplitterSupportedCl(const TensorInfo& input,
const ViewsDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsFakeQuantizationSupportedCl(const TensorInfo& input,
const FakeQuantizationDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsReshapeSupportedCl(const TensorInfo& input,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsFloorSupportedCl(const TensorInfo& input,
const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
-
-bool IsMeanSupportedCl(const TensorInfo& input,
- const TensorInfo& output,
- const MeanDescriptor& descriptor,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsConvertFp16ToFp32SupportedCl(const TensorInfo& input,
const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
bool IsConvertFp32ToFp16SupportedCl(const TensorInfo& input,
const TensorInfo& output,
- std::string* reasonIfUnsupported = nullptr);
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
+
+bool IsMeanSupportedCl(const TensorInfo& input,
+ const TensorInfo& output,
+ const MeanDescriptor& descriptor,
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
+
+bool IsPadSupportedCl(const TensorInfo& input,
+ const TensorInfo& output,
+ const PadDescriptor& descriptor,
+ Optional<std::string&> reasonIfUnsupported = EmptyOptional());
}
diff --git a/src/backends/cl/workloads/ClAdditionWorkload.cpp b/src/backends/cl/workloads/ClAdditionWorkload.cpp
index aa032e872c..ec0dd30275 100644
--- a/src/backends/cl/workloads/ClAdditionWorkload.cpp
+++ b/src/backends/cl/workloads/ClAdditionWorkload.cpp
@@ -37,10 +37,9 @@ void ClAdditionWorkload<T...>::Execute() const
m_Layer.run();
}
-bool ClAdditionValidate(const TensorInfo& input0,
- const TensorInfo& input1,
- const TensorInfo& output,
- std::string* reasonIfUnsupported)
+arm_compute::Status ClAdditionValidate(const TensorInfo& input0,
+ const TensorInfo& input1,
+ const TensorInfo& output)
{
const arm_compute::TensorInfo aclInput0Info = BuildArmComputeTensorInfo(input0);
const arm_compute::TensorInfo aclInput1Info = BuildArmComputeTensorInfo(input1);
@@ -51,13 +50,7 @@ bool ClAdditionValidate(const TensorInfo& input0,
&aclOutputInfo,
g_AclConvertPolicy);
- const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK);
- if (!supported && reasonIfUnsupported)
- {
- *reasonIfUnsupported = aclStatus.error_description();
- }
-
- return supported;
+ return aclStatus;
}
} //namespace armnn
diff --git a/src/backends/cl/workloads/ClAdditionWorkload.hpp b/src/backends/cl/workloads/ClAdditionWorkload.hpp
index 3e4ee26793..c5e6affac8 100644
--- a/src/backends/cl/workloads/ClAdditionWorkload.hpp
+++ b/src/backends/cl/workloads/ClAdditionWorkload.hpp
@@ -24,8 +24,7 @@ private:
mutable arm_compute::CLArithmeticAddition m_Layer;
};
-bool ClAdditionValidate(const TensorInfo& input0,
- const TensorInfo& input1,
- const TensorInfo& output,
- std::string* reasonIfUnsupported);
+arm_compute::Status ClAdditionValidate(const TensorInfo& input0,
+ const TensorInfo& input1,
+ const TensorInfo& output);
} //namespace armnn
diff --git a/src/backends/cl/workloads/ClConvertFp16ToFp32Workload.cpp b/src/backends/cl/workloads/ClConvertFp16ToFp32Workload.cpp
index e7663b4ca4..2c9a0e1fc2 100644
--- a/src/backends/cl/workloads/ClConvertFp16ToFp32Workload.cpp
+++ b/src/backends/cl/workloads/ClConvertFp16ToFp32Workload.cpp
@@ -32,19 +32,15 @@ void ClConvertFp16ToFp32Workload::Execute() const
m_Layer.run();
}
-arm_compute::Status ClConvertFp16ToFp32WorkloadValidate(const TensorInfo& input,
- const TensorInfo& output,
- std::string* reasonIfUnsupported)
+arm_compute::Status ClConvertFp16ToFp32WorkloadValidate(const TensorInfo& input, const TensorInfo& output)
{
if (input.GetDataType() != DataType::Float16)
{
- *reasonIfUnsupported = "Input should be Float16";
- return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, *reasonIfUnsupported);
+ return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, "Input should be Float16");
}
if (output.GetDataType() != DataType::Float32)
{
- *reasonIfUnsupported = "Output should be Float32";
- return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, *reasonIfUnsupported);
+ return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, "Output should be Float32");
}
const arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(input);
@@ -53,12 +49,6 @@ arm_compute::Status ClConvertFp16ToFp32WorkloadValidate(const TensorInfo& input,
const arm_compute::Status aclStatus = arm_compute::CLDepthConvertLayer::validate(
&aclInputInfo, &aclOutputInfo, g_AclConvertPolicy, 0);
- const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK);
- if (!supported && reasonIfUnsupported)
- {
- *reasonIfUnsupported = aclStatus.error_description();
- }
-
return aclStatus;
}
diff --git a/src/backends/cl/workloads/ClConvertFp16ToFp32Workload.hpp b/src/backends/cl/workloads/ClConvertFp16ToFp32Workload.hpp
index b6447488f7..f5f230d869 100644
--- a/src/backends/cl/workloads/ClConvertFp16ToFp32Workload.hpp
+++ b/src/backends/cl/workloads/ClConvertFp16ToFp32Workload.hpp
@@ -23,8 +23,6 @@ private:
mutable arm_compute::CLDepthConvertLayer m_Layer;
};
-arm_compute::Status ClConvertFp16ToFp32WorkloadValidate(const TensorInfo& input,
- const TensorInfo& output,
- std::string* reasonIfUnsupported);
+arm_compute::Status ClConvertFp16ToFp32WorkloadValidate(const TensorInfo& input, const TensorInfo& output);
} //namespace armnn
diff --git a/src/backends/cl/workloads/ClConvertFp32ToFp16Workload.cpp b/src/backends/cl/workloads/ClConvertFp32ToFp16Workload.cpp
index 2ae4adc424..6758180a6e 100644
--- a/src/backends/cl/workloads/ClConvertFp32ToFp16Workload.cpp
+++ b/src/backends/cl/workloads/ClConvertFp32ToFp16Workload.cpp
@@ -32,19 +32,15 @@ void ClConvertFp32ToFp16Workload::Execute() const
m_Layer.run();
}
-arm_compute::Status ClConvertFp32ToFp16WorkloadValidate(const TensorInfo& input,
- const TensorInfo& output,
- std::string* reasonIfUnsupported)
+arm_compute::Status ClConvertFp32ToFp16WorkloadValidate(const TensorInfo& input, const TensorInfo& output)
{
if (input.GetDataType() != DataType::Float32)
{
- *reasonIfUnsupported = "Input should be Float32";
- return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, *reasonIfUnsupported);
+ return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, "Input should be Float32");
}
if (output.GetDataType() != DataType::Float16)
{
- *reasonIfUnsupported = "Output should be Float16";
- return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, *reasonIfUnsupported);
+ return arm_compute::Status(arm_compute::ErrorCode::RUNTIME_ERROR, "Output should be Float16");
}
const arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(input);
@@ -53,12 +49,6 @@ arm_compute::Status ClConvertFp32ToFp16WorkloadValidate(const TensorInfo& input,
const arm_compute::Status aclStatus = arm_compute::CLDepthConvertLayer::validate(
&aclInputInfo, &aclOutputInfo, g_AclConvertPolicy, 0);
- const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK);
- if (!supported && reasonIfUnsupported)
- {
- *reasonIfUnsupported = aclStatus.error_description();
- }
-
return aclStatus;
}
diff --git a/src/backends/cl/workloads/ClConvertFp32ToFp16Workload.hpp b/src/backends/cl/workloads/ClConvertFp32ToFp16Workload.hpp
index 95d19905d7..28d0bfa9b0 100644
--- a/src/backends/cl/workloads/ClConvertFp32ToFp16Workload.hpp
+++ b/src/backends/cl/workloads/ClConvertFp32ToFp16Workload.hpp
@@ -23,8 +23,6 @@ private:
mutable arm_compute::CLDepthConvertLayer m_Layer;
};
-arm_compute::Status ClConvertFp32ToFp16WorkloadValidate(const TensorInfo& input,
- const TensorInfo& output,
- std::string* reasonIfUnsupported);
+arm_compute::Status ClConvertFp32ToFp16WorkloadValidate(const TensorInfo& input, const TensorInfo& output);
} //namespace armnn
diff --git a/src/backends/cl/workloads/ClPadWorkload.cpp b/src/backends/cl/workloads/ClPadWorkload.cpp
index 45dc5e8be7..89b0d8fde9 100644
--- a/src/backends/cl/workloads/ClPadWorkload.cpp
+++ b/src/backends/cl/workloads/ClPadWorkload.cpp
@@ -35,10 +35,9 @@ void ClPadWorkload<T...>::Execute() const
m_Layer.run();
}
-bool ClPadValidate(const TensorInfo& input,
- const TensorInfo& output,
- const PadDescriptor& descriptor,
- std::string* reasonIfUnsupported)
+arm_compute::Status ClPadValidate(const TensorInfo& input,
+ const TensorInfo& output,
+ const PadDescriptor& descriptor)
{
const arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(input);
const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output);
@@ -48,13 +47,7 @@ bool ClPadValidate(const TensorInfo& input,
&aclOutputInfo,
padList);
- const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK);
- if (!supported && reasonIfUnsupported)
- {
- *reasonIfUnsupported = aclStatus.error_description();
- }
-
- return supported;
+ return aclStatus;
}
} // namespace armnn
diff --git a/src/backends/cl/workloads/ClPadWorkload.hpp b/src/backends/cl/workloads/ClPadWorkload.hpp
index a7ad6670a7..97f57fdeac 100644
--- a/src/backends/cl/workloads/ClPadWorkload.hpp
+++ b/src/backends/cl/workloads/ClPadWorkload.hpp
@@ -23,10 +23,9 @@ private:
mutable arm_compute::CLPadLayer m_Layer;
};
-bool ClPadValidate(const TensorInfo& input,
- const TensorInfo& output,
- const PadDescriptor& descriptor,
- std::string* reasonIfUnsupported);
+arm_compute::Status ClPadValidate(const TensorInfo& input,
+ const TensorInfo& output,
+ const PadDescriptor& descriptor);
} //namespace armnn
diff --git a/src/backends/cl/workloads/ClSubtractionWorkload.cpp b/src/backends/cl/workloads/ClSubtractionWorkload.cpp
index 8efed94293..1967fae354 100644
--- a/src/backends/cl/workloads/ClSubtractionWorkload.cpp
+++ b/src/backends/cl/workloads/ClSubtractionWorkload.cpp
@@ -35,10 +35,9 @@ void ClSubtractionWorkload::Execute() const
m_Layer.run();
}
-bool ClSubtractionValidate(const TensorInfo& input0,
- const TensorInfo& input1,
- const TensorInfo& output,
- std::string* reasonIfUnsupported)
+arm_compute::Status ClSubtractionValidate(const TensorInfo& input0,
+ const TensorInfo& input1,
+ const TensorInfo& output)
{
const arm_compute::TensorInfo aclInput0Info = BuildArmComputeTensorInfo(input0);
const arm_compute::TensorInfo aclInput1Info = BuildArmComputeTensorInfo(input1);
@@ -49,13 +48,7 @@ bool ClSubtractionValidate(const TensorInfo& input0,
&aclOutputInfo,
g_AclConvertPolicy);
- const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK);
- if (!supported && reasonIfUnsupported)
- {
- *reasonIfUnsupported = aclStatus.error_description();
- }
-
- return supported;
+ return aclStatus;
}
} //namespace armnn
diff --git a/src/backends/cl/workloads/ClSubtractionWorkload.hpp b/src/backends/cl/workloads/ClSubtractionWorkload.hpp
index 7dd608bf8a..3a4210da07 100644
--- a/src/backends/cl/workloads/ClSubtractionWorkload.hpp
+++ b/src/backends/cl/workloads/ClSubtractionWorkload.hpp
@@ -23,8 +23,7 @@ private:
mutable arm_compute::CLArithmeticSubtraction m_Layer;
};
-bool ClSubtractionValidate(const TensorInfo& input0,
- const TensorInfo& input1,
- const TensorInfo& output,
- std::string* reasonIfUnsupported);
+arm_compute::Status ClSubtractionValidate(const TensorInfo& input0,
+ const TensorInfo& input1,
+ const TensorInfo& output);
} //namespace armnn
generated by cgit v1.2.1 (git 2.23.0) at 2024-06-29 22:51:12 +0000