IVGCVSW-5842 Remove cross-wiring in depthwise
* Reading tensor infos won't allow a permutation vector anymore.
The permutation only changed the quantization dimension not the
shape and was therefore misleading
* The permutation of the full tensor info is now performed in
armnnUtils::Permuted
* Changed TfLite Parser depthwise parsing function
* Added unit tests to TfLite Parser with more random data
* Changed TfLite Delegate depthwise parsing function
* Added unit test to the delegate with per channel quantization
!android-nn-driver:5412
Signed-off-by: Jan Eilers <jan.eilers@arm.com>
Change-Id: I1f985ee69547bcaf16a72201e00a6b6fe1ef9a97
diff --git a/src/armnn/test/UtilsTests.cpp b/src/armnn/test/UtilsTests.cpp
index f0198cb..a813fea 100644
--- a/src/armnn/test/UtilsTests.cpp
+++ b/src/armnn/test/UtilsTests.cpp
@@ -249,22 +249,24 @@
BOOST_AUTO_TEST_CASE(PermuteQuantizationDim)
{
- std::vector<float> scales;
+ std::vector<float> scales {1.0f, 1.0f};
// Set QuantizationDim to be index 1
- const armnn::TensorInfo info({ 1, 2, 3, 4 }, armnn::DataType::Float32, scales, 1U);
- BOOST_CHECK(info.GetQuantizationDim().value() == 1U);
+ const armnn::TensorInfo perChannelInfo({ 1, 2, 3, 4 }, armnn::DataType::Float32, scales, 1U);
+ BOOST_CHECK(perChannelInfo.GetQuantizationDim().value() == 1U);
// Permute so that index 1 moves to final index i.e. index 3
armnn::PermutationVector mappings({ 0, 3, 2, 1 });
- auto permutedPerChannel = armnnUtils::Permuted(info, mappings, true);
- auto permuted = armnnUtils::Permuted(info, mappings);
+ auto permutedPerChannel = armnnUtils::Permuted(perChannelInfo, mappings);
// Check that QuantizationDim is in index 3
BOOST_CHECK(permutedPerChannel.GetQuantizationDim().value() == 3U);
- // Check previous implementation unchanged
- BOOST_CHECK(permuted.GetQuantizationDim().value() == 1U);
+ // Even if there is only a single scale the quantization dim still exists and needs to be permuted
+ std::vector<float> scale {1.0f};
+ const armnn::TensorInfo perChannelInfo1({ 1, 2, 3, 4 }, armnn::DataType::Float32, scale, 1U);
+ auto permuted = armnnUtils::Permuted(perChannelInfo1, mappings);
+ BOOST_CHECK(permuted.GetQuantizationDim().value() == 3U);
}
#if defined(ARMNNREF_ENABLED)
diff --git a/src/armnnTfLiteParser/TfLiteParser.cpp b/src/armnnTfLiteParser/TfLiteParser.cpp
index a68839c..9b1fa90 100644
--- a/src/armnnTfLiteParser/TfLiteParser.cpp
+++ b/src/armnnTfLiteParser/TfLiteParser.cpp
@@ -359,7 +359,6 @@
armnn::TensorInfo ToTensorInfo(TfLiteParserImpl::TensorRawPtr tensorPtr,
const std::vector<unsigned int>& shapes,
- const armnn::PermutationVector& dimensionMappings = {0, 1, 2, 3},
const bool outputTensor = false)
{
armnn::DataType type;
@@ -472,8 +471,7 @@
armnn::TensorInfo result(tensorShape,
type,
quantizationScales,
- dimensionMappings[armnn::numeric_cast<unsigned int>(
- tensorPtr->quantization->quantized_dimension)]);
+ armnn::numeric_cast<unsigned int>(tensorPtr->quantization->quantized_dimension));
return result;
}
}
@@ -493,19 +491,17 @@
}
}
-armnn::TensorInfo ToTensorInfo(TfLiteParserImpl::TensorRawPtr tensorPtr,
- const armnn::PermutationVector& dimensionMappings = {0, 1, 2, 3})
+armnn::TensorInfo ToTensorInfo(TfLiteParserImpl::TensorRawPtr tensorPtr)
{
auto const & dimensions = AsUnsignedVector(tensorPtr->shape);
- return ToTensorInfo(tensorPtr, dimensions, dimensionMappings);
+ return ToTensorInfo(tensorPtr, dimensions);
}
armnn::TensorInfo ToTensorInfo(TfLiteParserImpl::TensorRawPtr tensorPtr,
const bool outputTensor)
{
auto const & dimensions = AsUnsignedVector(tensorPtr->shape);
- const armnn::PermutationVector& dimensionMappings = {0, 1, 2, 3};
- return ToTensorInfo(tensorPtr, dimensions, dimensionMappings, outputTensor);
+ return ToTensorInfo(tensorPtr, dimensions, outputTensor);
}
template<typename T>
@@ -1013,7 +1009,7 @@
PermutationVector permutationVector{ 2, 3, 1, 0 }; // [H, W, I, M] -> [M, I, H, W]
armnn::TensorInfo inputTensorInfo = ToTensorInfo(inputs[0]);
- armnn::TensorInfo filterTensorInfo = ToTensorInfo(inputs[1], permutationVector);
+ armnn::TensorInfo filterTensorInfo = ToTensorInfo(inputs[1]);
// Assuming input is NHWC
unsigned int inputHeight = inputTensorInfo.GetShape()[1];
diff --git a/src/armnnTfLiteParser/test/DepthwiseConvolution2D.cpp b/src/armnnTfLiteParser/test/DepthwiseConvolution2D.cpp
index 7380d88..95ad2d5 100644
--- a/src/armnnTfLiteParser/test/DepthwiseConvolution2D.cpp
+++ b/src/armnnTfLiteParser/test/DepthwiseConvolution2D.cpp
@@ -225,19 +225,19 @@
struct DepthwiseConvolution2dFixture2 : public ParserFlatbuffersFixture
{
explicit DepthwiseConvolution2dFixture2(const std::string& inputShape,
- const std::string& outputShape,
- const std::string& filterShape,
- const std::string& filterData,
- const std::string& strides,
- const std::string& paddingType,
- const std::string biasShape = "",
- const std::string biasData = "",
- const std::string filter_quant_min = "[ 0.0 ]",
- const std::string filter_quant_max = "[ 255.0 ]",
- const std::string filter_quant_scale = "[ 1.0 ]",
- const std::string filter_quant_zero_point = "[ 0 ]",
- const std::string filter_quant_axis = ""
- )
+ const std::string& outputShape,
+ const std::string& filterShape,
+ const std::string& filterData,
+ const std::string& strides,
+ const std::string& paddingType,
+ const std::string biasShape = "",
+ const std::string biasData = "",
+ const std::string filter_quant_min = "[ 0.0 ]",
+ const std::string filter_quant_max = "[ 255.0 ]",
+ const std::string filter_quant_scale = "[ 1.0 ]",
+ const std::string filter_quant_zero_point = "[ 0 ]",
+ const std::string filter_quant_axis = "",
+ const std::string output_scale = "[ 1.0 ]")
{
std::string inputTensors = "[ 0, 2 ]";
std::string biasTensor = "";
@@ -301,7 +301,7 @@
"quantization": {
"min": [ 0.0 ],
"max": [ 511.0 ],
- "scale": [ 1.0 ],
+ "scale": )" + output_scale + R"(,
"zero_point": [ 0 ],
}
},
@@ -381,12 +381,12 @@
: DepthwiseConvolution2dFixture2("[ 1, 3, 3, 3 ]", // inputShape
"[ 1, 3, 3, 3 ]", // outputShape
"[ 1, 3, 3, 3 ]", // filterShape
- "[ 9,8,7, 6,5,4, 3,2,1, 9,8,7, 6,5,4, 3,2,1, 9,8,7, 6,5,4, 3,2,1 ]", // filterData
+ "[ 9,8,7, 6,5,4, 3,2,1, 9,8,7, 6,5,4, 3,2,1, 9,8,7, 6,5,4, 3,2,1 ]", //filterData
"1", // stride w and h
"SAME", // padding type
"", // bias shape
"", // bias data
- "[ 0.0 ]", // filter quantization min values
+ "[ 0.0 ]", // filter quantization min values
"[ 255.0 ]", // filter quantization max values
"[ 1.0, 1.0, 1.0]", // filter quantization scales
"[ 0, 0, 0]", // filter quantization zero-points
@@ -582,4 +582,396 @@
9, 8, 7, 6, 5, 4, 3, 2, 1, 9, 8, 7, 6, 5, 4, 3});
}
+
+struct DepthwiseConvolution2dWeightsPerChannelQuant6Fixture : DepthwiseConvolution2dFixture2
+{
+ DepthwiseConvolution2dWeightsPerChannelQuant6Fixture()
+ : DepthwiseConvolution2dFixture2("[ 1, 4, 4, 4 ]", // inputShape
+ "[ 1, 4, 4, 16 ]", // outputShape
+ "[ 1, 2, 2, 16 ]", // filterShape
+ // filter data is [ 3,4,1,1,1,3,3,2,1,4,3,4,1,2,2,4,
+ // 2,0,3,1,0,2,4,3,4,3,0,1,3,4,4,1,
+ // 3,3,2,0,0,0,1,3,3,2,4,4,3,1,1,3,
+ // 1,0,0,2,3,0,1,1,4,2,2,1,2,3,2,0]
+ // quantized per channel with q_dim=3
+ "[12,20,10, 3, 4,15,30, 6, 4,20,30,12, 4,10,20,12,"
+ " 8, 0,30, 3, 0,10,40, 9,16,15, 0, 3,12,20,40, 3,"
+ " 12,15,20, 0, 0, 0,10, 9,12,10,40,12,12, 5,10, 9,"
+ " 4, 0, 0, 6,12, 0,10, 3,16,10,20, 3, 8,15,20, 0]",
+ "1", // stride w and h
+ "SAME", // padding type
+ "", // bias shape
+ "", // bias data
+ "[ 0.0 ]", // filter quantization min values
+ "[ 255.0 ]", // filter quantization max values
+ "[ 0.25, 0.2, 0.1, 0.333333333,"
+ "0.25, 0.2, 0.1, 0.333333333,"
+ "0.25, 0.2, 0.1, 0.333333333,"
+ "0.25, 0.2, 0.1, 0.333333333]", // filter quantization scales
+ "[ 0, 0, 0, 0]", // filter quantization zero-points
+ "3" // filter quantized axis
+ // (in case of per channel quantization)
+ )
+ {}
+};
+
+
+BOOST_FIXTURE_TEST_CASE(ParseDepthwiseConv2DFilterWeightsPerChannelQuant6,
+ DepthwiseConvolution2dWeightsPerChannelQuant6Fixture)
+{
+ RunTest<4, armnn::DataType::QAsymmS8>(
+ 0,
+ { 1,0,1,2,0,4,4,0,2,1,2,0,1,3,3,0,
+ 1,2,2,3,3,4,1,1,2,4,1,3,4,2,0,2,
+ 0,3,1,3,4,3,2,0,1,2,3,3,0,2,4,2,
+ 1,2,1,4,3,4,1,3,1,0,2,3,1,3,2,0},
+ { 9, 7, 3, 7,12, 8,22,22,27,22,13,17,13,10, 9,17,
+ 15, 9,12, 6,16,14,24,27,19,26,18,23, 9,10, 7, 3,
+ 18,14, 9,11, 7, 9,21,25,17,19,10,15,13, 9, 7, 9,
+ 15,16, 9, 1, 3, 9,11,12, 3,12, 9,12, 6, 2, 2, 6,
+ 13, 4,10,12,11,14,28,28,17,17,14,15,15,13,13,22,
+ 26,24,17, 7,10,20,33,31,23,17,17,16,16,23,20, 7,
+ 17,11,16, 6,10,16,24,22,26,18,23,20,22,23,21,23,
+ 12,16, 4, 4, 2, 6, 8,10,12, 8,16,16, 8, 6, 6,14,
+ 14, 3,14,10,15,15,27,25,16,14, 9,11,21,19,16,24,
+ 24,25,13, 7, 3,13,21,24,25,23,14,17,24,24,21,12,
+ 7, 7, 3, 3,11,10,17,13,33,32,21,26,18,17,17,23,
+ 3, 3, 2, 0, 2, 6, 9,13,10,20,20,24, 2, 4, 4, 8,
+ 9, 4,10, 4, 2,14,22,16, 5, 7, 3, 5,13,20,20,19,
+ 11,12, 6, 4, 4,12,12, 8, 9,10, 3, 6,12,18,18,15,
+ 5, 4, 4, 2, 0, 6,12, 9,10,14, 6,10, 3, 6, 6,12,
+ 3, 4, 1, 1, 3, 9, 9, 6, 2, 8, 6, 8, 0, 0, 0, 0});
+}
+
+
+struct DepthwiseConvolution2dWeightsPerChannelQuant1_1Fixture : DepthwiseConvolution2dFixture2
+{
+ DepthwiseConvolution2dWeightsPerChannelQuant1_1Fixture()
+ : DepthwiseConvolution2dFixture2("[ 1, 3, 3, 3 ]", // inputShape
+ "[ 1, 3, 3, 3 ]", // outputShape
+ "[ 1, 3, 3, 3 ]", // filterShape
+ // filterData is [ 1,4,0,2,4,3,1,0,1,
+ // 3,0,4,0,1,3,4,2,4,
+ // 3,0,3,4,4,0,3,4,2]
+ // quantized per channel with q_dim=3
+ "[ 4,20, 0, 8,20,30, 4, 0,10,12,"
+ " 0,40, 0, 5,30,16,10,40,12, 0,"
+ "30,16,20, 0,12,20,20]",
+ "1", // stride w and h
+ "SAME", // padding type
+ "", // bias shape
+ "", // bias data
+ "[ 0.0 ]", // filter quantization min values
+ "[ 255.0 ]", // filter quantization max values
+ "[ 0.25, 0.2, 0.1]", // filter quantization scales
+ "[ 0, 0, 0]", // filter quantization zero-points
+ "3" // filter quantized axis
+ // (in case of per channel quantization)
+ )
+ {}
+};
+
+
+BOOST_FIXTURE_TEST_CASE(ParseDepthwiseConv2DFilterWeightsPerChannelQuant1_1,
+ DepthwiseConvolution2dWeightsPerChannelQuant1_1Fixture)
+{
+ RunTest<4, armnn::DataType::QAsymmS8>(
+ 0,
+ { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1},
+ { 11,11, 9,17,11,16,10, 5,10,
+ 14,15,13,21,19,20,13,13,13,
+ 7, 7,11,11,11,15, 6, 9,10});
+}
+
+// Same with input different to 1
+struct DepthwiseConvolution2dWeightsPerChannelQuant1_2Fixture : DepthwiseConvolution2dFixture2
+{
+ DepthwiseConvolution2dWeightsPerChannelQuant1_2Fixture()
+ : DepthwiseConvolution2dFixture2("[ 1, 3, 3, 3 ]", // inputShape
+ "[ 1, 3, 3, 3 ]", // outputShape
+ "[ 1, 3, 3, 3 ]", // filterShape
+ // filterData is [ 1,4,0,2,4,3,1,0,1,
+ // 3,0,4,0,1,3,4,2,4,
+ // 3,0,3,4,4,0,3,4,2]
+ // quantized per channel with q_dim=3
+ "[ 4,20, 0, 8,20,30, 4, 0,10,12,"
+ " 0,40, 0, 5,30,16,10,40,12, 0,"
+ "30,16,20, 0,12,20,20]",
+ "1", // stride w and h
+ "SAME", // padding type
+ "", // bias shape
+ "", // bias data
+ "[ 0.0 ]", // filter quantization min values
+ "[ 255.0 ]", // filter quantization max values
+ "[ 0.25, 0.2, 0.1]", // filter quantization scales
+ "[ 0, 0, 0]", // filter quantization zero-points
+ "3" // filter quantized axis
+ // (in case of per channel quantization)
+ )
+ {}
+};
+
+
+BOOST_FIXTURE_TEST_CASE(ParseDepthwiseConv2DFilterWeightsPerChannelQuant1_2,
+ DepthwiseConvolution2dWeightsPerChannelQuant1_2Fixture)
+{
+ RunTest<4, armnn::DataType::QAsymmS8>(
+ 0,
+ { 3,2,0,0,4,3,0,1,2,
+ 0,1,3,0,4,2,2,2,3,
+ 2,4,3,2,0,4,3,4,0},
+ { 0,30,16,15,30,32, 8, 9,24,
+ 20,33,28,34,48,50,18,38,35,
+ 8, 8,36,20,28,33,10,28,25});
+}
+
+
+struct DepthwiseConvolution2dWeightsPerChannelQuant4_1Fixture : DepthwiseConvolution2dFixture2
+{
+ DepthwiseConvolution2dWeightsPerChannelQuant4_1Fixture()
+ : DepthwiseConvolution2dFixture2("[ 1, 4, 4, 4 ]", // inputShape
+ "[ 1, 4, 4, 16 ]", // outputShape
+ "[ 1, 2, 2, 16 ]", // filterShape
+ // filter data is [ 3,4,1,1,1,3,3,2,1,4,3,4,1,2,2,4,
+ // 2,0,3,1,0,2,4,3,4,3,0,1,3,4,4,1,
+ // 3,3,2,0,0,0,1,3,3,2,4,4,3,1,1,3,
+ // 1,0,0,2,3,0,1,1,4,2,2,1,2,3,2,0 ]
+ // quantized per channel with q_dim=3
+ "[12,20,10, 3, 4,15,30, 6, 4,20,30,13, 4,10,20,13,"
+ " 8, 0,30, 3, 0,10,40,10,16,15, 0, 3,12,20,40, 3,"
+ " 12,15,20, 0, 0, 0,10,10,12,10,40,13,12, 5,10,10,"
+ " 4, 0, 0, 6,12, 0,10, 3,16,10,20, 3, 8,15,20, 0]",
+ "1", // stride w and h
+ "SAME", // padding type
+ "", // bias shape
+ "", // bias data
+ "[ 0.0 ]", // filter quantization min values
+ "[ 255.0 ]", // filter quantization max values
+ "[ 0.25, 0.2, 0.1, 0.3,"
+ "0.25, 0.2, 0.1, 0.3,"
+ "0.25, 0.2, 0.1, 0.3,"
+ "0.25, 0.2, 0.1, 0.3]", // filter quantization scales
+ "[ 0, 0, 0, 0]", // filter quantization zero-points
+ "3" // filter quantized axis
+ // (in case of per channel quantization)
+ )
+ {}
+};
+
+
+BOOST_FIXTURE_TEST_CASE(ParseDepthwiseConv2DFilterWeightsPerChannelQuant4_1,
+ DepthwiseConvolution2dWeightsPerChannelQuant4_1Fixture)
+{
+ RunTest<4, armnn::DataType::QAsymmS8>(
+ 0,
+ { 1,1,1,1, 1,1,1,1, 1,1,1,1, 1,1,1,1,
+ 1,1,1,1, 1,1,1,1, 1,1,1,1, 1,1,1,1,
+ 1,1,1,1, 1,1,1,1, 1,1,1,1, 1,1,1,1,
+ 1,1,1,1, 1,1,1,1, 1,1,1,1, 1,1,1,1},
+ { 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 6, 7, 3, 1, 1, 3, 4, 5, 4, 6, 7, 8, 4, 3, 3, 7,
+ 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 6, 7, 3, 1, 1, 3, 4, 5, 4, 6, 7, 8, 4, 3, 3, 7,
+ 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 9, 7, 6, 4, 4, 5, 9, 9,12,11, 9,10, 9,10, 9, 8,
+ 6, 7, 3, 1, 1, 3, 4, 5, 4, 6, 7, 8, 4, 3, 3, 7,
+ 5, 4, 4, 2, 1, 5, 7, 5, 5, 7, 3, 5, 4, 6, 6, 5,
+ 5, 4, 4, 2, 1, 5, 7, 5, 5, 7, 3, 5, 4, 6, 6, 5,
+ 5, 4, 4, 2, 1, 5, 7, 5, 5, 7, 3, 5, 4, 6, 6, 5,
+ 3, 4, 1, 1, 1, 3, 3, 2, 1, 4, 3, 4, 1, 2, 2, 4});
+}
+
+
+
+struct DepthwiseConvolution2dWeightsPerChannelQuant4_2Fixture : DepthwiseConvolution2dFixture2
+{
+ DepthwiseConvolution2dWeightsPerChannelQuant4_2Fixture()
+ : DepthwiseConvolution2dFixture2("[ 1, 4, 4, 4 ]", // inputShape
+ "[ 1, 4, 4, 16 ]", // outputShape
+ "[ 1, 2, 2, 16 ]", // filterShape
+ // filter data is [ 3,4,1,1,1,3,3,2,1,4,3,4,1,2,2,4,
+ // 2,0,3,1,0,2,4,3,4,3,0,1,3,4,4,1,
+ // 3,3,2,0,0,0,1,3,3,2,4,4,3,1,1,3,
+ // 1,0,0,2,3,0,1,1,4,2,2,1,2,3,2,0 ]
+ // quantized per channel with q_dim=3
+ "[12,20,10, 3, 4,15,30, 6, 4,20,30,13, 4,10,20,13,"
+ " 8, 0,30, 3, 0,10,40,10,16,15, 0, 3,12,20,40, 3,"
+ " 12,15,20, 0, 0, 0,10,10,12,10,40,13,12, 5,10,10,"
+ " 4, 0, 0, 6,12, 0,10, 3,16,10,20, 3, 8,15,20, 0]",
+ "1", // stride w and h
+ "SAME", // padding type
+ "", // bias shape
+ "", // bias data
+ "[ 0.0 ]", // filter quantization min values
+ "[ 255.0 ]", // filter quantization max values
+ "[ 0.25, 0.2, 0.1, 0.3,"
+ "0.25, 0.2, 0.1, 0.3,"
+ "0.25, 0.2, 0.1, 0.3,"
+ "0.25, 0.2, 0.1, 0.3]", // filter quantization scales
+ "[ 0, 0, 0, 0]", // filter quantization zero-points
+ "3" // filter quantized axis
+ // (in case of per channel quantization)
+ )
+ {}
+};
+
+
+BOOST_FIXTURE_TEST_CASE(ParseDepthwiseConv2DFilterWeightsPerChannelQuant4_2,
+ DepthwiseConvolution2dWeightsPerChannelQuant4_2Fixture)
+{
+ RunTest<4, armnn::DataType::QAsymmS8>(
+ 0,
+ { 3,3,3,4, 4,4,0,0, 0,3,4,3, 0,2,2,3,
+ 3,0,3,0, 0,3,2,1, 4,1,2,2, 0,0,0,4,
+ 3,2,2,2, 2,1,0,4, 4,3,2,4, 3,2,0,0,
+ 4,1,4,4, 1,0,4,3, 3,2,0,3, 1,1,0,2},
+ { 26,21,21, 7,12,17,28,21,20,22,25,26, 6,11,10,16,
+ 16,16, 4,12, 7,18,28,27,30,20,12,14,16,19,17, 6,
+ 12,12, 8, 0, 3,13,18,15,18,26,20,26,26,32,28,21,
+ 0, 0, 0, 0, 2, 6, 6, 4, 2, 8, 6, 8,15,10,10,24,
+ 20,21, 9, 7, 3, 6,15,16,17,22,17,22,17,18,14, 7,
+ 18, 6,16,12,12,11,17,15,18,18,10,12,27,26,22,18,
+ 27,28,12,10, 7, 3, 8,13, 8,12,14,16,26,24,24,24,
+ 9, 9, 6, 0, 0, 0, 2, 6, 0, 0, 0, 0, 4, 8, 8,16,
+ 26,24,17, 7, 2, 8,11,10,30,24,30,28,32,33,30,24,
+ 20,11,16,12, 7, 9,17,13,20,14,16,18,31,36,33,29,
+ 28,25,19, 9, 6,13,20,19, 2, 8, 6, 8,17,17,15,25,
+ 12,15, 5, 3, 2, 6, 7, 7, 0, 0, 0, 0, 6, 2, 2, 6,
+ 14,16, 7, 5, 1, 3, 3, 2,20,28,12,20,13,20,20,19,
+ 9, 4,10, 4, 0, 4, 8, 6, 4,16,12,16,12,18,18,15,
+ 11,12, 6, 4, 2, 8,10, 7, 0, 0, 0, 0, 9,14,14,14,
+ 3, 4, 1, 1, 1, 3, 3, 2, 0, 0, 0, 0, 2, 4, 4, 8});
+}
+
+
+struct DepthwiseConvolution2dWeightsPerChannelQuant4_5Fixture : DepthwiseConvolution2dFixture2
+{
+ DepthwiseConvolution2dWeightsPerChannelQuant4_5Fixture()
+ : DepthwiseConvolution2dFixture2("[ 1, 4, 4, 4 ]", // inputShape
+ "[ 1, 4, 4, 16 ]", // outputShape
+ "[ 1, 2, 2, 16 ]", // filterShape
+ // filter data is [ 1, 4, 9, 16, 25, 36,
+ // 49, 64, 81, 100, 121, 144,
+ // 169, 196, 225, 256, 17, 36,
+ // 57, 80, 105, 132, 161, 192,
+ // 225, 260, 297, 336, 377, 420,
+ // 465, 512, 33, 68, 105, 144,
+ // 185, 228, 273, 320, 369, 420,
+ // 473, 528, 585, 644, 705, 768,
+ // 49, 100, 153, 208, 265, 324,
+ // 385, 448, 513, 580, 649, 720,
+ // 793, 868, 945,1024 ]
+ // quantized per channel with q_dim=3
+ "[ 1, 2, 3, 4, 5, 6, 7, 8, 9,10,11,12,13,14,15,16,"
+ " 17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,"
+ " 33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,"
+ "49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64]",
+ "1", // stride w and h
+ "SAME", // padding type
+ "", // bias shape
+ "", // bias data
+ "[ 0.0 ]", // filter quantization min values
+ "[ 255.0 ]", // filter quantization max values
+ "[1, 2, 3, 4, 5, 6, 7, 8, 9, 10,11,12,13,14,15,16]", // filter quantization scales
+ "[ 0, 0, 0, 0]", // filter quantization zero-points
+ "3", // filter quantized axis
+ // (in case of per channel quantization)
+ "[ 100.0 ]" // output scale
+ )
+ {}
+};
+
+// Test for depthwise_multiplier different to one (M > 1)
+BOOST_FIXTURE_TEST_CASE(ParseDepthwiseConv2DFilterWeightsPerChannelQuant4_5,
+ DepthwiseConvolution2dWeightsPerChannelQuant4_5Fixture)
+{
+ RunTest<4, armnn::DataType::QAsymmS8>(
+ 0,
+ { 1,1,1,2,2,2,1,2,1,2,2,1,2,2,1,1,1,1,1,1,1,2,2,2,
+ 1,2,2,2,1,1,1,2,1,1,1,1,2,1,2,1,2,1,1,2,1,2,1,1,
+ 1,2,2,1,2,2,1,1,2,1,2,1,1,2,1,2},
+ { 1, 2, 3, 5, 9,11,14,16,17,19,21,24,32,36,39,43,
+ 1, 2, 3, 4,11,14,17,20,22,26,29,33,34,38,42,46,
+ 1, 2, 3, 5, 8,11,13,16,16,18,21,24,33,36,39,43,
+ 0, 0, 1, 1, 2, 3, 3, 4, 4, 5, 5, 6,13,14,16,17,
+ 1, 3, 4, 6, 6, 8,10,12,19,22,24,27,23,25,28,30,
+ 1, 3, 5, 8, 7, 8,10,12,18,21,24,27,32,36,39,43,
+ 1, 2, 4, 5, 8,10,13,15,12,14,16,18,30,33,37,40,
+ 0, 0, 1, 1, 3, 4, 5, 7, 4, 5, 5, 6, 9,10,11,12,
+ 1, 3, 5, 7,10,12,15,17,17,20,23,25,19,21,23,25,
+ 2, 4, 6, 8, 7, 9,11,13,17,20,23,25,23,25,28,30,
+ 1, 2, 4, 6, 9,11,14,16,15,17,20,22,28,31,35,38,
+ 0, 0, 1, 1, 4, 5, 6, 7, 4, 5, 5, 6,13,14,16,17,
+ 0, 0, 1, 1, 2, 3, 4, 5, 3, 4, 5, 6, 5, 6, 6, 7,
+ 0, 0, 1, 1, 1, 2, 2, 3, 5, 6, 7, 8, 5, 6, 6, 7,
+ 0, 0, 0, 1, 2, 3, 3, 4, 3, 4, 5, 6, 9,10,11,12,
+ 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 1, 3, 3, 4, 5});
+}
+
+
+struct DepthwiseConvolution2dWeightsPerChannelQuant4_3_1Fixture : DepthwiseConvolution2dFixture2
+{
+ DepthwiseConvolution2dWeightsPerChannelQuant4_3_1Fixture()
+ : DepthwiseConvolution2dFixture2("[ 1, 4, 4, 4 ]", // inputShape
+ "[ 1, 4, 4, 16 ]", // outputShape
+ "[ 1, 2, 2, 16 ]", // filterShape
+ // filter data is [ 3,4,1,1,1,3,3,2,1,4,3,4,1,2,2,4,
+ // 2,0,3,1,0,2,4,3,4,3,0,1,3,4,4,1,
+ // 3,3,2,0,0,0,1,3,3,2,4,4,3,1,1,3,
+ // 1,0,0,2,3,0,1,1,4,2,2,1,2,3,2,0 ]
+ // quantized per channel with q_dim=3
+ "[12,20,10, 3, 2,24, 9,10, 5,16,30,12, 3,10, 4,32,"
+ " 8, 0,30, 3, 0,16,12,15,20,12, 0, 3, 9,20, 8, 8,"
+ " 12,15,20, 0, 0, 0, 3,15,15, 8,40,12, 9, 5, 2,24,"
+ " 4, 0, 0, 6, 6, 0, 3, 5,20, 8,20, 3, 6,15, 4, 0]",
+ "1", // stride w and h
+ "SAME", // padding type
+ "", // bias shape
+ "", // bias data
+ "[ 0.0 ]", // filter quantization min values
+ "[ 255.0 ]", // filter quantization max values
+ "[0.25, 0.2, 0.1, 0.3333333333, "
+ "0.5, 0.125, 0.33333333, 0.2, "
+ "0.2, 0.25, 0.1, 0.333333333, "
+ "0.3333333333, 0.2, 0.5, 0.125]", // filter quantization scales
+ "[ 0, 0, 0, 0]", // filter quantization zero-points
+ "3" // filter quantized axis
+ // (in case of per channel quantization)
+ )
+ {}
+};
+
+// Test for depthwise_multiplier different to one (M > 1)
+BOOST_FIXTURE_TEST_CASE(ParseDepthwiseConv2DFilterWeightsPerChannelQuant4_3_1,
+ DepthwiseConvolution2dWeightsPerChannelQuant4_3_1Fixture)
+{
+ RunTest<4, armnn::DataType::QAsymmS8>(
+ 0,
+ { 3,3,3,4, 4,4,0,0, 0,3,4,3, 0,2,2,3,
+ 3,0,3,0, 0,3,2,1, 4,1,2,2, 0,0,0,4,
+ 3,2,2,2, 2,1,0,4, 4,3,2,4, 3,2,0,0,
+ 4,1,4,4, 1,0,4,3, 3,2,0,3, 1,1,0,2},
+ { 26,21,21, 7,12,17,28,21,20,22,25,26, 6,11,10,16,
+ 16,16, 4,12, 7,18,28,27,30,20,12,14,16,19,17, 6,
+ 12,12, 8, 0, 3,13,18,15,18,26,20,26,26,32,28,21,
+ 0, 0, 0, 0, 2, 6, 6, 4, 2, 8, 6, 8,15,10,10,24,
+ 20,21, 9, 7, 3, 6,15,16,17,22,17,22,17,18,14, 7,
+ 18, 6,16,12,12,11,17,15,18,18,10,12,27,26,22,18,
+ 27,28,12,10, 7, 3, 8,13, 8,12,14,16,26,24,24,24,
+ 9, 9, 6, 0, 0, 0, 2, 6, 0, 0, 0, 0, 4, 8, 8,16,
+ 26,24,17, 7, 2, 8,11,10,30,24,30,28,32,33,30,24,
+ 20,11,16,12, 7, 9,17,13,20,14,16,18,31,36,33,29,
+ 28,25,19, 9, 6,13,20,19, 2, 8, 6, 8,17,17,15,25,
+ 12,15, 5, 3, 2, 6, 7, 7, 0, 0, 0, 0, 6, 2, 2, 6,
+ 14,16, 7, 5, 1, 3, 3, 2,20,28,12,20,13,20,20,19,
+ 9, 4,10, 4, 0, 4, 8, 6, 4,16,12,16,12,18,18,15,
+ 11,12, 6, 4, 2, 8,10, 7, 0, 0, 0, 0, 9,14,14,14,
+ 3, 4, 1, 1, 1, 3, 3, 2, 0, 0, 0, 0, 2, 4, 4, 8});
+}
+
BOOST_AUTO_TEST_SUITE_END()
diff --git a/src/armnnUtils/Permute.cpp b/src/armnnUtils/Permute.cpp
index 3770463..7d15f3c 100644
--- a/src/armnnUtils/Permute.cpp
+++ b/src/armnnUtils/Permute.cpp
@@ -113,14 +113,14 @@
}
armnn::TensorInfo Permuted(const armnn::TensorInfo& info,
- const armnn::PermutationVector& mappings,
- bool perChannelPermute)
+ const armnn::PermutationVector& mappings)
{
armnn::TensorInfo outInfo(info);
outInfo.SetShape(Permuted(info.GetShape(), mappings));
- // If TensorInfo has Per-Axis Quantization then permute QuantizationDim to mapping
- if (info.HasPerAxisQuantization() && perChannelPermute)
+ // If TensorInfo has Per-Axis Quantization then it also has a QuantizationDim which needs to
+ // be permuted according to the mapping
+ if (info.GetQuantizationDim().has_value())
{
outInfo.SetQuantizationDim(mappings[info.GetQuantizationDim().value()]);
}