COMPMID-1822: (Nightly) : 'CL/ArithmeticDivision mismatches

Change-Id: I14cea30ffa9ca735941b559bb272b8c476814a34
Reviewed-on: https://review.mlplatform.org/338
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <Anthony.barbier@arm.com>
Reviewed-by: Isabella Gottardi <isabella.gottardi@arm.com>

2 files changed, 94 insertions, 149 deletions

diff --git a/src/core/CL/kernels/CLElementwiseOperationKernel.cpp b/src/core/CL/kernels/CLElementwiseOperationKernel.cpp
index 5dc5b7e13f..37eeeb78bf 100644
--- a/src/core/CL/kernels/CLElementwiseOperationKernel.cpp
+++ b/src/core/CL/kernels/CLElementwiseOperationKernel.cpp
@@ -64,6 +64,29 @@ std::string generate_id_for_tuning_common(const std::string &kernel_name, const
     return config_id;
 }
 
+Status validate_arguments_with_division_rules(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(&input1, &input2, &output);
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&input1);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input1, 1, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&input1, &input2);
+
+    const TensorShape out_shape = TensorShape::broadcast_shape(input1.tensor_shape(), input2.tensor_shape());
+
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
+
+    // Validate in case of configured output
+    if(output.total_size() > 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&output, 1, DataType::F16, DataType::F32);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&input1, &output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, output.tensor_shape(), 0),
+                                        "Wrong shape for output");
+    }
+
+    return Status{};
+}
+
 Status validate_arguments_with_arithmetic_rules(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(&input1);
@@ -119,6 +142,26 @@ CLBuildOptions generate_build_options_with_arithmetic_rules(const ITensorInfo &i
     return build_opts;
 }
 
+std::pair<Status, Window> configure_window_arithmetic_common(const ValidRegion &valid_region, ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
+{
+    Window win        = calculate_max_window(valid_region, Steps(num_elems_processed_per_iteration));
+    Window win_input1 = win.broadcast_if_dimension_le_one(input1);
+    Window win_input2 = win.broadcast_if_dimension_le_one(input2);
+
+    AccessWindowHorizontal input1_access(&input1, 0, num_elems_processed_per_iteration);
+    AccessWindowHorizontal input2_access(&input2, 0, num_elems_processed_per_iteration);
+    AccessWindowHorizontal output_access(&output, 0, num_elems_processed_per_iteration);
+
+    bool window_changed = update_window_and_padding(win_input1, input1_access)
+                          || update_window_and_padding(win_input2, input2_access)
+                          || update_window_and_padding(win, output_access);
+
+    output_access.set_valid_region(win, valid_region);
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+
 std::pair<Status, Window> validate_and_configure_window_for_arithmetic_operators(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
 {
     const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(input1, input2);
@@ -140,22 +183,16 @@ std::pair<Status, Window> validate_and_configure_window_for_arithmetic_operators
         set_format_if_unknown(output, Format::F32);
     }
 
-    Window win        = calculate_max_window(valid_region, Steps(num_elems_processed_per_iteration));
-    Window win_input1 = win.broadcast_if_dimension_le_one(input1);
-    Window win_input2 = win.broadcast_if_dimension_le_one(input2);
-
-    AccessWindowHorizontal input1_access(&input1, 0, num_elems_processed_per_iteration);
-    AccessWindowHorizontal input2_access(&input2, 0, num_elems_processed_per_iteration);
-    AccessWindowHorizontal output_access(&output, 0, num_elems_processed_per_iteration);
-
-    bool window_changed = update_window_and_padding(win_input1, input1_access)
-                          || update_window_and_padding(win_input2, input2_access)
-                          || update_window_and_padding(win, output_access);
-
-    output_access.set_valid_region(win, valid_region);
+    return configure_window_arithmetic_common(valid_region, input1, input2, output);
+}
 
-    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
-    return std::make_pair(err, win);
+std::pair<Status, Window> validate_and_configure_window_for_division(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
+{
+    const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(input1, input2);
+    const TensorShape &out_shape    = broadcast_pair.first;
+    const ValidRegion &valid_region = broadcast_pair.second;
+    auto_init_if_empty(output, out_shape, 1, input1.data_type());
+    return configure_window_arithmetic_common(valid_region, input1, input2, output);
 }
 } // namespace
 
@@ -306,19 +343,44 @@ void CLArithmeticOperationKernel::configure(ArithmeticOperation op, const ICLTen
 
 Status CLArithmeticOperationKernel::validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output)
 {
-    ARM_COMPUTE_UNUSED(op);
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*input1, *input2, *output));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*input1->clone(), *input2->clone(), *output->clone()).first);
+    if(op == ArithmeticOperation::DIV)
+    {
+        // Division doesn't support integer arithmetic
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_division_rules(*input1, *input2, *output));
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_division(*input1->clone(), *input2->clone(), *output->clone()).first);
+    }
+    else
+    {
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*input1, *input2, *output));
+        ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*input1->clone(), *input2->clone(), *output->clone()).first);
+    }
+
     return Status{};
 }
 std::pair<Status, Window> CLArithmeticOperationKernel::validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
 {
-    return validate_and_configure_window_for_arithmetic_operators(input1, input2, output);
+    if(_op == ArithmeticOperation::DIV)
+    {
+        // Division doesn't support integer arithmetic
+        return validate_and_configure_window_for_division(input1, input2, output);
+    }
+    else
+    {
+        return validate_and_configure_window_for_arithmetic_operators(input1, input2, output);
+    }
 }
 Status CLArithmeticOperationKernel::validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
 {
-    return validate_arguments_with_arithmetic_rules(input1, input2, output);
+    if(_op == ArithmeticOperation::DIV)
+    {
+        // Division doesn't support integer arithmetic
+        return validate_arguments_with_division_rules(input1, input2, output);
+    }
+    else
+    {
+        return validate_arguments_with_arithmetic_rules(input1, input2, output);
+    }
 }
 
 CLBuildOptions CLArithmeticOperationKernel::generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
diff --git a/tests/validation/CL/ArithmeticDivision.cpp b/tests/validation/CL/ArithmeticDivision.cpp
index 87039d775f..330a999c24 100644
--- a/tests/validation/CL/ArithmeticDivision.cpp
+++ b/tests/validation/CL/ArithmeticDivision.cpp
@@ -48,13 +48,6 @@ RelativeTolerance<float> tolerance_fp16(0.001f);
 
 constexpr unsigned int num_elems_processed_per_iteration = 16;
 /** Input data sets **/
-const auto ArithmeticDivisionU8Dataset = combine(combine(framework::dataset::make("DataType", DataType::U8), framework::dataset::make("DataType", DataType::U8)), framework::dataset::make("DataType",
-                                                 DataType::U8));
-const auto ArithmeticDivisionQASYMM8Dataset = combine(combine(framework::dataset::make("DataType", DataType::QASYMM8), framework::dataset::make("DataType", DataType::QASYMM8)),
-                                                      framework::dataset::make("DataType",
-                                                                               DataType::QASYMM8));
-const auto ArithmeticDivisionS16Dataset = combine(combine(framework::dataset::make("DataType", { DataType::U8, DataType::S16 }), framework::dataset::make("DataType", DataType::S16)),
-                                                  framework::dataset::make("DataType", DataType::S16));
 const auto ArithmeticDivisionFP16Dataset = combine(combine(framework::dataset::make("DataType", DataType::F16), framework::dataset::make("DataType", DataType::F16)),
                                                    framework::dataset::make("DataType", DataType::F16));
 const auto ArithmeticDivisionFP32Dataset = combine(combine(framework::dataset::make("DataType", DataType::F32), framework::dataset::make("DataType", DataType::F32)),
@@ -67,25 +60,25 @@ TEST_SUITE(ArithmeticDivision)
 // *INDENT-OFF*
 // clang-format off
 DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
-               framework::dataset::make("Input1Info", { TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
-                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
-                                                        TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::U8),      // Window shrink
-                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),      // Invalid data type combination
+               framework::dataset::make("Input1Info", { TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32),      // Window shrink
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),      // Invalid data type combination
                                                         TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),     // Mismatching shapes
                                                       }),
-               framework::dataset::make("Input2Info",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
-                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
-                                                       TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::U8),
+               framework::dataset::make("Input2Info",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32),
                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),
                                                        TensorInfo(TensorShape(48U, 11U, 2U), 1, DataType::F32),
                                                      })),
-               framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),
-                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
-                                                       TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::U8),
-                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+               framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(27U, 13U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
                                                        TensorInfo(TensorShape(48U, 11U, 2U), 1, DataType::F32),
                                                      })),
-               framework::dataset::make("Expected", { true, true, false, false, false})),
+               framework::dataset::make("Expected", { true, false, false, false, false})),
                input1_info, input2_info, output_info, expected)
 {
     ARM_COMPUTE_EXPECT(bool(CLArithmeticDivision::validate(&input1_info.clone()->set_is_resizable(false), &input2_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false))) == expected, framework::LogLevel::ERRORS);
@@ -96,116 +89,6 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
 template <typename T>
 using CLArithmeticDivisionFixture = ArithmeticDivisionValidationFixture<CLTensor, CLAccessor, CLArithmeticDivision, T>;
 
-TEST_SUITE(U8)
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, framework::dataset::concat(datasets::SmallShapes(), datasets::LargeShapes()),
-               shape)
-{
-    // Create tensors
-    CLTensor ref_src1 = create_tensor<CLTensor>(shape, DataType::U8);
-    CLTensor ref_src2 = create_tensor<CLTensor>(shape, DataType::U8);
-    CLTensor dst      = create_tensor<CLTensor>(shape, DataType::U8);
-
-    // Create and Configure function
-    CLArithmeticDivision add;
-    add.configure(&ref_src1, &ref_src2, &dst);
-
-    // Validate valid region
-    const ValidRegion valid_region = shape_to_valid_region(shape);
-    validate(dst.info()->valid_region(), valid_region);
-
-    // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
-    validate(ref_src1.info()->padding(), padding);
-    validate(ref_src2.info()->padding(), padding);
-    validate(dst.info()->padding(), padding);
-}
-
-FIXTURE_DATA_TEST_CASE(RunSmall, CLArithmeticDivisionFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), ArithmeticDivisionU8Dataset))
-{
-    // Validate output
-    validate(CLAccessor(_target), _reference);
-}
-TEST_SUITE_END()
-
-template <typename T>
-using CLArithmeticDivisionQuantizedFixture = ArithmeticDivisionValidationQuantizedFixture<CLTensor, CLAccessor, CLArithmeticDivision, T>;
-
-TEST_SUITE(Quantized)
-TEST_SUITE(QASYMM8)
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, framework::dataset::concat(datasets::SmallShapes(), datasets::LargeShapes()),
-               shape)
-{
-    // Create tensors
-    CLTensor ref_src1 = create_tensor<CLTensor>(shape, DataType::QASYMM8);
-    CLTensor ref_src2 = create_tensor<CLTensor>(shape, DataType::QASYMM8);
-    CLTensor dst      = create_tensor<CLTensor>(shape, DataType::QASYMM8);
-
-    // Create and Configure function
-    CLArithmeticDivision add;
-    add.configure(&ref_src1, &ref_src2, &dst);
-
-    // Validate valid region
-    const ValidRegion valid_region = shape_to_valid_region(shape);
-    validate(dst.info()->valid_region(), valid_region);
-
-    // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
-    validate(ref_src1.info()->padding(), padding);
-    validate(ref_src2.info()->padding(), padding);
-    validate(dst.info()->padding(), padding);
-}
-
-FIXTURE_DATA_TEST_CASE(RunSmall, CLArithmeticDivisionQuantizedFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(combine(datasets::SmallShapes(),
-                       ArithmeticDivisionQASYMM8Dataset),
-                       framework::dataset::make("QuantizationInfo", { QuantizationInfo(5.f / 255.f, 20) })),
-                       framework::dataset::make("QuantizationInfo", { QuantizationInfo(2.f / 255.f, 10) })),
-                       framework::dataset::make("QuantizationInfo", { QuantizationInfo(1.f / 255.f, 5) }))
-
-                      )
-{
-    // Validate output
-    validate(CLAccessor(_target), _reference, tolerance_fp32, 0.01);
-}
-TEST_SUITE_END()
-TEST_SUITE_END()
-
-TEST_SUITE(S16)
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(framework::dataset::concat(datasets::SmallShapes(), datasets::LargeShapes()), framework::dataset::make("DataType", { DataType::U8, DataType::S16 })),
-               shape, data_type)
-{
-    // Create tensors
-    CLTensor ref_src1 = create_tensor<CLTensor>(shape, data_type);
-    CLTensor ref_src2 = create_tensor<CLTensor>(shape, DataType::S16);
-    CLTensor dst      = create_tensor<CLTensor>(shape, DataType::S16);
-
-    // Create and Configure function
-    CLArithmeticDivision add;
-    add.configure(&ref_src1, &ref_src2, &dst);
-
-    // Validate valid region
-    const ValidRegion valid_region = shape_to_valid_region(shape);
-    validate(dst.info()->valid_region(), valid_region);
-
-    // Validate padding
-    const PaddingSize padding = PaddingCalculator(shape.x(), num_elems_processed_per_iteration).required_padding();
-    validate(ref_src1.info()->padding(), padding);
-    validate(ref_src2.info()->padding(), padding);
-    validate(dst.info()->padding(), padding);
-}
-
-FIXTURE_DATA_TEST_CASE(RunSmall, CLArithmeticDivisionFixture<int16_t>, framework::DatasetMode::PRECOMMIT, combine(datasets::SmallShapes(), ArithmeticDivisionS16Dataset))
-{
-    // Validate output
-    validate(CLAccessor(_target), _reference);
-}
-
-FIXTURE_DATA_TEST_CASE(RunLarge, CLArithmeticDivisionFixture<int16_t>, framework::DatasetMode::NIGHTLY, combine(datasets::LargeShapes(), ArithmeticDivisionS16Dataset))
-{
-    // Validate output
-    validate(CLAccessor(_target), _reference);
-}
-TEST_SUITE_END()
-
 TEST_SUITE(Float)
 TEST_SUITE(FP16)
 FIXTURE_DATA_TEST_CASE(RunSmall, CLArithmeticDivisionFixture<half>, framework::DatasetMode::ALL, combine(datasets::SmallShapes(), ArithmeticDivisionFP16Dataset))