COMPMID-2406: Create a new GEMMLowpQuantizeDownInt32ToInt16ScaleKernel for NEON

Change-Id: I3f3e247728fd6dafca066e41835f0ef9442d9b7a
Signed-off-by: giuros01 <giuseppe.rossini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1379
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
diff --git a/tests/validation/NEON/GEMMLowp.cpp b/tests/validation/NEON/GEMMLowp.cpp
index f0460b4..2f604c9 100644
--- a/tests/validation/NEON/GEMMLowp.cpp
+++ b/tests/validation/NEON/GEMMLowp.cpp
@@ -294,6 +294,9 @@
 using NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointFixture =
     GEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointValidationFixture<Tensor, Accessor, NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint>;
 
+using NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointFixture =
+    GEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointValidationFixture<Tensor, Accessor, NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint>;
+
 // *INDENT-OFF*
 // clang-format off
 DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(
@@ -404,6 +407,117 @@
 TEST_SUITE_END() // BoundedReLu
 
 TEST_SUITE_END() // QuantizeDownInt32ToUint8ScaleByFixedPoint
+
+TEST_SUITE(QuantizeDownInt32ToInt16ScaleByFixedPoint)
+
+const auto quantize_down_int32_to_int16_scale_by_fixedpoint_cases = framework::dataset::make("result_fixedpoint_multiplier", 254601600, 254601602) * framework::dataset::make("result_shift", 1,
+                                                                    2)
+                                                                    * framework::dataset::make("min", 0) * framework::dataset::make("max", 0) * framework::dataset::make("addBias", { false, true });
+
+const auto quantize_down_int32_to_int16_scale_by_fixedpoint_relu_cases = framework::dataset::make("result_fixedpoint_multiplier", 254601600, 254601602) * framework::dataset::make("result_shift", 1,
+                                                                         2)
+                                                                         * framework::dataset::make("min", -2, 0) * framework::dataset::make("max", 1, 3) * framework::dataset::make("addBias", { false, true });
+
+using NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointFixture =
+    GEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointValidationFixture<Tensor, Accessor, NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint>;
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(zip(
+    framework::dataset::make("InputAInfo", { TensorInfo(TensorShape(21U, 13U), 1, DataType::S32), // Input not a multiple of 16
+                                             TensorInfo(TensorShape(21U, 13U), 1, DataType::S32), // Invalid min and max
+                                             TensorInfo(TensorShape(20U, 13U), 1, DataType::S32), // Wrong output data type
+                                          }),
+    framework::dataset::make("InputBInfo",{ TensorInfo(TensorShape(21U), 1, DataType::S32),
+                                            TensorInfo(TensorShape(21U), 1, DataType::S32),
+                                            TensorInfo(TensorShape(20U), 1, DataType::S32),
+                                          })),
+    framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(21U, 13U), 1, DataType::QSYMM16),
+                                            TensorInfo(TensorShape(21U, 13U), 1, DataType::QSYMM16),
+                                            TensorInfo(TensorShape(20U, 13U), 1, DataType::S32),
+                                           })),
+    framework::dataset::make("Min",{        -205,
+                                            -60000,
+                                            -180,
+                                           })),
+    framework::dataset::make("Max",{        205,
+                                            60000,
+                                            180,
+                                           })),
+    framework::dataset::make("Expected", { true, false, false })),
+    a_info, b_info, output_info, min, max, expected)
+{
+    // Lock tensors
+    Status status =  NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint::validate(&a_info.clone()->set_is_resizable(false),
+                                                                                 &b_info.clone()->set_is_resizable(false),
+                                                                                 &output_info.clone()->set_is_resizable(false),
+                                                                                 min,
+                                                                                 max);
+    ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(datasets::SmallShapes(),
+                                                                   quantize_down_int32_to_int16_scale_by_fixedpoint_cases),
+               shape, result_fixedpoint_multiplier, result_shift, min, max, add_bias)
+{
+    TensorShape shape_bias(shape[0]);
+
+    // Create tensors
+    Tensor in   = create_tensor<Tensor>(shape, DataType::S32);
+    Tensor bias = create_tensor<Tensor>(shape_bias, DataType::S32);
+    Tensor out  = create_tensor<Tensor>(shape, DataType::QSYMM16);
+
+    ARM_COMPUTE_EXPECT(in.info()->is_resizable(), framework::LogLevel::ERRORS);
+    ARM_COMPUTE_EXPECT(bias.info()->is_resizable(), framework::LogLevel::ERRORS);
+    ARM_COMPUTE_EXPECT(out.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+    // Create and configure function
+    NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint output_stage;
+    output_stage.configure(&in, add_bias ? &bias : nullptr, &out, result_fixedpoint_multiplier, result_shift, min, max);
+
+    // Validate valid region input and output
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+    validate(in.info()->valid_region(), valid_region);
+    validate(out.info()->valid_region(), valid_region);
+
+    // Validate valid region bias
+    if(add_bias)
+    {
+        const ValidRegion valid_region_bias = shape_to_valid_region(shape_bias);
+        validate(bias.info()->valid_region(), valid_region_bias);
+    }
+
+    // Validate padding
+    const PaddingSize padding(0);
+    validate(in.info()->padding(), padding);
+    validate(out.info()->padding(), padding);
+
+    if(add_bias)
+    {
+        validate(bias.info()->padding(), padding);
+    }
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointFixture, framework::DatasetMode::ALL, combine(datasets::SmallShapes(),
+                       quantize_down_int32_to_int16_scale_by_fixedpoint_cases))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+TEST_SUITE(BoundedReLu)
+FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointFixture, framework::DatasetMode::ALL, combine(datasets::SmallShapes(),
+                       quantize_down_int32_to_int16_scale_by_fixedpoint_relu_cases))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+TEST_SUITE_END() // BoundedReLu
+
+TEST_SUITE_END() // QuantizeDownInt32ToInt16ScaleByFixedPoint
+
 TEST_SUITE_END() // OutputStage
 
 TEST_SUITE_END() // GEMMLowp