COMPMID-3365: Add support for U8 datatype to CropResize on CL

Change-Id: I5164db1fa4cef003875647ed01879c5dbdfb9b35
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3286
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

5 files changed, 25 insertions, 16 deletions

diff --git a/arm_compute/core/CL/kernels/CLCropKernel.h b/arm_compute/core/CL/kernels/CLCropKernel.h
index a1c6f901eb..cbc2338940 100644
--- a/arm_compute/core/CL/kernels/CLCropKernel.h
+++ b/arm_compute/core/CL/kernels/CLCropKernel.h
@@ -49,7 +49,7 @@ public:
      *
      * @note Supported tensor rank: up to 4
      *
-     * @param[in]  input               Source tensor. Data type supported: U16/S16/U32/S32/F16/F32. Data layouts supported: NHWC.
+     * @param[in]  input               Source tensor. Data type supported: All. Data layouts supported: NHWC.
      * @param[out] output              Destination tensor. Data type supported: F32
      * @param[in]  start               Coordinates of where to start cropping the image.
      * @param[in]  end                 Coordinates of where to end cropping the image.
@@ -63,7 +63,7 @@ public:
      * @note Supported tensor rank: up to 4
      *
      * @param[in]  compile_context     The compile context to be used.
-     * @param[in]  input               Source tensor. Data type supported: U16/S16/U32/S32/F16/F32. Data layouts supported: NHWC.
+     * @param[in]  input               Source tensor. Data type supported: All. Data layouts supported: NHWC.
      * @param[out] output              Destination tensor. Data type supported: F32
      * @param[in]  start               Coordinates of where to start cropping the image.
      * @param[in]  end                 Coordinates of where to end cropping the image.
@@ -78,7 +78,7 @@ public:
      *
      * @note Supported tensor rank: up to 4
      *
-     * @param[in] input               Source tensor info. Data type supported: U16/S16/U32/S32/F16/F32. Data layouts supported: NHWC.
+     * @param[in] input               Source tensor info. Data type supported: All. Data layouts supported: NHWC.
      * @param[in] output              Destination tensor info. Data type supported: F32
      * @param[in] start               Coordinates of where to start cropping the image.
      * @param[in] end                 Coordinates of where to end cropping the image.
diff --git a/arm_compute/runtime/CL/functions/CLCropResize.h b/arm_compute/runtime/CL/functions/CLCropResize.h
index 86df0d46d1..ea50fe51c7 100644
--- a/arm_compute/runtime/CL/functions/CLCropResize.h
+++ b/arm_compute/runtime/CL/functions/CLCropResize.h
@@ -62,7 +62,7 @@ public:
      * @note Box indices may be outside of the bounds, in which case @p extrapolation_value is used.
      * @note Start and end indices of boxes are inclusive.
      *
-     * @param[in]  input               Source tensor containing N batches of 3D images to be cropped. Data type supported: : U16/S16/U32/S32/F16/F32
+     * @param[in]  input               Source tensor containing N batches of 3D images to be cropped. Data type supported: All
      * @param[in]  boxes               Tensor containing the boxes used to crop the images. It has to be known before configuration. Data type supported: F32
      * @param[in]  box_ind             One dimensional tensor containing the batch index of the 3D image in @p input that the corresponding
      *                                 box in @p boxes will be applied to. It has to be known before configuration. Data type supported: F32
@@ -80,7 +80,7 @@ public:
      * @note Start and end indices of boxes are inclusive.
      *
      * @param[in]  compile_context     The compile context to be used.
-     * @param[in]  input               Source tensor containing N batches of 3D images to be cropped. Data type supported: U16/S16/U32/S32/F16/F32
+     * @param[in]  input               Source tensor containing N batches of 3D images to be cropped. Data type supported: All
      * @param[in]  boxes               Tensor containing the boxes used to crop the images. It has to be known before configuration. Data type supported: F32
      * @param[in]  box_ind             One dimensional tensor containing the batch index of the 3D image in @p input that the corresponding
      *                                 box in @p boxes will be applied to. It has to be known before configuration. Data type supported: F32
@@ -98,7 +98,7 @@ public:
      * @note Box indices may be outside of the bounds, in which case @p extrapolation_value is used.
      * @note Start and end indices of boxes are inclusive.
      *
-     * @param[in] input               Source tensor info containing N batches of 3D images to be cropped. Data type supported: U16/S16/U32/S32/F16/F32
+     * @param[in] input               Source tensor info containing N batches of 3D images to be cropped. Data type supported: All
      * @param[in] boxes               Tensor info for the tensor containing the boxes used to crop the images. Data type supported: F32
      * @param[in] box_ind             Tensor info for the one dimensional tensor containing the batch index of the 3D image in @p input
      *                                that the corresponding box in @p boxes will be applied to. Data type supported: F32
diff --git a/src/core/CL/cl_kernels/crop_tensor.cl b/src/core/CL/cl_kernels/crop_tensor.cl
index 55f8544a10..efc3eabc82 100644
--- a/src/core/CL/cl_kernels/crop_tensor.cl
+++ b/src/core/CL/cl_kernels/crop_tensor.cl
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 ARM Limited.
+ * Copyright (c) 2019-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -27,7 +27,7 @@
 
 /** Performs a copy of input tensor to the output tensor.
  *
- * @param[in]  in_ptr                            Pointer to the source tensor. Supported data types: U16/S16/F16/U32/S32/F32
+ * @param[in]  in_ptr                            Pointer to the source tensor. Supported data types: All
  * @param[in]  in_stride_x                       Stride of the source tensor in X dimension (in bytes)
  * @param[in]  in_step_x                         input_stride_x * number of elements along X processed per workitem(in bytes)
  * @param[in]  in_stride_y                       Stride of the source tensor in Y dimension (in bytes)
@@ -35,7 +35,7 @@
  * @param[in]  in_stride_z                       Stride of the source tensor in Z dimension (in bytes)
  * @param[in]  in_step_z                         input_stride_z * number of elements along Z processed per workitem(in bytes)
  * @param[in]  in_offset_first_element_in_bytes  The offset of the first element in the source tensor
- * @param[out] out_ptr                           Pointer to the destination tensor. Supported data types: same as @p in_ptr
+ * @param[out] out_ptr                           Pointer to the destination tensor. Supported data types: F32
  * @param[in]  out_stride_x                      Stride of the destination tensor in X dimension (in bytes)
  * @param[in]  out_step_x                        output_stride_x * number of elements along X processed per workitem(in bytes)
  * @param[in]  out_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
diff --git a/src/core/CL/kernels/CLCropKernel.cpp b/src/core/CL/kernels/CLCropKernel.cpp
index eb1ab7aebb..29a97bbfa4 100644
--- a/src/core/CL/kernels/CLCropKernel.cpp
+++ b/src/core/CL/kernels/CLCropKernel.cpp
@@ -99,7 +99,7 @@ Status CLCropKernel::validate(const ITensorInfo *input, const ITensorInfo *outpu
 {
     ARM_COMPUTE_UNUSED(extrapolation_value, output_window);
     ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U16, DataType::S16, DataType::F16, DataType::U32, DataType::S32, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(input, DataLayout::NHWC);
     ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape().num_dimensions() > 4);
     ARM_COMPUTE_RETURN_ERROR_ON(start.x < 0 || start.y < 0 || end.x < 0 || end.y < 0);
diff --git a/tests/validation/CL/CropResize.cpp b/tests/validation/CL/CropResize.cpp
index cacf405c96..379597ee67 100644
--- a/tests/validation/CL/CropResize.cpp
+++ b/tests/validation/CL/CropResize.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 ARM Limited.
+ * Copyright (c) 2019-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -52,7 +52,6 @@ using CLCropResizeFixture = CropResizeFixture<CLTensor, CLAccessor, CLCropResize
 // clang-format off
 DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
                framework::dataset::make("InputInfo", { TensorInfo(TensorShape(15U, 30U, 40U, 10U), 1, DataType::S32),
-                                                       TensorInfo(TensorShape(15U, 30U, 40U, 10U), 1, DataType::U8),  // Invalid input data type.
                                                        TensorInfo(TensorShape(15U, 30U, 40U, 10U), 1, DataType::S32), // Invalid box_ind shape.
                                                        TensorInfo(TensorShape(15U, 30U, 40U, 10U), 1, DataType::S32), // Invalid output shape.
                                                        TensorInfo(TensorShape(15U, 30U, 40U, 10U), 1, DataType::S32), // Invalid output data type.
@@ -64,11 +63,9 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
                                                        TensorInfo(TensorShape(4, 20), 1, DataType::F32),
                                                        TensorInfo(TensorShape(4, 20), 1, DataType::F32),
                                                        TensorInfo(TensorShape(4, 20), 1, DataType::F32),
-                                                       TensorInfo(TensorShape(4, 20), 1, DataType::F32),
                                                        TensorInfo(TensorShape(3, 20), 1, DataType::F32),
                                                      })),
                framework::dataset::make("BoxIndInfo",{ TensorInfo(TensorShape(20), 1, DataType::S32),
-                                                       TensorInfo(TensorShape(20), 1, DataType::S32),
                                                        TensorInfo(TensorShape(10), 1, DataType::S32),
                                                        TensorInfo(TensorShape(20), 1, DataType::S32),
                                                        TensorInfo(TensorShape(20), 1, DataType::S32),
@@ -77,13 +74,12 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
                                                      })),
                framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(15U, 5, 5, 20U), 1, DataType::F32),
                                                        TensorInfo(TensorShape(15U, 5, 5, 20U), 1, DataType::F32),
-                                                       TensorInfo(TensorShape(15U, 5, 5, 20U), 1, DataType::F32),
                                                        TensorInfo(TensorShape(15U, 5, 5, 10U), 1, DataType::F32),
                                                        TensorInfo(TensorShape(15U, 5, 5, 20U), 1, DataType::S32),
                                                        TensorInfo(TensorShape(5U, 5, 5, 20U), 1, DataType::F32),
                                                        TensorInfo(TensorShape(15U, 5, 5, 20U), 1, DataType::F32),
                                                      })),
-               framework::dataset::make("Expected", { true, false, false, false, false, false, false})),
+               framework::dataset::make("Expected", { true, false, false, false, false, false})),
                input, boxes, box_ind, output, expected)
 {
     ARM_COMPUTE_EXPECT(bool(CLCropResize::validate(&input.clone()->set_data_layout(DataLayout::NHWC).set_is_resizable(false),
@@ -125,6 +121,19 @@ FIXTURE_DATA_TEST_CASE(RunSmall,
 TEST_SUITE_END() // F32
 TEST_SUITE_END() // Float
 
+TEST_SUITE(U8)
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       CLCropResizeFixture<uint8_t>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(datasets::SmallCropResizeDataset(),
+                               combine(framework::dataset::make("IsOutOfBounds", { true, false }),
+                                       framework::dataset::make("DataType", DataType::U8))))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_fp32, 0.01);
+}
+TEST_SUITE_END() // U8
+
 TEST_SUITE(U16)
 FIXTURE_DATA_TEST_CASE(RunSmall,
                        CLCropResizeFixture<uint16_t>,