COMPMID-3178: Remove padding from NETransposeKernel

Change-Id: I7d59f3f44bce0ab45ddaf00fd4347b632e07e2d6
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2803
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 28 insertions, 85 deletions

diff --git a/src/core/NEON/kernels/NETransposeKernel.cpp b/src/core/NEON/kernels/NETransposeKernel.cpp
index 8fc24949ed..2951a16d0b 100644
--- a/src/core/NEON/kernels/NETransposeKernel.cpp
+++ b/src/core/NEON/kernels/NETransposeKernel.cpp
@@ -54,25 +54,6 @@ TensorShape transposed_tensor_shape(const TensorShape &in)
     return output_shape;
 }
 
-unsigned int num_elems_processed(size_t element_size)
-{
-    switch(element_size)
-    {
-        case 1:
-            return 8;
-            break;
-        case 2:
-            return 4;
-            break;
-        case 4:
-            return 4;
-            break;
-        default:
-            break;
-    }
-    ARM_COMPUTE_ERROR("Element size not supported");
-}
-
 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input);
@@ -90,32 +71,20 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
 
     return Status{};
 }
-
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
+unsigned int num_elems_processed(size_t element_size)
 {
-    // Note: This kernel performs 16 elements per iteration.
-    // However, since we use a left-over for loop on both dimensions (X and Y), we cannot have any read or write out of memory
-    // For this reason num_elems_processed_per_iteration_x is set to 1
-    const unsigned int num_elems_processed_per_iteration_x = 1;
-    const unsigned int num_elems_processed_per_iteration_y = num_elems_processed(input->element_size());
-
-    // Configure kernel window
-    Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
-
-    AccessWindowRectangle input_access(input, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
-    bool                  window_changed = update_window_and_padding(win, input_access);
-
-    if(output->total_size() != 0)
+    switch(element_size)
     {
-        AccessWindowTranspose output_access(output, 0, 0, num_elems_processed_per_iteration_y, num_elems_processed_per_iteration_x);
-
-        window_changed = window_changed || update_window_and_padding(win, output_access);
-
-        output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
+        case 1:
+            return 8;
+        case 2:
+        case 4:
+            return 4;
+        default:
+            break;
     }
 
-    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
-    return std::make_pair(err, win);
+    ARM_COMPUTE_ERROR("Element size not supported");
 }
 
 void transpose_8bit_elements(const ITensor *in, ITensor *out, const Window &window)
@@ -487,7 +456,6 @@ Status NETransposeKernel::validate(const ITensorInfo *input, const ITensorInfo *
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);
     return Status{};
 }
 
@@ -525,9 +493,20 @@ void NETransposeKernel::configure(const ITensor *input, ITensor *output)
     }
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), output->info());
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    INEKernel::configure(win_config.second);
+    Coordinates coord;
+    coord.set_num_dimensions(output->info()->num_dimensions());
+    output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
+
+    // Note: This kernel performs 16 elements per iteration.
+    // However, since we use a left-over for loop on both dimensions (X and Y), we cannot have any read or write out of memory
+    // For this reason num_elems_processed_per_iteration_x is set to 1
+    const unsigned int num_elems_processed_per_iteration_x = 1;
+    const unsigned int num_elems_processed_per_iteration_y = num_elems_processed(input->info()->element_size());
+
+    // Configure kernel window
+    Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
+
+    INEKernel::configure(win);
 }
 
 void NETransposeKernel::run(const Window &window, const ThreadInfo &info)
diff --git a/tests/validation/NEON/Transpose.cpp b/tests/validation/NEON/Transpose.cpp
index c9ebdd5182..1f38fcc677 100644
--- a/tests/validation/NEON/Transpose.cpp
+++ b/tests/validation/NEON/Transpose.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -46,21 +46,17 @@ TEST_SUITE(Transpose)
 // *INDENT-OFF*
 // clang-format off
 DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(
-    framework::dataset::make("InputInfo", { TensorInfo(TensorShape(21U, 13U), 1, DataType::U8),  // Input not a multiple of 8
-                                            TensorInfo(TensorShape(21U, 13U), 1, DataType::U16), // Invalid shape
-                                            TensorInfo(TensorShape(20U, 13U), 1, DataType::U32), // Window shrink
+    framework::dataset::make("InputInfo", { TensorInfo(TensorShape(21U, 13U), 1, DataType::U16), // Invalid shape
                                             TensorInfo(TensorShape(20U, 13U), 1, DataType::U8),  // Wrong data type
                                             TensorInfo(TensorShape(20U, 16U), 1, DataType::U16),
                                             TensorInfo(TensorShape(20U, 16U), 1, DataType::U32),
                                           }),
-    framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(13U, 21U), 1, DataType::U8),
-                                            TensorInfo(TensorShape(21U, 13U), 1, DataType::U16),
-                                            TensorInfo(TensorShape(13U, 20U), 1, DataType::U32),
+    framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(21U, 13U), 1, DataType::U16),
                                             TensorInfo(TensorShape(31U, 20U), 1, DataType::U16),
                                             TensorInfo(TensorShape(16U, 20U), 1, DataType::U16),
                                             TensorInfo(TensorShape(16U, 20U), 1, DataType::U32),
                                            })),
-    framework::dataset::make("Expected", { false, false, false, false, true, true })),
+    framework::dataset::make("Expected", { false, false, true, true })),
     a_info, output_info, expected)
 {
     // Lock tensors
@@ -71,38 +67,6 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(
 // clang-format on
 // *INDENT-ON*
 
-DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, combine(datasets::Small2DShapes(), framework::dataset::make("DataType", { DataType::S8, DataType::U8, DataType::S16, DataType::U16, DataType::U32, DataType::S32, DataType::F16, DataType::F32 })),
-               shape, data_type)
-{
-    // Make rows the columns of the original shape
-    TensorShape output_shape{ shape[1], shape[0] };
-
-    // Create tensors
-    Tensor src = create_tensor<Tensor>(shape, data_type);
-    Tensor dst = create_tensor<Tensor>(output_shape, data_type);
-
-    // Create and Configure function
-    NETranspose trans;
-    trans.configure(&src, &dst);
-
-    // Validate valid region
-    const ValidRegion valid_region = shape_to_valid_region(output_shape);
-    validate(dst.info()->valid_region(), valid_region);
-
-    // Validate padding
-    const unsigned int num_elems_processed_per_iteration_x = 1;
-    const unsigned int num_elems_processed_per_iteration_y = std::max(4, static_cast<int>(8 / src.info()->element_size()));
-    const unsigned int max_in_x                            = ceil_to_multiple(shape[0], num_elems_processed_per_iteration_x);
-    const unsigned int max_in_y                            = ceil_to_multiple(shape[1], num_elems_processed_per_iteration_y);
-    const unsigned int max_out_x                           = ceil_to_multiple(output_shape[0], num_elems_processed_per_iteration_y);
-    const unsigned int max_out_y                           = ceil_to_multiple(output_shape[1], num_elems_processed_per_iteration_x);
-
-    const PaddingSize in_padding(0, max_in_x - shape[0], max_in_y - shape[1], 0);
-    const PaddingSize out_padding(0, max_out_x - output_shape[0], max_out_y - output_shape[1], 0);
-    validate(src.info()->padding(), in_padding);
-    validate(dst.info()->padding(), out_padding);
-}
-
 template <typename T>
 using NETransposeFixture = TransposeValidationFixture<Tensor, Accessor, NETranspose, T>;