COMPMID-617: Add validation to NEON functions.

Adds validation to:
- NECol2Im
- NEIm2Col

Change-Id: I346298583a6985ea793f71bb4527aa216a5cd4b2
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/110835
Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

6 files changed, 78 insertions, 22 deletions

diff --git a/src/core/NEON/kernels/NECol2ImKernel.cpp b/src/core/NEON/kernels/NECol2ImKernel.cpp
index 68fc50ced6..ca769f73cc 100644
--- a/src/core/NEON/kernels/NECol2ImKernel.cpp
+++ b/src/core/NEON/kernels/NECol2ImKernel.cpp
@@ -36,6 +36,37 @@
 
 using namespace arm_compute;
 
+namespace
+{
+TensorShape get_output_shape(const ITensorInfo *input, const Size2D &convolved_dims)
+{
+    TensorShape output_shape = input->tensor_shape();
+    output_shape.set(0, convolved_dims.width);
+    output_shape.set(1, convolved_dims.height);
+    output_shape.set(2, input->tensor_shape()[0]);
+
+    return output_shape;
+}
+
+Error validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &convolved_dims)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QS8, DataType::QASYMM8,
+                                                         DataType::U16, DataType::S16, DataType::QS16,
+                                                         DataType::U32, DataType::S32,
+                                                         DataType::F16, DataType::F32);
+
+    // Validate configured output
+    if(output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), get_output_shape(input, convolved_dims));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+    }
+
+    return Error{};
+}
+} // namespace
+
 template <typename T>
 void NECol2ImKernel::run_col2im(const Window &window)
 {
@@ -55,7 +86,7 @@ void NECol2ImKernel::run_col2im(const Window &window)
     execute_window_loop(window, [&](const Coordinates & id)
     {
         const int hidx = id.y();
-        const int idx  = id.x() * output_stride_z + (hidx / _convolved_dims.first) * output_stride_y + (hidx % _convolved_dims.first) * output_stride_x;
+        const int idx  = id.x() * output_stride_z + (hidx / _convolved_dims.width) * output_stride_y + (hidx % _convolved_dims.width) * output_stride_x;
 
         *(reinterpret_cast<T *>(out.ptr() + idx)) = *(reinterpret_cast<const T *>(in.ptr()));
     },
@@ -67,24 +98,15 @@ NECol2ImKernel::NECol2ImKernel()
 {
 }
 
-void NECol2ImKernel::configure(const ITensor *input, ITensor *output, std::pair<unsigned int, unsigned int> convolved_dims)
+void NECol2ImKernel::configure(const ITensor *input, ITensor *output, const Size2D &convolved_dims)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QS8, DataType::QASYMM8, DataType::U16, DataType::S16, DataType::QS16, DataType::U32, DataType::S32,
-                                                  DataType::F16,
-                                                  DataType::F32);
-    ARM_COMPUTE_ERROR_ON_NULLPTR(output);
-
-    TensorShape output_shape = input->info()->tensor_shape();
-    output_shape.set(0, convolved_dims.first);
-    output_shape.set(1, convolved_dims.second);
-    output_shape.set(2, input->info()->tensor_shape()[0]);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
     // Output auto inizialitation if not yet initialized
-    auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type(), input->info()->fixed_point_position());
+    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(get_output_shape(input->info(), convolved_dims)));
 
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+    // Perform validation step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), convolved_dims));
 
     _input          = input;
     _output         = output;
@@ -117,6 +139,12 @@ void NECol2ImKernel::configure(const ITensor *input, ITensor *output, std::pair<
     INEKernel::configure(win);
 }
 
+Error NECol2ImKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &convolved_dims)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, convolved_dims));
+    return Error{};
+}
+
 void NECol2ImKernel::run(const Window &window, const ThreadInfo &info)
 {
     ARM_COMPUTE_UNUSED(info);
diff --git a/src/core/NEON/kernels/NEIm2ColKernel.cpp b/src/core/NEON/kernels/NEIm2ColKernel.cpp
index 090e554834..7a9b06ebda 100644
--- a/src/core/NEON/kernels/NEIm2ColKernel.cpp
+++ b/src/core/NEON/kernels/NEIm2ColKernel.cpp
@@ -42,6 +42,18 @@ using namespace arm_compute;
 
 namespace
 {
+Error validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::QS16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::QASYMM8 && has_bias);
+    ARM_COMPUTE_UNUSED(kernel_dims);
+    ARM_COMPUTE_UNUSED(conv_info);
+
+    return Error{};
+}
+
 template <typename T, bool has_pads>
 inline void linearize_volume(const uint8_t *const in_ptr,
                              T                   *out_ptr,
@@ -278,10 +290,10 @@ NEIm2ColKernel::NEIm2ColKernel()
 
 void NEIm2ColKernel::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32, DataType::QS8, DataType::QS16, DataType::QASYMM8);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-    ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, output);
-    ARM_COMPUTE_ERROR_ON(input->info()->data_type() == DataType::QASYMM8 && has_bias);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    // Perform validation step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), kernel_dims, conv_info, has_bias));
 
     _input          = input;
     _output         = output;
@@ -361,6 +373,12 @@ void NEIm2ColKernel::configure(const ITensor *input, ITensor *output, const Size
     IKernel::configure(window);
 }
 
+Error NEIm2ColKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, kernel_dims, conv_info, has_bias));
+    return Error{};
+}
+
 void NEIm2ColKernel::run(const Window &window, const ThreadInfo &info)
 {
     ARM_COMPUTE_UNUSED(info);
diff --git a/src/runtime/NEON/functions/NECol2Im.cpp b/src/runtime/NEON/functions/NECol2Im.cpp
index 2dc4ebd49a..2a923f3730 100644
--- a/src/runtime/NEON/functions/NECol2Im.cpp
+++ b/src/runtime/NEON/functions/NECol2Im.cpp
@@ -28,9 +28,14 @@
 
 using namespace arm_compute;
 
-void NECol2Im::configure(const ITensor *input, ITensor *output, std::pair<unsigned int, unsigned int> convolved_dims)
+void NECol2Im::configure(const ITensor *input, ITensor *output, const Size2D &convolved_dims)
 {
     auto k = arm_compute::support::cpp14::make_unique<NECol2ImKernel>();
     k->configure(input, output, convolved_dims);
     _kernel = std::move(k);
 }
+
+Error NECol2Im::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &convolved_dims)
+{
+    return NECol2ImKernel::validate(input, output, convolved_dims);
+}
diff --git a/src/runtime/NEON/functions/NEConvolutionLayer.cpp b/src/runtime/NEON/functions/NEConvolutionLayer.cpp
index 155f4e561a..865672e525 100644
--- a/src/runtime/NEON/functions/NEConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEConvolutionLayer.cpp
@@ -292,7 +292,7 @@ void NEConvolutionLayer::configure(const ITensor *input, const ITensor *weights,
     }
 
     _input_im2col_reshaped.allocator()->allocate();
-    _output_col2im_kernel.configure(&_gemm_output, output, std::make_pair(conv_w, conv_h));
+    _output_col2im_kernel.configure(&_gemm_output, output, Size2D(conv_w, conv_h));
     _gemm_output.allocator()->allocate();
 
     ARM_COMPUTE_ERROR_ON_MSG((output->info()->dimension(0) != conv_w) || (output->info()->dimension(1) != conv_h), "Output shape does not match the expected one");
diff --git a/src/runtime/NEON/functions/NEIm2Col.cpp b/src/runtime/NEON/functions/NEIm2Col.cpp
index 4ed591523b..354415daa3 100644
--- a/src/runtime/NEON/functions/NEIm2Col.cpp
+++ b/src/runtime/NEON/functions/NEIm2Col.cpp
@@ -34,3 +34,8 @@ void NEIm2Col::configure(const ITensor *input, ITensor *output, const Size2D &ke
     k->configure(input, output, kernel_dims, conv_info, has_bias);
     _kernel = std::move(k);
 }
+
+Error NEIm2Col::validate(const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+{
+    return NEIm2ColKernel::validate(input, output, kernel_dims, conv_info, has_bias);
+}
diff --git a/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp b/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
index cb48598921..b29b796276 100644
--- a/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
+++ b/src/runtime/NEON/functions/NELocallyConnectedLayer.cpp
@@ -111,7 +111,7 @@ void NELocallyConnectedLayer::configure(const ITensor *input, const ITensor *wei
     _input_im2col_kernel.configure(input, &_input_im2col_reshaped, Size2D(kernel_width, kernel_height), conv_info, _has_bias);
     _weights_reshape_kernel.configure(weights, biases, &_weights_reshaped);
     _mm_kernel.configure(&_input_im2col_reshaped, &_weights_reshaped, &_gemm_output);
-    _output_col2im_kernel.configure(&_gemm_output, output, std::make_pair(conv_w, conv_h));
+    _output_col2im_kernel.configure(&_gemm_output, output, Size2D(conv_w, conv_h));
 
     // Allocate intermediate tensors
     _weights_reshaped.allocator()->allocate();