COMPMID-667: Add validation static method to NEON GEMMlowp

Change-Id: I8a470cc1351593ad8eeaf4ec92e04865e83d4f3c
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/96147
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

1 files changed, 57 insertions, 19 deletions

diff --git a/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp b/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp
index a29b661a00..1f4d9b176e 100644
--- a/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp
@@ -40,6 +40,50 @@ using namespace arm_compute;
 
 namespace
 {
+Error validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::U8, DataType::S8,
+                                                         DataType::QS16, DataType::U16, DataType::S16, DataType::U32, DataType::S32,
+                                                         DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+
+    if(output->total_size() != 0)
+    {
+        TensorShape output_shape = input->tensor_shape();
+        output_shape.set(0, input->dimension(0) * 4);
+        output_shape.set(1, std::ceil(input->dimension(1) / 4.0f));
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), output_shape);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+    }
+
+    return Error{};
+}
+
+std::pair<Error, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
+{
+    unsigned int           num_elems_processed_per_iteration_x = (input->element_size() == 1) ? 8 : 4;
+    constexpr unsigned int num_elems_processed_per_iteration_y = 4;
+    bool                   window_changed                      = false;
+
+    // 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);
+    window_changed = window_changed || update_window_and_padding(win, input_access);
+
+    // Configure window in case of configured output
+    if(output->total_size() != 0)
+    {
+        AccessWindowRectangle output_access(output, 0, 0, num_elems_processed_per_iteration_x * num_elems_processed_per_iteration_y, 1, 4.0f, 0.25f);
+        window_changed = window_changed || update_window_and_padding(win, output_access);
+        output_access.set_valid_region(win, input->valid_region());
+    }
+
+    Error err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Error{};
+    return std::make_pair(err, win);
+}
+
 void gemm_interleave_8bit_elements(const ITensor *input, ITensor *output, const Window &window)
 {
     const size_t in_stride = input->info()->strides_in_bytes()[1];
@@ -132,10 +176,7 @@ NEGEMMInterleave4x4Kernel::NEGEMMInterleave4x4Kernel()
 
 void NEGEMMInterleave4x4Kernel::configure(const ITensor *input, ITensor *output)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8, DataType::QS8, DataType::QS16, DataType::U8, DataType::S8, DataType::U16, DataType::S16, DataType::U32, DataType::S32,
-                                                  DataType::F16,
-                                                  DataType::F32);
-    ARM_COMPUTE_ERROR_ON_NULLPTR(output);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
     TensorShape output_shape = input->info()->tensor_shape();
     output_shape.set(0, input->info()->dimension(0) * 4);
@@ -144,21 +185,16 @@ void NEGEMMInterleave4x4Kernel::configure(const ITensor *input, ITensor *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());
 
-    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 validate step
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info()));
 
     _input  = input;
     _output = output;
 
-    unsigned int           num_elems_processed_per_iteration_x = 4;
-    constexpr unsigned int num_elems_processed_per_iteration_y = 4;
-
     switch(input->info()->element_size())
     {
         case 1:
-            num_elems_processed_per_iteration_x = 8;
-            _func                               = &gemm_interleave_8bit_elements;
+            _func = &gemm_interleave_8bit_elements;
             break;
         case 2:
             _func = &gemm_interleave_16bit_elements;
@@ -172,15 +208,17 @@ void NEGEMMInterleave4x4Kernel::configure(const ITensor *input, ITensor *output)
     }
 
     // Configure kernel window
-    Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
-
-    AccessWindowRectangle output_access(output->info(), 0, 0, num_elems_processed_per_iteration_x * num_elems_processed_per_iteration_y, 1, 4.0f, 0.25f);
-    AccessWindowRectangle input_access(input->info(), 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
-    update_window_and_padding(win, output_access, input_access);
+    auto win_config = validate_and_configure_window(input->info(), output->info());
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    INEKernel::configure(win_config.second);
+}
 
-    output_access.set_valid_region(win, input->info()->valid_region());
+Error NEGEMMInterleave4x4Kernel::validate(const ITensorInfo *input, const ITensorInfo *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);
 
-    INEKernel::configure(win);
+    return Error{};
 }
 
 void NEGEMMInterleave4x4Kernel::run(const Window &window, const ThreadInfo &info)