COMPMID-417: Auto initialize for SoftmaxLayer NEON/CL.
Change-Id: I6f35ac7a15fecab93deec4c6266e5c9632e599f0
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/79628
Reviewed-by: Moritz Pflanzer <moritz.pflanzer@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
diff --git a/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp b/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp
index 3608a17..3bca938 100644
--- a/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp
+++ b/src/core/CL/kernels/CLSoftmaxLayerKernel.cpp
@@ -42,8 +42,18 @@
void CLLogits1DMaxKernel::configure(const ICLTensor *input, ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::F16, DataType::F32);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(output);
+
+ // Softmax across the x dimension
+ TensorShape output_shape{ input->info()->tensor_shape() };
+ output_shape.set(0, 1);
+
+ // Output auto initialization 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_DATA_TYPES(input, output);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, output);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape);
_input = input;
_output = output;
@@ -94,9 +104,16 @@
void CLLogits1DShiftExpSumKernel::configure(const ICLTensor *input, const ICLTensor *max, ICLTensor *output, ICLTensor *sum)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::F16, DataType::F32);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(max, sum, output);
+
+ // Output auto initialization if not yet initialized
+ auto_init_if_empty(*sum->info(), max->info()->tensor_shape(), 1, input->info()->data_type(), input->info()->fixed_point_position());
+ auto_init_if_empty(*output->info(), input->info()->tensor_shape(), 1, input->info()->data_type(), input->info()->fixed_point_position());
+
ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output, max, sum);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, output, max, sum);
ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(input, output);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output, max, sum);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(max, sum);
_input = input;
_max = max;
@@ -171,9 +188,14 @@
void CLLogits1DNormKernel::configure(const ICLTensor *input, const ICLTensor *sum, ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::F16, DataType::F32);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(sum, output);
+
+ // Output auto initialization if not yet initialized
+ auto_init_if_empty(*output->info(), input->info()->tensor_shape(), 1, input->info()->data_type(), input->info()->fixed_point_position());
+
ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, sum, output);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, sum, output);
ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(input, output);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, sum, output);
_input = input;
_sum = sum;
diff --git a/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp b/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp
index 942662e..854fd84 100644
--- a/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp
+++ b/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp
@@ -120,9 +120,19 @@
void NELogits1DMaxKernel::configure(const ITensor *input, ITensor *output)
{
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::QS8);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32, DataType::QS8);
+ ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::F32);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(output);
+
+ // Softmax across the x dimension
+ TensorShape output_shape{ input->info()->tensor_shape() };
+ output_shape.set(0, 1);
+
+ // Output auto initialization 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_DATA_TYPES(input, output);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, output);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape);
const int input_width = input->info()->valid_region().shape.x();
unsigned int num_elems_processed_per_iteration = 0;
@@ -302,11 +312,16 @@
void NELogits1DShiftExpSumKernel::configure(const ITensor *input, const ITensor *max, ITensor *output, ITensor *sum)
{
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::QS8);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(max, 1, DataType::F32, DataType::QS8);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::F32, DataType::QS8);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, max, output);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, max, output);
+ ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::F32);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(max, sum, output);
+
+ // Output auto initialization if not yet initialized
+ auto_init_if_empty(*sum->info(), max->info()->tensor_shape(), 1, input->info()->data_type(), input->info()->fixed_point_position());
+ auto_init_if_empty(*output->info(), input->info()->tensor_shape(), 1, input->info()->data_type(), input->info()->fixed_point_position());
+
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output, max, sum);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, output, max, sum);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(input, output);
ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(max, sum);
unsigned int num_elems_processed_per_iteration = input->info()->valid_region().shape.x();
@@ -426,8 +441,13 @@
void NELogits1DNormKernel::configure(const ITensor *input, const ITensor *sum, ITensor *output)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32, DataType::QS8);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output, sum);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output, sum);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(sum, output);
+
+ // Output auto initialization if not yet initialized
+ auto_init_if_empty(*output->info(), input->info()->tensor_shape(), 1, input->info()->data_type(), input->info()->fixed_point_position());
+
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, sum, output);
+ ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, sum, output);
ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(input, output);
_input = input;
diff --git a/src/runtime/CL/functions/CLSoftmaxLayer.cpp b/src/runtime/CL/functions/CLSoftmaxLayer.cpp
index 32dc8ff..0f20af0 100644
--- a/src/runtime/CL/functions/CLSoftmaxLayer.cpp
+++ b/src/runtime/CL/functions/CLSoftmaxLayer.cpp
@@ -37,7 +37,6 @@
void CLSoftmaxLayer::configure(const ICLTensor *input, ICLTensor *output)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::F16, DataType::F32);
- ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
// Create intermediate tensors shapes
_tmp.allocator()->init(TensorInfo(input->info()->tensor_shape(), input->info()->num_channels(), input->info()->data_type(), input->info()->fixed_point_position()));
diff --git a/src/runtime/NEON/functions/NESoftmaxLayer.cpp b/src/runtime/NEON/functions/NESoftmaxLayer.cpp
index 0651eab..ca81b95 100644
--- a/src/runtime/NEON/functions/NESoftmaxLayer.cpp
+++ b/src/runtime/NEON/functions/NESoftmaxLayer.cpp
@@ -39,7 +39,6 @@
void NESoftmaxLayer::configure(ITensor *input, ITensor *output)
{
ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::F32);
- ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::QS8, DataType::F32);
// Create intermediate tensors shapes
TensorInfo tensor_info_tmp(input->info()->tensor_shape(), input->info()->num_channels(), input->info()->data_type(), input->info()->fixed_point_position());