diff options
author | Pablo Marquez Tello <pablo.tello@arm.com> | 2021-09-15 10:14:20 +0100 |
---|---|---|
committer | Sheri Zhang <sheri.zhang@arm.com> | 2021-09-16 09:59:54 +0000 |
commit | 9ac7b99106e24ab426d8a948dd5243defb753b32 (patch) | |
tree | 1eedc0be5bba626afceecfb9d5d4436e3ac38502 /src/cpu/operators/CpuFullyConnected.cpp | |
parent | cc171f9e4520e16b5e1b9c483562ed022d9151fa (diff) | |
download | ComputeLibrary-9ac7b99106e24ab426d8a948dd5243defb753b32.tar.gz |
Revert "Add support for non-constant weights and biases in CpuFullyConnected"
This reverts commit aed63ee175e0d64c934389e9d1b2edd0cb1a5cdd.
* Resolves COMPMID-4812
Change-Id: I16919e2f3b22c868ae146d0d10dae97a80e1ba46
Signed-off-by: Pablo Marquez Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6266
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sheri Zhang <sheri.zhang@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/cpu/operators/CpuFullyConnected.cpp')
-rw-r--r-- | src/cpu/operators/CpuFullyConnected.cpp | 26 |
1 files changed, 5 insertions, 21 deletions
diff --git a/src/cpu/operators/CpuFullyConnected.cpp b/src/cpu/operators/CpuFullyConnected.cpp index 57094cb0b4..cafb3484b6 100644 --- a/src/cpu/operators/CpuFullyConnected.cpp +++ b/src/cpu/operators/CpuFullyConnected.cpp @@ -312,13 +312,9 @@ void CpuFullyConnected::configure(const ITensorInfo *src, const ITensorInfo *wei if(_aux_mem[Pretranspose].size > 0) { - // Release permuted weights at the end of prepare as they are further transposed by the assembly dispatch - // Do not release them if biases are dynamic and data type is quantized, since the weights tensor will be used for biases offset calculation - _aux_mem[TransposedWeights] = MemoryInfo(offset_int_vec(TransposedWeights), (_is_quantized_asymmetric && biases && !(biases->are_values_constant())) ? - MemoryLifetime::Persistent : - MemoryLifetime::Prepare, - _reshaped_weights.total_size()); - _aux_mem[ConvertedWeights] = MemoryInfo(offset_int_vec(ConvertedWeights), MemoryLifetime::Prepare, _converted_weights.total_size()); + // Release permuted weights at the of prepare as they are further transposed by the assembly dispatch + _aux_mem[TransposedWeights] = MemoryInfo(offset_int_vec(TransposedWeights), MemoryLifetime::Prepare, _reshaped_weights.total_size()); + _aux_mem[ConvertedWeights] = MemoryInfo(offset_int_vec(ConvertedWeights), MemoryLifetime::Prepare, _converted_weights.total_size()); } else { @@ -336,9 +332,10 @@ Status CpuFullyConnected::validate(const ITensorInfo *src, const ITensorInfo *we ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(src, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32); ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, weights, dst); ARM_COMPUTE_RETURN_ERROR_ON(weights->num_dimensions() > 2); + ARM_COMPUTE_RETURN_ERROR_ON(biases != nullptr && biases->num_dimensions() > 1); ARM_COMPUTE_RETURN_ERROR_ON(fc_info.activation_info.enabled() && is_data_type_quantized(src->data_type()) && fc_info.activation_info.activation() != ActivationLayerInfo::ActivationFunction::RELU && fc_info.activation_info.activation() != ActivationLayerInfo::ActivationFunction::BOUNDED_RELU && fc_info.activation_info.activation() != ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU); - ARM_COMPUTE_RETURN_ERROR_ON(!weights->are_values_constant() && (!fc_info.are_weights_reshaped || fc_info.transpose_weights)); + ARM_COMPUTE_RETURN_ERROR_ON_MSG(!fc_info.constant_weights, "Non-constant weights are currently not supported"); bool weights_reshaped = fc_info.transpose_weights ? fc_info.are_weights_reshaped : true; bool is_fc_after_conv = true; @@ -359,19 +356,6 @@ Status CpuFullyConnected::validate(const ITensorInfo *src, const ITensorInfo *we // Check if we have a fully connected layer with batches const bool is_batched_fc_layer = dst->dimension(1) > 1; - if(biases != nullptr) - { - ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1); - if(is_data_type_quantized(src->data_type())) - { - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(biases, 1, DataType::S32); - } - else - { - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, biases); - } - } - if(is_batched_fc_layer) { is_fc_after_conv = (TensorShape::num_max_dimensions >= 4) && (std::equal(src->tensor_shape().cbegin() + 3, |