COMPMID-3236: Implement CLQLSTMLayer
COMPMID-3081: Extend CLQLSTMLayer with enhancements
Change-Id: Idb7aaaacdba957e5ad61e94edeab2e898730a109
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3057
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
index ef17f11..3465da9 100644
--- a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
+++ b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
@@ -303,11 +303,9 @@
{
ARM_COMPUTE_ERROR_ON_NULLPTR(a, b, output);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(a, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED);
- //DataType::QSYMM8_PER_CHANNEL supported only for weights
- if(b->data_type() != DataType::QSYMM8_PER_CHANNEL)
- {
- ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(a, b);
- }
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(b, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QSYMM8, DataType::QSYMM8_PER_CHANNEL);
+ ARM_COMPUTE_RETURN_ERROR_ON(a->data_type() == DataType::QASYMM8 && b->data_type() == DataType::QASYMM8_SIGNED);
+ ARM_COMPUTE_RETURN_ERROR_ON(a->data_type() == DataType::QASYMM8_SIGNED && b->data_type() == DataType::QASYMM8);
ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.is_a_reshaped(), "Matrix A already reshaped is not supported");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.is_b_reshaped(), "Matrix B already reshaped is not supported");
diff --git a/src/runtime/CL/functions/CLGEMMLowpOutputStage.cpp b/src/runtime/CL/functions/CLGEMMLowpOutputStage.cpp
index 2114d39..aff7f54 100644
--- a/src/runtime/CL/functions/CLGEMMLowpOutputStage.cpp
+++ b/src/runtime/CL/functions/CLGEMMLowpOutputStage.cpp
@@ -149,6 +149,13 @@
_kernel = std::move(k);
break;
}
+ case DataType::QSYMM16:
+ {
+ auto k = arm_compute::support::cpp14::make_unique<CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel>();
+ k->configure(input, bias, output, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
+ _kernel = std::move(k);
+ break;
+ }
default:
ARM_COMPUTE_ERROR("Unsupported output data type.");
}
@@ -188,6 +195,8 @@
return CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::validate(input, bias, output, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
case DataType::QASYMM8_SIGNED:
return CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::validate(input, bias, output, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
+ case DataType::QSYMM16:
+ return CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::validate(input, bias, output, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
default:
return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported output data type.");
}
diff --git a/src/runtime/CL/functions/CLQLSTMLayer.cpp b/src/runtime/CL/functions/CLQLSTMLayer.cpp
new file mode 100644
index 0000000..4b994d4
--- /dev/null
+++ b/src/runtime/CL/functions/CLQLSTMLayer.cpp
@@ -0,0 +1,853 @@
+/*
+ * Copyright (c) 2020 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/CL/functions/CLQLSTMLayer.h"
+
+#include "arm_compute/core/KernelDescriptors.h"
+#include "arm_compute/core/QuantizationInfo.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/core/utils/misc/InfoHelpers.h"
+#include "arm_compute/core/utils/quantization/AsymmHelpers.h"
+#include "arm_compute/runtime/CL/CLScheduler.h"
+
+namespace arm_compute
+{
+using namespace arm_compute::utils::info_helpers;
+namespace
+{
+Status validate_mm(GEMMLowpOutputStageInfo &gemmlowp_info, const ITensorInfo *mm_input, const ITensorInfo *mm_weights, const ITensorInfo *bias,
+ float gemmlowp_scale, const TensorInfo *mm_res_info, const TensorInfo *outstage_tensor_info)
+{
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixMultiplyCore::validate(mm_input, mm_weights, nullptr, mm_res_info));
+ ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(gemmlowp_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpOutputStage::validate(mm_res_info, bias, outstage_tensor_info, gemmlowp_info));
+ return Status{};
+}
+} // namespace
+
+CLQLSTMLayer::CLQLSTMLayer(std::shared_ptr<IMemoryManager> memory_manager)
+{
+ _memory_group = MemoryGroup(std::move(memory_manager));
+}
+
+void CLQLSTMLayer::configure_mm(CLGEMMLowpMatrixMultiplyCore &mm, CLGEMMLowpOutputStage &outstage, GEMMLowpOutputStageInfo &gemmlowp_info,
+ const ICLTensor *mm_input, const ICLTensor *mm_weights, const ICLTensor *bias,
+ CLTensor *mm_res, CLTensor *outstage_res, float gemmlowp_scale,
+ const TensorInfo &mm_res_info, const TensorInfo &outstage_tensor_info)
+{
+ _memory_group.manage(mm_res);
+ _memory_group.manage(outstage_res);
+
+ mm_res->allocator()->init(mm_res_info);
+ outstage_res->allocator()->init(outstage_tensor_info);
+
+ // Configure matrix-multiplication
+ mm.configure(mm_input, mm_weights, nullptr, mm_res);
+
+ // Configure output stage
+ quantization::calculate_quantized_multiplier(gemmlowp_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift);
+ outstage.configure(mm_res, bias, outstage_res, gemmlowp_info);
+ mm_res->allocator()->allocate();
+}
+
+void CLQLSTMLayer::configure(const ICLTensor *input,
+ const ICLTensor *input_to_forget_weights, const ICLTensor *input_to_cell_weights, const ICLTensor *input_to_output_weights,
+ const ICLTensor *recurrent_to_forget_weights, const ICLTensor *recurrent_to_cell_weights, const ICLTensor *recurrent_to_output_weights,
+ const ICLTensor *forget_gate_bias, const ICLTensor *cell_bias, const ICLTensor *output_gate_bias,
+ const ICLTensor *cell_state_in, const ICLTensor *output_state_in,
+ ICLTensor *cell_state_out, ICLTensor *output_state_out,
+ const LSTMParams<ICLTensor> &lstm_params)
+{
+ ARM_COMPUTE_UNUSED(forget_gate_bias);
+ ARM_COMPUTE_UNUSED(cell_bias);
+ ARM_COMPUTE_UNUSED(output_gate_bias);
+ ARM_COMPUTE_ERROR_ON_NULLPTR(input, input_to_forget_weights, input_to_cell_weights, input_to_output_weights,
+ recurrent_to_forget_weights, recurrent_to_cell_weights, recurrent_to_output_weights,
+ forget_gate_bias, cell_bias, output_gate_bias, cell_state_in, output_state_in, cell_state_out, output_state_out);
+
+ // Set lstm parameters
+ LSTMParams<ITensorInfo> lstm_params_info{};
+ build_lstm_params_tensor_info(lstm_params, &lstm_params_info);
+
+ // Validate
+ ARM_COMPUTE_ERROR_THROW_ON(CLQLSTMLayer::validate(input->info(), input_to_forget_weights->info(), input_to_cell_weights->info(), input_to_output_weights->info(),
+ recurrent_to_forget_weights->info(), recurrent_to_cell_weights->info(), recurrent_to_output_weights->info(),
+ forget_gate_bias->info(), cell_bias->info(), output_gate_bias->info(),
+ cell_state_in->info(), output_state_in->info(), cell_state_out->info(), output_state_out->info(), lstm_params_info));
+
+ const int batch_size = input->info()->dimension(1);
+ const int num_units = input_to_output_weights->info()->dimension(1);
+
+ const UniformQuantizationInfo qinput = input->info()->quantization_info().uniform();
+ const UniformQuantizationInfo qcell_state_in = cell_state_in->info()->quantization_info().uniform();
+ const UniformQuantizationInfo qoutput_state_in = output_state_in->info()->quantization_info().uniform();
+
+ _projection_bias = lstm_params.projection_bias();
+ _input_to_forget_weights = input_to_forget_weights;
+ _input_to_cell_weights = input_to_cell_weights;
+ _input_to_output_weights = input_to_output_weights;
+ _recurrent_to_forget_weights = recurrent_to_forget_weights;
+ _recurrent_to_cell_weights = recurrent_to_cell_weights;
+ _recurrent_to_output_weights = recurrent_to_output_weights;
+ _projection_weights = lstm_params.projection_weights();
+
+ _has_cifg = lstm_params.has_cifg_opt();
+ _has_projection = lstm_params.has_projection();
+ _has_peephole = lstm_params.has_peephole_opt();
+
+ // Calculate and decompose effective scales for optimizing matmul calculation
+ const int32_t cell_shift = log2(qcell_state_in.scale);
+
+ // Calculate quantized parameters for clipping.
+ int16_t quantized_cell_clip = 0;
+ if(lstm_params.cell_clip() > 0.0f)
+ {
+ quantized_cell_clip = quantize_qsymm16(lstm_params.cell_clip(), qcell_state_in);
+ }
+ _has_cell_clipping = quantized_cell_clip > 0;
+
+ // Precompute effective bias for optimizing the matmul computations.
+ if(!_has_cifg)
+ {
+ _input_to_input_weights = lstm_params.input_to_input_weights();
+ _recurrent_to_input_weights = lstm_params.recurrent_to_input_weights();
+
+ _input_to_input_reduction.configure(_input_to_input_weights, &_input_to_input_eff_bias, GEMMLowpReductionKernelInfo(num_units, false, -qinput.offset, true));
+ _recurrent_to_input_reduction.configure(_recurrent_to_input_weights, &_recurrent_to_input_eff_bias, GEMMLowpReductionKernelInfo(num_units, false, -qoutput_state_in.offset, true));
+ }
+ _input_to_forget_reduction.configure(input_to_forget_weights, &_input_to_forget_eff_bias, GEMMLowpReductionKernelInfo(num_units, false, -qinput.offset, true));
+ _recurrent_to_forget_reduction.configure(recurrent_to_forget_weights, &_recurrent_to_forget_eff_bias, GEMMLowpReductionKernelInfo(num_units, false, -qoutput_state_in.offset, true));
+ _input_to_cell_reduction.configure(input_to_cell_weights, &_input_to_cell_eff_bias, GEMMLowpReductionKernelInfo(num_units, false, -qinput.offset, true));
+ _recurrent_to_cell_reduction.configure(recurrent_to_cell_weights, &_recurrent_to_cell_eff_bias, GEMMLowpReductionKernelInfo(num_units, false, -qoutput_state_in.offset, true));
+ _input_to_output_reduction.configure(input_to_output_weights, &_input_to_output_eff_bias, GEMMLowpReductionKernelInfo(num_units, false, -qinput.offset, true));
+ _recurrent_to_output_reduction.configure(recurrent_to_output_weights, &_recurrent_to_output_eff_bias, GEMMLowpReductionKernelInfo(num_units, false, -qoutput_state_in.offset, true));
+ if(_projection_bias != nullptr)
+ {
+ _projection_reduction.configure(_projection_weights, &_projection_reduction_res, GEMMLowpReductionKernelInfo(num_units, false, lstm_params.hidden_state_zero(), true));
+ _projection_bias_add.configure(ArithmeticOperation::ADD, _projection_bias, &_projection_reduction_res, &_projection_eff_bias, ConvertPolicy::SATURATE);
+ }
+
+ // Pre-transpose weights to be used in GEMM.
+ _transpose_input_to_forget_weights.configure(input_to_forget_weights, &_input_to_forget_weights_transposed);
+ _transpose_input_to_cell_weights.configure(input_to_cell_weights, &_input_to_cell_weights_transposed);
+ _transpose_input_to_output_weights.configure(input_to_output_weights, &_input_to_output_weights_transposed);
+ _transpose_recurrent_to_forget_weights.configure(recurrent_to_forget_weights, &_recurrent_to_forget_weights_transposed);
+ _transpose_recurrent_to_cell_weights.configure(recurrent_to_cell_weights, &_recurrent_to_cell_weights_transposed);
+ _transpose_recurrent_to_output_weights.configure(recurrent_to_output_weights, &_recurrent_to_output_weights_transposed);
+ if(!_has_cifg)
+ {
+ _transpose_input_to_input_weights.configure(lstm_params.input_to_input_weights(), &_input_to_input_weights_transposed);
+ _transpose_recurrent_to_input_weights.configure(lstm_params.recurrent_to_input_weights(), &_recurrent_to_input_weights_transposed);
+ }
+ if(_has_projection)
+ {
+ _transpose_projection_weights.configure(_projection_weights, &_projection_weights_transposed);
+ }
+
+ GEMMLowpOutputStageInfo gemmlowp_info;
+ gemmlowp_info.type = GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT;
+ gemmlowp_info.gemmlowp_min_bound = std::numeric_limits<int16_t>::lowest();
+ gemmlowp_info.gemmlowp_max_bound = std::numeric_limits<int16_t>::max();
+ gemmlowp_info.output_data_type = DataType::QSYMM16;
+
+ const TensorInfo mm_out_info(TensorShape(num_units, batch_size), 1, DataType::S32);
+ // Forget gate.
+ const TensorInfo forget_gate_outstage_info(mm_out_info.tensor_shape(), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.forget_intermediate_scale(), 0));
+ const float input_to_forget_scale = input_to_forget_weights->info()->quantization_info().uniform().scale * qinput.scale / lstm_params.forget_intermediate_scale();
+ configure_mm(_mm_input_to_forget, _input_to_forget_outstage, gemmlowp_info,
+ input, &_input_to_forget_weights_transposed, &_input_to_forget_eff_bias,
+ &_mm_input_to_forget_res, &_input_to_forget_outstage_res, input_to_forget_scale,
+ mm_out_info, forget_gate_outstage_info);
+
+ const float recurrent_to_forget_scale = recurrent_to_forget_weights->info()->quantization_info().uniform().scale * qoutput_state_in.scale / lstm_params.forget_intermediate_scale();
+ configure_mm(_mm_recurrent_to_forget, _recurrent_to_forget_outstage, gemmlowp_info,
+ output_state_in, &_recurrent_to_forget_weights_transposed, &_recurrent_to_forget_eff_bias,
+ &_mm_recurrent_to_forget_res, &_recurrent_to_forget_outstage_res, recurrent_to_forget_scale,
+ mm_out_info, forget_gate_outstage_info);
+
+ _accumulate_input_recurrent_forget.configure(ArithmeticOperation::ADD, &_input_to_forget_outstage_res, &_recurrent_to_forget_outstage_res, &_recurrent_to_forget_outstage_res, ConvertPolicy::SATURATE);
+ _input_to_forget_outstage_res.allocator()->allocate();
+
+ if(_has_peephole)
+ {
+ _memory_group.manage(&_mul_cell_to_forget_res);
+ _pixelwise_mul_cell_to_forget.configure(cell_state_in, lstm_params.cell_to_forget_weights(), &_mul_cell_to_forget_res, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+ _cell_to_forget_outstage_res.allocator()->init(TensorInfo(_mul_cell_to_forget_res.info()->tensor_shape(), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.forget_intermediate_scale(), 0)));
+ _memory_group.manage(&_cell_to_forget_outstage_res);
+ const float cell_to_forget_scale = std::pow(2, cell_shift) * lstm_params.cell_to_forget_weights()->info()->quantization_info().uniform().scale / lstm_params.forget_intermediate_scale();
+ quantization::calculate_quantized_multiplier(cell_to_forget_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift);
+ _cell_to_forget_outstage.configure(&_mul_cell_to_forget_res, nullptr, &_cell_to_forget_outstage_res, gemmlowp_info);
+ _mul_cell_to_forget_res.allocator()->allocate();
+ _accumulate_cell_forget.configure(ArithmeticOperation::ADD, &_recurrent_to_forget_outstage_res, &_cell_to_forget_outstage_res, &_recurrent_to_forget_outstage_res, ConvertPolicy::SATURATE);
+ _cell_to_forget_outstage_res.allocator()->allocate();
+ }
+
+ // Output quantization info of Sigmoid and Tanh activations
+ const QuantizationInfo sigmoid_tanh_outqinfo(1.f / 32768.f, 0);
+
+ const TensorInfo forget_gate_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, sigmoid_tanh_outqinfo);
+ _memory_group.manage(&_forget_gate);
+ _forget_gate.allocator()->init(forget_gate_info);
+ _forget_gate_sigmoid.configure(&_recurrent_to_forget_outstage_res, &_forget_gate, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC));
+ _recurrent_to_forget_outstage_res.allocator()->allocate();
+
+ // Modulation gate.
+ const TensorInfo cell_outstage_info(mm_out_info.tensor_shape(), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.cell_intermediate_scale(), 0));
+ const float input_to_cell_scale = input_to_cell_weights->info()->quantization_info().uniform().scale * qinput.scale / lstm_params.cell_intermediate_scale();
+ configure_mm(_mm_input_to_cell, _input_to_cell_outstage, gemmlowp_info,
+ input, &_input_to_cell_weights_transposed, &_input_to_cell_eff_bias,
+ &_mm_input_to_cell_res, &_input_to_cell_outstage_res, input_to_cell_scale,
+ mm_out_info, cell_outstage_info);
+
+ const float recurrent_to_cell_scale = recurrent_to_cell_weights->info()->quantization_info().uniform().scale * qoutput_state_in.scale / lstm_params.cell_intermediate_scale();
+ configure_mm(_mm_recurrent_to_cell, _recurrent_to_cell_outstage, gemmlowp_info,
+ output_state_in, &_recurrent_to_cell_weights_transposed, &_recurrent_to_cell_eff_bias,
+ &_mm_recurrent_to_cell_res, &_recurrent_to_cell_outstage_res, recurrent_to_cell_scale,
+ mm_out_info, cell_outstage_info);
+
+ _accumulate_input_recurrent_modulation.configure(ArithmeticOperation::ADD, &_input_to_cell_outstage_res, &_recurrent_to_cell_outstage_res, &_recurrent_to_cell_outstage_res, ConvertPolicy::SATURATE);
+ _input_to_cell_outstage_res.allocator()->allocate();
+
+ const TensorInfo cell_gate_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, sigmoid_tanh_outqinfo);
+ _memory_group.manage(&_cell_gate);
+ _cell_gate.allocator()->init(cell_gate_info);
+ _cell_gate_tanh.configure(&_recurrent_to_cell_outstage_res, &_cell_gate, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::TANH, 1.f, 1.f));
+ _recurrent_to_cell_outstage_res.allocator()->allocate();
+
+ // Input gate.
+ const TensorInfo input_gate_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, sigmoid_tanh_outqinfo);
+ _input_gate.allocator()->init(input_gate_info);
+ _memory_group.manage(&_input_gate);
+ if(_has_cifg)
+ {
+ _ones.allocator()->init(*_forget_gate.info());
+ _input_gate_sub.configure(ArithmeticOperation::SUB, &_ones, &_forget_gate, &_input_gate, ConvertPolicy::SATURATE);
+ _ones.allocator()->allocate();
+ }
+ else
+ {
+ const TensorInfo input_outstage_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.input_intermediate_scale(), 0));
+ const float input_to_input_scale = _input_to_input_weights->info()->quantization_info().uniform().scale * qinput.scale / lstm_params.input_intermediate_scale();
+ configure_mm(_mm_input_to_input, _input_to_input_outstage, gemmlowp_info,
+ input, &_input_to_input_weights_transposed, &_input_to_input_eff_bias,
+ &_mm_input_to_input_res, &_input_to_input_outstage_res, input_to_input_scale,
+ mm_out_info, input_outstage_info);
+
+ const float recurrent_to_input_scale = _recurrent_to_input_weights->info()->quantization_info().uniform().scale * qoutput_state_in.scale / lstm_params.input_intermediate_scale();
+ configure_mm(_mm_recurrent_to_input, _recurrent_to_input_outstage, gemmlowp_info,
+ input, &_recurrent_to_input_weights_transposed, &_recurrent_to_input_eff_bias,
+ &_mm_recurrent_to_input_res, &_recurrent_to_input_outstage_res, recurrent_to_input_scale,
+ mm_out_info, input_outstage_info);
+ _accumulate_input_recurrent_input.configure(ArithmeticOperation::ADD, &_input_to_input_outstage_res, &_recurrent_to_input_outstage_res, &_recurrent_to_input_outstage_res, ConvertPolicy::SATURATE);
+ _input_to_input_outstage_res.allocator()->allocate();
+
+ if(_has_peephole)
+ {
+ _memory_group.manage(&_mul_cell_to_input_res);
+ _pixelwise_mul_cell_to_input.configure(cell_state_in, lstm_params.cell_to_input_weights(), &_mul_cell_to_input_res, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+ const float cell_to_input_scale = std::pow(2, cell_shift) * lstm_params.cell_to_input_weights()->info()->quantization_info().uniform().scale / lstm_params.input_intermediate_scale();
+ quantization::calculate_quantized_multiplier(cell_to_input_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift);
+ _cell_to_input_outstage_res.allocator()->init(TensorInfo(_mul_cell_to_input_res.info()->tensor_shape(), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.input_intermediate_scale(), 0)));
+ _memory_group.manage(&_cell_to_input_outstage_res);
+ _cell_to_input_outstage.configure(&_mul_cell_to_input_res, nullptr, &_cell_to_input_outstage_res, gemmlowp_info);
+ _mul_cell_to_input_res.allocator()->allocate();
+ _accumulate_cell_input.configure(ArithmeticOperation::ADD, &_recurrent_to_input_outstage_res, &_cell_to_input_outstage_res, &_recurrent_to_input_outstage_res, ConvertPolicy::SATURATE);
+ _cell_to_input_outstage_res.allocator()->allocate();
+ }
+
+ _input_gate_tanh.configure(&_recurrent_to_input_outstage_res, &_input_gate, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::TANH, 1.f, 1.f));
+ _recurrent_to_input_outstage_res.allocator()->allocate();
+ }
+ // Cell.
+ // TODO(COMPMID-3396): Perform multiplication in the quantized domain in CLPixelWiseMultiplicationKernel
+ _pixelwise_mul_forget_cell.configure(&_forget_gate, cell_state_in, &_forget_gate, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+ const float cell_gate_scale = _cell_gate.info()->quantization_info().uniform().scale;
+ const float mul_input_cell_scale = cell_gate_scale * std::pow(2, 15 + cell_shift);
+ const TensorInfo mul_input_cell_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, QuantizationInfo(mul_input_cell_scale, 0));
+ _memory_group.manage(&_mul_input_cell_res);
+ _mul_input_cell_res.allocator()->init(mul_input_cell_info);
+ _pixelwise_mul_input_cell.configure(&_input_gate, &_cell_gate, &_mul_input_cell_res, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+ _cell_gate.allocator()->allocate();
+ _add_forget_cell.configure(ArithmeticOperation::ADD, &_forget_gate, &_mul_input_cell_res, cell_state_out, ConvertPolicy::SATURATE);
+ _mul_input_cell_res.allocator()->allocate();
+ _forget_gate.allocator()->allocate();
+ if(_has_cell_clipping)
+ {
+ _cell_clip.configure(cell_state_out, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -quantized_cell_clip, quantized_cell_clip));
+ }
+ // Output gate.
+ const TensorInfo output_outstage_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.output_intermediate_scale(), 0));
+ const float input_to_output_scale = input_to_output_weights->info()->quantization_info().uniform().scale * qinput.scale / lstm_params.output_intermediate_scale();
+ configure_mm(_mm_input_to_output, _input_to_output_outstage, gemmlowp_info,
+ input, &_input_to_output_weights_transposed, &_input_to_output_eff_bias,
+ &_mm_input_to_output_res, &_input_to_output_outstage_res, input_to_output_scale,
+ mm_out_info, output_outstage_info);
+
+ const float recurrent_to_output_scale = recurrent_to_output_weights->info()->quantization_info().uniform().scale * qoutput_state_in.scale / lstm_params.output_intermediate_scale();
+ configure_mm(_mm_recurrent_to_output, _recurrent_to_output_outstage, gemmlowp_info,
+ output_state_in, &_recurrent_to_output_weights_transposed, &_recurrent_to_output_eff_bias,
+ &_mm_recurrent_to_output_res, &_recurrent_to_output_outstage_res, recurrent_to_output_scale,
+ mm_out_info, output_outstage_info);
+
+ _accumulate_input_recurrent_output.configure(ArithmeticOperation::ADD, &_recurrent_to_output_outstage_res, &_input_to_output_outstage_res, &_recurrent_to_output_outstage_res, ConvertPolicy::SATURATE);
+ _input_to_output_outstage_res.allocator()->allocate();
+
+ if(_has_peephole)
+ {
+ // TODO(COMPMID-3396): Perform multiplication in the quantized domain in CLPixelWiseMultiplicationKernel
+ // Here we are not using the output stage because all operations are done in float
+ // const float cell_to_output_scale = std::pow(2, cell_shift) * lstm_params.cell_to_output_weights()->info()->quantization_info().uniform().scale / lstm_params.output_intermediate_scale();
+ // quantization::calculate_quantized_multiplier(cell_to_output_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift);
+ _memory_group.manage(&_mul_cell_to_output_res);
+ _pixelwise_mul_cell_to_output.configure(cell_state_out, lstm_params.cell_to_output_weights(), &_mul_cell_to_output_res, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+ _accumulate_cell_to_output.configure(ArithmeticOperation::ADD, &_recurrent_to_output_outstage_res, &_mul_cell_to_output_res, &_recurrent_to_output_outstage_res, ConvertPolicy::SATURATE);
+ _mul_cell_to_output_res.allocator()->allocate();
+ }
+
+ const TensorInfo output_gate_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, sigmoid_tanh_outqinfo);
+ _memory_group.manage(&_output_gate);
+ _output_gate.allocator()->init(output_gate_info);
+ _output_gate_sigmoid.configure(&_recurrent_to_output_outstage_res, &_output_gate, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC));
+ _recurrent_to_output_outstage_res.allocator()->allocate();
+
+ // Hidden.
+ _hidden_tanh.configure(cell_state_out, &_input_gate, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::TANH, 1.f, 1.f));
+ // TODO(COMPMID-3396): Perform multiplication in the quantized domain in CLPixelWiseMultiplicationKernel
+ _memory_group.manage(&_hidden_mul_res);
+ const TensorInfo hidden_mul_res(_input_gate.info()->tensor_shape(), 1, DataType::S32);
+ _hidden_mul_res.allocator()->init(hidden_mul_res);
+ _pixelwise_mul_hidden.configure(&_output_gate, &_input_gate, &_hidden_mul_res, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO);
+ _output_gate.allocator()->allocate();
+ _input_gate.allocator()->allocate();
+ const float hidden_state_scale = std::pow(2, -15) / lstm_params.hidden_state_scale() * std::pow(2, -15);
+ quantization::calculate_quantized_multiplier(hidden_state_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift, /* ignore_epsilon */ true);
+ gemmlowp_info.gemmlowp_offset = lstm_params.hidden_state_zero();
+ gemmlowp_info.output_data_type = output_state_in->info()->data_type();
+ _hidden_outstage.configure(&_hidden_mul_res, nullptr, output_state_out, gemmlowp_info);
+ _hidden_mul_res.allocator()->allocate();
+
+ // Projection.
+ if(_has_projection)
+ {
+ const TensorInfo projection_outstage_info(*output_state_out->info());
+ const UniformQuantizationInfo qprojection = _projection_weights->info()->quantization_info().uniform();
+ const float projection_scale = qprojection.scale * lstm_params.hidden_state_scale() / qoutput_state_in.scale;
+ gemmlowp_info.gemmlowp_offset = qoutput_state_in.offset;
+ gemmlowp_info.gemmlowp_min_bound = std::numeric_limits<int8_t>::lowest();
+ gemmlowp_info.gemmlowp_max_bound = std::numeric_limits<int8_t>::max();
+ gemmlowp_info.output_data_type = DataType::QASYMM8_SIGNED;
+
+ configure_mm(_mm_projection, _projection_outstage, gemmlowp_info,
+ output_state_out, &_projection_weights_transposed, &_projection_eff_bias,
+ &_mm_projection_res, &_projection_outstage_res, projection_scale,
+ mm_out_info, projection_outstage_info);
+
+ _accumulate_projection.configure(ArithmeticOperation::ADD, &_projection_outstage_res, output_state_out, output_state_out, ConvertPolicy::SATURATE);
+ _projection_outstage_res.allocator()->allocate();
+
+ int8_t quantized_projection_clip{ 0 };
+ if(lstm_params.projection_clip() > 0.0f)
+ {
+ quantized_projection_clip = utility::clamp<int8_t>(lstm_params.projection_clip() / qprojection.scale, -128, 127);
+ }
+
+ if(quantized_projection_clip > 0)
+ {
+ _projection_clip.configure(output_state_out, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -quantized_projection_clip, quantized_projection_clip));
+ _has_projection_clipping = true;
+ }
+ }
+}
+
+Status CLQLSTMLayer::validate(const ITensorInfo *input,
+ const ITensorInfo *input_to_forget_weights, const ITensorInfo *input_to_cell_weights, const ITensorInfo *input_to_output_weights,
+ const ITensorInfo *recurrent_to_forget_weights, const ITensorInfo *recurrent_to_cell_weights, const ITensorInfo *recurrent_to_output_weights,
+ const ITensorInfo *forget_gate_bias, const ITensorInfo *cell_bias, const ITensorInfo *output_gate_bias,
+ const ITensorInfo *cell_state_in, const ITensorInfo *output_state_in,
+ const ITensorInfo *cell_state_out, const ITensorInfo *output_state_out,
+ const LSTMParams<ITensorInfo> &lstm_params)
+{
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, input_to_forget_weights, input_to_cell_weights, input_to_output_weights, recurrent_to_forget_weights, recurrent_to_cell_weights,
+ recurrent_to_output_weights, forget_gate_bias, cell_bias, output_gate_bias, cell_state_in, output_state_in, cell_state_out, output_state_out);
+
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QASYMM8_SIGNED);
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->num_dimensions() != 2, "Input must have exactly 2 dimensions");
+
+ const unsigned int input_size = input->dimension(0);
+ const unsigned int batch_size = input->dimension(1);
+ const unsigned int num_units = input_to_output_weights->dimension(1);
+ const unsigned int output_size = recurrent_to_output_weights->dimension(0);
+
+ ARM_COMPUTE_RETURN_ERROR_ON(input_to_output_weights->num_dimensions() != 2);
+ ARM_COMPUTE_RETURN_ERROR_ON(input_to_output_weights->dimension(0) != input_size);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input_to_output_weights, input_to_forget_weights, input_to_cell_weights);
+ ARM_COMPUTE_RETURN_ERROR_ON(recurrent_to_output_weights->num_dimensions() != 2);
+ ARM_COMPUTE_RETURN_ERROR_ON(recurrent_to_output_weights->dimension(1) != num_units);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(recurrent_to_output_weights, recurrent_to_forget_weights, recurrent_to_cell_weights);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input_to_forget_weights, 1, DataType::QSYMM8);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input_to_forget_weights, input_to_cell_weights, input_to_output_weights,
+ recurrent_to_forget_weights, recurrent_to_cell_weights, recurrent_to_output_weights);
+
+ ARM_COMPUTE_RETURN_ERROR_ON(forget_gate_bias->num_dimensions() != 1);
+ ARM_COMPUTE_RETURN_ERROR_ON(forget_gate_bias->dimension(0) != num_units);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(forget_gate_bias, cell_bias, output_gate_bias);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(forget_gate_bias, 1, DataType::S32);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(forget_gate_bias, cell_bias, output_gate_bias);
+
+ ARM_COMPUTE_RETURN_ERROR_ON(cell_state_in->num_dimensions() != 2);
+ ARM_COMPUTE_RETURN_ERROR_ON(cell_state_in->dimension(0) != num_units);
+ ARM_COMPUTE_RETURN_ERROR_ON(cell_state_in->dimension(1) != batch_size);
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(cell_state_in, 1, DataType::QSYMM16);
+
+ ARM_COMPUTE_RETURN_ERROR_ON(output_state_in->num_dimensions() != 2);
+ ARM_COMPUTE_RETURN_ERROR_ON(output_state_in->dimension(0) != output_size);
+ ARM_COMPUTE_RETURN_ERROR_ON(output_state_in->dimension(1) != batch_size);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output_state_in);
+
+ // Check whether peephole weights are all there or none
+ if(lstm_params.has_peephole_opt())
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lstm_params.cell_to_forget_weights(), lstm_params.cell_to_output_weights());
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lstm_params.cell_to_forget_weights(), 1, DataType::QSYMM16);
+ ARM_COMPUTE_RETURN_ERROR_ON(lstm_params.cell_to_forget_weights()->num_dimensions() != 1);
+ ARM_COMPUTE_RETURN_ERROR_ON(lstm_params.cell_to_forget_weights()->dimension(0) != num_units);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(lstm_params.cell_to_forget_weights(), lstm_params.cell_to_output_weights());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(lstm_params.cell_to_forget_weights(), lstm_params.cell_to_output_weights());
+
+ if(!lstm_params.has_cifg_opt())
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lstm_params.cell_to_input_weights());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(lstm_params.cell_to_forget_weights(), lstm_params.cell_to_input_weights());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(lstm_params.cell_to_forget_weights(), lstm_params.cell_to_input_weights());
+ }
+ }
+
+ const UniformQuantizationInfo qinput = input->quantization_info().uniform();
+ const UniformQuantizationInfo qcell_state_in = cell_state_in->quantization_info().uniform();
+ const UniformQuantizationInfo qoutput_state_in = output_state_in->quantization_info().uniform();
+
+ // Calculate and decompose effective scales for optimizing matmul calculation
+ const int32_t cell_shift = log2(qcell_state_in.scale);
+ ARM_COMPUTE_RETURN_ERROR_ON(cell_shift > -9);
+
+ // Calculate quantized parameters for clipping.
+ int16_t quantized_cell_clip = 0;
+ if(lstm_params.cell_clip() > 0.0f)
+ {
+ quantized_cell_clip = quantize_qsymm16(lstm_params.cell_clip(), qcell_state_in);
+ }
+
+ // Precompute effective bias for optimizing the matmul computations.
+ const TensorInfo eff_bias_info(TensorShape(num_units), 1, DataType::S32);
+ if(!lstm_params.has_cifg_opt())
+ {
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(lstm_params.input_to_input_weights(), &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, -qinput.offset, true)));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(lstm_params.recurrent_to_input_weights(), &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, -qoutput_state_in.offset,
+ true)));
+ }
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(input_to_forget_weights, &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, -qinput.offset, true)));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(recurrent_to_forget_weights, &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, -qoutput_state_in.offset, true)));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(input_to_cell_weights, &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, -qinput.offset, true)));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(recurrent_to_cell_weights, &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, -qoutput_state_in.offset, true)));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(input_to_output_weights, &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, -qinput.offset, true)));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(recurrent_to_output_weights, &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, -qoutput_state_in.offset, true)));
+ if(lstm_params.projection_bias() != nullptr)
+ {
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixAReductionKernel::validate(lstm_params.projection_weights(), &eff_bias_info, GEMMLowpReductionKernelInfo(num_units, false, lstm_params.hidden_state_zero(),
+ true)));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, lstm_params.projection_bias(), &eff_bias_info, &eff_bias_info, ConvertPolicy::SATURATE));
+ }
+
+ const TensorInfo input_weights_transposed(TensorShape(num_units, input_size), 1, input_to_forget_weights->data_type(), input_to_forget_weights->quantization_info());
+ const TensorInfo recurrent_weights_transposed(TensorShape(num_units, output_size), 1, recurrent_to_forget_weights->data_type(), recurrent_to_forget_weights->quantization_info());
+
+ // Validate weights transpose
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(input_to_forget_weights, &input_weights_transposed));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(input_to_cell_weights, &input_weights_transposed));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(input_to_output_weights, &input_weights_transposed));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(recurrent_to_forget_weights, &recurrent_weights_transposed));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(recurrent_to_cell_weights, &recurrent_weights_transposed));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(recurrent_to_output_weights, &recurrent_weights_transposed));
+ if(!lstm_params.has_cifg_opt())
+ {
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(lstm_params.input_to_input_weights(), &input_weights_transposed));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(lstm_params.recurrent_to_input_weights(), &recurrent_weights_transposed));
+ }
+ if(lstm_params.has_projection())
+ {
+ ARM_COMPUTE_RETURN_ON_ERROR(CLTranspose::validate(lstm_params.projection_weights(), &recurrent_weights_transposed));
+ }
+
+ GEMMLowpOutputStageInfo gemmlowp_info;
+ gemmlowp_info.type = GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT;
+ gemmlowp_info.gemmlowp_min_bound = std::numeric_limits<int16_t>::lowest();
+ gemmlowp_info.gemmlowp_max_bound = std::numeric_limits<int16_t>::max();
+ gemmlowp_info.output_data_type = DataType::QSYMM16;
+
+ // Forget gate.
+ const TensorInfo forget_outstage_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.forget_intermediate_scale(), 0));
+ const TensorInfo mm_out_info(TensorShape(num_units, batch_size), 1, DataType::S32);
+ const float input_to_forget_scale = input_to_forget_weights->quantization_info().uniform().scale * qinput.scale / lstm_params.forget_intermediate_scale();
+ validate_mm(gemmlowp_info, input, &input_weights_transposed, &eff_bias_info, input_to_forget_scale, &mm_out_info, &forget_outstage_info);
+
+ const float recurrent_to_forget_scale = recurrent_to_forget_weights->quantization_info().uniform().scale * qoutput_state_in.scale / lstm_params.forget_intermediate_scale();
+ validate_mm(gemmlowp_info, input, &recurrent_weights_transposed, &eff_bias_info, recurrent_to_forget_scale, &mm_out_info, &forget_outstage_info);
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, &forget_outstage_info, &forget_outstage_info, &forget_outstage_info, ConvertPolicy::SATURATE));
+
+ if(lstm_params.has_peephole_opt())
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lstm_params.cell_to_forget_weights(), 1, DataType::QSYMM16);
+ ARM_COMPUTE_RETURN_ON_ERROR(CLPixelWiseMultiplicationKernel::validate(cell_state_in, lstm_params.cell_to_forget_weights(), &mm_out_info, 1.f, ConvertPolicy::SATURATE,
+ RoundingPolicy::TO_ZERO));
+ const float cell_to_forget_scale = std::pow(2, cell_shift) * lstm_params.cell_to_forget_weights()->quantization_info().uniform().scale / lstm_params.forget_intermediate_scale();
+ ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(cell_to_forget_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpOutputStage::validate(&mm_out_info, nullptr, &forget_outstage_info, gemmlowp_info));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, &forget_outstage_info, &forget_outstage_info, &forget_outstage_info, ConvertPolicy::SATURATE));
+ }
+
+ // Output quantization info of Sigmoid and Tanh activations
+ const QuantizationInfo sigmoid_tanh_outqinfo(1.f / 32768.f, 0);
+
+ const TensorInfo forget_gate_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, sigmoid_tanh_outqinfo);
+ ARM_COMPUTE_RETURN_ON_ERROR(CLActivationLayer::validate(&forget_outstage_info, &forget_gate_info, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
+
+ // Modulation gate.
+ const TensorInfo cell_outstage_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.cell_intermediate_scale(), 0));
+ const float input_to_cell_scale = input_to_cell_weights->quantization_info().uniform().scale * qinput.scale / lstm_params.cell_intermediate_scale();
+ validate_mm(gemmlowp_info, input, &input_weights_transposed, &eff_bias_info, input_to_cell_scale, &mm_out_info, &cell_outstage_info);
+
+ const float recurrent_to_cell_scale = recurrent_to_cell_weights->quantization_info().uniform().scale * qoutput_state_in.scale / lstm_params.cell_intermediate_scale();
+ validate_mm(gemmlowp_info, input, &input_weights_transposed, &eff_bias_info, recurrent_to_cell_scale, &mm_out_info, &cell_outstage_info);
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, &cell_outstage_info, &cell_outstage_info, &cell_outstage_info, ConvertPolicy::SATURATE));
+
+ const TensorInfo cell_gate_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, sigmoid_tanh_outqinfo);
+ ARM_COMPUTE_RETURN_ON_ERROR(CLActivationLayer::validate(&cell_outstage_info, &cell_gate_info, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::TANH, 1.f, 1.f)));
+
+ // Input gate.
+ const TensorInfo input_gate_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, sigmoid_tanh_outqinfo);
+ if(lstm_params.has_cifg_opt())
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_MSG(lstm_params.input_gate_bias() != nullptr, "Input gate bias must not be present when CIFG is used");
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::SUB, &input_gate_info, &forget_gate_info, &forget_gate_info, ConvertPolicy::SATURATE));
+ }
+ else
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lstm_params.input_to_input_weights(), lstm_params.recurrent_to_input_weights(), lstm_params.input_gate_bias());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input_to_forget_weights, lstm_params.input_to_input_weights(), lstm_params.recurrent_to_input_weights());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input_to_forget_weights, lstm_params.input_to_input_weights());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(recurrent_to_forget_weights, lstm_params.recurrent_to_input_weights());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(forget_gate_bias, lstm_params.input_gate_bias());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(forget_gate_bias, lstm_params.input_gate_bias());
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpMatrixMultiplyCore::validate(input, lstm_params.input_to_input_weights(), nullptr, &mm_out_info));
+ const TensorInfo input_outstage_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.input_intermediate_scale(), 0));
+ const float input_to_input_scale = lstm_params.input_to_input_weights()->quantization_info().uniform().scale * qinput.scale / lstm_params.input_intermediate_scale();
+ validate_mm(gemmlowp_info, input, lstm_params.input_to_input_weights(), &eff_bias_info, input_to_input_scale, &mm_out_info, &input_outstage_info);
+
+ const float recurrent_to_input_scale = lstm_params.recurrent_to_input_weights()->quantization_info().uniform().scale * qoutput_state_in.scale / lstm_params.input_intermediate_scale();
+ validate_mm(gemmlowp_info, input, lstm_params.recurrent_to_input_weights(), &eff_bias_info, recurrent_to_input_scale, &mm_out_info, &input_outstage_info);
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, &input_outstage_info, &input_outstage_info, &input_outstage_info, ConvertPolicy::SATURATE));
+
+ if(lstm_params.has_peephole_opt())
+ {
+ ARM_COMPUTE_RETURN_ON_ERROR(CLPixelWiseMultiplicationKernel::validate(cell_state_in, lstm_params.cell_to_input_weights(), &input_outstage_info, 1.f, ConvertPolicy::SATURATE,
+ RoundingPolicy::TO_ZERO));
+ const float cell_to_input_scale = std::pow(2, cell_shift) * lstm_params.cell_to_input_weights()->quantization_info().uniform().scale / lstm_params.input_intermediate_scale();
+ ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(cell_to_input_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpOutputStage::validate(&input_outstage_info, &eff_bias_info, &input_outstage_info, gemmlowp_info));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, &input_outstage_info, &input_outstage_info, &input_outstage_info, ConvertPolicy::SATURATE));
+ }
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLActivationLayer::validate(&input_outstage_info, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::TANH, 1.f, 1.f)));
+ }
+ // Cell.
+ ARM_COMPUTE_RETURN_ON_ERROR(CLPixelWiseMultiplicationKernel::validate(&forget_gate_info, cell_state_in, &forget_gate_info, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLPixelWiseMultiplicationKernel::validate(&input_gate_info, cell_state_in, &cell_gate_info, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, &forget_gate_info, &cell_gate_info, cell_state_out, ConvertPolicy::SATURATE));
+ if(quantized_cell_clip > 0)
+ {
+ ARM_COMPUTE_RETURN_ON_ERROR(CLActivationLayer::validate(cell_state_out, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -quantized_cell_clip,
+ quantized_cell_clip)));
+ }
+ // Output gate.
+ const TensorInfo output_outstage_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, QuantizationInfo(lstm_params.output_intermediate_scale(), 0));
+ const float input_to_output_scale = input_to_output_weights->quantization_info().uniform().scale * qinput.scale / lstm_params.output_intermediate_scale();
+ validate_mm(gemmlowp_info, input, &input_weights_transposed, &eff_bias_info, input_to_output_scale, &mm_out_info, &output_outstage_info);
+
+ const float recurrent_to_output_scale = recurrent_to_output_weights->quantization_info().uniform().scale * qoutput_state_in.scale / lstm_params.output_intermediate_scale();
+ validate_mm(gemmlowp_info, output_state_in, &recurrent_weights_transposed, &eff_bias_info, recurrent_to_output_scale, &mm_out_info, &output_outstage_info);
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, &output_outstage_info, &output_outstage_info, &output_outstage_info, ConvertPolicy::SATURATE));
+ if(lstm_params.has_peephole_opt())
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lstm_params.cell_to_output_weights(), 1, DataType::QSYMM16);
+ // TODO(COMPMID-3395): Perform multiplication in the quantized domain in NEPixelWiseMultiplicationKernel
+ // Here we are not using the output stage because all operations are done in float
+ // const float cell_to_output_scale = std::pow(2, cell_shift) * lstm_params.cell_to_output_weights()->quantization_info().uniform().scale / lstm_params.output_intermediate_scale();
+ // ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(cell_to_output_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLPixelWiseMultiplicationKernel::validate(cell_state_out, lstm_params.cell_to_output_weights(), &output_outstage_info, 1.f, ConvertPolicy::SATURATE,
+ RoundingPolicy::TO_ZERO));
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, &output_outstage_info, &output_outstage_info, &output_outstage_info, ConvertPolicy::SATURATE));
+ }
+
+ const TensorInfo output_gate_info(TensorShape(num_units, batch_size), 1, DataType::QSYMM16, sigmoid_tanh_outqinfo);
+ ARM_COMPUTE_RETURN_ON_ERROR(CLActivationLayer::validate(&output_outstage_info, &output_gate_info, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
+
+ // Hidden.
+ ARM_COMPUTE_RETURN_ON_ERROR(CLActivationLayer::validate(cell_state_out, &input_gate_info, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::TANH, 1.f, 1.f)));
+ const TensorInfo hidden_mul_res(TensorShape(num_units, batch_size), 1, DataType::S32);
+ ARM_COMPUTE_RETURN_ON_ERROR(CLPixelWiseMultiplicationKernel::validate(&output_gate_info, &input_gate_info, &hidden_mul_res, 1.f, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
+ const float hidden_state_scale = std::pow(2, -15) / lstm_params.hidden_state_scale() * std::pow(2, -15);
+ ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(hidden_state_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift, /* ignore_epsilon */ true));
+ gemmlowp_info.gemmlowp_offset = lstm_params.hidden_state_zero();
+ ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpOutputStage::validate(&hidden_mul_res, nullptr, output_state_out, gemmlowp_info));
+
+ // Projection.
+ if(lstm_params.has_projection())
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(recurrent_to_forget_weights, lstm_params.projection_weights());
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(forget_gate_bias, lstm_params.projection_bias());
+
+ const UniformQuantizationInfo qprojection = lstm_params.projection_weights()->quantization_info().uniform();
+ const float projection_scale = qprojection.scale * lstm_params.hidden_state_scale() / qoutput_state_in.scale;
+ ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(projection_scale, &gemmlowp_info.gemmlowp_multiplier, &gemmlowp_info.gemmlowp_shift));
+ gemmlowp_info.gemmlowp_offset = qoutput_state_in.offset;
+ gemmlowp_info.gemmlowp_min_bound = std::numeric_limits<int8_t>::lowest();
+ gemmlowp_info.gemmlowp_max_bound = std::numeric_limits<int8_t>::max();
+ gemmlowp_info.output_data_type = DataType::QASYMM8_SIGNED;
+
+ const TensorInfo projection_outstage_info(*output_state_out);
+ validate_mm(gemmlowp_info, output_state_out, &recurrent_weights_transposed, &eff_bias_info, input_to_output_scale, &mm_out_info, &projection_outstage_info);
+
+ ARM_COMPUTE_RETURN_ON_ERROR(CLSaturatedArithmeticOperationKernel::validate(ArithmeticOperation::ADD, output_state_out, output_state_out, output_state_out, ConvertPolicy::SATURATE));
+
+ int8_t quantized_projection_clip{ 0 };
+ if(lstm_params.projection_clip() > 0.0f)
+ {
+ quantized_projection_clip = quantize_qasymm8_signed(lstm_params.projection_clip(), qprojection);
+ }
+
+ if(quantized_projection_clip > 0)
+ {
+ ARM_COMPUTE_RETURN_ON_ERROR(CLActivationLayer::validate(output_state_out, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -quantized_projection_clip,
+ quantized_projection_clip)));
+ }
+ }
+
+ if(cell_state_out->total_size() > 0)
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(cell_state_in, cell_state_out);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(cell_state_in, cell_state_out);
+ }
+
+ if(output_state_out->total_size() > 0)
+ {
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output_state_out);
+ ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output_state_in, output_state_out);
+ }
+
+ return Status{};
+}
+
+void CLQLSTMLayer::run()
+{
+ prepare();
+
+ // Acquire all the temporaries
+ MemoryGroupResourceScope scope_mg(_memory_group);
+
+ // Forget gate.
+ _mm_input_to_forget.run();
+ _input_to_forget_outstage.run();
+
+ _mm_recurrent_to_forget.run();
+ _recurrent_to_forget_outstage.run();
+ CLScheduler::get().enqueue(_accumulate_input_recurrent_forget);
+
+ if(_has_peephole)
+ {
+ CLScheduler::get().enqueue(_pixelwise_mul_cell_to_forget);
+ _cell_to_forget_outstage.run();
+ CLScheduler::get().enqueue(_accumulate_cell_forget);
+ }
+
+ _forget_gate_sigmoid.run();
+
+ // Modulation gate.
+ _mm_input_to_cell.run();
+ _input_to_cell_outstage.run();
+
+ _mm_recurrent_to_cell.run();
+ _recurrent_to_cell_outstage.run();
+ CLScheduler::get().enqueue(_accumulate_input_recurrent_modulation);
+
+ _cell_gate_tanh.run();
+
+ // Input gate
+ if(_has_cifg)
+ {
+ CLScheduler::get().enqueue(_input_gate_sub);
+ }
+ else
+ {
+ _mm_input_to_input.run();
+ _input_to_input_outstage.run();
+ _mm_recurrent_to_input.run();
+ _recurrent_to_input_outstage.run();
+ CLScheduler::get().enqueue(_accumulate_input_recurrent_input);
+
+ if(_has_peephole)
+ {
+ CLScheduler::get().enqueue(_pixelwise_mul_cell_to_input);
+ _cell_to_input_outstage.run();
+ CLScheduler::get().enqueue(_accumulate_cell_input);
+ }
+
+ _input_gate_tanh.run();
+ }
+
+ // Cell.
+ CLScheduler::get().enqueue(_pixelwise_mul_forget_cell);
+ CLScheduler::get().enqueue(_pixelwise_mul_input_cell);
+ CLScheduler::get().enqueue(_add_forget_cell);
+ if(_has_cell_clipping)
+ {
+ _cell_clip.run();
+ }
+
+ // Output gate.
+ _mm_input_to_output.run();
+ _input_to_output_outstage.run();
+ _mm_recurrent_to_output.run();
+ _recurrent_to_output_outstage.run();
+ CLScheduler::get().enqueue(_accumulate_input_recurrent_output);
+ if(_has_peephole)
+ {
+ CLScheduler::get().enqueue(_pixelwise_mul_cell_to_output);
+ CLScheduler::get().enqueue(_accumulate_cell_to_output);
+ }
+
+ _output_gate_sigmoid.run();
+
+ // Hidden.
+ _hidden_tanh.run();
+ CLScheduler::get().enqueue(_pixelwise_mul_hidden);
+ _hidden_outstage.run();
+
+ // Projection.
+ if(_has_projection)
+ {
+ _mm_projection.run();
+ _projection_outstage.run();
+ CLScheduler::get().enqueue(_accumulate_projection);
+ if(_has_projection_clipping)
+ {
+ _projection_clip.run();
+ }
+ }
+}
+
+void CLQLSTMLayer::prepare()
+{
+ if(!_is_prepared)
+ {
+ // Pre-transpose weights to be used in GEMM.
+ _input_to_forget_weights_transposed.allocator()->allocate();
+ _input_to_cell_weights_transposed.allocator()->allocate();
+ _input_to_output_weights_transposed.allocator()->allocate();
+ _recurrent_to_forget_weights_transposed.allocator()->allocate();
+ _recurrent_to_cell_weights_transposed.allocator()->allocate();
+ _recurrent_to_output_weights_transposed.allocator()->allocate();
+ _transpose_input_to_forget_weights.run();
+ _transpose_input_to_cell_weights.run();
+ _transpose_input_to_output_weights.run();
+ _transpose_recurrent_to_forget_weights.run();
+ _transpose_recurrent_to_cell_weights.run();
+ _transpose_recurrent_to_output_weights.run();
+
+ // Precompute effective biases
+ if(_has_cifg)
+ {
+ _ones.map(true);
+ std::fill_n(reinterpret_cast<int16_t *>(_ones.buffer()), _ones.info()->total_size() / _ones.info()->element_size(), 32767);
+ _ones.unmap();
+ }
+ else
+ {
+ _input_to_input_eff_bias.allocator()->allocate();
+ _recurrent_to_input_eff_bias.allocator()->allocate();
+ CLScheduler::get().enqueue(_input_to_input_reduction);
+ CLScheduler::get().enqueue(_recurrent_to_input_reduction);
+
+ _input_to_input_weights_transposed.allocator()->allocate();
+ _recurrent_to_input_weights_transposed.allocator()->allocate();
+ _transpose_input_to_input_weights.run();
+ _transpose_recurrent_to_input_weights.run();
+ _input_to_input_weights->mark_as_unused();
+ _recurrent_to_input_weights->mark_as_unused();
+ }
+ _input_to_forget_eff_bias.allocator()->allocate();
+ _recurrent_to_forget_eff_bias.allocator()->allocate();
+ _input_to_cell_eff_bias.allocator()->allocate();
+ _recurrent_to_cell_eff_bias.allocator()->allocate();
+ _input_to_output_eff_bias.allocator()->allocate();
+ _recurrent_to_output_eff_bias.allocator()->allocate();
+ CLScheduler::get().enqueue(_input_to_forget_reduction);
+ CLScheduler::get().enqueue(_recurrent_to_forget_reduction);
+ CLScheduler::get().enqueue(_input_to_cell_reduction);
+ CLScheduler::get().enqueue(_recurrent_to_cell_reduction);
+ CLScheduler::get().enqueue(_input_to_output_reduction);
+ CLScheduler::get().enqueue(_recurrent_to_output_reduction);
+
+ if(_has_projection)
+ {
+ if(_projection_bias != nullptr)
+ {
+ _projection_eff_bias.allocator()->allocate();
+ CLScheduler::get().enqueue(_projection_reduction);
+ _projection_bias->mark_as_unused();
+ }
+
+ _projection_weights_transposed.allocator()->allocate();
+ _transpose_projection_weights.run();
+ _projection_weights->mark_as_unused();
+ }
+
+ // Mark weights as unused
+ _input_to_forget_weights->mark_as_unused();
+ _input_to_cell_weights->mark_as_unused();
+ _input_to_output_weights->mark_as_unused();
+ _recurrent_to_forget_weights->mark_as_unused();
+ _recurrent_to_cell_weights->mark_as_unused();
+ _recurrent_to_output_weights->mark_as_unused();
+
+ CLScheduler::get().queue().finish();
+ _is_prepared = true;
+ }
+}
+
+} // namespace arm_compute