diff options
| author | John Kesapides <john.kesapides@arm.com> | 2019-02-04 12:37:29 +0000 |
|---|---|---|
| committer | Pablo Marquez <pablo.tello@arm.com> | 2019-03-14 09:29:15 +0000 |
| commit | 917959c88361e8148696c156453f69c6ae0c95c0 (patch) | |
| tree | fbdadd65904b446edcc275f9e85874d2dfb7f8d3 /arm_compute/runtime/NEON/functions/NELSTMLayer.h | |
| parent | 894066de8cc26d1a3aca62dcaa6b30a2a1116028 (diff) | |
| download | ComputeLibrary-917959c88361e8148696c156453f69c6ae0c95c0.tar.gz | |
COMPMID-1281 Investigate concatenation for RNN/LSTM NEON
Change-Id: I7f099348a361a6f2d4efb30618f58bd44dd41e6c
Signed-off-by: John Kesapides <john.kesapides@arm.com>
Reviewed-on: https://review.mlplatform.org/c/712
Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'arm_compute/runtime/NEON/functions/NELSTMLayer.h')
| -rw-r--r-- | arm_compute/runtime/NEON/functions/NELSTMLayer.h | 11 |
1 files changed, 8 insertions, 3 deletions
diff --git a/arm_compute/runtime/NEON/functions/NELSTMLayer.h b/arm_compute/runtime/NEON/functions/NELSTMLayer.h index b98e74d969..f3a1aa7c75 100644 --- a/arm_compute/runtime/NEON/functions/NELSTMLayer.h +++ b/arm_compute/runtime/NEON/functions/NELSTMLayer.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018 ARM Limited. + * Copyright (c) 2018-2019 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -132,6 +132,7 @@ public: // Inherited methods overridden: void run() override; + void prepare() override; private: MemoryGroup _memory_group; @@ -176,16 +177,20 @@ private: NECopyKernel _copy_cell_state; NECopyKernel _copy_output; NEWidthConcatenateLayer _concat_scratch_buffer; + NEWidthConcatenateLayer _concat_inputs_forget_gate; + NEWidthConcatenateLayer _concat_weights_forget_gate; + NEWidthConcatenateLayer _concat_weights_input_gate; + NEWidthConcatenateLayer _concat_weights_output; Tensor _input_gate_out1; Tensor _input_gate_out2; Tensor _input_gate_out3; Tensor _input_gate_out4; - Tensor _input_gate_out5; Tensor _forget_gate_out1; Tensor _forget_gate_out2; Tensor _forget_gate_out3; Tensor _forget_gate_out4; Tensor _forget_gate_out5; + Tensor _forget_gate_out6; Tensor _cell_state_out1; Tensor _cell_state_out2; Tensor _cell_state_out3; @@ -195,7 +200,6 @@ private: Tensor _output2; Tensor _output3; Tensor _output4; - Tensor _output5; Tensor _cell_state_activation; Tensor _output_state1; Tensor _ones; @@ -204,6 +208,7 @@ private: bool _perform_cell_clipping; bool _has_projection_weights; bool _perform_projection_clipping; + bool _is_prepared; }; } // namespace arm_compute #endif /* __ARM_COMPUTE_NELSTMLAYER_H__ */ |