diff options
author | Gian Marco <gianmarco.iodice@arm.com> | 2017-11-08 12:24:09 +0000 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:35:24 +0000 |
commit | e75a02b60736f37c34388c23c0ccee230f65da59 (patch) | |
tree | f8e9423e40589e99bd8be6c1e740b17792e2058e /arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h | |
parent | 6c0348f4cbf6e30a715780f50aebf6dd0a2a8fc3 (diff) | |
download | ComputeLibrary-e75a02b60736f37c34388c23c0ccee230f65da59.tar.gz |
COMPMID-675 - Reworked NEGEMMLowp interface/function
The new interface makes NEGEMMLowp able to work with ASYMM8 data types.
Implemented 2 new functions:
- NEGEMMLowpMatrixMultiplyCore
- NEGEMMLowpOutputStage
These functions should make the integration in android NN doable
For more information about GEMMLowp:
https://github.com/google/gemmlowp/blob/master/doc/low-precision.md
Change-Id: Ie2c775f45234f68ca53dba644b3a912b997fd890
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/95504
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Diffstat (limited to 'arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h')
-rw-r--r-- | arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h | 69 |
1 files changed, 69 insertions, 0 deletions
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h new file mode 100644 index 0000000000..8557ef42e1 --- /dev/null +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h @@ -0,0 +1,69 @@ +/* + * Copyright (c) 2017 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. + */ +#ifndef __ARM_COMPUTE_NEGEMMLOWPOUTPUTSTAGE_H__ +#define __ARM_COMPUTE_NEGEMMLOWPOUTPUTSTAGE_H__ + +#include "arm_compute/runtime/NEON/INESimpleFunction.h" + +/** This file contains all available output stages for GEMMLowp on NEON. + * + * In gemmlowp, the "output stage" is the process that takes a final int32 accumulator value (the output of @ref NEGEMMLowpMatrixMultiplyCore), + * and processes it to obtain the final ASYMM8 value. + * + * More information about the GEMMLowp output stage can be found at https://github.com/google/gemmlowp/blob/master/doc/output.md + */ + +namespace arm_compute +{ +class ITensor; + +/** Basic function to execute NEGEMMLowpQuantizeDownInt32ToUint8Scale on NEON. + * + * NEGEMMLowpQuantizeDownInt32ToUint8Scale depends on 3 parameters: result_offset, result_mult_int, result_shift + * The final result is: + * + * ((input[i][k] + result_offset) * result_mult_int + rounding) >> result_shift + * + * where rounding = (result_shift < 1) ? 0 : (1 << (result_shift - 1)) + * + * This function calls the following NEON kernels: + * + * -# @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleKernel + * +*/ +class NEGEMMLowpQuantizeDownInt32ToUint8Scale : public INESimpleFunction +{ +public: + /** Initialise the kernel's inputs, output + * + * @param[in] input Input tensor. It is the output of @ref NEGEMMLowpMatrixMultiplyCore function. Data type supported: S32 + * @param[out] output Output tensor. Data type supported: Data type supported: QASYMM8 + * @param[in] result_offset Offset to be added to each element of the input matrix + * @param[in] result_mult_int Value to be multiplied to each element of the input matrix when once the result_offset has been add + * @param[in] result_shift Number of bits to shift right the result before converting back to QASYMM8 + */ + void configure(const ITensor *input, ITensor *output, int result_offset, int result_mult_int, int result_shift); +}; +} +#endif /*__ARM_COMPUTE_NEGEMMLOWPOUTPUTSTAGE_H__ */
\ No newline at end of file |