blob: ebd3f602809a3f20cc4717c39e1abb8135cbe64d [file] [log] [blame]
Manuel Bottiniae58bdf2021-06-17 17:18:45 +01001/*
2 * Copyright (c) 2021 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
Georgios Pinitas7891a732021-08-20 21:39:25 +010024#include "src/cpu/operators/CpuGemmLowpOutputStage.h"
Manuel Bottiniae58bdf2021-06-17 17:18:45 +010025
26#include "arm_compute/core/ITensor.h"
27#include "arm_compute/core/Validate.h"
28#include "arm_compute/runtime/NEON/NEScheduler.h"
Georgios Pinitas7891a732021-08-20 21:39:25 +010029#include "src/cpu/kernels/CpuGemmLowpQuantizeDownInt32ScaleKernel.h"
30#include "src/cpu/kernels/CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h"
31#include "src/cpu/kernels/CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h"
32#include "src/cpu/kernels/CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h"
Manuel Bottiniae58bdf2021-06-17 17:18:45 +010033
34namespace arm_compute
35{
36namespace cpu
37{
38void CpuGemmLowpOutputStage::configure(ITensorInfo *src, ITensorInfo *bias, ITensorInfo *dst, const GEMMLowpOutputStageInfo &info)
39{
40 // Perform validate step
41 ARM_COMPUTE_ERROR_THROW_ON(CpuGemmLowpOutputStage::validate(src, bias, dst, info));
42
43 switch(info.type)
44 {
45 case GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT:
46 {
47 switch(info.output_data_type)
48 {
49 case DataType::QASYMM8:
50 {
51 auto k = std::make_unique<kernels::CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel>();
52 k->configure(src, bias, dst, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_offset, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
53 _kernel = std::move(k);
54 break;
55 }
56 case DataType::QASYMM8_SIGNED:
57 {
58 auto k = std::make_unique<kernels::CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel>();
59 k->configure(src, bias, dst, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_offset, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
60 _kernel = std::move(k);
61 break;
62 }
63 case DataType::QSYMM16:
64 {
65 auto k = std::make_unique<kernels::CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel>();
66 k->configure(src, bias, dst, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
67 _kernel = std::move(k);
68 break;
69 }
70 default:
71 {
72 ARM_COMPUTE_ERROR("Unsupported output data type.");
73 break;
74 }
75 }
76 break;
77 }
78 case GEMMLowpOutputStageType::QUANTIZE_DOWN:
79 {
80 switch(info.output_data_type)
81 {
82 case DataType::QASYMM8:
83 case DataType::QASYMM8_SIGNED:
84 {
85 auto k = std::make_unique<kernels::CpuGemmLowpQuantizeDownInt32ScaleKernel>();
86 k->configure(src, bias, dst, &info);
87 _kernel = std::move(k);
88 break;
89 }
90 default:
91 {
92 ARM_COMPUTE_ERROR("Unsupported output data type.");
93 break;
94 }
95 }
96 break;
97 }
98 default:
99 ARM_COMPUTE_ERROR("Unsupported GEMMLowpOutputStage type.");
100 }
101}
102
103Status CpuGemmLowpOutputStage::validate(const ITensorInfo *src, const ITensorInfo *bias, const ITensorInfo *dst, const GEMMLowpOutputStageInfo &info)
104{
105 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
106 ARM_COMPUTE_RETURN_ERROR_ON_MSG(dst->data_type() == DataType::UNKNOWN, "CpuGemmLowpOutputStage cannot be used with UNKNOWN output data type.");
107 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(dst, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QSYMM16);
108 ARM_COMPUTE_RETURN_ERROR_ON((info.type != GEMMLowpOutputStageType::QUANTIZE_DOWN) && (info.type != GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT));
109
110 switch(info.type)
111 {
112 case GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT:
113 {
114 switch(dst->data_type())
115 {
116 case DataType::QASYMM8:
117 return kernels::CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::validate(src, bias, dst, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
118 case DataType::QASYMM8_SIGNED:
119 return kernels::CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::validate(src, bias, dst, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
120 case DataType::QSYMM16:
121 return kernels::CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::validate(src, bias, dst, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
122 default:
123 return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported output data type.");
124 }
125 }
126 case GEMMLowpOutputStageType::QUANTIZE_DOWN:
127 {
128 switch(dst->data_type())
129 {
130 case DataType::QASYMM8:
131 case DataType::QASYMM8_SIGNED:
132 return kernels::CpuGemmLowpQuantizeDownInt32ScaleKernel::validate(src, bias, dst, &info);
133 default:
134 return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported output data type.");
135 }
136 }
137 default:
138 return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported GEMMLowpOutputStage type.");
139 }
140}
141
142void CpuGemmLowpOutputStage::run(ITensorPack &tensors)
143{
144 NEScheduler::get().schedule_op(_kernel.get(), Window::DimY, _kernel->window(), tensors);
145}
146} // namespace cpu
147} // namespace arm_compute