1 /*
2 * Copyright (c) 2017-2021 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 */
24 #include "src/gpu/cl/operators/ClGemmLowpOutputStage.h"
25
26 #include "arm_compute/core/CL/ICLTensor.h"
27 #include "arm_compute/core/Types.h"
28 #include "arm_compute/runtime/CL/CLScheduler.h"
29
30 #include "src/gpu/cl/kernels/ClGemmLowpQuantizeDownInt32ScaleByFixedPointKernel.h"
31 #include "src/gpu/cl/kernels/ClGemmLowpQuantizeDownInt32ScaleByFloatKernel.h"
32 #include "src/gpu/cl/kernels/ClGemmLowpQuantizeDownInt32ScaleKernel.h"
33
34 #include "src/common/utils/Log.h"
35
36 namespace arm_compute
37 {
38 namespace opencl
39 {
configure(const CLCompileContext & compile_context,const ITensorInfo * src,const ITensorInfo * bias,ITensorInfo * dst,const GEMMLowpOutputStageInfo & info)40 void ClGemmLowpOutputStage::configure(const CLCompileContext &compile_context, const ITensorInfo *src, const ITensorInfo *bias, ITensorInfo *dst, const GEMMLowpOutputStageInfo &info)
41 {
42 ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
43 ARM_COMPUTE_LOG_PARAMS(src, bias, dst, info);
44
45 switch(info.type)
46 {
47 case GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT:
48 {
49 auto k = std::make_unique<opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleByFixedPointKernel>();
50 k->configure(compile_context, src, bias, dst, &info);
51 _kernel = std::move(k);
52 break;
53 }
54 case GEMMLowpOutputStageType::QUANTIZE_DOWN:
55 {
56 auto k = std::make_unique<opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleKernel>();
57 k->configure(compile_context, src, bias, dst, &info);
58 _kernel = std::move(k);
59 break;
60 }
61 case GEMMLowpOutputStageType::QUANTIZE_DOWN_FLOAT:
62 {
63 auto k = std::make_unique<opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleByFloatKernel>();
64 k->configure(compile_context, src, bias, dst, &info);
65 _kernel = std::move(k);
66 break;
67 }
68 default:
69 ARM_COMPUTE_ERROR("Unsupported GEMMLowpOutputStage type.");
70 }
71 }
72
validate(const ITensorInfo * src,const ITensorInfo * bias,const ITensorInfo * dst,const GEMMLowpOutputStageInfo & info)73 Status ClGemmLowpOutputStage::validate(const ITensorInfo *src, const ITensorInfo *bias, const ITensorInfo *dst, const GEMMLowpOutputStageInfo &info)
74 {
75 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(dst);
76 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(dst, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QSYMM16);
77
78 switch(info.type)
79 {
80 case GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT:
81 return opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleByFixedPointKernel::validate(src, bias, dst, &info);
82 case GEMMLowpOutputStageType::QUANTIZE_DOWN:
83 return opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleKernel::validate(src, bias, dst, &info);
84 case GEMMLowpOutputStageType::QUANTIZE_DOWN_FLOAT:
85 return opencl::kernels::ClGemmLowpQuantizeDownInt32ScaleByFloatKernel::validate(src, bias, dst, &info);
86 default:
87 return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported GEMMLowpOutputStage type.");
88 }
89 }
90
run(ITensorPack & tensors)91 void ClGemmLowpOutputStage::run(ITensorPack &tensors)
92 {
93 const ITensor *src = tensors.get_const_tensor(ACL_SRC);
94 const ITensor *bias = tensors.get_const_tensor(ACL_BIAS);
95 ITensor *dst = tensors.get_tensor(ACL_DST);
96
97 ITensorPack pack{ { ACL_SRC, src }, { ACL_BIAS, bias }, { ACL_DST, dst } };
98 CLScheduler::get().enqueue_op(*_kernel, pack, true);
99 }
100 } // namespace opencl
101 } // namespace arm_compute
102