1 /*
2 * Copyright (c) 2022 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
25 #include "arm_compute/core/Helpers.h"
26 #include "arm_compute/core/utils/misc/Traits.h"
27 #include "src/core/NEON/wrapper/intrinsics/intrinsics.h"
28
29 namespace arm_compute
30 {
31 namespace cpu
32 {
33 template <typename ScalarType, typename VectorType>
elementwise_op_sve_imp(svbool_t pg,ElementWiseUnary op,const VectorType & a)34 inline typename std::enable_if<utils::traits::is_floating_point<ScalarType>::value, VectorType>::type elementwise_op_sve_imp(svbool_t pg, ElementWiseUnary op, const VectorType &a)
35 {
36 switch(op)
37 {
38 case ElementWiseUnary::RSQRT:
39 return svinvsqrt(pg, a);
40 case ElementWiseUnary::EXP:
41 return wrapper::svexp_z(pg, a);
42 case ElementWiseUnary::NEG:
43 return svneg_z(pg, a);
44 case ElementWiseUnary::LOG:
45 return wrapper::svlog_z(pg, a);
46 case ElementWiseUnary::ABS:
47 return svabs_z(pg, a);
48 case ElementWiseUnary::ROUND:
49 return svrintn_z(pg, a);
50 case ElementWiseUnary::SIN:
51 return wrapper::svsin_z(pg, a);
52 default:
53 ARM_COMPUTE_ERROR("NOT_SUPPORTED");
54 }
55 }
56
57 template <typename ScalarType, typename VectorType>
elementwise_op_sve_imp(svbool_t pg,ElementWiseUnary op,const VectorType & a)58 inline typename std::enable_if<std::is_integral<ScalarType>::value, VectorType>::type elementwise_op_sve_imp(svbool_t pg, ElementWiseUnary op, const VectorType &a)
59 {
60 switch(op)
61 {
62 case ElementWiseUnary::NEG:
63 return svneg_z(pg, a);
64 case ElementWiseUnary::ABS:
65 return svabs_z(pg, a);
66 default:
67 ARM_COMPUTE_ERROR("NOT_SUPPORTED");
68 }
69 }
70
71 template <typename ScalarType>
elementwise_sve_op(const ITensor * in,ITensor * out,const Window & window,ElementWiseUnary op)72 void elementwise_sve_op(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op)
73 {
74 const auto all_true_pg = wrapper::svptrue<ScalarType>();
75 const auto window_start_x = static_cast<int>(window.x().start());
76 const auto window_end_x = static_cast<int>(window.x().end());
77
78 Window win = window;
79 win.set(Window::DimX, Window::Dimension(0, 1, 1));
80
81 Iterator input(in, win);
82 Iterator output(out, win);
83
84 execute_window_loop(win, [&](const Coordinates &)
85 {
86 auto output_ptr = reinterpret_cast<ScalarType *>(output.ptr());
87 const auto input_ptr = reinterpret_cast<const ScalarType *>(input.ptr());
88 int x = window_start_x;
89
90 svbool_t pg = wrapper::svwhilelt<ScalarType>(x, window_end_x);
91 do
92 {
93 const auto vin = svld1(pg, input_ptr + x);
94 svst1(pg, output_ptr + x, elementwise_op_sve_imp<ScalarType, decltype(vin)>(pg, op, vin));
95 x += wrapper::svcnt<ScalarType>();
96 pg = wrapper::svwhilelt<ScalarType>(x, window_end_x);
97 }
98 while(svptest_any(all_true_pg, pg));
99 },
100 input, output);
101 }
102
103 template void elementwise_sve_op<float16_t>(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op);
104 template void elementwise_sve_op<float32_t>(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op);
105 template void elementwise_sve_op<int32_t>(const ITensor *in, ITensor *out, const Window &window, ElementWiseUnary op);
106
107 } // namespace cpu
108 } // namespace arm_compute
109