• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2020 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 #if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS)
25 
26 #include "src/core/NEON/NEMath.h"
27 #include "src/core/common/StdTypes.h"
28 #include "src/core/common/Validate.h"
29 
30 #include <arm_neon.h>
31 #include <cmath>
32 #include <cstddef>
33 
34 namespace arm_compute
35 {
36 namespace cpu
37 {
38 constexpr int step = 8;
39 
fp16_neon_floor(const void * src,void * dst,int len)40 void fp16_neon_floor(const void *src, void *dst, int len)
41 {
42     ARM_COMPUTE_ASSERT_NOT_NULLPTR(src);
43     ARM_COMPUTE_ASSERT_NOT_NULLPTR(dst);
44     ARM_COMPUTE_ASSERT(len >= 0);
45 
46     auto psrc = static_cast<const f16 *>(src);
47     auto pdst = static_cast<f16 *>(dst);
48 
49     for(; len >= step; len -= step)
50     {
51         vst1q_f16(pdst, vfloorq_f16(vld1q_f16(psrc)));
52         psrc += step;
53         pdst += step;
54     }
55 
56     for(; len > 0; --len)
57     {
58         *pdst = std::floor(*psrc);
59         ++psrc;
60         ++pdst;
61     }
62 }
63 } // namespace cpu
64 } // namespace arm_compute
65 #endif /* defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && defined(ENABLE_FP16_KERNELS) */
66