1 /*
2 * Copyright (C) 2010 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 *
16 */
17 #include "helloneon-intrinsics.h"
18 #include <arm_neon.h>
19
20 /* this source file should only be compiled by Android.mk when targeting
21 * the armeabi-v7a ABI, and should be built in NEON mode
22 */
23 void
fir_filter_neon_intrinsics(short * output,const short * input,const short * kernel,int width,int kernelSize)24 fir_filter_neon_intrinsics(short *output, const short* input, const short* kernel, int width, int kernelSize)
25 {
26 #if 1
27 int nn, offset = -kernelSize/2;
28
29 for (nn = 0; nn < width; nn++)
30 {
31 int mm, sum = 0;
32 int32x4_t sum_vec = vdupq_n_s32(0);
33 for(mm = 0; mm < kernelSize/4; mm++)
34 {
35 int16x4_t kernel_vec = vld1_s16(kernel + mm*4);
36 int16x4_t input_vec = vld1_s16(input + (nn+offset+mm*4));
37 sum_vec = vmlal_s16(sum_vec, kernel_vec, input_vec);
38 }
39
40 sum += vgetq_lane_s32(sum_vec, 0);
41 sum += vgetq_lane_s32(sum_vec, 1);
42 sum += vgetq_lane_s32(sum_vec, 2);
43 sum += vgetq_lane_s32(sum_vec, 3);
44
45 if(kernelSize & 3)
46 {
47 for(mm = kernelSize - (kernelSize & 3); mm < kernelSize; mm++)
48 sum += kernel[mm] * input[nn+offset+mm];
49 }
50
51 output[nn] = (short)((sum + 0x8000) >> 16);
52 }
53 #else /* for comparison purposes only */
54 int nn, offset = -kernelSize/2;
55 for (nn = 0; nn < width; nn++) {
56 int sum = 0;
57 int mm;
58 for (mm = 0; mm < kernelSize; mm++) {
59 sum += kernel[mm]*input[nn+offset+mm];
60 }
61 output[n] = (short)((sum + 0x8000) >> 16);
62 }
63 #endif
64 }
65