1 /*
2 * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <arm_neon.h>
12 #include <assert.h>
13
14 #include "./vpx_dsp_rtcd.h"
15 #include "vpx_dsp/arm/idct_neon.h"
16 #include "vpx_dsp/arm/mem_neon.h"
17 #include "vpx_dsp/txfm_common.h"
18
vpx_idct4x4_16_add_neon(const tran_low_t * input,uint8_t * dest,int stride)19 void vpx_idct4x4_16_add_neon(const tran_low_t *input, uint8_t *dest,
20 int stride) {
21 const uint8_t *dst = dest;
22 uint32x2_t s32 = vdup_n_u32(0);
23 int16x8_t a[2];
24 uint8x8_t s, d[2];
25 uint16x8_t sum[2];
26
27 assert(!((intptr_t)dest % sizeof(uint32_t)));
28 assert(!(stride % sizeof(uint32_t)));
29
30 // Rows
31 a[0] = load_tran_low_to_s16q(input);
32 a[1] = load_tran_low_to_s16q(input + 8);
33 transpose_idct4x4_16_bd8(a);
34
35 // Columns
36 a[1] = vcombine_s16(vget_high_s16(a[1]), vget_low_s16(a[1]));
37 transpose_idct4x4_16_bd8(a);
38 a[0] = vrshrq_n_s16(a[0], 4);
39 a[1] = vrshrq_n_s16(a[1], 4);
40
41 s = load_u8(dst, stride);
42 dst += 2 * stride;
43 // The elements are loaded in reverse order.
44 s32 = vld1_lane_u32((const uint32_t *)dst, s32, 1);
45 dst += stride;
46 s32 = vld1_lane_u32((const uint32_t *)dst, s32, 0);
47
48 sum[0] = vaddw_u8(vreinterpretq_u16_s16(a[0]), s);
49 sum[1] = vaddw_u8(vreinterpretq_u16_s16(a[1]), vreinterpret_u8_u32(s32));
50 d[0] = vqmovun_s16(vreinterpretq_s16_u16(sum[0]));
51 d[1] = vqmovun_s16(vreinterpretq_s16_u16(sum[1]));
52
53 store_u8(dest, stride, d[0]);
54 dest += 2 * stride;
55 // The elements are stored in reverse order.
56 vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d[1]), 1);
57 dest += stride;
58 vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d[1]), 0);
59 }
60