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 const int16x4_t cospis = vld1_s16(kCospi);
23 uint8x8_t dest01_u8;
24 uint32x2_t dest32_u32 = vdup_n_u32(0);
25 int16x8_t a0, a1;
26 uint8x8_t d01, d32;
27 uint16x8_t d01_u16, d32_u16;
28
29 assert(!((intptr_t)dest % sizeof(uint32_t)));
30 assert(!(stride % sizeof(uint32_t)));
31
32 // Rows
33 a0 = load_tran_low_to_s16q(input);
34 a1 = load_tran_low_to_s16q(input + 8);
35 idct4x4_16_kernel_bd8(cospis, &a0, &a1);
36
37 // Columns
38 a1 = vcombine_s16(vget_high_s16(a1), vget_low_s16(a1));
39 idct4x4_16_kernel_bd8(cospis, &a0, &a1);
40 a0 = vrshrq_n_s16(a0, 4);
41 a1 = vrshrq_n_s16(a1, 4);
42
43 dest01_u8 = load_u8(dst, stride);
44 dst += 2 * stride;
45 // The elements are loaded in reverse order.
46 dest32_u32 = vld1_lane_u32((const uint32_t *)dst, dest32_u32, 1);
47 dst += stride;
48 dest32_u32 = vld1_lane_u32((const uint32_t *)dst, dest32_u32, 0);
49
50 d01_u16 = vaddw_u8(vreinterpretq_u16_s16(a0), dest01_u8);
51 d32_u16 =
52 vaddw_u8(vreinterpretq_u16_s16(a1), vreinterpret_u8_u32(dest32_u32));
53 d01 = vqmovun_s16(vreinterpretq_s16_u16(d01_u16));
54 d32 = vqmovun_s16(vreinterpretq_s16_u16(d32_u16));
55
56 store_u8(dest, stride, d01);
57 dest += 2 * stride;
58 // The elements are stored in reverse order.
59 vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d32), 1);
60 dest += stride;
61 vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d32), 0);
62 }
63