ref: 734c3d2b66b063e9ebcf0bc4ceb9ae3a9baacd31
dir: /vp8/common/arm/neon/dequant_idct_neon.c/
/*
* Copyright (c) 2014 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <arm_neon.h>
#include "./vp8_rtcd.h"
static const int16_t cospi8sqrt2minus1 = 20091;
// 35468 exceeds INT16_MAX and gets converted to a negative number. Because of
// the way it is used in vqdmulh, where the result is doubled, it can be divided
// by 2 beforehand. This saves compensating for the negative value as well as
// shifting the result.
static const int16_t sinpi8sqrt2 = 35468 >> 1;
void vp8_dequant_idct_add_neon(int16_t *input, int16_t *dq, unsigned char *dst,
int stride) {
unsigned char *dst0;
int32x2_t d14, d15;
int16x4_t d2, d3, d4, d5, d10, d11, d12, d13;
int16x8_t q1, q2, q3, q4, q5, q6;
int16x8_t qEmpty = vdupq_n_s16(0);
int32x2x2_t d2tmp0, d2tmp1;
int16x4x2_t d2tmp2, d2tmp3;
d14 = d15 = vdup_n_s32(0);
// load input
q3 = vld1q_s16(input);
vst1q_s16(input, qEmpty);
input += 8;
q4 = vld1q_s16(input);
vst1q_s16(input, qEmpty);
// load dq
q5 = vld1q_s16(dq);
dq += 8;
q6 = vld1q_s16(dq);
// load src from dst
dst0 = dst;
d14 = vld1_lane_s32((const int32_t *)dst0, d14, 0);
dst0 += stride;
d14 = vld1_lane_s32((const int32_t *)dst0, d14, 1);
dst0 += stride;
d15 = vld1_lane_s32((const int32_t *)dst0, d15, 0);
dst0 += stride;
d15 = vld1_lane_s32((const int32_t *)dst0, d15, 1);
q1 = vreinterpretq_s16_u16(
vmulq_u16(vreinterpretq_u16_s16(q3), vreinterpretq_u16_s16(q5)));
q2 = vreinterpretq_s16_u16(
vmulq_u16(vreinterpretq_u16_s16(q4), vreinterpretq_u16_s16(q6)));
d12 = vqadd_s16(vget_low_s16(q1), vget_low_s16(q2));
d13 = vqsub_s16(vget_low_s16(q1), vget_low_s16(q2));
q2 = vcombine_s16(vget_high_s16(q1), vget_high_s16(q2));
q3 = vqdmulhq_n_s16(q2, sinpi8sqrt2);
q4 = vqdmulhq_n_s16(q2, cospi8sqrt2minus1);
q4 = vshrq_n_s16(q4, 1);
q4 = vqaddq_s16(q4, q2);
d10 = vqsub_s16(vget_low_s16(q3), vget_high_s16(q4));
d11 = vqadd_s16(vget_high_s16(q3), vget_low_s16(q4));
d2 = vqadd_s16(d12, d11);
d3 = vqadd_s16(d13, d10);
d4 = vqsub_s16(d13, d10);
d5 = vqsub_s16(d12, d11);
d2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4));
d2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5));
d2tmp2 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[0]),
vreinterpret_s16_s32(d2tmp1.val[0]));
d2tmp3 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[1]),
vreinterpret_s16_s32(d2tmp1.val[1]));
// loop 2
q2 = vcombine_s16(d2tmp2.val[1], d2tmp3.val[1]);
q3 = vqdmulhq_n_s16(q2, sinpi8sqrt2);
q4 = vqdmulhq_n_s16(q2, cospi8sqrt2minus1);
d12 = vqadd_s16(d2tmp2.val[0], d2tmp3.val[0]);
d13 = vqsub_s16(d2tmp2.val[0], d2tmp3.val[0]);
q4 = vshrq_n_s16(q4, 1);
q4 = vqaddq_s16(q4, q2);
d10 = vqsub_s16(vget_low_s16(q3), vget_high_s16(q4));
d11 = vqadd_s16(vget_high_s16(q3), vget_low_s16(q4));
d2 = vqadd_s16(d12, d11);
d3 = vqadd_s16(d13, d10);
d4 = vqsub_s16(d13, d10);
d5 = vqsub_s16(d12, d11);
d2 = vrshr_n_s16(d2, 3);
d3 = vrshr_n_s16(d3, 3);
d4 = vrshr_n_s16(d4, 3);
d5 = vrshr_n_s16(d5, 3);
d2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4));
d2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5));
d2tmp2 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[0]),
vreinterpret_s16_s32(d2tmp1.val[0]));
d2tmp3 = vtrn_s16(vreinterpret_s16_s32(d2tmp0.val[1]),
vreinterpret_s16_s32(d2tmp1.val[1]));
q1 = vcombine_s16(d2tmp2.val[0], d2tmp2.val[1]);
q2 = vcombine_s16(d2tmp3.val[0], d2tmp3.val[1]);
q1 = vreinterpretq_s16_u16(
vaddw_u8(vreinterpretq_u16_s16(q1), vreinterpret_u8_s32(d14)));
q2 = vreinterpretq_s16_u16(
vaddw_u8(vreinterpretq_u16_s16(q2), vreinterpret_u8_s32(d15)));
d14 = vreinterpret_s32_u8(vqmovun_s16(q1));
d15 = vreinterpret_s32_u8(vqmovun_s16(q2));
dst0 = dst;
vst1_lane_s32((int32_t *)dst0, d14, 0);
dst0 += stride;
vst1_lane_s32((int32_t *)dst0, d14, 1);
dst0 += stride;
vst1_lane_s32((int32_t *)dst0, d15, 0);
dst0 += stride;
vst1_lane_s32((int32_t *)dst0, d15, 1);
return;
}