Coverage Report

Created: 2024-09-06 07:53

/src/libvpx/vpx_dsp/x86/quantize_sse2.c
Line
Count
Source (jump to first uncovered line)
1
/*
2
 *  Copyright (c) 2015 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 <assert.h>
12
#include <emmintrin.h>
13
#include <xmmintrin.h>
14
15
#include "./vpx_dsp_rtcd.h"
16
#include "vpx/vpx_integer.h"
17
#include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
18
#include "vpx_dsp/x86/quantize_sse2.h"
19
#include "vp9/common/vp9_scan.h"
20
21
void vpx_quantize_b_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
22
                         const struct macroblock_plane *const mb_plane,
23
                         tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
24
                         const int16_t *dequant_ptr, uint16_t *eob_ptr,
25
0
                         const struct ScanOrder *const scan_order) {
26
0
  const __m128i zero = _mm_setzero_si128();
27
0
  int index = 16;
28
0
  const int16_t *iscan = scan_order->iscan;
29
30
0
  __m128i zbin, round, quant, dequant, shift;
31
0
  __m128i coeff0, coeff1, coeff0_sign, coeff1_sign;
32
0
  __m128i qcoeff0, qcoeff1;
33
0
  __m128i cmp_mask0, cmp_mask1;
34
0
  __m128i eob, eob0;
35
36
  // Setup global values.
37
0
  load_b_values(mb_plane, &zbin, &round, &quant, dequant_ptr, &dequant, &shift);
38
39
  // Do DC and first 15 AC.
40
0
  coeff0 = load_tran_low(coeff_ptr);
41
0
  coeff1 = load_tran_low(coeff_ptr + 8);
42
43
  // Poor man's abs().
44
0
  coeff0_sign = _mm_srai_epi16(coeff0, 15);
45
0
  coeff1_sign = _mm_srai_epi16(coeff1, 15);
46
0
  qcoeff0 = invert_sign_sse2(coeff0, coeff0_sign);
47
0
  qcoeff1 = invert_sign_sse2(coeff1, coeff1_sign);
48
49
0
  cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
50
0
  zbin = _mm_unpackhi_epi64(zbin, zbin);  // Switch DC to AC
51
0
  cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
52
53
0
  calculate_qcoeff(&qcoeff0, round, quant, shift);
54
55
0
  round = _mm_unpackhi_epi64(round, round);
56
0
  quant = _mm_unpackhi_epi64(quant, quant);
57
0
  shift = _mm_unpackhi_epi64(shift, shift);
58
59
0
  calculate_qcoeff(&qcoeff1, round, quant, shift);
60
61
  // Reinsert signs
62
0
  qcoeff0 = invert_sign_sse2(qcoeff0, coeff0_sign);
63
0
  qcoeff1 = invert_sign_sse2(qcoeff1, coeff1_sign);
64
65
  // Mask out zbin threshold coeffs
66
0
  qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
67
0
  qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
68
69
0
  store_tran_low(qcoeff0, qcoeff_ptr);
70
0
  store_tran_low(qcoeff1, qcoeff_ptr + 8);
71
72
0
  calculate_dqcoeff_and_store(qcoeff0, dequant, dqcoeff_ptr);
73
0
  dequant = _mm_unpackhi_epi64(dequant, dequant);
74
0
  calculate_dqcoeff_and_store(qcoeff1, dequant, dqcoeff_ptr + 8);
75
76
0
  eob = scan_for_eob(&qcoeff0, &qcoeff1, iscan, 0, zero);
77
78
  // AC only loop.
79
0
  while (index < n_coeffs) {
80
0
    coeff0 = load_tran_low(coeff_ptr + index);
81
0
    coeff1 = load_tran_low(coeff_ptr + index + 8);
82
83
0
    coeff0_sign = _mm_srai_epi16(coeff0, 15);
84
0
    coeff1_sign = _mm_srai_epi16(coeff1, 15);
85
0
    qcoeff0 = invert_sign_sse2(coeff0, coeff0_sign);
86
0
    qcoeff1 = invert_sign_sse2(coeff1, coeff1_sign);
87
88
0
    cmp_mask0 = _mm_cmpgt_epi16(qcoeff0, zbin);
89
0
    cmp_mask1 = _mm_cmpgt_epi16(qcoeff1, zbin);
90
91
0
    calculate_qcoeff(&qcoeff0, round, quant, shift);
92
0
    calculate_qcoeff(&qcoeff1, round, quant, shift);
93
94
0
    qcoeff0 = invert_sign_sse2(qcoeff0, coeff0_sign);
95
0
    qcoeff1 = invert_sign_sse2(qcoeff1, coeff1_sign);
96
97
0
    qcoeff0 = _mm_and_si128(qcoeff0, cmp_mask0);
98
0
    qcoeff1 = _mm_and_si128(qcoeff1, cmp_mask1);
99
100
0
    store_tran_low(qcoeff0, qcoeff_ptr + index);
101
0
    store_tran_low(qcoeff1, qcoeff_ptr + index + 8);
102
103
0
    calculate_dqcoeff_and_store(qcoeff0, dequant, dqcoeff_ptr + index);
104
0
    calculate_dqcoeff_and_store(qcoeff1, dequant, dqcoeff_ptr + index + 8);
105
106
0
    eob0 = scan_for_eob(&qcoeff0, &qcoeff1, iscan, index, zero);
107
0
    eob = _mm_max_epi16(eob, eob0);
108
109
0
    index += 16;
110
0
  }
111
112
0
  *eob_ptr = accumulate_eob(eob);
113
0
}