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 <assert.h>
12 #include <emmintrin.h>
13 #include <xmmintrin.h>
14
15 #include "./vp9_rtcd.h"
16 #include "vpx/vpx_integer.h"
17 #include "vpx_dsp/vpx_dsp_common.h"
18 #include "vpx_dsp/x86/bitdepth_conversion_sse2.h"
19 #include "vpx_dsp/x86/quantize_sse2.h"
20 #include "vp9/common/vp9_scan.h"
21 #include "vp9/encoder/vp9_block.h"
22
vp9_quantize_fp_sse2(const tran_low_t * coeff_ptr,intptr_t n_coeffs,const struct macroblock_plane * const mb_plane,tran_low_t * qcoeff_ptr,tran_low_t * dqcoeff_ptr,const int16_t * dequant_ptr,uint16_t * eob_ptr,const struct ScanOrder * const scan_order)23 void vp9_quantize_fp_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
24 const struct macroblock_plane *const mb_plane,
25 tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
26 const int16_t *dequant_ptr, uint16_t *eob_ptr,
27 const struct ScanOrder *const scan_order) {
28 const __m128i zero = _mm_setzero_si128();
29 __m128i thr;
30 int nzflag;
31 int index = 16;
32 __m128i round, quant, dequant;
33 __m128i coeff0, coeff1, coeff0_sign, coeff1_sign;
34 __m128i qcoeff0, qcoeff1;
35 __m128i eob;
36 const int16_t *iscan = scan_order->iscan;
37
38 // Setup global values.
39 load_fp_values(mb_plane, &round, &quant, dequant_ptr, &dequant);
40
41 // Do DC and first 15 AC.
42 coeff0 = load_tran_low(coeff_ptr);
43 coeff1 = load_tran_low(coeff_ptr + 8);
44
45 // Poor man's abs().
46 coeff0_sign = _mm_srai_epi16(coeff0, 15);
47 coeff1_sign = _mm_srai_epi16(coeff1, 15);
48 qcoeff0 = invert_sign_sse2(coeff0, coeff0_sign);
49 qcoeff1 = invert_sign_sse2(coeff1, coeff1_sign);
50
51 qcoeff0 = _mm_adds_epi16(qcoeff0, round);
52 qcoeff0 = _mm_mulhi_epi16(qcoeff0, quant);
53
54 round = _mm_unpackhi_epi64(round, round);
55 quant = _mm_unpackhi_epi64(quant, quant);
56
57 qcoeff1 = _mm_adds_epi16(qcoeff1, round);
58 qcoeff1 = _mm_mulhi_epi16(qcoeff1, quant);
59
60 // Reinsert signs.
61 qcoeff0 = invert_sign_sse2(qcoeff0, coeff0_sign);
62 qcoeff1 = invert_sign_sse2(qcoeff1, coeff1_sign);
63
64 store_tran_low(qcoeff0, qcoeff_ptr);
65 store_tran_low(qcoeff1, qcoeff_ptr + 8);
66
67 qcoeff0 = _mm_mullo_epi16(qcoeff0, dequant);
68 dequant = _mm_unpackhi_epi64(dequant, dequant);
69 qcoeff1 = _mm_mullo_epi16(qcoeff1, dequant);
70
71 store_tran_low(qcoeff0, dqcoeff_ptr);
72 store_tran_low(qcoeff1, dqcoeff_ptr + 8);
73
74 eob = scan_for_eob(&qcoeff0, &qcoeff1, iscan, 0, zero);
75
76 thr = _mm_srai_epi16(dequant, 1);
77
78 // AC only loop.
79 while (index < n_coeffs) {
80 coeff0 = load_tran_low(coeff_ptr + index);
81 coeff1 = load_tran_low(coeff_ptr + index + 8);
82
83 // Poor man's abs().
84 coeff0_sign = _mm_srai_epi16(coeff0, 15);
85 coeff1_sign = _mm_srai_epi16(coeff1, 15);
86 qcoeff0 = invert_sign_sse2(coeff0, coeff0_sign);
87 qcoeff1 = invert_sign_sse2(coeff1, coeff1_sign);
88
89 nzflag = _mm_movemask_epi8(_mm_cmpgt_epi16(qcoeff0, thr)) |
90 _mm_movemask_epi8(_mm_cmpgt_epi16(qcoeff1, thr));
91
92 if (nzflag) {
93 __m128i eob0;
94 qcoeff0 = _mm_adds_epi16(qcoeff0, round);
95 qcoeff1 = _mm_adds_epi16(qcoeff1, round);
96 qcoeff0 = _mm_mulhi_epi16(qcoeff0, quant);
97 qcoeff1 = _mm_mulhi_epi16(qcoeff1, quant);
98
99 // Reinsert signs.
100 qcoeff0 = invert_sign_sse2(qcoeff0, coeff0_sign);
101 qcoeff1 = invert_sign_sse2(qcoeff1, coeff1_sign);
102
103 store_tran_low(qcoeff0, qcoeff_ptr + index);
104 store_tran_low(qcoeff1, qcoeff_ptr + index + 8);
105
106 qcoeff0 = _mm_mullo_epi16(qcoeff0, dequant);
107 qcoeff1 = _mm_mullo_epi16(qcoeff1, dequant);
108
109 store_tran_low(qcoeff0, dqcoeff_ptr + index);
110 store_tran_low(qcoeff1, dqcoeff_ptr + index + 8);
111
112 eob0 = scan_for_eob(&qcoeff0, &qcoeff1, iscan, index, zero);
113 eob = _mm_max_epi16(eob, eob0);
114 } else {
115 store_zero_tran_low(qcoeff_ptr + index);
116 store_zero_tran_low(qcoeff_ptr + index + 8);
117
118 store_zero_tran_low(dqcoeff_ptr + index);
119 store_zero_tran_low(dqcoeff_ptr + index + 8);
120 }
121
122 index += 16;
123 }
124
125 *eob_ptr = accumulate_eob(eob);
126 }
127