1 /*
2 * Copyright (c) 2017 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 #ifndef VPX_VPX_DSP_X86_HIGHBD_INV_TXFM_SSE4_H_
12 #define VPX_VPX_DSP_X86_HIGHBD_INV_TXFM_SSE4_H_
13
14 #include <smmintrin.h> // SSE4.1
15
16 #include "./vpx_config.h"
17 #include "vpx_dsp/x86/highbd_inv_txfm_sse2.h"
18
multiplication_round_shift_sse4_1(const __m128i * const in,const int c)19 static INLINE __m128i multiplication_round_shift_sse4_1(
20 const __m128i *const in /*in[2]*/, const int c) {
21 const __m128i pair_c = pair_set_epi32(c * 4, 0);
22 __m128i t0, t1;
23
24 t0 = _mm_mul_epi32(in[0], pair_c);
25 t1 = _mm_mul_epi32(in[1], pair_c);
26 t0 = dct_const_round_shift_64bit(t0);
27 t1 = dct_const_round_shift_64bit(t1);
28
29 return pack_4(t0, t1);
30 }
31
highbd_butterfly_sse4_1(const __m128i in0,const __m128i in1,const int c0,const int c1,__m128i * const out0,__m128i * const out1)32 static INLINE void highbd_butterfly_sse4_1(const __m128i in0, const __m128i in1,
33 const int c0, const int c1,
34 __m128i *const out0,
35 __m128i *const out1) {
36 const __m128i pair_c0 = pair_set_epi32(4 * c0, 0);
37 const __m128i pair_c1 = pair_set_epi32(4 * c1, 0);
38 __m128i temp1[4], temp2[4];
39
40 extend_64bit(in0, temp1);
41 extend_64bit(in1, temp2);
42 temp1[2] = _mm_mul_epi32(temp1[0], pair_c1);
43 temp1[3] = _mm_mul_epi32(temp1[1], pair_c1);
44 temp1[0] = _mm_mul_epi32(temp1[0], pair_c0);
45 temp1[1] = _mm_mul_epi32(temp1[1], pair_c0);
46 temp2[2] = _mm_mul_epi32(temp2[0], pair_c0);
47 temp2[3] = _mm_mul_epi32(temp2[1], pair_c0);
48 temp2[0] = _mm_mul_epi32(temp2[0], pair_c1);
49 temp2[1] = _mm_mul_epi32(temp2[1], pair_c1);
50 temp1[0] = _mm_sub_epi64(temp1[0], temp2[0]);
51 temp1[1] = _mm_sub_epi64(temp1[1], temp2[1]);
52 temp2[0] = _mm_add_epi64(temp1[2], temp2[2]);
53 temp2[1] = _mm_add_epi64(temp1[3], temp2[3]);
54 temp1[0] = dct_const_round_shift_64bit(temp1[0]);
55 temp1[1] = dct_const_round_shift_64bit(temp1[1]);
56 temp2[0] = dct_const_round_shift_64bit(temp2[0]);
57 temp2[1] = dct_const_round_shift_64bit(temp2[1]);
58 *out0 = pack_4(temp1[0], temp1[1]);
59 *out1 = pack_4(temp2[0], temp2[1]);
60 }
61
highbd_butterfly_cospi16_sse4_1(const __m128i in0,const __m128i in1,__m128i * const out0,__m128i * const out1)62 static INLINE void highbd_butterfly_cospi16_sse4_1(const __m128i in0,
63 const __m128i in1,
64 __m128i *const out0,
65 __m128i *const out1) {
66 __m128i temp1[2], temp2;
67
68 temp2 = _mm_add_epi32(in0, in1);
69 extend_64bit(temp2, temp1);
70 *out0 = multiplication_round_shift_sse4_1(temp1, cospi_16_64);
71 temp2 = _mm_sub_epi32(in0, in1);
72 extend_64bit(temp2, temp1);
73 *out1 = multiplication_round_shift_sse4_1(temp1, cospi_16_64);
74 }
75
highbd_partial_butterfly_sse4_1(const __m128i in,const int c0,const int c1,__m128i * const out0,__m128i * const out1)76 static INLINE void highbd_partial_butterfly_sse4_1(const __m128i in,
77 const int c0, const int c1,
78 __m128i *const out0,
79 __m128i *const out1) {
80 __m128i temp[2];
81
82 extend_64bit(in, temp);
83 *out0 = multiplication_round_shift_sse4_1(temp, c0);
84 *out1 = multiplication_round_shift_sse4_1(temp, c1);
85 }
86
highbd_idct4_sse4_1(__m128i * const io)87 static INLINE void highbd_idct4_sse4_1(__m128i *const io) {
88 __m128i temp[2], step[4];
89
90 transpose_32bit_4x4(io, io);
91
92 // stage 1
93 temp[0] = _mm_add_epi32(io[0], io[2]); // input[0] + input[2]
94 extend_64bit(temp[0], temp);
95 step[0] = multiplication_round_shift_sse4_1(temp, cospi_16_64);
96 temp[0] = _mm_sub_epi32(io[0], io[2]); // input[0] - input[2]
97 extend_64bit(temp[0], temp);
98 step[1] = multiplication_round_shift_sse4_1(temp, cospi_16_64);
99 highbd_butterfly_sse4_1(io[1], io[3], cospi_24_64, cospi_8_64, &step[2],
100 &step[3]);
101
102 // stage 2
103 io[0] = _mm_add_epi32(step[0], step[3]); // step[0] + step[3]
104 io[1] = _mm_add_epi32(step[1], step[2]); // step[1] + step[2]
105 io[2] = _mm_sub_epi32(step[1], step[2]); // step[1] - step[2]
106 io[3] = _mm_sub_epi32(step[0], step[3]); // step[0] - step[3]
107 }
108
109 void vpx_highbd_idct8x8_half1d_sse4_1(__m128i *const io);
110 void vpx_highbd_idct16_4col_sse4_1(__m128i *const io /*io[16]*/);
111
112 #endif // VPX_VPX_DSP_X86_HIGHBD_INV_TXFM_SSE4_H_
113