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_INV_TXFM_SSSE3_H_
12 #define VPX_VPX_DSP_X86_INV_TXFM_SSSE3_H_
13
14 #include <tmmintrin.h>
15
16 #include "./vpx_dsp_rtcd.h"
17 #include "vpx_dsp/x86/inv_txfm_sse2.h"
18 #include "vpx_dsp/x86/transpose_sse2.h"
19 #include "vpx_dsp/x86/txfm_common_sse2.h"
20
idct8x8_12_add_kernel_ssse3(__m128i * const io)21 static INLINE void idct8x8_12_add_kernel_ssse3(__m128i *const io /* io[8] */) {
22 const __m128i cp_28d_4d = dual_set_epi16(2 * cospi_28_64, 2 * cospi_4_64);
23 const __m128i cp_n20d_12d = dual_set_epi16(-2 * cospi_20_64, 2 * cospi_12_64);
24 const __m128i cp_8d_24d = dual_set_epi16(2 * cospi_8_64, 2 * cospi_24_64);
25 const __m128i cp_16_16 = _mm_set1_epi16(cospi_16_64);
26 const __m128i cp_16_n16 = pair_set_epi16(cospi_16_64, -cospi_16_64);
27 const __m128i cospi_16_64d = _mm_set1_epi16((int16_t)(2 * cospi_16_64));
28 const __m128i cospi_28_64d = _mm_set1_epi16((int16_t)(2 * cospi_28_64));
29 const __m128i cospi_4_64d = _mm_set1_epi16((int16_t)(2 * cospi_4_64));
30 const __m128i cospi_n20_64d = _mm_set1_epi16((int16_t)(-2 * cospi_20_64));
31 const __m128i cospi_12_64d = _mm_set1_epi16((int16_t)(2 * cospi_12_64));
32 const __m128i cospi_24_64d = _mm_set1_epi16((int16_t)(2 * cospi_24_64));
33 const __m128i cospi_8_64d = _mm_set1_epi16((int16_t)(2 * cospi_8_64));
34 __m128i step1[8], step2[8], tmp[4];
35
36 // pass 1
37
38 transpose_16bit_4x4(io, io);
39 // io[0]: 00 10 20 30 01 11 21 31
40 // io[1]: 02 12 22 32 03 13 23 33
41
42 // stage 1
43 tmp[0] = _mm_unpacklo_epi64(io[0], io[0]);
44 tmp[1] = _mm_unpackhi_epi64(io[0], io[0]);
45 tmp[2] = _mm_unpacklo_epi64(io[1], io[1]);
46 tmp[3] = _mm_unpackhi_epi64(io[1], io[1]);
47 step1[4] = _mm_mulhrs_epi16(tmp[1], cp_28d_4d); // step1 4&7
48 step1[5] = _mm_mulhrs_epi16(tmp[3], cp_n20d_12d); // step1 5&6
49
50 // stage 2
51 step2[0] = _mm_mulhrs_epi16(tmp[0], cospi_16_64d); // step2 0&1
52 step2[2] = _mm_mulhrs_epi16(tmp[2], cp_8d_24d); // step2 3&2
53 step2[4] = _mm_add_epi16(step1[4], step1[5]); // step2 4&7
54 step2[5] = _mm_sub_epi16(step1[4], step1[5]); // step2 5&6
55 step2[6] = _mm_unpackhi_epi64(step2[5], step2[5]); // step2 6
56
57 // stage 3
58 tmp[0] = _mm_unpacklo_epi16(step2[6], step2[5]);
59 step1[5] = idct_calc_wraplow_sse2(cp_16_n16, cp_16_16, tmp[0]); // step1 5&6
60 tmp[0] = _mm_add_epi16(step2[0], step2[2]); // step1 0&1
61 tmp[1] = _mm_sub_epi16(step2[0], step2[2]); // step1 3&2
62 step1[2] = _mm_unpackhi_epi64(tmp[1], tmp[0]); // step1 2&1
63 step1[3] = _mm_unpacklo_epi64(tmp[1], tmp[0]); // step1 3&0
64
65 // stage 4
66 tmp[0] = _mm_add_epi16(step1[3], step2[4]); // output 3&0
67 tmp[1] = _mm_add_epi16(step1[2], step1[5]); // output 2&1
68 tmp[2] = _mm_sub_epi16(step1[3], step2[4]); // output 4&7
69 tmp[3] = _mm_sub_epi16(step1[2], step1[5]); // output 5&6
70
71 // pass 2
72
73 idct8x8_12_transpose_16bit_4x8(tmp, io);
74
75 // stage 1
76 step1[4] = _mm_mulhrs_epi16(io[1], cospi_28_64d);
77 step1[7] = _mm_mulhrs_epi16(io[1], cospi_4_64d);
78 step1[5] = _mm_mulhrs_epi16(io[3], cospi_n20_64d);
79 step1[6] = _mm_mulhrs_epi16(io[3], cospi_12_64d);
80
81 // stage 2
82 step2[0] = _mm_mulhrs_epi16(io[0], cospi_16_64d); // step2[1] = step2[0]
83 step2[2] = _mm_mulhrs_epi16(io[2], cospi_24_64d);
84 step2[3] = _mm_mulhrs_epi16(io[2], cospi_8_64d);
85 step2[4] = _mm_add_epi16(step1[4], step1[5]);
86 step2[5] = _mm_sub_epi16(step1[4], step1[5]);
87 step2[6] = _mm_sub_epi16(step1[7], step1[6]);
88 step2[7] = _mm_add_epi16(step1[7], step1[6]);
89
90 // stage 3
91 step1[0] = _mm_add_epi16(step2[0], step2[3]);
92 step1[1] = _mm_add_epi16(step2[0], step2[2]);
93 step1[2] = _mm_sub_epi16(step2[0], step2[2]);
94 step1[3] = _mm_sub_epi16(step2[0], step2[3]);
95 butterfly(step2[6], step2[5], cospi_16_64, cospi_16_64, &step1[5], &step1[6]);
96
97 // stage 4
98 io[0] = _mm_add_epi16(step1[0], step2[7]);
99 io[1] = _mm_add_epi16(step1[1], step1[6]);
100 io[2] = _mm_add_epi16(step1[2], step1[5]);
101 io[3] = _mm_add_epi16(step1[3], step2[4]);
102 io[4] = _mm_sub_epi16(step1[3], step2[4]);
103 io[5] = _mm_sub_epi16(step1[2], step1[5]);
104 io[6] = _mm_sub_epi16(step1[1], step1[6]);
105 io[7] = _mm_sub_epi16(step1[0], step2[7]);
106 }
107
108 void idct32_135_8x32_ssse3(const __m128i *const in, __m128i *const out);
109
110 #endif // VPX_VPX_DSP_X86_INV_TXFM_SSSE3_H_
111