idct4x4_msa.c revision 7ce0a1d1337c01056ba24006efab21f00e179e04 
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 "vpx_dsp/mips/inv_txfm_msa.h"
12
13void vpx_iwht4x4_16_add_msa(const int16_t *input, uint8_t *dst,
14                            int32_t dst_stride) {
15  v8i16 in0, in1, in2, in3;
16  v4i32 in0_r, in1_r, in2_r, in3_r, in4_r;
17
18  /* load vector elements of 4x4 block */
19  LD4x4_SH(input, in0, in2, in3, in1);
20  TRANSPOSE4x4_SH_SH(in0, in2, in3, in1, in0, in2, in3, in1);
21  UNPCK_R_SH_SW(in0, in0_r);
22  UNPCK_R_SH_SW(in2, in2_r);
23  UNPCK_R_SH_SW(in3, in3_r);
24  UNPCK_R_SH_SW(in1, in1_r);
25  SRA_4V(in0_r, in1_r, in2_r, in3_r, UNIT_QUANT_SHIFT);
26
27  in0_r += in2_r;
28  in3_r -= in1_r;
29  in4_r = (in0_r - in3_r) >> 1;
30  in1_r = in4_r - in1_r;
31  in2_r = in4_r - in2_r;
32  in0_r -= in1_r;
33  in3_r += in2_r;
34
35  TRANSPOSE4x4_SW_SW(in0_r, in1_r, in2_r, in3_r, in0_r, in1_r, in2_r, in3_r);
36
37  in0_r += in1_r;
38  in2_r -= in3_r;
39  in4_r = (in0_r - in2_r) >> 1;
40  in3_r = in4_r - in3_r;
41  in1_r = in4_r - in1_r;
42  in0_r -= in3_r;
43  in2_r += in1_r;
44
45  PCKEV_H4_SH(in0_r, in0_r, in1_r, in1_r, in2_r, in2_r, in3_r, in3_r,
46              in0, in1, in2, in3);
47  ADDBLK_ST4x4_UB(in0, in3, in1, in2, dst, dst_stride);
48}
49
50void vpx_iwht4x4_1_add_msa(const int16_t *input, uint8_t *dst,
51                           int32_t dst_stride) {
52  int16_t a1, e1;
53  v8i16 in1, in0 = { 0 };
54
55  a1 = input[0] >> UNIT_QUANT_SHIFT;
56  e1 = a1 >> 1;
57  a1 -= e1;
58
59  in0 = __msa_insert_h(in0, 0, a1);
60  in0 = __msa_insert_h(in0, 1, e1);
61  in0 = __msa_insert_h(in0, 2, e1);
62  in0 = __msa_insert_h(in0, 3, e1);
63
64  in1 = in0 >> 1;
65  in0 -= in1;
66
67  ADDBLK_ST4x4_UB(in0, in1, in1, in1, dst, dst_stride);
68}
69
70void vpx_idct4x4_16_add_msa(const int16_t *input, uint8_t *dst,
71                            int32_t dst_stride) {
72  v8i16 in0, in1, in2, in3;
73
74  /* load vector elements of 4x4 block */
75  LD4x4_SH(input, in0, in1, in2, in3);
76  /* rows */
77  TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
78  VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
79  /* columns */
80  TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
81  VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3);
82  /* rounding (add 2^3, divide by 2^4) */
83  SRARI_H4_SH(in0, in1, in2, in3, 4);
84  ADDBLK_ST4x4_UB(in0, in1, in2, in3, dst, dst_stride);
85}
86
87void vpx_idct4x4_1_add_msa(const int16_t *input, uint8_t *dst,
88                           int32_t dst_stride) {
89  int16_t out;
90  v8i16 vec;
91
92  out = ROUND_POWER_OF_TWO((input[0] * cospi_16_64), DCT_CONST_BITS);
93  out = ROUND_POWER_OF_TWO((out * cospi_16_64), DCT_CONST_BITS);
94  out = ROUND_POWER_OF_TWO(out, 4);
95  vec = __msa_fill_h(out);
96
97  ADDBLK_ST4x4_UB(vec, vec, vec, vec, dst, dst_stride);
98}
99