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_rtcd.h"
12 #include "vpx_dsp/mips/inv_txfm_msa.h"
13
vpx_idct8x8_64_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)14 void vpx_idct8x8_64_add_msa(const int16_t *input, uint8_t *dst,
15 int32_t dst_stride) {
16 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
17
18 /* load vector elements of 8x8 block */
19 LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
20
21 /* rows transform */
22 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
23 in4, in5, in6, in7);
24 /* 1D idct8x8 */
25 VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
26 in4, in5, in6, in7);
27 /* columns transform */
28 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
29 in4, in5, in6, in7);
30 /* 1D idct8x8 */
31 VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
32 in4, in5, in6, in7);
33 /* final rounding (add 2^4, divide by 2^5) and shift */
34 SRARI_H4_SH(in0, in1, in2, in3, 5);
35 SRARI_H4_SH(in4, in5, in6, in7, 5);
36 /* add block and store 8x8 */
37 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3);
38 dst += (4 * dst_stride);
39 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7);
40 }
41
vpx_idct8x8_12_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)42 void vpx_idct8x8_12_add_msa(const int16_t *input, uint8_t *dst,
43 int32_t dst_stride) {
44 v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
45 v8i16 s0, s1, s2, s3, s4, s5, s6, s7, k0, k1, k2, k3, m0, m1, m2, m3;
46 v4i32 tmp0, tmp1, tmp2, tmp3;
47 v8i16 zero = { 0 };
48
49 /* load vector elements of 8x8 block */
50 LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
51 TRANSPOSE8X4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
52
53 /* stage1 */
54 ILVL_H2_SH(in3, in0, in2, in1, s0, s1);
55 k0 = VP9_SET_COSPI_PAIR(cospi_28_64, -cospi_4_64);
56 k1 = VP9_SET_COSPI_PAIR(cospi_4_64, cospi_28_64);
57 k2 = VP9_SET_COSPI_PAIR(-cospi_20_64, cospi_12_64);
58 k3 = VP9_SET_COSPI_PAIR(cospi_12_64, cospi_20_64);
59 DOTP_SH4_SW(s0, s0, s1, s1, k0, k1, k2, k3, tmp0, tmp1, tmp2, tmp3);
60 SRARI_W4_SW(tmp0, tmp1, tmp2, tmp3, DCT_CONST_BITS);
61 PCKEV_H2_SH(zero, tmp0, zero, tmp1, s0, s1);
62 PCKEV_H2_SH(zero, tmp2, zero, tmp3, s2, s3);
63 BUTTERFLY_4(s0, s1, s3, s2, s4, s7, s6, s5);
64
65 /* stage2 */
66 ILVR_H2_SH(in3, in1, in2, in0, s1, s0);
67 k0 = VP9_SET_COSPI_PAIR(cospi_16_64, cospi_16_64);
68 k1 = VP9_SET_COSPI_PAIR(cospi_16_64, -cospi_16_64);
69 k2 = VP9_SET_COSPI_PAIR(cospi_24_64, -cospi_8_64);
70 k3 = VP9_SET_COSPI_PAIR(cospi_8_64, cospi_24_64);
71 DOTP_SH4_SW(s0, s0, s1, s1, k0, k1, k2, k3, tmp0, tmp1, tmp2, tmp3);
72 SRARI_W4_SW(tmp0, tmp1, tmp2, tmp3, DCT_CONST_BITS);
73 PCKEV_H2_SH(zero, tmp0, zero, tmp1, s0, s1);
74 PCKEV_H2_SH(zero, tmp2, zero, tmp3, s2, s3);
75 BUTTERFLY_4(s0, s1, s2, s3, m0, m1, m2, m3);
76
77 /* stage3 */
78 s0 = __msa_ilvr_h(s6, s5);
79
80 k1 = VP9_SET_COSPI_PAIR(-cospi_16_64, cospi_16_64);
81 DOTP_SH2_SW(s0, s0, k1, k0, tmp0, tmp1);
82 SRARI_W2_SW(tmp0, tmp1, DCT_CONST_BITS);
83 PCKEV_H2_SH(zero, tmp0, zero, tmp1, s2, s3);
84
85 /* stage4 */
86 BUTTERFLY_8(m0, m1, m2, m3, s4, s2, s3, s7, in0, in1, in2, in3, in4, in5, in6,
87 in7);
88 TRANSPOSE4X8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
89 in4, in5, in6, in7);
90 VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
91 in4, in5, in6, in7);
92
93 /* final rounding (add 2^4, divide by 2^5) and shift */
94 SRARI_H4_SH(in0, in1, in2, in3, 5);
95 SRARI_H4_SH(in4, in5, in6, in7, 5);
96
97 /* add block and store 8x8 */
98 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3);
99 dst += (4 * dst_stride);
100 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7);
101 }
102
vpx_idct8x8_1_add_msa(const int16_t * input,uint8_t * dst,int32_t dst_stride)103 void vpx_idct8x8_1_add_msa(const int16_t *input, uint8_t *dst,
104 int32_t dst_stride) {
105 int16_t out;
106 int32_t val;
107 v8i16 vec;
108
109 out = ROUND_POWER_OF_TWO((input[0] * cospi_16_64), DCT_CONST_BITS);
110 out = ROUND_POWER_OF_TWO((out * cospi_16_64), DCT_CONST_BITS);
111 val = ROUND_POWER_OF_TWO(out, 5);
112 vec = __msa_fill_h(val);
113
114 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, vec, vec, vec, vec);
115 dst += (4 * dst_stride);
116 VP9_ADDBLK_ST8x4_UB(dst, dst_stride, vec, vec, vec, vec);
117 }
118