1 | // Copyright 2016 Google Inc. All Rights Reserved. |
2 | // |
3 | // Use of this source code is governed by a BSD-style license |
4 | // that can be found in the COPYING file in the root of the source |
5 | // tree. An additional intellectual property rights grant can be found |
6 | // in the file PATENTS. All contributing project authors may |
7 | // be found in the AUTHORS file in the root of the source tree. |
8 | // ----------------------------------------------------------------------------- |
9 | // |
10 | // MSA variant of Image transform methods for lossless encoder. |
11 | // |
12 | // Authors: Prashant Patil (Prashant.Patil@imgtec.com) |
13 | |
14 | #include "src/dsp/dsp.h" |
15 | |
16 | #if defined(WEBP_USE_MSA) |
17 | |
18 | #include "src/dsp/lossless.h" |
19 | #include "src/dsp/msa_macro.h" |
20 | |
21 | #define TRANSFORM_COLOR_8(src0, src1, dst0, dst1, c0, c1, mask0, mask1) do { \ |
22 | v8i16 g0, g1, t0, t1, t2, t3; \ |
23 | v4i32 t4, t5; \ |
24 | VSHF_B2_SH(src0, src0, src1, src1, mask0, mask0, g0, g1); \ |
25 | DOTP_SB2_SH(g0, g1, c0, c0, t0, t1); \ |
26 | SRAI_H2_SH(t0, t1, 5); \ |
27 | t0 = __msa_subv_h((v8i16)src0, t0); \ |
28 | t1 = __msa_subv_h((v8i16)src1, t1); \ |
29 | t4 = __msa_srli_w((v4i32)src0, 16); \ |
30 | t5 = __msa_srli_w((v4i32)src1, 16); \ |
31 | DOTP_SB2_SH(t4, t5, c1, c1, t2, t3); \ |
32 | SRAI_H2_SH(t2, t3, 5); \ |
33 | SUB2(t0, t2, t1, t3, t0, t1); \ |
34 | VSHF_B2_UB(src0, t0, src1, t1, mask1, mask1, dst0, dst1); \ |
35 | } while (0) |
36 | |
37 | #define TRANSFORM_COLOR_4(src, dst, c0, c1, mask0, mask1) do { \ |
38 | const v16i8 g0 = VSHF_SB(src, src, mask0); \ |
39 | v8i16 t0 = __msa_dotp_s_h(c0, g0); \ |
40 | v8i16 t1; \ |
41 | v4i32 t2; \ |
42 | t0 = SRAI_H(t0, 5); \ |
43 | t0 = __msa_subv_h((v8i16)src, t0); \ |
44 | t2 = __msa_srli_w((v4i32)src, 16); \ |
45 | t1 = __msa_dotp_s_h(c1, (v16i8)t2); \ |
46 | t1 = SRAI_H(t1, 5); \ |
47 | t0 = t0 - t1; \ |
48 | dst = VSHF_UB(src, t0, mask1); \ |
49 | } while (0) |
50 | |
51 | static void TransformColor_MSA(const VP8LMultipliers* const m, uint32_t* data, |
52 | int num_pixels) { |
53 | v16u8 src0, dst0; |
54 | const v16i8 g2br = (v16i8)__msa_fill_w(m->green_to_blue_ | |
55 | (m->green_to_red_ << 16)); |
56 | const v16i8 r2b = (v16i8)__msa_fill_w(m->red_to_blue_); |
57 | const v16u8 mask0 = { 1, 255, 1, 255, 5, 255, 5, 255, 9, 255, 9, 255, |
58 | 13, 255, 13, 255 }; |
59 | const v16u8 mask1 = { 16, 1, 18, 3, 20, 5, 22, 7, 24, 9, 26, 11, |
60 | 28, 13, 30, 15 }; |
61 | |
62 | while (num_pixels >= 8) { |
63 | v16u8 src1, dst1; |
64 | LD_UB2(data, 4, src0, src1); |
65 | TRANSFORM_COLOR_8(src0, src1, dst0, dst1, g2br, r2b, mask0, mask1); |
66 | ST_UB2(dst0, dst1, data, 4); |
67 | data += 8; |
68 | num_pixels -= 8; |
69 | } |
70 | if (num_pixels > 0) { |
71 | if (num_pixels >= 4) { |
72 | src0 = LD_UB(data); |
73 | TRANSFORM_COLOR_4(src0, dst0, g2br, r2b, mask0, mask1); |
74 | ST_UB(dst0, data); |
75 | data += 4; |
76 | num_pixels -= 4; |
77 | } |
78 | if (num_pixels > 0) { |
79 | src0 = LD_UB(data); |
80 | TRANSFORM_COLOR_4(src0, dst0, g2br, r2b, mask0, mask1); |
81 | if (num_pixels == 3) { |
82 | const uint64_t pix_d = __msa_copy_s_d((v2i64)dst0, 0); |
83 | const uint32_t pix_w = __msa_copy_s_w((v4i32)dst0, 2); |
84 | SD(pix_d, data + 0); |
85 | SW(pix_w, data + 2); |
86 | } else if (num_pixels == 2) { |
87 | const uint64_t pix_d = __msa_copy_s_d((v2i64)dst0, 0); |
88 | SD(pix_d, data); |
89 | } else { |
90 | const uint32_t pix_w = __msa_copy_s_w((v4i32)dst0, 0); |
91 | SW(pix_w, data); |
92 | } |
93 | } |
94 | } |
95 | } |
96 | |
97 | static void SubtractGreenFromBlueAndRed_MSA(uint32_t* argb_data, |
98 | int num_pixels) { |
99 | int i; |
100 | uint8_t* ptemp_data = (uint8_t*)argb_data; |
101 | v16u8 src0, dst0, tmp0; |
102 | const v16u8 mask = { 1, 255, 1, 255, 5, 255, 5, 255, 9, 255, 9, 255, |
103 | 13, 255, 13, 255 }; |
104 | |
105 | while (num_pixels >= 8) { |
106 | v16u8 src1, dst1, tmp1; |
107 | LD_UB2(ptemp_data, 16, src0, src1); |
108 | VSHF_B2_UB(src0, src1, src1, src0, mask, mask, tmp0, tmp1); |
109 | SUB2(src0, tmp0, src1, tmp1, dst0, dst1); |
110 | ST_UB2(dst0, dst1, ptemp_data, 16); |
111 | ptemp_data += 8 * 4; |
112 | num_pixels -= 8; |
113 | } |
114 | if (num_pixels > 0) { |
115 | if (num_pixels >= 4) { |
116 | src0 = LD_UB(ptemp_data); |
117 | tmp0 = VSHF_UB(src0, src0, mask); |
118 | dst0 = src0 - tmp0; |
119 | ST_UB(dst0, ptemp_data); |
120 | ptemp_data += 4 * 4; |
121 | num_pixels -= 4; |
122 | } |
123 | for (i = 0; i < num_pixels; i++) { |
124 | const uint8_t b = ptemp_data[0]; |
125 | const uint8_t g = ptemp_data[1]; |
126 | const uint8_t r = ptemp_data[2]; |
127 | ptemp_data[0] = (b - g) & 0xff; |
128 | ptemp_data[2] = (r - g) & 0xff; |
129 | ptemp_data += 4; |
130 | } |
131 | } |
132 | } |
133 | |
134 | //------------------------------------------------------------------------------ |
135 | // Entry point |
136 | |
137 | extern void VP8LEncDspInitMSA(void); |
138 | |
139 | WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitMSA(void) { |
140 | VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed_MSA; |
141 | VP8LTransformColor = TransformColor_MSA; |
142 | } |
143 | |
144 | #else // !WEBP_USE_MSA |
145 | |
146 | WEBP_DSP_INIT_STUB(VP8LEncDspInitMSA) |
147 | |
148 | #endif // WEBP_USE_MSA |
149 | |