• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 version of YUV to RGB upsampling functions.
11 //
12 // Author: Prashant Patil (prashant.patil@imgtec.com)
13 
14 #include <string.h>
15 #include "src/dsp/dsp.h"
16 
17 #if defined(WEBP_USE_MSA)
18 
19 #include "src/dsp/msa_macro.h"
20 #include "src/dsp/yuv.h"
21 
22 #ifdef FANCY_UPSAMPLING
23 
24 #define ILVR_UW2(in, out0, out1) do {                            \
25   const v8i16 t0 = (v8i16)__msa_ilvr_b((v16i8)zero, (v16i8)in);  \
26   out0 = (v4u32)__msa_ilvr_h((v8i16)zero, t0);                   \
27   out1 = (v4u32)__msa_ilvl_h((v8i16)zero, t0);                   \
28 } while (0)
29 
30 #define ILVRL_UW4(in, out0, out1, out2, out3) do {  \
31   v16u8 t0, t1;                                     \
32   ILVRL_B2_UB(zero, in, t0, t1);                    \
33   ILVRL_H2_UW(zero, t0, out0, out1);                \
34   ILVRL_H2_UW(zero, t1, out2, out3);                \
35 } while (0)
36 
37 #define MULTHI_16(in0, in1, in2, in3, cnst, out0, out1) do {   \
38   const v4i32 const0 = (v4i32)__msa_fill_w(cnst * 256);        \
39   v4u32 temp0, temp1, temp2, temp3;                            \
40   MUL4(in0, const0, in1, const0, in2, const0, in3, const0,     \
41        temp0, temp1, temp2, temp3);                            \
42   PCKOD_H2_UH(temp1, temp0, temp3, temp2, out0, out1);         \
43 } while (0)
44 
45 #define MULTHI_8(in0, in1, cnst, out0) do {                 \
46   const v4i32 const0 = (v4i32)__msa_fill_w(cnst * 256);     \
47   v4u32 temp0, temp1;                                       \
48   MUL2(in0, const0, in1, const0, temp0, temp1);             \
49   out0 = (v8u16)__msa_pckod_h((v8i16)temp1, (v8i16)temp0);  \
50 } while (0)
51 
52 #define CALC_R16(y0, y1, v0, v1, dst) do {                \
53   const v8i16 const_a = (v8i16)__msa_fill_h(14234);       \
54   const v8i16 a0 = __msa_adds_s_h((v8i16)y0, (v8i16)v0);  \
55   const v8i16 a1 = __msa_adds_s_h((v8i16)y1, (v8i16)v1);  \
56   v8i16 b0 = __msa_subs_s_h(a0, const_a);                 \
57   v8i16 b1 = __msa_subs_s_h(a1, const_a);                 \
58   SRAI_H2_SH(b0, b1, 6);                                  \
59   CLIP_SH2_0_255(b0, b1);                                 \
60   dst = (v16u8)__msa_pckev_b((v16i8)b1, (v16i8)b0);       \
61 } while (0)
62 
63 #define CALC_R8(y0, v0, dst) do {                         \
64   const v8i16 const_a = (v8i16)__msa_fill_h(14234);       \
65   const v8i16 a0 = __msa_adds_s_h((v8i16)y0, (v8i16)v0);  \
66   v8i16 b0 = __msa_subs_s_h(a0, const_a);                 \
67   b0 = SRAI_H(b0, 6);                                     \
68   CLIP_SH_0_255(b0);                                      \
69   dst = (v16u8)__msa_pckev_b((v16i8)b0, (v16i8)b0);       \
70 } while (0)
71 
72 #define CALC_G16(y0, y1, u0, u1, v0, v1, dst) do {   \
73   const v8i16 const_a = (v8i16)__msa_fill_h(8708);   \
74   v8i16 a0 = __msa_subs_s_h((v8i16)y0, (v8i16)u0);   \
75   v8i16 a1 = __msa_subs_s_h((v8i16)y1, (v8i16)u1);   \
76   const v8i16 b0 = __msa_subs_s_h(a0, (v8i16)v0);    \
77   const v8i16 b1 = __msa_subs_s_h(a1, (v8i16)v1);    \
78   a0 = __msa_adds_s_h(b0, const_a);                  \
79   a1 = __msa_adds_s_h(b1, const_a);                  \
80   SRAI_H2_SH(a0, a1, 6);                             \
81   CLIP_SH2_0_255(a0, a1);                            \
82   dst = (v16u8)__msa_pckev_b((v16i8)a1, (v16i8)a0);  \
83 } while (0)
84 
85 #define CALC_G8(y0, u0, v0, dst) do {                \
86   const v8i16 const_a = (v8i16)__msa_fill_h(8708);   \
87   v8i16 a0 = __msa_subs_s_h((v8i16)y0, (v8i16)u0);   \
88   const v8i16 b0 = __msa_subs_s_h(a0, (v8i16)v0);    \
89   a0 = __msa_adds_s_h(b0, const_a);                  \
90   a0 = SRAI_H(a0, 6);                                \
91   CLIP_SH_0_255(a0);                                 \
92   dst = (v16u8)__msa_pckev_b((v16i8)a0, (v16i8)a0);  \
93 } while (0)
94 
95 #define CALC_B16(y0, y1, u0, u1, dst) do {           \
96   const v8u16 const_a = (v8u16)__msa_fill_h(17685);  \
97   const v8u16 a0 = __msa_adds_u_h((v8u16)y0, u0);    \
98   const v8u16 a1 = __msa_adds_u_h((v8u16)y1, u1);    \
99   v8u16 b0 = __msa_subs_u_h(a0, const_a);            \
100   v8u16 b1 = __msa_subs_u_h(a1, const_a);            \
101   SRAI_H2_UH(b0, b1, 6);                             \
102   CLIP_UH2_0_255(b0, b1);                            \
103   dst = (v16u8)__msa_pckev_b((v16i8)b1, (v16i8)b0);  \
104 } while (0)
105 
106 #define CALC_B8(y0, u0, dst) do {                    \
107   const v8u16 const_a = (v8u16)__msa_fill_h(17685);  \
108   const v8u16 a0 = __msa_adds_u_h((v8u16)y0, u0);    \
109   v8u16 b0 = __msa_subs_u_h(a0, const_a);            \
110   b0 = SRAI_H(b0, 6);                                \
111   CLIP_UH_0_255(b0);                                 \
112   dst = (v16u8)__msa_pckev_b((v16i8)b0, (v16i8)b0);  \
113 } while (0)
114 
115 #define CALC_RGB16(y, u, v, R, G, B) do {    \
116   const v16u8 zero = { 0 };                  \
117   v8u16 y0, y1, u0, u1, v0, v1;              \
118   v4u32 p0, p1, p2, p3;                      \
119   const v16u8 in_y = LD_UB(y);               \
120   const v16u8 in_u = LD_UB(u);               \
121   const v16u8 in_v = LD_UB(v);               \
122   ILVRL_UW4(in_y, p0, p1, p2, p3);           \
123   MULTHI_16(p0, p1, p2, p3, 19077, y0, y1);  \
124   ILVRL_UW4(in_v, p0, p1, p2, p3);           \
125   MULTHI_16(p0, p1, p2, p3, 26149, v0, v1);  \
126   CALC_R16(y0, y1, v0, v1, R);               \
127   MULTHI_16(p0, p1, p2, p3, 13320, v0, v1);  \
128   ILVRL_UW4(in_u, p0, p1, p2, p3);           \
129   MULTHI_16(p0, p1, p2, p3, 6419, u0, u1);   \
130   CALC_G16(y0, y1, u0, u1, v0, v1, G);       \
131   MULTHI_16(p0, p1, p2, p3, 33050, u0, u1);  \
132   CALC_B16(y0, y1, u0, u1, B);               \
133 } while (0)
134 
135 #define CALC_RGB8(y, u, v, R, G, B) do {  \
136   const v16u8 zero = { 0 };               \
137   v8u16 y0, u0, v0;                       \
138   v4u32 p0, p1;                           \
139   const v16u8 in_y = LD_UB(y);            \
140   const v16u8 in_u = LD_UB(u);            \
141   const v16u8 in_v = LD_UB(v);            \
142   ILVR_UW2(in_y, p0, p1);                 \
143   MULTHI_8(p0, p1, 19077, y0);            \
144   ILVR_UW2(in_v, p0, p1);                 \
145   MULTHI_8(p0, p1, 26149, v0);            \
146   CALC_R8(y0, v0, R);                     \
147   MULTHI_8(p0, p1, 13320, v0);            \
148   ILVR_UW2(in_u, p0, p1);                 \
149   MULTHI_8(p0, p1, 6419, u0);             \
150   CALC_G8(y0, u0, v0, G);                 \
151   MULTHI_8(p0, p1, 33050, u0);            \
152   CALC_B8(y0, u0, B);                     \
153 } while (0)
154 
155 #define STORE16_3(a0, a1, a2, dst) do {                          \
156   const v16u8 mask0 = { 0, 1, 16, 2, 3, 17, 4, 5, 18, 6, 7, 19,  \
157                         8, 9, 20, 10 };                          \
158   const v16u8 mask1 = { 0, 21, 1, 2, 22, 3, 4, 23, 5, 6, 24, 7,  \
159                         8, 25, 9, 10 };                          \
160   const v16u8 mask2 = { 26, 0, 1, 27, 2, 3, 28, 4, 5, 29, 6, 7,  \
161                         30, 8, 9, 31 };                          \
162   v16u8 out0, out1, out2, tmp0, tmp1, tmp2;                      \
163   ILVRL_B2_UB(a1, a0, tmp0, tmp1);                               \
164   out0 = VSHF_UB(tmp0, a2, mask0);                               \
165   tmp2 = SLDI_UB(tmp1, tmp0, 11);                                \
166   out1 = VSHF_UB(tmp2, a2, mask1);                               \
167   tmp2 = SLDI_UB(tmp1, tmp1, 6);                                 \
168   out2 = VSHF_UB(tmp2, a2, mask2);                               \
169   ST_UB(out0, dst +  0);                                         \
170   ST_UB(out1, dst + 16);                                         \
171   ST_UB(out2, dst + 32);                                         \
172 } while (0)
173 
174 #define STORE8_3(a0, a1, a2, dst) do {                             \
175   int64_t out_m;                                                   \
176   const v16u8 mask0 = { 0, 1, 16, 2, 3, 17, 4, 5, 18, 6, 7, 19,    \
177                         8, 9, 20, 10 };                            \
178   const v16u8 mask1 = { 11, 21, 12, 13, 22, 14, 15, 23,            \
179                         255, 255, 255, 255, 255, 255, 255, 255 };  \
180   const v16u8 tmp0 = (v16u8)__msa_ilvr_b((v16i8)a1, (v16i8)a0);    \
181   v16u8 out0, out1;                                                \
182   VSHF_B2_UB(tmp0, a2, tmp0, a2, mask0, mask1, out0, out1);        \
183   ST_UB(out0, dst);                                                \
184   out_m = __msa_copy_s_d((v2i64)out1, 0);                          \
185   SD(out_m, dst + 16);                                             \
186 } while (0)
187 
188 #define STORE16_4(a0, a1, a2, a3, dst) do {  \
189   v16u8 tmp0, tmp1, tmp2, tmp3;              \
190   v16u8 out0, out1, out2, out3;              \
191   ILVRL_B2_UB(a1, a0, tmp0, tmp1);           \
192   ILVRL_B2_UB(a3, a2, tmp2, tmp3);           \
193   ILVRL_H2_UB(tmp2, tmp0, out0, out1);       \
194   ILVRL_H2_UB(tmp3, tmp1, out2, out3);       \
195   ST_UB(out0, dst +  0);                     \
196   ST_UB(out1, dst + 16);                     \
197   ST_UB(out2, dst + 32);                     \
198   ST_UB(out3, dst + 48);                     \
199 } while (0)
200 
201 #define STORE8_4(a0, a1, a2, a3, dst) do {  \
202   v16u8 tmp0, tmp1, tmp2, tmp3;             \
203   ILVR_B2_UB(a1, a0, a3, a2, tmp0, tmp1);   \
204   ILVRL_H2_UB(tmp1, tmp0, tmp2, tmp3);      \
205   ST_UB(tmp2, dst +  0);                    \
206   ST_UB(tmp3, dst + 16);                    \
207 } while (0)
208 
209 #define STORE2_16(a0, a1, dst) do {  \
210   v16u8 out0, out1;                  \
211   ILVRL_B2_UB(a1, a0, out0, out1);   \
212   ST_UB(out0, dst +  0);             \
213   ST_UB(out1, dst + 16);             \
214 } while (0)
215 
216 #define STORE2_8(a0, a1, dst) do {                               \
217   const v16u8 out0 = (v16u8)__msa_ilvr_b((v16i8)a1, (v16i8)a0);  \
218   ST_UB(out0, dst);                                              \
219 } while (0)
220 
221 #define CALC_RGBA4444(y, u, v, out0, out1, N, dst) do {  \
222   CALC_RGB##N(y, u, v, R, G, B);                         \
223   tmp0 = ANDI_B(R, 0xf0);                                \
224   tmp1 = SRAI_B(G, 4);                                   \
225   RG = tmp0 | tmp1;                                      \
226   tmp0 = ANDI_B(B, 0xf0);                                \
227   BA = ORI_B(tmp0, 0x0f);                                \
228   STORE2_##N(out0, out1, dst);                           \
229 } while (0)
230 
231 #define CALC_RGB565(y, u, v, out0, out1, N, dst) do {  \
232   CALC_RGB##N(y, u, v, R, G, B);                       \
233   tmp0 = ANDI_B(R, 0xf8);                              \
234   tmp1 = SRAI_B(G, 5);                                 \
235   RG = tmp0 | tmp1;                                    \
236   tmp0 = SLLI_B(G, 3);                                 \
237   tmp1 = ANDI_B(tmp0, 0xe0);                           \
238   tmp0 = SRAI_B(B, 3);                                 \
239   GB = tmp0 | tmp1;                                    \
240   STORE2_##N(out0, out1, dst);                         \
241 } while (0)
242 
Clip8(int v)243 static WEBP_INLINE int Clip8(int v) {
244   return v < 0 ? 0 : v > 255 ? 255 : v;
245 }
246 
YuvToRgb(int y,int u,int v,uint8_t * const rgb)247 static void YuvToRgb(int y, int u, int v, uint8_t* const rgb) {
248   const int y1 = MultHi(y, 19077);
249   const int r1 = y1 + MultHi(v, 26149) - 14234;
250   const int g1 = y1 - MultHi(u, 6419) - MultHi(v, 13320) + 8708;
251   const int b1 = y1 + MultHi(u, 33050) - 17685;
252   rgb[0] = Clip8(r1 >> 6);
253   rgb[1] = Clip8(g1 >> 6);
254   rgb[2] = Clip8(b1 >> 6);
255 }
256 
YuvToBgr(int y,int u,int v,uint8_t * const bgr)257 static void YuvToBgr(int y, int u, int v, uint8_t* const bgr) {
258   const int y1 = MultHi(y, 19077);
259   const int r1 = y1 + MultHi(v, 26149) - 14234;
260   const int g1 = y1 - MultHi(u, 6419) - MultHi(v, 13320) + 8708;
261   const int b1 = y1 + MultHi(u, 33050) - 17685;
262   bgr[0] = Clip8(b1 >> 6);
263   bgr[1] = Clip8(g1 >> 6);
264   bgr[2] = Clip8(r1 >> 6);
265 }
266 
267 #if !defined(WEBP_REDUCE_CSP)
YuvToRgb565(int y,int u,int v,uint8_t * const rgb)268 static void YuvToRgb565(int y, int u, int v, uint8_t* const rgb) {
269   const int y1 = MultHi(y, 19077);
270   const int r1 = y1 + MultHi(v, 26149) - 14234;
271   const int g1 = y1 - MultHi(u, 6419) - MultHi(v, 13320) + 8708;
272   const int b1 = y1 + MultHi(u, 33050) - 17685;
273   const int r = Clip8(r1 >> 6);
274   const int g = Clip8(g1 >> 6);
275   const int b = Clip8(b1 >> 6);
276   const int rg = (r & 0xf8) | (g >> 5);
277   const int gb = ((g << 3) & 0xe0) | (b >> 3);
278 #if (WEBP_SWAP_16BIT_CSP == 1)
279   rgb[0] = gb;
280   rgb[1] = rg;
281 #else
282   rgb[0] = rg;
283   rgb[1] = gb;
284 #endif
285 }
286 
YuvToRgba4444(int y,int u,int v,uint8_t * const argb)287 static void YuvToRgba4444(int y, int u, int v, uint8_t* const argb) {
288   const int y1 = MultHi(y, 19077);
289   const int r1 = y1 + MultHi(v, 26149) - 14234;
290   const int g1 = y1 - MultHi(u, 6419) - MultHi(v, 13320) + 8708;
291   const int b1 = y1 + MultHi(u, 33050) - 17685;
292   const int r = Clip8(r1 >> 6);
293   const int g = Clip8(g1 >> 6);
294   const int b = Clip8(b1 >> 6);
295   const int rg = (r & 0xf0) | (g >> 4);
296   const int ba = (b & 0xf0) | 0x0f;     // overwrite the lower 4 bits
297 #if (WEBP_SWAP_16BIT_CSP == 1)
298   argb[0] = ba;
299   argb[1] = rg;
300 #else
301   argb[0] = rg;
302   argb[1] = ba;
303 #endif
304 }
305 
YuvToArgb(uint8_t y,uint8_t u,uint8_t v,uint8_t * const argb)306 static void YuvToArgb(uint8_t y, uint8_t u, uint8_t v, uint8_t* const argb) {
307   argb[0] = 0xff;
308   YuvToRgb(y, u, v, argb + 1);
309 }
310 #endif  // WEBP_REDUCE_CSP
311 
YuvToBgra(uint8_t y,uint8_t u,uint8_t v,uint8_t * const bgra)312 static void YuvToBgra(uint8_t y, uint8_t u, uint8_t v, uint8_t* const bgra) {
313   YuvToBgr(y, u, v, bgra);
314   bgra[3] = 0xff;
315 }
316 
YuvToRgba(uint8_t y,uint8_t u,uint8_t v,uint8_t * const rgba)317 static void YuvToRgba(uint8_t y, uint8_t u, uint8_t v, uint8_t* const rgba) {
318   YuvToRgb(y, u, v, rgba);
319   rgba[3] = 0xff;
320 }
321 
322 #if !defined(WEBP_REDUCE_CSP)
YuvToRgbLine(const uint8_t * y,const uint8_t * u,const uint8_t * v,uint8_t * dst,int length)323 static void YuvToRgbLine(const uint8_t* y, const uint8_t* u,
324                          const uint8_t* v, uint8_t* dst, int length) {
325   v16u8 R, G, B;
326   while (length >= 16) {
327     CALC_RGB16(y, u, v, R, G, B);
328     STORE16_3(R, G, B, dst);
329     y      += 16;
330     u      += 16;
331     v      += 16;
332     dst    += 16 * 3;
333     length -= 16;
334   }
335   if (length > 8) {
336     uint8_t temp[3 * 16] = { 0 };
337     memcpy(temp, y, length * sizeof(*temp));
338     CALC_RGB16(temp, u, v, R, G, B);
339     STORE16_3(R, G, B, temp);
340     memcpy(dst, temp, length * 3 * sizeof(*dst));
341   } else if (length > 0) {
342     uint8_t temp[3 * 8] = { 0 };
343     memcpy(temp, y, length * sizeof(*temp));
344     CALC_RGB8(temp, u, v, R, G, B);
345     STORE8_3(R, G, B, temp);
346     memcpy(dst, temp, length * 3 * sizeof(*dst));
347   }
348 }
349 
YuvToBgrLine(const uint8_t * y,const uint8_t * u,const uint8_t * v,uint8_t * dst,int length)350 static void YuvToBgrLine(const uint8_t* y, const uint8_t* u,
351                          const uint8_t* v, uint8_t* dst, int length) {
352   v16u8 R, G, B;
353   while (length >= 16) {
354     CALC_RGB16(y, u, v, R, G, B);
355     STORE16_3(B, G, R, dst);
356     y      += 16;
357     u      += 16;
358     v      += 16;
359     dst    += 16 * 3;
360     length -= 16;
361   }
362   if (length > 8) {
363     uint8_t temp[3 * 16] = { 0 };
364     memcpy(temp, y, length * sizeof(*temp));
365     CALC_RGB16(temp, u, v, R, G, B);
366     STORE16_3(B, G, R, temp);
367     memcpy(dst, temp, length * 3 * sizeof(*dst));
368   } else if (length > 0) {
369     uint8_t temp[3 * 8] = { 0 };
370     memcpy(temp, y, length * sizeof(*temp));
371     CALC_RGB8(temp, u, v, R, G, B);
372     STORE8_3(B, G, R, temp);
373     memcpy(dst, temp, length * 3 * sizeof(*dst));
374   }
375 }
376 #endif  // WEBP_REDUCE_CSP
377 
YuvToRgbaLine(const uint8_t * y,const uint8_t * u,const uint8_t * v,uint8_t * dst,int length)378 static void YuvToRgbaLine(const uint8_t* y, const uint8_t* u,
379                           const uint8_t* v, uint8_t* dst, int length) {
380   v16u8 R, G, B;
381   const v16u8 A = (v16u8)__msa_ldi_b(ALPHAVAL);
382   while (length >= 16) {
383     CALC_RGB16(y, u, v, R, G, B);
384     STORE16_4(R, G, B, A, dst);
385     y      += 16;
386     u      += 16;
387     v      += 16;
388     dst    += 16 * 4;
389     length -= 16;
390   }
391   if (length > 8) {
392     uint8_t temp[4 * 16] = { 0 };
393     memcpy(temp, y, length * sizeof(*temp));
394     CALC_RGB16(&temp[0], u, v, R, G, B);
395     STORE16_4(R, G, B, A, temp);
396     memcpy(dst, temp, length * 4 * sizeof(*dst));
397   } else if (length > 0) {
398     uint8_t temp[4 * 8] = { 0 };
399     memcpy(temp, y, length * sizeof(*temp));
400     CALC_RGB8(temp, u, v, R, G, B);
401     STORE8_4(R, G, B, A, temp);
402     memcpy(dst, temp, length * 4 * sizeof(*dst));
403   }
404 }
405 
YuvToBgraLine(const uint8_t * y,const uint8_t * u,const uint8_t * v,uint8_t * dst,int length)406 static void YuvToBgraLine(const uint8_t* y, const uint8_t* u,
407                           const uint8_t* v, uint8_t* dst, int length) {
408   v16u8 R, G, B;
409   const v16u8 A = (v16u8)__msa_ldi_b(ALPHAVAL);
410   while (length >= 16) {
411     CALC_RGB16(y, u, v, R, G, B);
412     STORE16_4(B, G, R, A, dst);
413     y      += 16;
414     u      += 16;
415     v      += 16;
416     dst    += 16 * 4;
417     length -= 16;
418   }
419   if (length > 8) {
420     uint8_t temp[4 * 16] = { 0 };
421     memcpy(temp, y, length * sizeof(*temp));
422     CALC_RGB16(temp, u, v, R, G, B);
423     STORE16_4(B, G, R, A, temp);
424     memcpy(dst, temp, length * 4 * sizeof(*dst));
425   } else if (length > 0) {
426     uint8_t temp[4 * 8] = { 0 };
427     memcpy(temp, y, length * sizeof(*temp));
428     CALC_RGB8(temp, u, v, R, G, B);
429     STORE8_4(B, G, R, A, temp);
430     memcpy(dst, temp, length * 4 * sizeof(*dst));
431   }
432 }
433 
434 #if !defined(WEBP_REDUCE_CSP)
YuvToArgbLine(const uint8_t * y,const uint8_t * u,const uint8_t * v,uint8_t * dst,int length)435 static void YuvToArgbLine(const uint8_t* y, const uint8_t* u,
436                           const uint8_t* v, uint8_t* dst, int length) {
437   v16u8 R, G, B;
438   const v16u8 A = (v16u8)__msa_ldi_b(ALPHAVAL);
439   while (length >= 16) {
440     CALC_RGB16(y, u, v, R, G, B);
441     STORE16_4(A, R, G, B, dst);
442     y      += 16;
443     u      += 16;
444     v      += 16;
445     dst    += 16 * 4;
446     length -= 16;
447   }
448   if (length > 8) {
449     uint8_t temp[4 * 16] = { 0 };
450     memcpy(temp, y, length * sizeof(*temp));
451     CALC_RGB16(temp, u, v, R, G, B);
452     STORE16_4(A, R, G, B, temp);
453     memcpy(dst, temp, length * 4 * sizeof(*dst));
454   } else if (length > 0) {
455     uint8_t temp[4 * 8] = { 0 };
456     memcpy(temp, y, length * sizeof(*temp));
457     CALC_RGB8(temp, u, v, R, G, B);
458     STORE8_4(A, R, G, B, temp);
459     memcpy(dst, temp, length * 4 * sizeof(*dst));
460   }
461 }
462 
YuvToRgba4444Line(const uint8_t * y,const uint8_t * u,const uint8_t * v,uint8_t * dst,int length)463 static void YuvToRgba4444Line(const uint8_t* y, const uint8_t* u,
464                               const uint8_t* v, uint8_t* dst, int length) {
465   v16u8 R, G, B, RG, BA, tmp0, tmp1;
466   while (length >= 16) {
467 #if (WEBP_SWAP_16BIT_CSP == 1)
468     CALC_RGBA4444(y, u, v, BA, RG, 16, dst);
469 #else
470     CALC_RGBA4444(y, u, v, RG, BA, 16, dst);
471 #endif
472     y      += 16;
473     u      += 16;
474     v      += 16;
475     dst    += 16 * 2;
476     length -= 16;
477   }
478   if (length > 8) {
479     uint8_t temp[2 * 16] = { 0 };
480     memcpy(temp, y, length * sizeof(*temp));
481 #if (WEBP_SWAP_16BIT_CSP == 1)
482     CALC_RGBA4444(temp, u, v, BA, RG, 16, temp);
483 #else
484     CALC_RGBA4444(temp, u, v, RG, BA, 16, temp);
485 #endif
486     memcpy(dst, temp, length * 2 * sizeof(*dst));
487   } else if (length > 0) {
488     uint8_t temp[2 * 8] = { 0 };
489     memcpy(temp, y, length * sizeof(*temp));
490 #if (WEBP_SWAP_16BIT_CSP == 1)
491     CALC_RGBA4444(temp, u, v, BA, RG, 8, temp);
492 #else
493     CALC_RGBA4444(temp, u, v, RG, BA, 8, temp);
494 #endif
495     memcpy(dst, temp, length * 2 * sizeof(*dst));
496   }
497 }
498 
YuvToRgb565Line(const uint8_t * y,const uint8_t * u,const uint8_t * v,uint8_t * dst,int length)499 static void YuvToRgb565Line(const uint8_t* y, const uint8_t* u,
500                             const uint8_t* v, uint8_t* dst, int length) {
501   v16u8 R, G, B, RG, GB, tmp0, tmp1;
502   while (length >= 16) {
503 #if (WEBP_SWAP_16BIT_CSP == 1)
504     CALC_RGB565(y, u, v, GB, RG, 16, dst);
505 #else
506     CALC_RGB565(y, u, v, RG, GB, 16, dst);
507 #endif
508     y      += 16;
509     u      += 16;
510     v      += 16;
511     dst    += 16 * 2;
512     length -= 16;
513   }
514   if (length > 8) {
515     uint8_t temp[2 * 16] = { 0 };
516     memcpy(temp, y, length * sizeof(*temp));
517 #if (WEBP_SWAP_16BIT_CSP == 1)
518     CALC_RGB565(temp, u, v, GB, RG, 16, temp);
519 #else
520     CALC_RGB565(temp, u, v, RG, GB, 16, temp);
521 #endif
522     memcpy(dst, temp, length * 2 * sizeof(*dst));
523   } else if (length > 0) {
524     uint8_t temp[2 * 8] = { 0 };
525     memcpy(temp, y, length * sizeof(*temp));
526 #if (WEBP_SWAP_16BIT_CSP == 1)
527     CALC_RGB565(temp, u, v, GB, RG, 8, temp);
528 #else
529     CALC_RGB565(temp, u, v, RG, GB, 8, temp);
530 #endif
531     memcpy(dst, temp, length * 2 * sizeof(*dst));
532   }
533 }
534 #endif  // WEBP_REDUCE_CSP
535 
536 #define UPSAMPLE_32PIXELS(a, b, c, d) do {    \
537   v16u8 s = __msa_aver_u_b(a, d);             \
538   v16u8 t = __msa_aver_u_b(b, c);             \
539   const v16u8 st = s ^ t;                     \
540   v16u8 ad = a ^ d;                           \
541   v16u8 bc = b ^ c;                           \
542   v16u8 t0 = ad | bc;                         \
543   v16u8 t1 = t0 | st;                         \
544   v16u8 t2 = ANDI_B(t1, 1);                   \
545   v16u8 t3 = __msa_aver_u_b(s, t);            \
546   const v16u8 k = t3 - t2;                    \
547   v16u8 diag1, diag2;                         \
548   AVER_UB2_UB(t, k, s, k, t0, t1);            \
549   bc = bc & st;                               \
550   ad = ad & st;                               \
551   t = t ^ k;                                  \
552   s = s ^ k;                                  \
553   t2 = bc | t;                                \
554   t3 = ad | s;                                \
555   t2 = ANDI_B(t2, 1);                         \
556   t3 = ANDI_B(t3, 1);                         \
557   SUB2(t0, t2, t1, t3, diag1, diag2);         \
558   AVER_UB2_UB(a, diag1, b, diag2, t0, t1);    \
559   ILVRL_B2_UB(t1, t0, a, b);                  \
560   if (pbot_y != NULL) {                       \
561     AVER_UB2_UB(c, diag2, d, diag1, t0, t1);  \
562     ILVRL_B2_UB(t1, t0, c, d);                \
563   }                                           \
564 } while (0)
565 
566 #define UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP)                            \
567 static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y,        \
568                       const uint8_t* top_u, const uint8_t* top_v,        \
569                       const uint8_t* cur_u, const uint8_t* cur_v,        \
570                       uint8_t* top_dst, uint8_t* bot_dst, int len)       \
571 {                                                                        \
572   int size = (len - 1) >> 1;                                             \
573   uint8_t temp_u[64];                                                    \
574   uint8_t temp_v[64];                                                    \
575   const uint32_t tl_uv = ((top_u[0]) | ((top_v[0]) << 16));              \
576   const uint32_t l_uv = ((cur_u[0]) | ((cur_v[0]) << 16));               \
577   const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2;            \
578   const uint8_t* ptop_y = &top_y[1];                                     \
579   uint8_t* ptop_dst = top_dst + XSTEP;                                   \
580   const uint8_t* pbot_y = &bot_y[1];                                     \
581   uint8_t* pbot_dst = bot_dst + XSTEP;                                   \
582                                                                          \
583   FUNC(top_y[0], uv0 & 0xff, (uv0 >> 16), top_dst);                      \
584   if (bot_y != NULL) {                                                   \
585     const uint32_t uv1 = (3 * l_uv + tl_uv + 0x00020002u) >> 2;          \
586     FUNC(bot_y[0], uv1 & 0xff, (uv1 >> 16), bot_dst);                    \
587   }                                                                      \
588   while (size >= 16) {                                                   \
589     v16u8 tu0, tu1, tv0, tv1, cu0, cu1, cv0, cv1;                        \
590     LD_UB2(top_u, 1, tu0, tu1);                                          \
591     LD_UB2(cur_u, 1, cu0, cu1);                                          \
592     LD_UB2(top_v, 1, tv0, tv1);                                          \
593     LD_UB2(cur_v, 1, cv0, cv1);                                          \
594     UPSAMPLE_32PIXELS(tu0, tu1, cu0, cu1);                               \
595     UPSAMPLE_32PIXELS(tv0, tv1, cv0, cv1);                               \
596     ST_UB4(tu0, tu1, cu0, cu1, &temp_u[0], 16);                          \
597     ST_UB4(tv0, tv1, cv0, cv1, &temp_v[0], 16);                          \
598     FUNC##Line(ptop_y, &temp_u[ 0], &temp_v[0], ptop_dst, 32);           \
599     if (bot_y != NULL) {                                                 \
600       FUNC##Line(pbot_y, &temp_u[32], &temp_v[32], pbot_dst, 32);        \
601     }                                                                    \
602     ptop_y   += 32;                                                      \
603     pbot_y   += 32;                                                      \
604     ptop_dst += XSTEP * 32;                                              \
605     pbot_dst += XSTEP * 32;                                              \
606     top_u    += 16;                                                      \
607     top_v    += 16;                                                      \
608     cur_u    += 16;                                                      \
609     cur_v    += 16;                                                      \
610     size     -= 16;                                                      \
611   }                                                                      \
612   if (size > 0) {                                                        \
613     v16u8 tu0, tu1, tv0, tv1, cu0, cu1, cv0, cv1;                        \
614     memcpy(&temp_u[ 0], top_u, 17 * sizeof(uint8_t));                    \
615     memcpy(&temp_u[32], cur_u, 17 * sizeof(uint8_t));                    \
616     memcpy(&temp_v[ 0], top_v, 17 * sizeof(uint8_t));                    \
617     memcpy(&temp_v[32], cur_v, 17 * sizeof(uint8_t));                    \
618     LD_UB2(&temp_u[ 0], 1, tu0, tu1);                                    \
619     LD_UB2(&temp_u[32], 1, cu0, cu1);                                    \
620     LD_UB2(&temp_v[ 0], 1, tv0, tv1);                                    \
621     LD_UB2(&temp_v[32], 1, cv0, cv1);                                    \
622     UPSAMPLE_32PIXELS(tu0, tu1, cu0, cu1);                               \
623     UPSAMPLE_32PIXELS(tv0, tv1, cv0, cv1);                               \
624     ST_UB4(tu0, tu1, cu0, cu1, &temp_u[0], 16);                          \
625     ST_UB4(tv0, tv1, cv0, cv1, &temp_v[0], 16);                          \
626     FUNC##Line(ptop_y, &temp_u[ 0], &temp_v[0], ptop_dst, size * 2);     \
627     if (bot_y != NULL) {                                                 \
628       FUNC##Line(pbot_y, &temp_u[32], &temp_v[32], pbot_dst, size * 2);  \
629     }                                                                    \
630     top_u += size;                                                       \
631     top_v += size;                                                       \
632     cur_u += size;                                                       \
633     cur_v += size;                                                       \
634   }                                                                      \
635   if (!(len & 1)) {                                                      \
636     const uint32_t t0 = ((top_u[0]) | ((top_v[0]) << 16));               \
637     const uint32_t c0  = ((cur_u[0]) | ((cur_v[0]) << 16));              \
638     const uint32_t tmp0 = (3 * t0 + c0 + 0x00020002u) >> 2;              \
639     FUNC(top_y[len - 1], tmp0 & 0xff, (tmp0 >> 16),                      \
640                 top_dst + (len - 1) * XSTEP);                            \
641     if (bot_y != NULL) {                                                 \
642       const uint32_t tmp1 = (3 * c0 + t0 + 0x00020002u) >> 2;            \
643       FUNC(bot_y[len - 1], tmp1 & 0xff, (tmp1 >> 16),                    \
644            bot_dst + (len - 1) * XSTEP);                                 \
645     }                                                                    \
646   }                                                                      \
647 }
648 
649 UPSAMPLE_FUNC(UpsampleRgbaLinePair,     YuvToRgba,     4)
650 UPSAMPLE_FUNC(UpsampleBgraLinePair,     YuvToBgra,     4)
651 #if !defined(WEBP_REDUCE_CSP)
652 UPSAMPLE_FUNC(UpsampleRgbLinePair,      YuvToRgb,      3)
653 UPSAMPLE_FUNC(UpsampleBgrLinePair,      YuvToBgr,      3)
654 UPSAMPLE_FUNC(UpsampleArgbLinePair,     YuvToArgb,     4)
655 UPSAMPLE_FUNC(UpsampleRgba4444LinePair, YuvToRgba4444, 2)
656 UPSAMPLE_FUNC(UpsampleRgb565LinePair,   YuvToRgb565,   2)
657 #endif   // WEBP_REDUCE_CSP
658 
659 //------------------------------------------------------------------------------
660 // Entry point
661 
662 extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
663 
664 extern void WebPInitUpsamplersMSA(void);
665 
WebPInitUpsamplersMSA(void)666 WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersMSA(void) {
667   WebPUpsamplers[MODE_RGBA]      = UpsampleRgbaLinePair;
668   WebPUpsamplers[MODE_BGRA]      = UpsampleBgraLinePair;
669   WebPUpsamplers[MODE_rgbA]      = UpsampleRgbaLinePair;
670   WebPUpsamplers[MODE_bgrA]      = UpsampleBgraLinePair;
671 #if !defined(WEBP_REDUCE_CSP)
672   WebPUpsamplers[MODE_RGB]       = UpsampleRgbLinePair;
673   WebPUpsamplers[MODE_BGR]       = UpsampleBgrLinePair;
674   WebPUpsamplers[MODE_ARGB]      = UpsampleArgbLinePair;
675   WebPUpsamplers[MODE_Argb]      = UpsampleArgbLinePair;
676   WebPUpsamplers[MODE_RGB_565]   = UpsampleRgb565LinePair;
677   WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
678   WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
679 #endif   // WEBP_REDUCE_CSP
680 }
681 
682 #endif  // FANCY_UPSAMPLING
683 
684 #endif  // WEBP_USE_MSA
685 
686 #if !(defined(FANCY_UPSAMPLING) && defined(WEBP_USE_MSA))
687 WEBP_DSP_INIT_STUB(WebPInitUpsamplersMSA)
688 #endif
689