1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
|
// Copyright 2016 Google Inc. All Rights Reserved.
//
// Use of this source code is governed by a BSD-style license
// that can be found in the COPYING file in the root of the source
// tree. An additional intellectual property rights grant can be found
// in the file PATENTS. All contributing project authors may
// be found in the AUTHORS file in the root of the source tree.
// -----------------------------------------------------------------------------
//
// MSA version of rescaling functions
//
// Author: Prashant Patil (prashant.patil@imgtec.com)
#include "src/dsp/dsp.h"
#if defined(WEBP_USE_MSA) && !defined(WEBP_REDUCE_SIZE)
#include <assert.h>
#include "src/utils/rescaler_utils.h"
#include "src/dsp/msa_macro.h"
#define ROUNDER (WEBP_RESCALER_ONE >> 1)
#define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
#define MULT_FIX_FLOOR(x, y) (((uint64_t)(x) * (y)) >> WEBP_RESCALER_RFIX)
#define CALC_MULT_FIX_16(in0, in1, in2, in3, scale, shift, dst) do { \
v4u32 tmp0, tmp1, tmp2, tmp3; \
v16u8 t0, t1, t2, t3, t4, t5; \
v2u64 out0, out1, out2, out3; \
ILVRL_W2_UW(zero, in0, tmp0, tmp1); \
ILVRL_W2_UW(zero, in1, tmp2, tmp3); \
DOTP_UW2_UD(tmp0, tmp1, scale, scale, out0, out1); \
DOTP_UW2_UD(tmp2, tmp3, scale, scale, out2, out3); \
SRAR_D4_UD(out0, out1, out2, out3, shift); \
PCKEV_B2_UB(out1, out0, out3, out2, t0, t1); \
ILVRL_W2_UW(zero, in2, tmp0, tmp1); \
ILVRL_W2_UW(zero, in3, tmp2, tmp3); \
DOTP_UW2_UD(tmp0, tmp1, scale, scale, out0, out1); \
DOTP_UW2_UD(tmp2, tmp3, scale, scale, out2, out3); \
SRAR_D4_UD(out0, out1, out2, out3, shift); \
PCKEV_B2_UB(out1, out0, out3, out2, t2, t3); \
PCKEV_B2_UB(t1, t0, t3, t2, t4, t5); \
dst = (v16u8)__msa_pckev_b((v16i8)t5, (v16i8)t4); \
} while (0)
#define CALC_MULT_FIX_4(in0, scale, shift, dst) do { \
v4u32 tmp0, tmp1; \
v16i8 t0, t1; \
v2u64 out0, out1; \
ILVRL_W2_UW(zero, in0, tmp0, tmp1); \
DOTP_UW2_UD(tmp0, tmp1, scale, scale, out0, out1); \
SRAR_D2_UD(out0, out1, shift); \
t0 = __msa_pckev_b((v16i8)out1, (v16i8)out0); \
t1 = __msa_pckev_b(t0, t0); \
t0 = __msa_pckev_b(t1, t1); \
dst = __msa_copy_s_w((v4i32)t0, 0); \
} while (0)
#define CALC_MULT_FIX1_16(in0, in1, in2, in3, fyscale, shift, \
dst0, dst1, dst2, dst3) do { \
v4u32 tmp0, tmp1, tmp2, tmp3; \
v2u64 out0, out1, out2, out3; \
ILVRL_W2_UW(zero, in0, tmp0, tmp1); \
ILVRL_W2_UW(zero, in1, tmp2, tmp3); \
DOTP_UW2_UD(tmp0, tmp1, fyscale, fyscale, out0, out1); \
DOTP_UW2_UD(tmp2, tmp3, fyscale, fyscale, out2, out3); \
SRAR_D4_UD(out0, out1, out2, out3, shift); \
PCKEV_W2_UW(out1, out0, out3, out2, dst0, dst1); \
ILVRL_W2_UW(zero, in2, tmp0, tmp1); \
ILVRL_W2_UW(zero, in3, tmp2, tmp3); \
DOTP_UW2_UD(tmp0, tmp1, fyscale, fyscale, out0, out1); \
DOTP_UW2_UD(tmp2, tmp3, fyscale, fyscale, out2, out3); \
SRAR_D4_UD(out0, out1, out2, out3, shift); \
PCKEV_W2_UW(out1, out0, out3, out2, dst2, dst3); \
} while (0)
#define CALC_MULT_FIX1_4(in0, scale, shift, dst) do { \
v4u32 tmp0, tmp1; \
v2u64 out0, out1; \
ILVRL_W2_UW(zero, in0, tmp0, tmp1); \
DOTP_UW2_UD(tmp0, tmp1, scale, scale, out0, out1); \
SRAR_D2_UD(out0, out1, shift); \
dst = (v4u32)__msa_pckev_w((v4i32)out1, (v4i32)out0); \
} while (0)
#define CALC_MULT_FIX2_16(in0, in1, in2, in3, mult, scale, shift, \
dst0, dst1) do { \
v4u32 tmp0, tmp1, tmp2, tmp3; \
v2u64 out0, out1, out2, out3; \
ILVRL_W2_UW(in0, in2, tmp0, tmp1); \
ILVRL_W2_UW(in1, in3, tmp2, tmp3); \
DOTP_UW2_UD(tmp0, tmp1, mult, mult, out0, out1); \
DOTP_UW2_UD(tmp2, tmp3, mult, mult, out2, out3); \
SRAR_D4_UD(out0, out1, out2, out3, shift); \
DOTP_UW2_UD(out0, out1, scale, scale, out0, out1); \
DOTP_UW2_UD(out2, out3, scale, scale, out2, out3); \
SRAR_D4_UD(out0, out1, out2, out3, shift); \
PCKEV_B2_UB(out1, out0, out3, out2, dst0, dst1); \
} while (0)
#define CALC_MULT_FIX2_4(in0, in1, mult, scale, shift, dst) do { \
v4u32 tmp0, tmp1; \
v2u64 out0, out1; \
v16i8 t0, t1; \
ILVRL_W2_UW(in0, in1, tmp0, tmp1); \
DOTP_UW2_UD(tmp0, tmp1, mult, mult, out0, out1); \
SRAR_D2_UD(out0, out1, shift); \
DOTP_UW2_UD(out0, out1, scale, scale, out0, out1); \
SRAR_D2_UD(out0, out1, shift); \
t0 = __msa_pckev_b((v16i8)out1, (v16i8)out0); \
t1 = __msa_pckev_b(t0, t0); \
t0 = __msa_pckev_b(t1, t1); \
dst = __msa_copy_s_w((v4i32)t0, 0); \
} while (0)
static WEBP_INLINE void ExportRowExpand_0(const uint32_t* frow, uint8_t* dst,
int length,
WebPRescaler* const wrk) {
const v4u32 scale = (v4u32)__msa_fill_w(wrk->fy_scale);
const v4u32 shift = (v4u32)__msa_fill_w(WEBP_RESCALER_RFIX);
const v4i32 zero = { 0 };
while (length >= 16) {
v4u32 src0, src1, src2, src3;
v16u8 out;
LD_UW4(frow, 4, src0, src1, src2, src3);
CALC_MULT_FIX_16(src0, src1, src2, src3, scale, shift, out);
ST_UB(out, dst);
length -= 16;
frow += 16;
dst += 16;
}
if (length > 0) {
int x_out;
if (length >= 12) {
uint32_t val0_m, val1_m, val2_m;
v4u32 src0, src1, src2;
LD_UW3(frow, 4, src0, src1, src2);
CALC_MULT_FIX_4(src0, scale, shift, val0_m);
CALC_MULT_FIX_4(src1, scale, shift, val1_m);
CALC_MULT_FIX_4(src2, scale, shift, val2_m);
SW3(val0_m, val1_m, val2_m, dst, 4);
length -= 12;
frow += 12;
dst += 12;
} else if (length >= 8) {
uint32_t val0_m, val1_m;
v4u32 src0, src1;
LD_UW2(frow, 4, src0, src1);
CALC_MULT_FIX_4(src0, scale, shift, val0_m);
CALC_MULT_FIX_4(src1, scale, shift, val1_m);
SW2(val0_m, val1_m, dst, 4);
length -= 8;
frow += 8;
dst += 8;
} else if (length >= 4) {
uint32_t val0_m;
const v4u32 src0 = LD_UW(frow);
CALC_MULT_FIX_4(src0, scale, shift, val0_m);
SW(val0_m, dst);
length -= 4;
frow += 4;
dst += 4;
}
for (x_out = 0; x_out < length; ++x_out) {
const uint32_t J = frow[x_out];
const int v = (int)MULT_FIX(J, wrk->fy_scale);
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
}
}
}
static WEBP_INLINE void ExportRowExpand_1(const uint32_t* frow, uint32_t* irow,
uint8_t* dst, int length,
WebPRescaler* const wrk) {
const uint32_t B = WEBP_RESCALER_FRAC(-wrk->y_accum, wrk->y_sub);
const uint32_t A = (uint32_t)(WEBP_RESCALER_ONE - B);
const v4i32 B1 = __msa_fill_w(B);
const v4i32 A1 = __msa_fill_w(A);
const v4i32 AB = __msa_ilvr_w(A1, B1);
const v4u32 scale = (v4u32)__msa_fill_w(wrk->fy_scale);
const v4u32 shift = (v4u32)__msa_fill_w(WEBP_RESCALER_RFIX);
while (length >= 16) {
v4u32 frow0, frow1, frow2, frow3, irow0, irow1, irow2, irow3;
v16u8 t0, t1, t2, t3, t4, t5;
LD_UW4(frow, 4, frow0, frow1, frow2, frow3);
LD_UW4(irow, 4, irow0, irow1, irow2, irow3);
CALC_MULT_FIX2_16(frow0, frow1, irow0, irow1, AB, scale, shift, t0, t1);
CALC_MULT_FIX2_16(frow2, frow3, irow2, irow3, AB, scale, shift, t2, t3);
PCKEV_B2_UB(t1, t0, t3, t2, t4, t5);
t0 = (v16u8)__msa_pckev_b((v16i8)t5, (v16i8)t4);
ST_UB(t0, dst);
frow += 16;
irow += 16;
dst += 16;
length -= 16;
}
if (length > 0) {
int x_out;
if (length >= 12) {
uint32_t val0_m, val1_m, val2_m;
v4u32 frow0, frow1, frow2, irow0, irow1, irow2;
LD_UW3(frow, 4, frow0, frow1, frow2);
LD_UW3(irow, 4, irow0, irow1, irow2);
CALC_MULT_FIX2_4(frow0, irow0, AB, scale, shift, val0_m);
CALC_MULT_FIX2_4(frow1, irow1, AB, scale, shift, val1_m);
CALC_MULT_FIX2_4(frow2, irow2, AB, scale, shift, val2_m);
SW3(val0_m, val1_m, val2_m, dst, 4);
frow += 12;
irow += 12;
dst += 12;
length -= 12;
} else if (length >= 8) {
uint32_t val0_m, val1_m;
v4u32 frow0, frow1, irow0, irow1;
LD_UW2(frow, 4, frow0, frow1);
LD_UW2(irow, 4, irow0, irow1);
CALC_MULT_FIX2_4(frow0, irow0, AB, scale, shift, val0_m);
CALC_MULT_FIX2_4(frow1, irow1, AB, scale, shift, val1_m);
SW2(val0_m, val1_m, dst, 4);
frow += 4;
irow += 4;
dst += 4;
length -= 4;
} else if (length >= 4) {
uint32_t val0_m;
const v4u32 frow0 = LD_UW(frow + 0);
const v4u32 irow0 = LD_UW(irow + 0);
CALC_MULT_FIX2_4(frow0, irow0, AB, scale, shift, val0_m);
SW(val0_m, dst);
frow += 4;
irow += 4;
dst += 4;
length -= 4;
}
for (x_out = 0; x_out < length; ++x_out) {
const uint64_t I = (uint64_t)A * frow[x_out]
+ (uint64_t)B * irow[x_out];
const uint32_t J = (uint32_t)((I + ROUNDER) >> WEBP_RESCALER_RFIX);
const int v = (int)MULT_FIX(J, wrk->fy_scale);
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
}
}
}
static void RescalerExportRowExpand_MIPSdspR2(WebPRescaler* const wrk) {
uint8_t* dst = wrk->dst;
rescaler_t* irow = wrk->irow;
const int x_out_max = wrk->dst_width * wrk->num_channels;
const rescaler_t* frow = wrk->frow;
assert(!WebPRescalerOutputDone(wrk));
assert(wrk->y_accum <= 0);
assert(wrk->y_expand);
assert(wrk->y_sub != 0);
if (wrk->y_accum == 0) {
ExportRowExpand_0(frow, dst, x_out_max, wrk);
} else {
ExportRowExpand_1(frow, irow, dst, x_out_max, wrk);
}
}
#if 0 // disabled for now. TODO(skal): make match the C-code
static WEBP_INLINE void ExportRowShrink_0(const uint32_t* frow, uint32_t* irow,
uint8_t* dst, int length,
const uint32_t yscale,
WebPRescaler* const wrk) {
const v4u32 y_scale = (v4u32)__msa_fill_w(yscale);
const v4u32 fxyscale = (v4u32)__msa_fill_w(wrk->fxy_scale);
const v4u32 shiftval = (v4u32)__msa_fill_w(WEBP_RESCALER_RFIX);
const v4i32 zero = { 0 };
while (length >= 16) {
v4u32 src0, src1, src2, src3, frac0, frac1, frac2, frac3;
v16u8 out;
LD_UW4(frow, 4, src0, src1, src2, src3);
CALC_MULT_FIX1_16(src0, src1, src2, src3, y_scale, shiftval,
frac0, frac1, frac2, frac3);
LD_UW4(irow, 4, src0, src1, src2, src3);
SUB4(src0, frac0, src1, frac1, src2, frac2, src3, frac3,
src0, src1, src2, src3);
CALC_MULT_FIX_16(src0, src1, src2, src3, fxyscale, shiftval, out);
ST_UB(out, dst);
ST_UW4(frac0, frac1, frac2, frac3, irow, 4);
frow += 16;
irow += 16;
dst += 16;
length -= 16;
}
if (length > 0) {
int x_out;
if (length >= 12) {
uint32_t val0_m, val1_m, val2_m;
v4u32 src0, src1, src2, frac0, frac1, frac2;
LD_UW3(frow, 4, src0, src1, src2);
CALC_MULT_FIX1_4(src0, y_scale, shiftval, frac0);
CALC_MULT_FIX1_4(src1, y_scale, shiftval, frac1);
CALC_MULT_FIX1_4(src2, y_scale, shiftval, frac2);
LD_UW3(irow, 4, src0, src1, src2);
SUB3(src0, frac0, src1, frac1, src2, frac2, src0, src1, src2);
CALC_MULT_FIX_4(src0, fxyscale, shiftval, val0_m);
CALC_MULT_FIX_4(src1, fxyscale, shiftval, val1_m);
CALC_MULT_FIX_4(src2, fxyscale, shiftval, val2_m);
SW3(val0_m, val1_m, val2_m, dst, 4);
ST_UW3(frac0, frac1, frac2, irow, 4);
frow += 12;
irow += 12;
dst += 12;
length -= 12;
} else if (length >= 8) {
uint32_t val0_m, val1_m;
v4u32 src0, src1, frac0, frac1;
LD_UW2(frow, 4, src0, src1);
CALC_MULT_FIX1_4(src0, y_scale, shiftval, frac0);
CALC_MULT_FIX1_4(src1, y_scale, shiftval, frac1);
LD_UW2(irow, 4, src0, src1);
SUB2(src0, frac0, src1, frac1, src0, src1);
CALC_MULT_FIX_4(src0, fxyscale, shiftval, val0_m);
CALC_MULT_FIX_4(src1, fxyscale, shiftval, val1_m);
SW2(val0_m, val1_m, dst, 4);
ST_UW2(frac0, frac1, irow, 4);
frow += 8;
irow += 8;
dst += 8;
length -= 8;
} else if (length >= 4) {
uint32_t val0_m;
v4u32 frac0;
v4u32 src0 = LD_UW(frow);
CALC_MULT_FIX1_4(src0, y_scale, shiftval, frac0);
src0 = LD_UW(irow);
src0 = src0 - frac0;
CALC_MULT_FIX_4(src0, fxyscale, shiftval, val0_m);
SW(val0_m, dst);
ST_UW(frac0, irow);
frow += 4;
irow += 4;
dst += 4;
length -= 4;
}
for (x_out = 0; x_out < length; ++x_out) {
const uint32_t frac = (uint32_t)MULT_FIX_FLOOR(frow[x_out], yscale);
const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
irow[x_out] = frac;
}
}
}
static WEBP_INLINE void ExportRowShrink_1(uint32_t* irow, uint8_t* dst,
int length,
WebPRescaler* const wrk) {
const v4u32 scale = (v4u32)__msa_fill_w(wrk->fxy_scale);
const v4u32 shift = (v4u32)__msa_fill_w(WEBP_RESCALER_RFIX);
const v4i32 zero = { 0 };
while (length >= 16) {
v4u32 src0, src1, src2, src3;
v16u8 dst0;
LD_UW4(irow, 4, src0, src1, src2, src3);
CALC_MULT_FIX_16(src0, src1, src2, src3, scale, shift, dst0);
ST_UB(dst0, dst);
ST_SW4(zero, zero, zero, zero, irow, 4);
length -= 16;
irow += 16;
dst += 16;
}
if (length > 0) {
int x_out;
if (length >= 12) {
uint32_t val0_m, val1_m, val2_m;
v4u32 src0, src1, src2;
LD_UW3(irow, 4, src0, src1, src2);
CALC_MULT_FIX_4(src0, scale, shift, val0_m);
CALC_MULT_FIX_4(src1, scale, shift, val1_m);
CALC_MULT_FIX_4(src2, scale, shift, val2_m);
SW3(val0_m, val1_m, val2_m, dst, 4);
ST_SW3(zero, zero, zero, irow, 4);
length -= 12;
irow += 12;
dst += 12;
} else if (length >= 8) {
uint32_t val0_m, val1_m;
v4u32 src0, src1;
LD_UW2(irow, 4, src0, src1);
CALC_MULT_FIX_4(src0, scale, shift, val0_m);
CALC_MULT_FIX_4(src1, scale, shift, val1_m);
SW2(val0_m, val1_m, dst, 4);
ST_SW2(zero, zero, irow, 4);
length -= 8;
irow += 8;
dst += 8;
} else if (length >= 4) {
uint32_t val0_m;
const v4u32 src0 = LD_UW(irow + 0);
CALC_MULT_FIX_4(src0, scale, shift, val0_m);
SW(val0_m, dst);
ST_SW(zero, irow);
length -= 4;
irow += 4;
dst += 4;
}
for (x_out = 0; x_out < length; ++x_out) {
const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
dst[x_out] = (v > 255) ? 255u : (uint8_t)v;
irow[x_out] = 0;
}
}
}
static void RescalerExportRowShrink_MIPSdspR2(WebPRescaler* const wrk) {
uint8_t* dst = wrk->dst;
rescaler_t* irow = wrk->irow;
const int x_out_max = wrk->dst_width * wrk->num_channels;
const rescaler_t* frow = wrk->frow;
const uint32_t yscale = wrk->fy_scale * (-wrk->y_accum);
assert(!WebPRescalerOutputDone(wrk));
assert(wrk->y_accum <= 0);
assert(!wrk->y_expand);
if (yscale) {
ExportRowShrink_0(frow, irow, dst, x_out_max, yscale, wrk);
} else {
ExportRowShrink_1(irow, dst, x_out_max, wrk);
}
}
#endif // 0
//------------------------------------------------------------------------------
// Entry point
extern void WebPRescalerDspInitMSA(void);
WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMSA(void) {
WebPRescalerExportRowExpand = RescalerExportRowExpand_MIPSdspR2;
// WebPRescalerExportRowShrink = RescalerExportRowShrink_MIPSdspR2;
}
#else // !WEBP_USE_MSA
WEBP_DSP_INIT_STUB(WebPRescalerDspInitMSA)
#endif // WEBP_USE_MSA
|