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
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
|
/*
* Copyright 2023 Google LLC
*
* Use of this source code is governed by a BSD-style license that can be
* found in the LICENSE file.
*/
#include "src/core/SkScanPriv.h"
#include "include/core/SkMatrix.h"
#include "include/core/SkPath.h"
#include "include/core/SkRegion.h"
#include "include/private/base/SkTo.h"
#include "src/core/SkAntiRun.h"
#include "src/core/SkBlitter.h"
#if defined(SK_FORCE_AAA)
void SkScan::SAAFillPath(const SkPath&, SkBlitter*, const SkIRect&, const SkIRect&, bool) {
SkDEBUGFAIL("SAA Disabled");
}
#else
#define SHIFT SK_SUPERSAMPLE_SHIFT
#define SCALE (1 << SHIFT)
#define MASK (SCALE - 1)
/** @file
We have two techniques for capturing the output of the supersampler:
- SUPERMASK, which records a large mask-bitmap
this is often faster for small, complex objects
- RLE, which records a rle-encoded scanline
this is often faster for large objects with big spans
These blitters use two coordinate systems:
- destination coordinates, scale equal to the output - often
abbreviated with 'i' or 'I' in variable names
- supersampled coordinates, scale equal to the output * SCALE
*/
//#define FORCE_SUPERMASK
//#define FORCE_RLE
///////////////////////////////////////////////////////////////////////////////
/// Base class for a single-pass supersampled blitter.
class BaseSuperBlitter : public SkBlitter {
public:
BaseSuperBlitter(SkBlitter* realBlitter, const SkIRect& ir,
const SkIRect& clipBounds, bool isInverse);
/// Must be explicitly defined on subclasses.
void blitAntiH(int x, int y, const SkAlpha antialias[], const int16_t runs[]) override {
SkDEBUGFAIL("How did I get here?");
}
/// May not be called on BaseSuperBlitter because it blits out of order.
void blitV(int x, int y, int height, SkAlpha alpha) override {
SkDEBUGFAIL("How did I get here?");
}
protected:
SkBlitter* fRealBlitter;
/// Current y coordinate, in destination coordinates.
int fCurrIY;
/// Widest row of region to be blitted, in destination coordinates.
int fWidth;
/// Leftmost x coordinate in any row, in destination coordinates.
int fLeft;
/// Leftmost x coordinate in any row, in supersampled coordinates.
int fSuperLeft;
SkDEBUGCODE(int fCurrX;)
/// Current y coordinate in supersampled coordinates.
int fCurrY;
/// Initial y coordinate (top of bounds).
int fTop;
SkIRect fSectBounds;
};
BaseSuperBlitter::BaseSuperBlitter(SkBlitter* realBlit, const SkIRect& ir,
const SkIRect& clipBounds, bool isInverse) {
fRealBlitter = realBlit;
SkIRect sectBounds;
if (isInverse) {
// We use the clip bounds instead of the ir, since we may be asked to
//draw outside of the rect when we're a inverse filltype
sectBounds = clipBounds;
} else {
if (!sectBounds.intersect(ir, clipBounds)) {
sectBounds.setEmpty();
}
}
const int left = sectBounds.left();
const int right = sectBounds.right();
fLeft = left;
fSuperLeft = SkLeftShift(left, SHIFT);
fWidth = right - left;
fTop = sectBounds.top();
fCurrIY = fTop - 1;
fCurrY = SkLeftShift(fTop, SHIFT) - 1;
SkDEBUGCODE(fCurrX = -1;)
}
/// Run-length-encoded supersampling antialiased blitter.
class SuperBlitter : public BaseSuperBlitter {
public:
SuperBlitter(SkBlitter* realBlitter, const SkIRect& ir, const SkIRect& clipBounds,
bool isInverse);
~SuperBlitter() override {
this->flush();
}
/// Once fRuns contains a complete supersampled row, flush() blits
/// it out through the wrapped blitter.
void flush();
/// Blits a row of pixels, with location and width specified
/// in supersampled coordinates.
void blitH(int x, int y, int width) override;
/// Blits a rectangle of pixels, with location and size specified
/// in supersampled coordinates.
void blitRect(int x, int y, int width, int height) override;
private:
// The next three variables are used to track a circular buffer that
// contains the values used in SkAlphaRuns. These variables should only
// ever be updated in advanceRuns(), and fRuns should always point to
// a valid SkAlphaRuns...
int fRunsToBuffer;
void* fRunsBuffer;
int fCurrentRun;
SkAlphaRuns fRuns;
// extra one to store the zero at the end
int getRunsSz() const { return (fWidth + 1 + (fWidth + 2)/2) * sizeof(int16_t); }
// This function updates the fRuns variable to point to the next buffer space
// with adequate storage for a SkAlphaRuns. It mostly just advances fCurrentRun
// and resets fRuns to point to an empty scanline.
void advanceRuns() {
const size_t kRunsSz = this->getRunsSz();
fCurrentRun = (fCurrentRun + 1) % fRunsToBuffer;
fRuns.fRuns = reinterpret_cast<int16_t*>(
reinterpret_cast<uint8_t*>(fRunsBuffer) + fCurrentRun * kRunsSz);
fRuns.fAlpha = reinterpret_cast<SkAlpha*>(fRuns.fRuns + fWidth + 1);
fRuns.reset(fWidth);
}
int fOffsetX;
};
SuperBlitter::SuperBlitter(SkBlitter* realBlitter, const SkIRect& ir, const SkIRect& clipBounds,
bool isInverse)
: BaseSuperBlitter(realBlitter, ir, clipBounds, isInverse)
{
fRunsToBuffer = realBlitter->requestRowsPreserved();
fRunsBuffer = realBlitter->allocBlitMemory(fRunsToBuffer * this->getRunsSz());
fCurrentRun = -1;
this->advanceRuns();
fOffsetX = 0;
}
void SuperBlitter::flush() {
if (fCurrIY >= fTop) {
SkASSERT(fCurrentRun < fRunsToBuffer);
if (!fRuns.empty()) {
// SkDEBUGCODE(fRuns.dump();)
fRealBlitter->blitAntiH(fLeft, fCurrIY, fRuns.fAlpha, fRuns.fRuns);
this->advanceRuns();
fOffsetX = 0;
}
fCurrIY = fTop - 1;
SkDEBUGCODE(fCurrX = -1;)
}
}
/** coverage_to_partial_alpha() is being used by SkAlphaRuns, which
*accumulates* SCALE pixels worth of "alpha" in [0,(256/SCALE)]
to produce a final value in [0, 255] and handles clamping 256->255
itself, with the same (alpha - (alpha >> 8)) correction as
coverage_to_exact_alpha().
*/
static inline int coverage_to_partial_alpha(int aa) {
aa <<= 8 - 2*SHIFT;
return aa;
}
/** coverage_to_exact_alpha() is being used by our blitter, which wants
a final value in [0, 255].
*/
static inline int coverage_to_exact_alpha(int aa) {
int alpha = (256 >> SHIFT) * aa;
// clamp 256->255
return alpha - (alpha >> 8);
}
void SuperBlitter::blitH(int x, int y, int width) {
SkASSERT(width > 0);
int iy = y >> SHIFT;
SkASSERT(iy >= fCurrIY);
x -= fSuperLeft;
// hack, until I figure out why my cubics (I think) go beyond the bounds
if (x < 0) {
width += x;
x = 0;
}
#ifdef SK_DEBUG
SkASSERT(y != fCurrY || x >= fCurrX);
#endif
SkASSERT(y >= fCurrY);
if (fCurrY != y) {
fOffsetX = 0;
fCurrY = y;
}
if (iy != fCurrIY) { // new scanline
this->flush();
fCurrIY = iy;
}
int start = x;
int stop = x + width;
SkASSERT(start >= 0 && stop > start);
// integer-pixel-aligned ends of blit, rounded out
int fb = start & MASK;
int fe = stop & MASK;
int n = (stop >> SHIFT) - (start >> SHIFT) - 1;
if (n < 0) {
fb = fe - fb;
n = 0;
fe = 0;
} else {
if (fb == 0) {
n += 1;
} else {
fb = SCALE - fb;
}
}
fOffsetX = fRuns.add(x >> SHIFT, coverage_to_partial_alpha(fb),
n, coverage_to_partial_alpha(fe),
(1 << (8 - SHIFT)) - (((y & MASK) + 1) >> SHIFT),
fOffsetX);
#ifdef SK_DEBUG
fRuns.assertValid(y & MASK, (1 << (8 - SHIFT)));
fCurrX = x + width;
#endif
}
#if 0 // UNUSED
static void set_left_rite_runs(SkAlphaRuns& runs, int ileft, U8CPU leftA,
int n, U8CPU riteA) {
SkASSERT(leftA <= 0xFF);
SkASSERT(riteA <= 0xFF);
int16_t* run = runs.fRuns;
uint8_t* aa = runs.fAlpha;
if (ileft > 0) {
run[0] = ileft;
aa[0] = 0;
run += ileft;
aa += ileft;
}
SkASSERT(leftA < 0xFF);
if (leftA > 0) {
*run++ = 1;
*aa++ = leftA;
}
if (n > 0) {
run[0] = n;
aa[0] = 0xFF;
run += n;
aa += n;
}
SkASSERT(riteA < 0xFF);
if (riteA > 0) {
*run++ = 1;
*aa++ = riteA;
}
run[0] = 0;
}
#endif
void SuperBlitter::blitRect(int x, int y, int width, int height) {
SkASSERT(width > 0);
SkASSERT(height > 0);
// blit leading rows
while ((y & MASK)) {
this->blitH(x, y++, width);
if (--height <= 0) {
return;
}
}
SkASSERT(height > 0);
// Since this is a rect, instead of blitting supersampled rows one at a
// time and then resolving to the destination canvas, we can blit
// directly to the destintion canvas one row per SCALE supersampled rows.
int start_y = y >> SHIFT;
int stop_y = (y + height) >> SHIFT;
int count = stop_y - start_y;
if (count > 0) {
y += count << SHIFT;
height -= count << SHIFT;
// save original X for our tail blitH() loop at the bottom
int origX = x;
x -= fSuperLeft;
// hack, until I figure out why my cubics (I think) go beyond the bounds
if (x < 0) {
width += x;
x = 0;
}
// There is always a left column, a middle, and a right column.
// ileft is the destination x of the first pixel of the entire rect.
// xleft is (SCALE - # of covered supersampled pixels) in that
// destination pixel.
int ileft = x >> SHIFT;
int xleft = x & MASK;
// irite is the destination x of the last pixel of the OPAQUE section.
// xrite is the number of supersampled pixels extending beyond irite;
// xrite/SCALE should give us alpha.
int irite = (x + width) >> SHIFT;
int xrite = (x + width) & MASK;
if (!xrite) {
xrite = SCALE;
irite--;
}
// Need to call flush() to clean up pending draws before we
// even consider blitV(), since otherwise it can look nonmonotonic.
SkASSERT(start_y > fCurrIY);
this->flush();
int n = irite - ileft - 1;
if (n < 0) {
// If n < 0, we'll only have a single partially-transparent column
// of pixels to render.
xleft = xrite - xleft;
SkASSERT(xleft <= SCALE);
SkASSERT(xleft > 0);
fRealBlitter->blitV(ileft + fLeft, start_y, count,
coverage_to_exact_alpha(xleft));
} else {
// With n = 0, we have two possibly-transparent columns of pixels
// to render; with n > 0, we have opaque columns between them.
xleft = SCALE - xleft;
// Using coverage_to_exact_alpha is not consistent with blitH()
const int coverageL = coverage_to_exact_alpha(xleft);
const int coverageR = coverage_to_exact_alpha(xrite);
SkASSERT(coverageL > 0 || n > 0 || coverageR > 0);
SkASSERT((coverageL != 0) + n + (coverageR != 0) <= fWidth);
fRealBlitter->blitAntiRect(ileft + fLeft, start_y, n, count,
coverageL, coverageR);
}
// preamble for our next call to blitH()
fCurrIY = stop_y - 1;
fOffsetX = 0;
fCurrY = y - 1;
fRuns.reset(fWidth);
x = origX;
}
// catch any remaining few rows
SkASSERT(height <= MASK);
while (--height >= 0) {
this->blitH(x, y++, width);
}
}
///////////////////////////////////////////////////////////////////////////////
/// Masked supersampling antialiased blitter.
class MaskSuperBlitter : public BaseSuperBlitter {
public:
MaskSuperBlitter(SkBlitter* realBlitter, const SkIRect& ir, const SkIRect&, bool isInverse);
~MaskSuperBlitter() override {
fRealBlitter->blitMask(fMask, fClipRect);
}
void blitH(int x, int y, int width) override;
static bool CanHandleRect(const SkIRect& bounds) {
#ifdef FORCE_RLE
return false;
#endif
int width = bounds.width();
int64_t rb = SkAlign4(width);
// use 64bits to detect overflow
int64_t storage = rb * bounds.height();
return (width <= MaskSuperBlitter::kMAX_WIDTH) &&
(storage <= MaskSuperBlitter::kMAX_STORAGE);
}
private:
enum {
#ifdef FORCE_SUPERMASK
kMAX_WIDTH = 2048,
kMAX_STORAGE = 1024 * 1024 * 2
#else
kMAX_WIDTH = 32, // so we don't try to do very wide things, where the RLE blitter would be faster
kMAX_STORAGE = 1024
#endif
};
SkMask fMask;
SkIRect fClipRect;
// we add 1 because add_aa_span can write (unchanged) 1 extra byte at the end, rather than
// perform a test to see if stopAlpha != 0
uint32_t fStorage[(kMAX_STORAGE >> 2) + 1];
};
MaskSuperBlitter::MaskSuperBlitter(SkBlitter* realBlitter, const SkIRect& ir,
const SkIRect& clipBounds, bool isInverse)
: BaseSuperBlitter(realBlitter, ir, clipBounds, isInverse)
{
SkASSERT(CanHandleRect(ir));
SkASSERT(!isInverse);
fMask.fImage = (uint8_t*)fStorage;
fMask.fBounds = ir;
fMask.fRowBytes = ir.width();
fMask.fFormat = SkMask::kA8_Format;
fClipRect = ir;
if (!fClipRect.intersect(clipBounds)) {
SkASSERT(0);
fClipRect.setEmpty();
}
// For valgrind, write 1 extra byte at the end so we don't read
// uninitialized memory. See comment in add_aa_span and fStorage[].
memset(fStorage, 0, fMask.fBounds.height() * fMask.fRowBytes + 1);
}
static void add_aa_span(uint8_t* alpha, U8CPU startAlpha) {
/* I should be able to just add alpha[x] + startAlpha.
However, if the trailing edge of the previous span and the leading
edge of the current span round to the same super-sampled x value,
I might overflow to 256 with this add, hence the funny subtract.
*/
unsigned tmp = *alpha + startAlpha;
SkASSERT(tmp <= 256);
*alpha = SkToU8(tmp - (tmp >> 8));
}
static inline uint32_t quadplicate_byte(U8CPU value) {
uint32_t pair = (value << 8) | value;
return (pair << 16) | pair;
}
// Perform this tricky subtract, to avoid overflowing to 256. Our caller should
// only ever call us with at most enough to hit 256 (never larger), so it is
// enough to just subtract the high-bit. Actually clamping with a branch would
// be slower (e.g. if (tmp > 255) tmp = 255;)
//
static inline void saturated_add(uint8_t* ptr, U8CPU add) {
unsigned tmp = *ptr + add;
SkASSERT(tmp <= 256);
*ptr = SkToU8(tmp - (tmp >> 8));
}
// minimum count before we want to setup an inner loop, adding 4-at-a-time
#define MIN_COUNT_FOR_QUAD_LOOP 16
static void add_aa_span(uint8_t* alpha, U8CPU startAlpha, int middleCount,
U8CPU stopAlpha, U8CPU maxValue) {
SkASSERT(middleCount >= 0);
saturated_add(alpha, startAlpha);
alpha += 1;
if (middleCount >= MIN_COUNT_FOR_QUAD_LOOP) {
// loop until we're quad-byte aligned
while (reinterpret_cast<intptr_t>(alpha) & 0x3) {
alpha[0] = SkToU8(alpha[0] + maxValue);
alpha += 1;
middleCount -= 1;
}
int bigCount = middleCount >> 2;
uint32_t* qptr = reinterpret_cast<uint32_t*>(alpha);
uint32_t qval = quadplicate_byte(maxValue);
do {
*qptr++ += qval;
} while (--bigCount > 0);
middleCount &= 3;
alpha = reinterpret_cast<uint8_t*> (qptr);
// fall through to the following while-loop
}
while (--middleCount >= 0) {
alpha[0] = SkToU8(alpha[0] + maxValue);
alpha += 1;
}
// potentially this can be off the end of our "legal" alpha values, but that
// only happens if stopAlpha is also 0. Rather than test for stopAlpha != 0
// every time (slow), we just do it, and ensure that we've allocated extra space
// (see the + 1 comment in fStorage[]
saturated_add(alpha, stopAlpha);
}
void MaskSuperBlitter::blitH(int x, int y, int width) {
int iy = (y >> SHIFT);
SkASSERT(iy >= fMask.fBounds.fTop && iy < fMask.fBounds.fBottom);
iy -= fMask.fBounds.fTop; // make it relative to 0
// This should never happen, but it does. Until the true cause is
// discovered, let's skip this span instead of crashing.
// See http://crbug.com/17569.
if (iy < 0) {
return;
}
#ifdef SK_DEBUG
{
int ix = x >> SHIFT;
SkASSERT(ix >= fMask.fBounds.fLeft && ix < fMask.fBounds.fRight);
}
#endif
x -= SkLeftShift(fMask.fBounds.fLeft, SHIFT);
// hack, until I figure out why my cubics (I think) go beyond the bounds
if (x < 0) {
width += x;
x = 0;
}
uint8_t* row = fMask.fImage + iy * fMask.fRowBytes + (x >> SHIFT);
int start = x;
int stop = x + width;
SkASSERT(start >= 0 && stop > start);
int fb = start & MASK;
int fe = stop & MASK;
int n = (stop >> SHIFT) - (start >> SHIFT) - 1;
if (n < 0) {
SkASSERT(row >= fMask.fImage);
SkASSERT(row < fMask.fImage + kMAX_STORAGE + 1);
add_aa_span(row, coverage_to_partial_alpha(fe - fb));
} else {
fb = SCALE - fb;
SkASSERT(row >= fMask.fImage);
SkASSERT(row + n + 1 < fMask.fImage + kMAX_STORAGE + 1);
add_aa_span(row, coverage_to_partial_alpha(fb),
n, coverage_to_partial_alpha(fe),
(1 << (8 - SHIFT)) - (((y & MASK) + 1) >> SHIFT));
}
#ifdef SK_DEBUG
fCurrX = x + width;
#endif
}
///////////////////////////////////////////////////////////////////////////////
void SkScan::SAAFillPath(const SkPath& path, SkBlitter* blitter, const SkIRect& ir,
const SkIRect& clipBounds, bool forceRLE) {
bool containedInClip = clipBounds.contains(ir);
bool isInverse = path.isInverseFillType();
// MaskSuperBlitter can't handle drawing outside of ir, so we can't use it
// if we're an inverse filltype
if (!isInverse && MaskSuperBlitter::CanHandleRect(ir) && !forceRLE) {
MaskSuperBlitter superBlit(blitter, ir, clipBounds, isInverse);
SkASSERT(SkIntToScalar(ir.fTop) <= path.getBounds().fTop);
sk_fill_path(path, clipBounds, &superBlit, ir.fTop, ir.fBottom, SHIFT, containedInClip);
} else {
SuperBlitter superBlit(blitter, ir, clipBounds, isInverse);
sk_fill_path(path, clipBounds, &superBlit, ir.fTop, ir.fBottom, SHIFT, containedInClip);
}
}
#endif // defined(SK_FORCE_AAA)
|