1/* libs/graphics/sgl/SkBlitter_RGB16.cpp
2**
3** Copyright 2006, The Android Open Source Project
4**
5** Licensed under the Apache License, Version 2.0 (the "License");
6** you may not use this file except in compliance with the License.
7** You may obtain a copy of the License at
8**
9**     http://www.apache.org/licenses/LICENSE-2.0
10**
11** Unless required by applicable law or agreed to in writing, software
12** distributed under the License is distributed on an "AS IS" BASIS,
13** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14** See the License for the specific language governing permissions and
15** limitations under the License.
16*/
17
18#include "SkBlitRow.h"
19#include "SkCoreBlitters.h"
20#include "SkColorPriv.h"
21#include "SkDither.h"
22#include "SkShader.h"
23#include "SkTemplatesPriv.h"
24#include "SkUtils.h"
25#include "SkXfermode.h"
26
27#if defined(__ARM_HAVE_NEON) && defined(SK_CPU_LENDIAN)
28    #define SK_USE_NEON
29    #include <arm_neon.h>
30#else
31    // if we don't have neon, then our black blitter is worth the extra code
32    #define USE_BLACK_BLITTER
33#endif
34
35void sk_dither_memset16(uint16_t dst[], uint16_t value, uint16_t other,
36                        int count) {
37    if (count > 0) {
38        // see if we need to write one short before we can cast to an 4byte ptr
39        // (we do this subtract rather than (unsigned)dst so we don't get warnings
40        //  on 64bit machines)
41        if (((char*)dst - (char*)0) & 2) {
42            *dst++ = value;
43            count -= 1;
44            SkTSwap(value, other);
45        }
46
47        // fast way to set [value,other] pairs
48#ifdef SK_CPU_BENDIAN
49        sk_memset32((uint32_t*)dst, (value << 16) | other, count >> 1);
50#else
51        sk_memset32((uint32_t*)dst, (other << 16) | value, count >> 1);
52#endif
53
54        if (count & 1) {
55            dst[count - 1] = value;
56        }
57    }
58}
59
60///////////////////////////////////////////////////////////////////////////////
61
62class SkRGB16_Blitter : public SkRasterBlitter {
63public:
64    SkRGB16_Blitter(const SkBitmap& device, const SkPaint& paint);
65    virtual void blitH(int x, int y, int width);
66    virtual void blitAntiH(int x, int y, const SkAlpha antialias[], const int16_t runs[]);
67    virtual void blitV(int x, int y, int height, SkAlpha alpha);
68    virtual void blitRect(int x, int y, int width, int height);
69    virtual void blitMask(const SkMask&, const SkIRect&);
70    virtual const SkBitmap* justAnOpaqueColor(uint32_t*);
71
72protected:
73    SkPMColor   fSrcColor32;
74    uint32_t    fExpandedRaw16;
75    unsigned    fScale;
76    uint16_t    fColor16;       // already scaled by fScale
77    uint16_t    fRawColor16;    // unscaled
78    uint16_t    fRawDither16;   // unscaled
79    SkBool8     fDoDither;
80
81    // illegal
82    SkRGB16_Blitter& operator=(const SkRGB16_Blitter&);
83
84    typedef SkRasterBlitter INHERITED;
85};
86
87class SkRGB16_Opaque_Blitter : public SkRGB16_Blitter {
88public:
89    SkRGB16_Opaque_Blitter(const SkBitmap& device, const SkPaint& paint);
90    virtual void blitH(int x, int y, int width);
91    virtual void blitAntiH(int x, int y, const SkAlpha antialias[], const int16_t runs[]);
92    virtual void blitV(int x, int y, int height, SkAlpha alpha);
93    virtual void blitRect(int x, int y, int width, int height);
94    virtual void blitMask(const SkMask&, const SkIRect&);
95
96private:
97    typedef SkRGB16_Blitter INHERITED;
98};
99
100#ifdef USE_BLACK_BLITTER
101class SkRGB16_Black_Blitter : public SkRGB16_Opaque_Blitter {
102public:
103    SkRGB16_Black_Blitter(const SkBitmap& device, const SkPaint& paint);
104    virtual void blitMask(const SkMask&, const SkIRect&);
105    virtual void blitAntiH(int x, int y, const SkAlpha antialias[], const int16_t runs[]);
106
107private:
108    typedef SkRGB16_Opaque_Blitter INHERITED;
109};
110#endif
111
112class SkRGB16_Shader_Blitter : public SkShaderBlitter {
113public:
114    SkRGB16_Shader_Blitter(const SkBitmap& device, const SkPaint& paint);
115    virtual ~SkRGB16_Shader_Blitter();
116    virtual void blitH(int x, int y, int width);
117    virtual void blitAntiH(int x, int y, const SkAlpha antialias[], const int16_t runs[]);
118    virtual void blitRect(int x, int y, int width, int height);
119
120protected:
121    SkPMColor*      fBuffer;
122    SkBlitRow::Proc fOpaqueProc;
123    SkBlitRow::Proc fAlphaProc;
124
125private:
126    // illegal
127    SkRGB16_Shader_Blitter& operator=(const SkRGB16_Shader_Blitter&);
128
129    typedef SkShaderBlitter INHERITED;
130};
131
132// used only if the shader can perform shadSpan16
133class SkRGB16_Shader16_Blitter : public SkRGB16_Shader_Blitter {
134public:
135    SkRGB16_Shader16_Blitter(const SkBitmap& device, const SkPaint& paint);
136    virtual void blitH(int x, int y, int width);
137    virtual void blitAntiH(int x, int y, const SkAlpha antialias[], const int16_t runs[]);
138    virtual void blitRect(int x, int y, int width, int height);
139
140private:
141    typedef SkRGB16_Shader_Blitter INHERITED;
142};
143
144class SkRGB16_Shader_Xfermode_Blitter : public SkShaderBlitter {
145public:
146    SkRGB16_Shader_Xfermode_Blitter(const SkBitmap& device, const SkPaint& paint);
147    virtual ~SkRGB16_Shader_Xfermode_Blitter();
148    virtual void blitH(int x, int y, int width);
149    virtual void blitAntiH(int x, int y, const SkAlpha antialias[], const int16_t runs[]);
150
151private:
152    SkXfermode* fXfermode;
153    SkPMColor*  fBuffer;
154    uint8_t*    fAAExpand;
155
156    // illegal
157    SkRGB16_Shader_Xfermode_Blitter& operator=(const SkRGB16_Shader_Xfermode_Blitter&);
158
159    typedef SkShaderBlitter INHERITED;
160};
161
162///////////////////////////////////////////////////////////////////////////////
163#ifdef USE_BLACK_BLITTER
164SkRGB16_Black_Blitter::SkRGB16_Black_Blitter(const SkBitmap& device, const SkPaint& paint)
165    : INHERITED(device, paint) {
166    SkASSERT(paint.getShader() == NULL);
167    SkASSERT(paint.getColorFilter() == NULL);
168    SkASSERT(paint.getXfermode() == NULL);
169    SkASSERT(paint.getColor() == SK_ColorBLACK);
170}
171
172#if 1
173#define black_8_pixels(mask, dst)       \
174    do {                                \
175        if (mask & 0x80) dst[0] = 0;    \
176        if (mask & 0x40) dst[1] = 0;    \
177        if (mask & 0x20) dst[2] = 0;    \
178        if (mask & 0x10) dst[3] = 0;    \
179        if (mask & 0x08) dst[4] = 0;    \
180        if (mask & 0x04) dst[5] = 0;    \
181        if (mask & 0x02) dst[6] = 0;    \
182        if (mask & 0x01) dst[7] = 0;    \
183    } while (0)
184#else
185static inline black_8_pixels(U8CPU mask, uint16_t dst[])
186{
187    if (mask & 0x80) dst[0] = 0;
188    if (mask & 0x40) dst[1] = 0;
189    if (mask & 0x20) dst[2] = 0;
190    if (mask & 0x10) dst[3] = 0;
191    if (mask & 0x08) dst[4] = 0;
192    if (mask & 0x04) dst[5] = 0;
193    if (mask & 0x02) dst[6] = 0;
194    if (mask & 0x01) dst[7] = 0;
195}
196#endif
197
198#define SK_BLITBWMASK_NAME                  SkRGB16_Black_BlitBW
199#define SK_BLITBWMASK_ARGS
200#define SK_BLITBWMASK_BLIT8(mask, dst)      black_8_pixels(mask, dst)
201#define SK_BLITBWMASK_GETADDR               getAddr16
202#define SK_BLITBWMASK_DEVTYPE               uint16_t
203#include "SkBlitBWMaskTemplate.h"
204
205void SkRGB16_Black_Blitter::blitMask(const SkMask& SK_RESTRICT mask,
206                                     const SkIRect& SK_RESTRICT clip)
207                                     SK_RESTRICT {
208    if (mask.fFormat == SkMask::kBW_Format) {
209        SkRGB16_Black_BlitBW(fDevice, mask, clip);
210    } else {
211        uint16_t* SK_RESTRICT device = fDevice.getAddr16(clip.fLeft, clip.fTop);
212        const uint8_t* SK_RESTRICT alpha = mask.getAddr(clip.fLeft, clip.fTop);
213        unsigned width = clip.width();
214        unsigned height = clip.height();
215        unsigned deviceRB = fDevice.rowBytes() - (width << 1);
216        unsigned maskRB = mask.fRowBytes - width;
217
218        SkASSERT((int)height > 0);
219        SkASSERT((int)width > 0);
220        SkASSERT((int)deviceRB >= 0);
221        SkASSERT((int)maskRB >= 0);
222
223        do {
224            unsigned w = width;
225            do {
226                unsigned aa = *alpha++;
227                *device = SkAlphaMulRGB16(*device, SkAlpha255To256(255 - aa));
228                device += 1;
229            } while (--w != 0);
230            device = (uint16_t*)((char*)device + deviceRB);
231            alpha += maskRB;
232        } while (--height != 0);
233    }
234}
235
236void SkRGB16_Black_Blitter::blitAntiH(int x, int y,
237                                      const SkAlpha* SK_RESTRICT antialias,
238                                      const int16_t* SK_RESTRICT runs)
239                                      SK_RESTRICT {
240    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
241
242    for (;;) {
243        int count = runs[0];
244        SkASSERT(count >= 0);
245        if (count <= 0) {
246            return;
247        }
248        runs += count;
249
250        unsigned aa = antialias[0];
251        antialias += count;
252        if (aa) {
253            if (aa == 255) {
254                memset(device, 0, count << 1);
255            } else {
256                aa = SkAlpha255To256(255 - aa);
257                do {
258                    *device = SkAlphaMulRGB16(*device, aa);
259                    device += 1;
260                } while (--count != 0);
261                continue;
262            }
263        }
264        device += count;
265    }
266}
267#endif
268
269///////////////////////////////////////////////////////////////////////////////
270///////////////////////////////////////////////////////////////////////////////
271
272SkRGB16_Opaque_Blitter::SkRGB16_Opaque_Blitter(const SkBitmap& device,
273                                               const SkPaint& paint)
274: INHERITED(device, paint) {}
275
276void SkRGB16_Opaque_Blitter::blitH(int x, int y, int width) SK_RESTRICT {
277    SkASSERT(width > 0);
278    SkASSERT(x + width <= fDevice.width());
279    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
280    uint16_t srcColor = fColor16;
281
282    SkASSERT(fRawColor16 == srcColor);
283    if (fDoDither) {
284        uint16_t ditherColor = fRawDither16;
285        if ((x ^ y) & 1) {
286            SkTSwap(ditherColor, srcColor);
287        }
288        sk_dither_memset16(device, srcColor, ditherColor, width);
289    } else {
290        sk_memset16(device, srcColor, width);
291    }
292}
293
294// return 1 or 0 from a bool
295static inline int Bool2Int(int value) {
296	return !!value;
297}
298
299void SkRGB16_Opaque_Blitter::blitAntiH(int x, int y,
300                                       const SkAlpha* SK_RESTRICT antialias,
301                                       const int16_t* SK_RESTRICT runs) SK_RESTRICT {
302    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
303    uint16_t    srcColor = fRawColor16;
304    uint32_t    srcExpanded = fExpandedRaw16;
305    int         ditherInt = Bool2Int(fDoDither);
306    uint16_t    ditherColor = fRawDither16;
307    // if we have no dithering, this will always fail
308    if ((x ^ y) & ditherInt) {
309        SkTSwap(ditherColor, srcColor);
310    }
311    for (;;) {
312        int count = runs[0];
313        SkASSERT(count >= 0);
314        if (count <= 0) {
315            return;
316        }
317        runs += count;
318
319        unsigned aa = antialias[0];
320        antialias += count;
321        if (aa) {
322            if (aa == 255) {
323                if (ditherInt) {
324                    sk_dither_memset16(device, srcColor,
325                                       ditherColor, count);
326                } else {
327                    sk_memset16(device, srcColor, count);
328                }
329            } else {
330                // TODO: respect fDoDither
331                unsigned scale5 = SkAlpha255To256(aa) >> 3;
332                uint32_t src32 = srcExpanded * scale5;
333                scale5 = 32 - scale5; // now we can use it on the device
334                int n = count;
335                do {
336                    uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
337                    *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
338                } while (--n != 0);
339                goto DONE;
340            }
341        }
342        device += count;
343
344    DONE:
345        // if we have no dithering, this will always fail
346        if (count & ditherInt) {
347            SkTSwap(ditherColor, srcColor);
348        }
349    }
350}
351
352#define solid_8_pixels(mask, dst, color)    \
353    do {                                    \
354        if (mask & 0x80) dst[0] = color;    \
355        if (mask & 0x40) dst[1] = color;    \
356        if (mask & 0x20) dst[2] = color;    \
357        if (mask & 0x10) dst[3] = color;    \
358        if (mask & 0x08) dst[4] = color;    \
359        if (mask & 0x04) dst[5] = color;    \
360        if (mask & 0x02) dst[6] = color;    \
361        if (mask & 0x01) dst[7] = color;    \
362    } while (0)
363
364#define SK_BLITBWMASK_NAME                  SkRGB16_BlitBW
365#define SK_BLITBWMASK_ARGS                  , uint16_t color
366#define SK_BLITBWMASK_BLIT8(mask, dst)      solid_8_pixels(mask, dst, color)
367#define SK_BLITBWMASK_GETADDR               getAddr16
368#define SK_BLITBWMASK_DEVTYPE               uint16_t
369#include "SkBlitBWMaskTemplate.h"
370
371static U16CPU blend_compact(uint32_t src32, uint32_t dst32, unsigned scale5) {
372    return SkCompact_rgb_16(dst32 + ((src32 - dst32) * scale5 >> 5));
373}
374
375void SkRGB16_Opaque_Blitter::blitMask(const SkMask& SK_RESTRICT mask,
376                                      const SkIRect& SK_RESTRICT clip) SK_RESTRICT {
377    if (mask.fFormat == SkMask::kBW_Format) {
378        SkRGB16_BlitBW(fDevice, mask, clip, fColor16);
379        return;
380    }
381
382    uint16_t* SK_RESTRICT device = fDevice.getAddr16(clip.fLeft, clip.fTop);
383    const uint8_t* SK_RESTRICT alpha = mask.getAddr(clip.fLeft, clip.fTop);
384    int width = clip.width();
385    int height = clip.height();
386    unsigned    deviceRB = fDevice.rowBytes() - (width << 1);
387    unsigned    maskRB = mask.fRowBytes - width;
388    uint32_t    expanded32 = fExpandedRaw16;
389
390#ifdef SK_USE_NEON
391#define	UNROLL	8
392    do {
393        int w = width;
394        if (w >= UNROLL) {
395            uint32x4_t color;		/* can use same one */
396            uint32x4_t dev_lo, dev_hi;
397            uint32x4_t t1, t2;
398            uint32x4_t wn1, wn2;
399            uint16x4_t odev_lo, odev_hi;
400            uint16x4_t alpha_lo, alpha_hi;
401            uint16x8_t  alpha_full;
402
403            color = vdupq_n_u32(expanded32);
404
405            do {
406                /* alpha is 8x8, widen and split to get pair of 16x4's */
407                alpha_full = vmovl_u8(vld1_u8(alpha));
408                alpha_full = vaddq_u16(alpha_full, vshrq_n_u16(alpha_full,7));
409                alpha_full = vshrq_n_u16(alpha_full, 3);
410                alpha_lo = vget_low_u16(alpha_full);
411                alpha_hi = vget_high_u16(alpha_full);
412
413                dev_lo = vmovl_u16(vld1_u16(device));
414                dev_hi = vmovl_u16(vld1_u16(device+4));
415
416                /* unpack in 32 bits */
417                dev_lo = vorrq_u32(
418                                   vandq_u32(dev_lo, vdupq_n_u32(0x0000F81F)),
419                                   vshlq_n_u32(vandq_u32(dev_lo,
420                                                         vdupq_n_u32(0x000007E0)),
421                                               16)
422                                   );
423                dev_hi = vorrq_u32(
424                                   vandq_u32(dev_hi, vdupq_n_u32(0x0000F81F)),
425                                   vshlq_n_u32(vandq_u32(dev_hi,
426                                                         vdupq_n_u32(0x000007E0)),
427                                               16)
428                                   );
429
430                /* blend the two */
431                t1 = vmulq_u32(vsubq_u32(color, dev_lo), vmovl_u16(alpha_lo));
432                t1 = vshrq_n_u32(t1, 5);
433                dev_lo = vaddq_u32(dev_lo, t1);
434
435                t1 = vmulq_u32(vsubq_u32(color, dev_hi), vmovl_u16(alpha_hi));
436                t1 = vshrq_n_u32(t1, 5);
437                dev_hi = vaddq_u32(dev_hi, t1);
438
439                /* re-compact and store */
440                wn1 = vandq_u32(dev_lo, vdupq_n_u32(0x0000F81F)),
441                wn2 = vshrq_n_u32(dev_lo, 16);
442                wn2 = vandq_u32(wn2, vdupq_n_u32(0x000007E0));
443                odev_lo = vmovn_u32(vorrq_u32(wn1, wn2));
444
445                wn1 = vandq_u32(dev_hi, vdupq_n_u32(0x0000F81F)),
446                wn2 = vshrq_n_u32(dev_hi, 16);
447                wn2 = vandq_u32(wn2, vdupq_n_u32(0x000007E0));
448                odev_hi = vmovn_u32(vorrq_u32(wn1, wn2));
449
450                vst1_u16(device, odev_lo);
451                vst1_u16(device+4, odev_hi);
452
453                device += UNROLL;
454                alpha += UNROLL;
455                w -= UNROLL;
456            } while (w >= UNROLL);
457        }
458
459        /* residuals (which is everything if we have no neon) */
460        while (w > 0) {
461            *device = blend_compact(expanded32, SkExpand_rgb_16(*device),
462                                    SkAlpha255To256(*alpha++) >> 3);
463            device += 1;
464            --w;
465        }
466        device = (uint16_t*)((char*)device + deviceRB);
467        alpha += maskRB;
468    } while (--height != 0);
469#undef	UNROLL
470#else   // non-neon code
471    do {
472        int w = width;
473        do {
474            *device = blend_compact(expanded32, SkExpand_rgb_16(*device),
475                                    SkAlpha255To256(*alpha++) >> 3);
476            device += 1;
477        } while (--w != 0);
478        device = (uint16_t*)((char*)device + deviceRB);
479        alpha += maskRB;
480    } while (--height != 0);
481#endif
482}
483
484void SkRGB16_Opaque_Blitter::blitV(int x, int y, int height, SkAlpha alpha) {
485    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
486    unsigned    deviceRB = fDevice.rowBytes();
487
488    // TODO: respect fDoDither
489    unsigned scale5 = SkAlpha255To256(alpha) >> 3;
490    uint32_t src32 =  fExpandedRaw16 * scale5;
491    scale5 = 32 - scale5;
492    do {
493        uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
494        *device = SkCompact_rgb_16((src32 + dst32) >> 5);
495        device = (uint16_t*)((char*)device + deviceRB);
496    } while (--height != 0);
497}
498
499void SkRGB16_Opaque_Blitter::blitRect(int x, int y, int width, int height) {
500    SkASSERT(x + width <= fDevice.width() && y + height <= fDevice.height());
501    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
502    unsigned    deviceRB = fDevice.rowBytes();
503    uint16_t    color16 = fColor16;
504
505    if (fDoDither) {
506        uint16_t ditherColor = fRawDither16;
507        if ((x ^ y) & 1) {
508            SkTSwap(ditherColor, color16);
509        }
510        while (--height >= 0) {
511            sk_dither_memset16(device, color16, ditherColor, width);
512            SkTSwap(ditherColor, color16);
513            device = (uint16_t*)((char*)device + deviceRB);
514        }
515    } else {  // no dither
516        while (--height >= 0) {
517            sk_memset16(device, color16, width);
518            device = (uint16_t*)((char*)device + deviceRB);
519        }
520    }
521}
522
523///////////////////////////////////////////////////////////////////////////////
524
525SkRGB16_Blitter::SkRGB16_Blitter(const SkBitmap& device, const SkPaint& paint)
526    : INHERITED(device) {
527    SkColor color = paint.getColor();
528
529    fSrcColor32 = SkPreMultiplyColor(color);
530    fScale = SkAlpha255To256(SkColorGetA(color));
531
532    int r = SkColorGetR(color);
533    int g = SkColorGetG(color);
534    int b = SkColorGetB(color);
535
536    fRawColor16 = fRawDither16 = SkPack888ToRGB16(r, g, b);
537    // if we're dithered, use fRawDither16 to hold that.
538    if ((fDoDither = paint.isDither()) != false) {
539        fRawDither16 = SkDitherPack888ToRGB16(r, g, b);
540    }
541
542    fExpandedRaw16 = SkExpand_rgb_16(fRawColor16);
543
544    fColor16 = SkPackRGB16( SkAlphaMul(r, fScale) >> (8 - SK_R16_BITS),
545                            SkAlphaMul(g, fScale) >> (8 - SK_G16_BITS),
546                            SkAlphaMul(b, fScale) >> (8 - SK_B16_BITS));
547}
548
549const SkBitmap* SkRGB16_Blitter::justAnOpaqueColor(uint32_t* value) {
550    if (!fDoDither && 256 == fScale) {
551        *value = fRawColor16;
552        return &fDevice;
553    }
554    return NULL;
555}
556
557static uint32_t pmcolor_to_expand16(SkPMColor c) {
558    unsigned r = SkGetPackedR32(c);
559    unsigned g = SkGetPackedG32(c);
560    unsigned b = SkGetPackedB32(c);
561    return (g << 24) | (r << 13) | (b << 2);
562}
563
564static inline void blend32_16_row(SkPMColor src, uint16_t dst[], int count) {
565    SkASSERT(count > 0);
566    uint32_t src_expand = pmcolor_to_expand16(src);
567    unsigned scale = SkAlpha255To256(0xFF - SkGetPackedA32(src)) >> 3;
568    do {
569        uint32_t dst_expand = SkExpand_rgb_16(*dst) * scale;
570        *dst = SkCompact_rgb_16((src_expand + dst_expand) >> 5);
571        dst += 1;
572    } while (--count != 0);
573}
574
575void SkRGB16_Blitter::blitH(int x, int y, int width) SK_RESTRICT {
576    SkASSERT(width > 0);
577    SkASSERT(x + width <= fDevice.width());
578    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
579
580    // TODO: respect fDoDither
581    blend32_16_row(fSrcColor32, device, width);
582}
583
584void SkRGB16_Blitter::blitAntiH(int x, int y,
585                                const SkAlpha* SK_RESTRICT antialias,
586                                const int16_t* SK_RESTRICT runs) SK_RESTRICT {
587    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
588    uint32_t    srcExpanded = fExpandedRaw16;
589    unsigned    scale = fScale;
590
591    // TODO: respect fDoDither
592    for (;;) {
593        int count = runs[0];
594        SkASSERT(count >= 0);
595        if (count <= 0) {
596            return;
597        }
598        runs += count;
599
600        unsigned aa = antialias[0];
601        antialias += count;
602        if (aa) {
603            unsigned scale5 = SkAlpha255To256(aa) * scale >> (8 + 3);
604            uint32_t src32 =  srcExpanded * scale5;
605            scale5 = 32 - scale5;
606            do {
607                uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
608                *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
609            } while (--count != 0);
610            continue;
611        }
612        device += count;
613    }
614}
615
616static inline void blend_8_pixels(U8CPU bw, uint16_t dst[], unsigned dst_scale,
617                                  U16CPU srcColor) {
618    if (bw & 0x80) dst[0] = srcColor + SkAlphaMulRGB16(dst[0], dst_scale);
619    if (bw & 0x40) dst[1] = srcColor + SkAlphaMulRGB16(dst[1], dst_scale);
620    if (bw & 0x20) dst[2] = srcColor + SkAlphaMulRGB16(dst[2], dst_scale);
621    if (bw & 0x10) dst[3] = srcColor + SkAlphaMulRGB16(dst[3], dst_scale);
622    if (bw & 0x08) dst[4] = srcColor + SkAlphaMulRGB16(dst[4], dst_scale);
623    if (bw & 0x04) dst[5] = srcColor + SkAlphaMulRGB16(dst[5], dst_scale);
624    if (bw & 0x02) dst[6] = srcColor + SkAlphaMulRGB16(dst[6], dst_scale);
625    if (bw & 0x01) dst[7] = srcColor + SkAlphaMulRGB16(dst[7], dst_scale);
626}
627
628#define SK_BLITBWMASK_NAME                  SkRGB16_BlendBW
629#define SK_BLITBWMASK_ARGS                  , unsigned dst_scale, U16CPU src_color
630#define SK_BLITBWMASK_BLIT8(mask, dst)      blend_8_pixels(mask, dst, dst_scale, src_color)
631#define SK_BLITBWMASK_GETADDR               getAddr16
632#define SK_BLITBWMASK_DEVTYPE               uint16_t
633#include "SkBlitBWMaskTemplate.h"
634
635void SkRGB16_Blitter::blitMask(const SkMask& SK_RESTRICT mask,
636                               const SkIRect& SK_RESTRICT clip) SK_RESTRICT {
637    if (mask.fFormat == SkMask::kBW_Format) {
638        SkRGB16_BlendBW(fDevice, mask, clip, 256 - fScale, fColor16);
639        return;
640    }
641
642    uint16_t* SK_RESTRICT device = fDevice.getAddr16(clip.fLeft, clip.fTop);
643    const uint8_t* SK_RESTRICT alpha = mask.getAddr(clip.fLeft, clip.fTop);
644    int width = clip.width();
645    int height = clip.height();
646    unsigned    deviceRB = fDevice.rowBytes() - (width << 1);
647    unsigned    maskRB = mask.fRowBytes - width;
648    uint32_t    color32 = fExpandedRaw16;
649
650    unsigned scale256 = fScale;
651    do {
652        int w = width;
653        do {
654            unsigned aa = *alpha++;
655            unsigned scale = SkAlpha255To256(aa) * scale256 >> (8 + 3);
656            uint32_t src32 = color32 * scale;
657            uint32_t dst32 = SkExpand_rgb_16(*device) * (32 - scale);
658            *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
659        } while (--w != 0);
660        device = (uint16_t*)((char*)device + deviceRB);
661        alpha += maskRB;
662    } while (--height != 0);
663}
664
665void SkRGB16_Blitter::blitV(int x, int y, int height, SkAlpha alpha) {
666    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
667    unsigned    deviceRB = fDevice.rowBytes();
668
669    // TODO: respect fDoDither
670    unsigned scale5 = SkAlpha255To256(alpha) * fScale >> (8 + 3);
671    uint32_t src32 =  fExpandedRaw16 * scale5;
672    scale5 = 32 - scale5;
673    do {
674        uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
675        *device = SkCompact_rgb_16((src32 + dst32) >> 5);
676        device = (uint16_t*)((char*)device + deviceRB);
677    } while (--height != 0);
678}
679
680void SkRGB16_Blitter::blitRect(int x, int y, int width, int height) {
681    SkASSERT(x + width <= fDevice.width() && y + height <= fDevice.height());
682    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
683    unsigned    deviceRB = fDevice.rowBytes();
684    SkPMColor src32 = fSrcColor32;
685
686    while (--height >= 0) {
687        blend32_16_row(src32, device, width);
688        device = (uint16_t*)((char*)device + deviceRB);
689    }
690}
691
692///////////////////////////////////////////////////////////////////////////////
693
694SkRGB16_Shader16_Blitter::SkRGB16_Shader16_Blitter(const SkBitmap& device,
695                                                   const SkPaint& paint)
696    : SkRGB16_Shader_Blitter(device, paint) {
697    SkASSERT(SkShader::CanCallShadeSpan16(fShaderFlags));
698}
699
700void SkRGB16_Shader16_Blitter::blitH(int x, int y, int width) SK_RESTRICT {
701    SkASSERT(x + width <= fDevice.width());
702
703    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
704    SkShader*   shader = fShader;
705
706    int alpha = shader->getSpan16Alpha();
707    if (0xFF == alpha) {
708        shader->shadeSpan16(x, y, device, width);
709    } else {
710        uint16_t* span16 = (uint16_t*)fBuffer;
711        shader->shadeSpan16(x, y, span16, width);
712        SkBlendRGB16(span16, device, SkAlpha255To256(alpha), width);
713    }
714}
715
716void SkRGB16_Shader16_Blitter::blitRect(int x, int y, int width, int height) {
717    SkShader*   shader = fShader;
718    uint16_t*   dst = fDevice.getAddr16(x, y);
719    size_t      dstRB = fDevice.rowBytes();
720    int         alpha = shader->getSpan16Alpha();
721
722    if (0xFF == alpha) {
723        if (fShaderFlags & SkShader::kConstInY16_Flag) {
724            // have the shader blit directly into the device the first time
725            shader->shadeSpan16(x, y, dst, width);
726            // and now just memcpy that line on the subsequent lines
727            if (--height > 0) {
728                const uint16_t* orig = dst;
729                do {
730                    dst = (uint16_t*)((char*)dst + dstRB);
731                    memcpy(dst, orig, width << 1);
732                } while (--height);
733            }
734        } else {    // need to call shadeSpan16 for every line
735            do {
736                shader->shadeSpan16(x, y, dst, width);
737                y += 1;
738                dst = (uint16_t*)((char*)dst + dstRB);
739            } while (--height);
740        }
741    } else {
742        int scale = SkAlpha255To256(alpha);
743        uint16_t* span16 = (uint16_t*)fBuffer;
744        if (fShaderFlags & SkShader::kConstInY16_Flag) {
745            shader->shadeSpan16(x, y, span16, width);
746            do {
747                SkBlendRGB16(span16, dst, scale, width);
748                dst = (uint16_t*)((char*)dst + dstRB);
749            } while (--height);
750        } else {
751            do {
752                shader->shadeSpan16(x, y, span16, width);
753                SkBlendRGB16(span16, dst, scale, width);
754                y += 1;
755                dst = (uint16_t*)((char*)dst + dstRB);
756            } while (--height);
757        }
758    }
759}
760
761void SkRGB16_Shader16_Blitter::blitAntiH(int x, int y,
762                                         const SkAlpha* SK_RESTRICT antialias,
763                                         const int16_t* SK_RESTRICT runs)
764                                         SK_RESTRICT {
765    SkShader*   shader = fShader;
766    SkPMColor* SK_RESTRICT span = fBuffer;
767    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
768
769    int alpha = shader->getSpan16Alpha();
770    uint16_t* span16 = (uint16_t*)span;
771
772    if (0xFF == alpha) {
773        for (;;) {
774            int count = *runs;
775            if (count <= 0) {
776                break;
777            }
778            SkASSERT(count <= fDevice.width()); // don't overrun fBuffer
779
780            int aa = *antialias;
781            if (aa == 255) {
782                // go direct to the device!
783                shader->shadeSpan16(x, y, device, count);
784            } else if (aa) {
785                shader->shadeSpan16(x, y, span16, count);
786                SkBlendRGB16(span16, device, SkAlpha255To256(aa), count);
787            }
788            device += count;
789            runs += count;
790            antialias += count;
791            x += count;
792        }
793    } else {  // span alpha is < 255
794        alpha = SkAlpha255To256(alpha);
795        for (;;) {
796            int count = *runs;
797            if (count <= 0) {
798                break;
799            }
800            SkASSERT(count <= fDevice.width()); // don't overrun fBuffer
801
802            int aa = SkAlphaMul(*antialias, alpha);
803            if (aa) {
804                shader->shadeSpan16(x, y, span16, count);
805                SkBlendRGB16(span16, device, SkAlpha255To256(aa), count);
806            }
807
808            device += count;
809            runs += count;
810            antialias += count;
811            x += count;
812        }
813    }
814}
815
816///////////////////////////////////////////////////////////////////////////////
817
818SkRGB16_Shader_Blitter::SkRGB16_Shader_Blitter(const SkBitmap& device,
819                                               const SkPaint& paint)
820: INHERITED(device, paint) {
821    SkASSERT(paint.getXfermode() == NULL);
822
823    fBuffer = (SkPMColor*)sk_malloc_throw(device.width() * sizeof(SkPMColor));
824
825    // compute SkBlitRow::Procs
826    unsigned flags = 0;
827
828    uint32_t shaderFlags = fShaderFlags;
829    // shaders take care of global alpha, so we never set it in SkBlitRow
830    if (!(shaderFlags & SkShader::kOpaqueAlpha_Flag)) {
831        flags |= SkBlitRow::kSrcPixelAlpha_Flag;
832        }
833    // don't dither if the shader is really 16bit
834    if (paint.isDither() && !(shaderFlags & SkShader::kIntrinsicly16_Flag)) {
835        flags |= SkBlitRow::kDither_Flag;
836    }
837    // used when we know our global alpha is 0xFF
838    fOpaqueProc = SkBlitRow::Factory(flags, SkBitmap::kRGB_565_Config);
839    // used when we know our global alpha is < 0xFF
840    fAlphaProc  = SkBlitRow::Factory(flags | SkBlitRow::kGlobalAlpha_Flag,
841                                     SkBitmap::kRGB_565_Config);
842}
843
844SkRGB16_Shader_Blitter::~SkRGB16_Shader_Blitter() {
845    sk_free(fBuffer);
846}
847
848void SkRGB16_Shader_Blitter::blitH(int x, int y, int width) {
849    SkASSERT(x + width <= fDevice.width());
850
851    fShader->shadeSpan(x, y, fBuffer, width);
852    // shaders take care of global alpha, so we pass 0xFF (should be ignored)
853    fOpaqueProc(fDevice.getAddr16(x, y), fBuffer, width, 0xFF, x, y);
854}
855
856void SkRGB16_Shader_Blitter::blitRect(int x, int y, int width, int height) {
857    SkShader*       shader = fShader;
858    SkBlitRow::Proc proc = fOpaqueProc;
859    SkPMColor*      buffer = fBuffer;
860    uint16_t*       dst = fDevice.getAddr16(x, y);
861    size_t          dstRB = fDevice.rowBytes();
862
863    if (fShaderFlags & SkShader::kConstInY32_Flag) {
864        shader->shadeSpan(x, y, buffer, width);
865        do {
866            proc(dst, buffer, width, 0xFF, x, y);
867            y += 1;
868            dst = (uint16_t*)((char*)dst + dstRB);
869        } while (--height);
870    } else {
871        do {
872            shader->shadeSpan(x, y, buffer, width);
873            proc(dst, buffer, width, 0xFF, x, y);
874            y += 1;
875            dst = (uint16_t*)((char*)dst + dstRB);
876        } while (--height);
877    }
878}
879
880static inline int count_nonzero_span(const int16_t runs[], const SkAlpha aa[]) {
881    int count = 0;
882    for (;;) {
883        int n = *runs;
884        if (n == 0 || *aa == 0) {
885            break;
886        }
887        runs += n;
888        aa += n;
889        count += n;
890    }
891    return count;
892}
893
894void SkRGB16_Shader_Blitter::blitAntiH(int x, int y,
895                                       const SkAlpha* SK_RESTRICT antialias,
896                                       const int16_t* SK_RESTRICT runs)
897                                       SK_RESTRICT {
898    SkShader*   shader = fShader;
899    SkPMColor* SK_RESTRICT span = fBuffer;
900    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
901
902    for (;;) {
903        int count = *runs;
904        if (count <= 0) {
905            break;
906        }
907        int aa = *antialias;
908        if (0 == aa) {
909            device += count;
910            runs += count;
911            antialias += count;
912            x += count;
913            continue;
914        }
915
916        int nonZeroCount = count + count_nonzero_span(runs + count, antialias + count);
917
918        SkASSERT(nonZeroCount <= fDevice.width()); // don't overrun fBuffer
919        shader->shadeSpan(x, y, span, nonZeroCount);
920
921        SkPMColor* localSpan = span;
922        for (;;) {
923            SkBlitRow::Proc proc = (aa == 0xFF) ? fOpaqueProc : fAlphaProc;
924            proc(device, localSpan, count, aa, x, y);
925
926            x += count;
927            device += count;
928            runs += count;
929            antialias += count;
930            nonZeroCount -= count;
931            if (nonZeroCount == 0) {
932                break;
933            }
934            localSpan += count;
935            SkASSERT(nonZeroCount > 0);
936            count = *runs;
937            SkASSERT(count > 0);
938            aa = *antialias;
939        }
940    }
941}
942
943///////////////////////////////////////////////////////////////////////
944
945SkRGB16_Shader_Xfermode_Blitter::SkRGB16_Shader_Xfermode_Blitter(
946                                const SkBitmap& device, const SkPaint& paint)
947: INHERITED(device, paint) {
948    fXfermode = paint.getXfermode();
949    SkASSERT(fXfermode);
950    fXfermode->ref();
951
952    int width = device.width();
953    fBuffer = (SkPMColor*)sk_malloc_throw((width + (SkAlign4(width) >> 2)) * sizeof(SkPMColor));
954    fAAExpand = (uint8_t*)(fBuffer + width);
955}
956
957SkRGB16_Shader_Xfermode_Blitter::~SkRGB16_Shader_Xfermode_Blitter() {
958    fXfermode->unref();
959    sk_free(fBuffer);
960}
961
962void SkRGB16_Shader_Xfermode_Blitter::blitH(int x, int y, int width) {
963    SkASSERT(x + width <= fDevice.width());
964
965    uint16_t*   device = fDevice.getAddr16(x, y);
966    SkPMColor*  span = fBuffer;
967
968    fShader->shadeSpan(x, y, span, width);
969    fXfermode->xfer16(device, span, width, NULL);
970}
971
972void SkRGB16_Shader_Xfermode_Blitter::blitAntiH(int x, int y,
973                                const SkAlpha* SK_RESTRICT antialias,
974                                const int16_t* SK_RESTRICT runs) SK_RESTRICT {
975    SkShader*   shader = fShader;
976    SkXfermode* mode = fXfermode;
977    SkPMColor* SK_RESTRICT span = fBuffer;
978    uint8_t* SK_RESTRICT aaExpand = fAAExpand;
979    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
980
981    for (;;) {
982        int count = *runs;
983        if (count <= 0) {
984            break;
985        }
986        int aa = *antialias;
987        if (0 == aa) {
988            device += count;
989            runs += count;
990            antialias += count;
991            x += count;
992            continue;
993        }
994
995        int nonZeroCount = count + count_nonzero_span(runs + count,
996                                                      antialias + count);
997
998        SkASSERT(nonZeroCount <= fDevice.width()); // don't overrun fBuffer
999        shader->shadeSpan(x, y, span, nonZeroCount);
1000
1001        x += nonZeroCount;
1002        SkPMColor* localSpan = span;
1003        for (;;) {
1004            if (aa == 0xFF) {
1005                mode->xfer16(device, localSpan, count, NULL);
1006            } else {
1007                SkASSERT(aa);
1008                memset(aaExpand, aa, count);
1009                mode->xfer16(device, localSpan, count, aaExpand);
1010            }
1011            device += count;
1012            runs += count;
1013            antialias += count;
1014            nonZeroCount -= count;
1015            if (nonZeroCount == 0) {
1016                break;
1017            }
1018            localSpan += count;
1019            SkASSERT(nonZeroCount > 0);
1020            count = *runs;
1021            SkASSERT(count > 0);
1022            aa = *antialias;
1023        }
1024    }
1025}
1026
1027///////////////////////////////////////////////////////////////////////////////
1028
1029SkBlitter* SkBlitter_ChooseD565(const SkBitmap& device, const SkPaint& paint,
1030                                void* storage, size_t storageSize) {
1031    SkBlitter* blitter;
1032    SkShader* shader = paint.getShader();
1033    SkXfermode* mode = paint.getXfermode();
1034
1035    // we require a shader if there is an xfermode, handled by our caller
1036    SkASSERT(NULL == mode || NULL != shader);
1037
1038    if (shader) {
1039        if (mode) {
1040            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Shader_Xfermode_Blitter,
1041                                  storage, storageSize, (device, paint));
1042        } else if (shader->canCallShadeSpan16()) {
1043            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Shader16_Blitter,
1044                                  storage, storageSize, (device, paint));
1045        } else {
1046            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Shader_Blitter,
1047                                  storage, storageSize, (device, paint));
1048        }
1049    } else {
1050        // no shader, no xfermode, (and we always ignore colorfilter)
1051        SkColor color = paint.getColor();
1052        if (0 == SkColorGetA(color)) {
1053            SK_PLACEMENT_NEW(blitter, SkNullBlitter, storage, storageSize);
1054#ifdef USE_BLACK_BLITTER
1055        } else if (SK_ColorBLACK == color) {
1056            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Black_Blitter, storage,
1057                                  storageSize, (device, paint));
1058#endif
1059        } else if (0xFF == SkColorGetA(color)) {
1060            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Opaque_Blitter, storage,
1061                                  storageSize, (device, paint));
1062        } else {
1063            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Blitter, storage,
1064                                  storageSize, (device, paint));
1065        }
1066    }
1067
1068    return blitter;
1069}
1070