SkBlitter_RGB16.cpp revision c2050e3a3ecfb8738b36e2add15c526e8e0f21fe
1
2/*
3 * Copyright 2006 The Android Open Source Project
4 *
5 * Use of this source code is governed by a BSD-style license that can be
6 * found in the LICENSE file.
7 */
8
9
10#include "SkBlitRow.h"
11#include "SkCoreBlitters.h"
12#include "SkColorPriv.h"
13#include "SkDither.h"
14#include "SkShader.h"
15#include "SkTemplatesPriv.h"
16#include "SkUtils.h"
17#include "SkXfermode.h"
18
19#if defined(__ARM_HAVE_NEON) && defined(SK_CPU_LENDIAN)
20    #define SK_USE_NEON
21    #include <arm_neon.h>
22#else
23    // if we don't have neon, then our black blitter is worth the extra code
24    #define USE_BLACK_BLITTER
25#endif
26
27void sk_dither_memset16(uint16_t dst[], uint16_t value, uint16_t other,
28                        int count) {
29    if (count > 0) {
30        // see if we need to write one short before we can cast to an 4byte ptr
31        // (we do this subtract rather than (unsigned)dst so we don't get warnings
32        //  on 64bit machines)
33        if (((char*)dst - (char*)0) & 2) {
34            *dst++ = value;
35            count -= 1;
36            SkTSwap(value, other);
37        }
38
39        // fast way to set [value,other] pairs
40#ifdef SK_CPU_BENDIAN
41        sk_memset32((uint32_t*)dst, (value << 16) | other, count >> 1);
42#else
43        sk_memset32((uint32_t*)dst, (other << 16) | value, count >> 1);
44#endif
45
46        if (count & 1) {
47            dst[count - 1] = value;
48        }
49    }
50}
51
52///////////////////////////////////////////////////////////////////////////////
53
54class SkRGB16_Blitter : public SkRasterBlitter {
55public:
56    SkRGB16_Blitter(const SkBitmap& device, const SkPaint& paint);
57    virtual void blitH(int x, int y, int width);
58    virtual void blitAntiH(int x, int y, const SkAlpha* antialias,
59                           const int16_t* runs);
60    virtual void blitV(int x, int y, int height, SkAlpha alpha);
61    virtual void blitRect(int x, int y, int width, int height);
62    virtual void blitMask(const SkMask&,
63                          const SkIRect&);
64    virtual const SkBitmap* justAnOpaqueColor(uint32_t*);
65
66protected:
67    SkPMColor   fSrcColor32;
68    uint32_t    fExpandedRaw16;
69    unsigned    fScale;
70    uint16_t    fColor16;       // already scaled by fScale
71    uint16_t    fRawColor16;    // unscaled
72    uint16_t    fRawDither16;   // unscaled
73    SkBool8     fDoDither;
74
75    // illegal
76    SkRGB16_Blitter& operator=(const SkRGB16_Blitter&);
77
78    typedef SkRasterBlitter INHERITED;
79};
80
81class SkRGB16_Opaque_Blitter : public SkRGB16_Blitter {
82public:
83    SkRGB16_Opaque_Blitter(const SkBitmap& device, const SkPaint& paint);
84    virtual void blitH(int x, int y, int width);
85    virtual void blitAntiH(int x, int y, const SkAlpha* antialias,
86                           const int16_t* runs);
87    virtual void blitV(int x, int y, int height, SkAlpha alpha);
88    virtual void blitRect(int x, int y, int width, int height);
89    virtual void blitMask(const SkMask&,
90                          const SkIRect&);
91
92private:
93    typedef SkRGB16_Blitter INHERITED;
94};
95
96#ifdef USE_BLACK_BLITTER
97class SkRGB16_Black_Blitter : public SkRGB16_Opaque_Blitter {
98public:
99    SkRGB16_Black_Blitter(const SkBitmap& device, const SkPaint& paint);
100    virtual void blitMask(const SkMask&, const SkIRect&);
101    virtual void blitAntiH(int x, int y, const SkAlpha* antialias,
102                           const int16_t* runs);
103
104private:
105    typedef SkRGB16_Opaque_Blitter INHERITED;
106};
107#endif
108
109class SkRGB16_Shader_Blitter : public SkShaderBlitter {
110public:
111    SkRGB16_Shader_Blitter(const SkBitmap& device, const SkPaint& paint);
112    virtual ~SkRGB16_Shader_Blitter();
113    virtual void blitH(int x, int y, int width);
114    virtual void blitAntiH(int x, int y, const SkAlpha* antialias,
115                           const int16_t* runs);
116    virtual void blitRect(int x, int y, int width, int height);
117
118protected:
119    SkPMColor*      fBuffer;
120    SkBlitRow::Proc fOpaqueProc;
121    SkBlitRow::Proc fAlphaProc;
122
123private:
124    // illegal
125    SkRGB16_Shader_Blitter& operator=(const SkRGB16_Shader_Blitter&);
126
127    typedef SkShaderBlitter INHERITED;
128};
129
130// used only if the shader can perform shadSpan16
131class SkRGB16_Shader16_Blitter : public SkRGB16_Shader_Blitter {
132public:
133    SkRGB16_Shader16_Blitter(const SkBitmap& device, const SkPaint& paint);
134    virtual void blitH(int x, int y, int width);
135    virtual void blitAntiH(int x, int y, const SkAlpha* antialias,
136                           const int16_t* runs);
137    virtual void blitRect(int x, int y, int width, int height);
138
139private:
140    typedef SkRGB16_Shader_Blitter INHERITED;
141};
142
143class SkRGB16_Shader_Xfermode_Blitter : public SkShaderBlitter {
144public:
145    SkRGB16_Shader_Xfermode_Blitter(const SkBitmap& device, const SkPaint& paint);
146    virtual ~SkRGB16_Shader_Xfermode_Blitter();
147    virtual void blitH(int x, int y, int width);
148    virtual void blitAntiH(int x, int y, const SkAlpha* antialias,
149                           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& mask,
206                                     const SkIRect& clip) {
207    if (mask.fFormat == SkMask::kBW_Format) {
208        SkRGB16_Black_BlitBW(fDevice, mask, clip);
209    } else {
210        uint16_t* SK_RESTRICT device = fDevice.getAddr16(clip.fLeft, clip.fTop);
211        const uint8_t* SK_RESTRICT alpha = mask.getAddr8(clip.fLeft, clip.fTop);
212        unsigned width = clip.width();
213        unsigned height = clip.height();
214        size_t deviceRB = fDevice.rowBytes() - (width << 1);
215        unsigned maskRB = mask.fRowBytes - width;
216
217        SkASSERT((int)height > 0);
218        SkASSERT((int)width > 0);
219        SkASSERT((int)deviceRB >= 0);
220        SkASSERT((int)maskRB >= 0);
221
222        do {
223            unsigned w = width;
224            do {
225                unsigned aa = *alpha++;
226                *device = SkAlphaMulRGB16(*device, SkAlpha255To256(255 - aa));
227                device += 1;
228            } while (--w != 0);
229            device = (uint16_t*)((char*)device + deviceRB);
230            alpha += maskRB;
231        } while (--height != 0);
232    }
233}
234
235void SkRGB16_Black_Blitter::blitAntiH(int x, int y,
236                                      const SkAlpha* SK_RESTRICT antialias,
237                                      const int16_t* SK_RESTRICT runs) {
238    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
239
240    for (;;) {
241        int count = runs[0];
242        SkASSERT(count >= 0);
243        if (count <= 0) {
244            return;
245        }
246        runs += count;
247
248        unsigned aa = antialias[0];
249        antialias += count;
250        if (aa) {
251            if (aa == 255) {
252                memset(device, 0, count << 1);
253            } else {
254                aa = SkAlpha255To256(255 - aa);
255                do {
256                    *device = SkAlphaMulRGB16(*device, aa);
257                    device += 1;
258                } while (--count != 0);
259                continue;
260            }
261        }
262        device += count;
263    }
264}
265#endif
266
267///////////////////////////////////////////////////////////////////////////////
268///////////////////////////////////////////////////////////////////////////////
269
270SkRGB16_Opaque_Blitter::SkRGB16_Opaque_Blitter(const SkBitmap& device,
271                                               const SkPaint& paint)
272: INHERITED(device, paint) {}
273
274void SkRGB16_Opaque_Blitter::blitH(int x, int y, int width) {
275    SkASSERT(width > 0);
276    SkASSERT(x + width <= fDevice.width());
277    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
278    uint16_t srcColor = fColor16;
279
280    SkASSERT(fRawColor16 == srcColor);
281    if (fDoDither) {
282        uint16_t ditherColor = fRawDither16;
283        if ((x ^ y) & 1) {
284            SkTSwap(ditherColor, srcColor);
285        }
286        sk_dither_memset16(device, srcColor, ditherColor, width);
287    } else {
288        sk_memset16(device, srcColor, width);
289    }
290}
291
292// return 1 or 0 from a bool
293static inline int Bool2Int(int value) {
294    return !!value;
295}
296
297void SkRGB16_Opaque_Blitter::blitAntiH(int x, int y,
298                                       const SkAlpha* SK_RESTRICT antialias,
299                                       const int16_t* SK_RESTRICT runs) {
300    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
301    uint16_t    srcColor = fRawColor16;
302    uint32_t    srcExpanded = fExpandedRaw16;
303    int         ditherInt = Bool2Int(fDoDither);
304    uint16_t    ditherColor = fRawDither16;
305    // if we have no dithering, this will always fail
306    if ((x ^ y) & ditherInt) {
307        SkTSwap(ditherColor, srcColor);
308    }
309    for (;;) {
310        int count = runs[0];
311        SkASSERT(count >= 0);
312        if (count <= 0) {
313            return;
314        }
315        runs += count;
316
317        unsigned aa = antialias[0];
318        antialias += count;
319        if (aa) {
320            if (aa == 255) {
321                if (ditherInt) {
322                    sk_dither_memset16(device, srcColor,
323                                       ditherColor, count);
324                } else {
325                    sk_memset16(device, srcColor, count);
326                }
327            } else {
328                // TODO: respect fDoDither
329                unsigned scale5 = SkAlpha255To256(aa) >> 3;
330                uint32_t src32 = srcExpanded * scale5;
331                scale5 = 32 - scale5; // now we can use it on the device
332                int n = count;
333                do {
334                    uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
335                    *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
336                } while (--n != 0);
337                goto DONE;
338            }
339        }
340        device += count;
341
342    DONE:
343        // if we have no dithering, this will always fail
344        if (count & ditherInt) {
345            SkTSwap(ditherColor, srcColor);
346        }
347    }
348}
349
350#define solid_8_pixels(mask, dst, color)    \
351    do {                                    \
352        if (mask & 0x80) dst[0] = color;    \
353        if (mask & 0x40) dst[1] = color;    \
354        if (mask & 0x20) dst[2] = color;    \
355        if (mask & 0x10) dst[3] = color;    \
356        if (mask & 0x08) dst[4] = color;    \
357        if (mask & 0x04) dst[5] = color;    \
358        if (mask & 0x02) dst[6] = color;    \
359        if (mask & 0x01) dst[7] = color;    \
360    } while (0)
361
362#define SK_BLITBWMASK_NAME                  SkRGB16_BlitBW
363#define SK_BLITBWMASK_ARGS                  , uint16_t color
364#define SK_BLITBWMASK_BLIT8(mask, dst)      solid_8_pixels(mask, dst, color)
365#define SK_BLITBWMASK_GETADDR               getAddr16
366#define SK_BLITBWMASK_DEVTYPE               uint16_t
367#include "SkBlitBWMaskTemplate.h"
368
369static U16CPU blend_compact(uint32_t src32, uint32_t dst32, unsigned scale5) {
370    return SkCompact_rgb_16(dst32 + ((src32 - dst32) * scale5 >> 5));
371}
372
373void SkRGB16_Opaque_Blitter::blitMask(const SkMask& mask,
374                                      const SkIRect& clip) {
375    if (mask.fFormat == SkMask::kBW_Format) {
376        SkRGB16_BlitBW(fDevice, mask, clip, fColor16);
377        return;
378    }
379
380    uint16_t* SK_RESTRICT device = fDevice.getAddr16(clip.fLeft, clip.fTop);
381    const uint8_t* SK_RESTRICT alpha = mask.getAddr8(clip.fLeft, clip.fTop);
382    int width = clip.width();
383    int height = clip.height();
384    size_t      deviceRB = fDevice.rowBytes() - (width << 1);
385    unsigned    maskRB = mask.fRowBytes - width;
386    uint32_t    expanded32 = fExpandedRaw16;
387
388#ifdef SK_USE_NEON
389#define    UNROLL    8
390    do {
391        int w = width;
392        if (w >= UNROLL) {
393            uint32x4_t color;        /* can use same one */
394            uint32x4_t dev_lo, dev_hi;
395            uint32x4_t t1;
396            uint32x4_t wn1, wn2;
397            uint16x4_t odev_lo, odev_hi;
398            uint16x4_t alpha_lo, alpha_hi;
399            uint16x8_t  alpha_full;
400
401            color = vdupq_n_u32(expanded32);
402
403            do {
404                /* alpha is 8x8, widen and split to get pair of 16x4's */
405                alpha_full = vmovl_u8(vld1_u8(alpha));
406                alpha_full = vaddq_u16(alpha_full, vshrq_n_u16(alpha_full,7));
407                alpha_full = vshrq_n_u16(alpha_full, 3);
408                alpha_lo = vget_low_u16(alpha_full);
409                alpha_hi = vget_high_u16(alpha_full);
410
411                dev_lo = vmovl_u16(vld1_u16(device));
412                dev_hi = vmovl_u16(vld1_u16(device+4));
413
414                /* unpack in 32 bits */
415                dev_lo = vorrq_u32(
416                                   vandq_u32(dev_lo, vdupq_n_u32(0x0000F81F)),
417                                   vshlq_n_u32(vandq_u32(dev_lo,
418                                                         vdupq_n_u32(0x000007E0)),
419                                               16)
420                                   );
421                dev_hi = vorrq_u32(
422                                   vandq_u32(dev_hi, vdupq_n_u32(0x0000F81F)),
423                                   vshlq_n_u32(vandq_u32(dev_hi,
424                                                         vdupq_n_u32(0x000007E0)),
425                                               16)
426                                   );
427
428                /* blend the two */
429                t1 = vmulq_u32(vsubq_u32(color, dev_lo), vmovl_u16(alpha_lo));
430                t1 = vshrq_n_u32(t1, 5);
431                dev_lo = vaddq_u32(dev_lo, t1);
432
433                t1 = vmulq_u32(vsubq_u32(color, dev_hi), vmovl_u16(alpha_hi));
434                t1 = vshrq_n_u32(t1, 5);
435                dev_hi = vaddq_u32(dev_hi, t1);
436
437                /* re-compact and store */
438                wn1 = vandq_u32(dev_lo, vdupq_n_u32(0x0000F81F)),
439                wn2 = vshrq_n_u32(dev_lo, 16);
440                wn2 = vandq_u32(wn2, vdupq_n_u32(0x000007E0));
441                odev_lo = vmovn_u32(vorrq_u32(wn1, wn2));
442
443                wn1 = vandq_u32(dev_hi, vdupq_n_u32(0x0000F81F)),
444                wn2 = vshrq_n_u32(dev_hi, 16);
445                wn2 = vandq_u32(wn2, vdupq_n_u32(0x000007E0));
446                odev_hi = vmovn_u32(vorrq_u32(wn1, wn2));
447
448                vst1_u16(device, odev_lo);
449                vst1_u16(device+4, odev_hi);
450
451                device += UNROLL;
452                alpha += UNROLL;
453                w -= UNROLL;
454            } while (w >= UNROLL);
455        }
456
457        /* residuals (which is everything if we have no neon) */
458        while (w > 0) {
459            *device = blend_compact(expanded32, SkExpand_rgb_16(*device),
460                                    SkAlpha255To256(*alpha++) >> 3);
461            device += 1;
462            --w;
463        }
464        device = (uint16_t*)((char*)device + deviceRB);
465        alpha += maskRB;
466    } while (--height != 0);
467#undef    UNROLL
468#else   // non-neon code
469    do {
470        int w = width;
471        do {
472            *device = blend_compact(expanded32, SkExpand_rgb_16(*device),
473                                    SkAlpha255To256(*alpha++) >> 3);
474            device += 1;
475        } while (--w != 0);
476        device = (uint16_t*)((char*)device + deviceRB);
477        alpha += maskRB;
478    } while (--height != 0);
479#endif
480}
481
482void SkRGB16_Opaque_Blitter::blitV(int x, int y, int height, SkAlpha alpha) {
483    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
484    size_t    deviceRB = fDevice.rowBytes();
485
486    // TODO: respect fDoDither
487    unsigned scale5 = SkAlpha255To256(alpha) >> 3;
488    uint32_t src32 =  fExpandedRaw16 * scale5;
489    scale5 = 32 - scale5;
490    do {
491        uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
492        *device = SkCompact_rgb_16((src32 + dst32) >> 5);
493        device = (uint16_t*)((char*)device + deviceRB);
494    } while (--height != 0);
495}
496
497void SkRGB16_Opaque_Blitter::blitRect(int x, int y, int width, int height) {
498    SkASSERT(x + width <= fDevice.width() && y + height <= fDevice.height());
499    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
500    size_t      deviceRB = fDevice.rowBytes();
501    uint16_t    color16 = fColor16;
502
503    if (fDoDither) {
504        uint16_t ditherColor = fRawDither16;
505        if ((x ^ y) & 1) {
506            SkTSwap(ditherColor, color16);
507        }
508        while (--height >= 0) {
509            sk_dither_memset16(device, color16, ditherColor, width);
510            SkTSwap(ditherColor, color16);
511            device = (uint16_t*)((char*)device + deviceRB);
512        }
513    } else {  // no dither
514        while (--height >= 0) {
515            sk_memset16(device, color16, width);
516            device = (uint16_t*)((char*)device + deviceRB);
517        }
518    }
519}
520
521///////////////////////////////////////////////////////////////////////////////
522
523SkRGB16_Blitter::SkRGB16_Blitter(const SkBitmap& device, const SkPaint& paint)
524    : INHERITED(device) {
525    SkColor color = paint.getColor();
526
527    fSrcColor32 = SkPreMultiplyColor(color);
528    fScale = SkAlpha255To256(SkColorGetA(color));
529
530    int r = SkColorGetR(color);
531    int g = SkColorGetG(color);
532    int b = SkColorGetB(color);
533
534    fRawColor16 = fRawDither16 = SkPack888ToRGB16(r, g, b);
535    // if we're dithered, use fRawDither16 to hold that.
536    if ((fDoDither = paint.isDither()) != false) {
537        fRawDither16 = SkDitherPack888ToRGB16(r, g, b);
538    }
539
540    fExpandedRaw16 = SkExpand_rgb_16(fRawColor16);
541
542    fColor16 = SkPackRGB16( SkAlphaMul(r, fScale) >> (8 - SK_R16_BITS),
543                            SkAlphaMul(g, fScale) >> (8 - SK_G16_BITS),
544                            SkAlphaMul(b, fScale) >> (8 - SK_B16_BITS));
545}
546
547const SkBitmap* SkRGB16_Blitter::justAnOpaqueColor(uint32_t* value) {
548    if (!fDoDither && 256 == fScale) {
549        *value = fRawColor16;
550        return &fDevice;
551    }
552    return NULL;
553}
554
555static uint32_t pmcolor_to_expand16(SkPMColor c) {
556    unsigned r = SkGetPackedR32(c);
557    unsigned g = SkGetPackedG32(c);
558    unsigned b = SkGetPackedB32(c);
559    return (g << 24) | (r << 13) | (b << 2);
560}
561
562static inline void blend32_16_row(SkPMColor src, uint16_t dst[], int count) {
563    SkASSERT(count > 0);
564    uint32_t src_expand = pmcolor_to_expand16(src);
565    unsigned scale = SkAlpha255To256(0xFF - SkGetPackedA32(src)) >> 3;
566    do {
567        uint32_t dst_expand = SkExpand_rgb_16(*dst) * scale;
568        *dst = SkCompact_rgb_16((src_expand + dst_expand) >> 5);
569        dst += 1;
570    } while (--count != 0);
571}
572
573void SkRGB16_Blitter::blitH(int x, int y, int width) {
574    SkASSERT(width > 0);
575    SkASSERT(x + width <= fDevice.width());
576    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
577
578    // TODO: respect fDoDither
579    blend32_16_row(fSrcColor32, device, width);
580}
581
582void SkRGB16_Blitter::blitAntiH(int x, int y,
583                                const SkAlpha* SK_RESTRICT antialias,
584                                const int16_t* SK_RESTRICT runs) {
585    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
586    uint32_t    srcExpanded = fExpandedRaw16;
587    unsigned    scale = fScale;
588
589    // TODO: respect fDoDither
590    for (;;) {
591        int count = runs[0];
592        SkASSERT(count >= 0);
593        if (count <= 0) {
594            return;
595        }
596        runs += count;
597
598        unsigned aa = antialias[0];
599        antialias += count;
600        if (aa) {
601            unsigned scale5 = SkAlpha255To256(aa) * scale >> (8 + 3);
602            uint32_t src32 =  srcExpanded * scale5;
603            scale5 = 32 - scale5;
604            do {
605                uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
606                *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
607            } while (--count != 0);
608            continue;
609        }
610        device += count;
611    }
612}
613
614static inline void blend_8_pixels(U8CPU bw, uint16_t dst[], unsigned dst_scale,
615                                  U16CPU srcColor) {
616    if (bw & 0x80) dst[0] = srcColor + SkAlphaMulRGB16(dst[0], dst_scale);
617    if (bw & 0x40) dst[1] = srcColor + SkAlphaMulRGB16(dst[1], dst_scale);
618    if (bw & 0x20) dst[2] = srcColor + SkAlphaMulRGB16(dst[2], dst_scale);
619    if (bw & 0x10) dst[3] = srcColor + SkAlphaMulRGB16(dst[3], dst_scale);
620    if (bw & 0x08) dst[4] = srcColor + SkAlphaMulRGB16(dst[4], dst_scale);
621    if (bw & 0x04) dst[5] = srcColor + SkAlphaMulRGB16(dst[5], dst_scale);
622    if (bw & 0x02) dst[6] = srcColor + SkAlphaMulRGB16(dst[6], dst_scale);
623    if (bw & 0x01) dst[7] = srcColor + SkAlphaMulRGB16(dst[7], dst_scale);
624}
625
626#define SK_BLITBWMASK_NAME                  SkRGB16_BlendBW
627#define SK_BLITBWMASK_ARGS                  , unsigned dst_scale, U16CPU src_color
628#define SK_BLITBWMASK_BLIT8(mask, dst)      blend_8_pixels(mask, dst, dst_scale, src_color)
629#define SK_BLITBWMASK_GETADDR               getAddr16
630#define SK_BLITBWMASK_DEVTYPE               uint16_t
631#include "SkBlitBWMaskTemplate.h"
632
633void SkRGB16_Blitter::blitMask(const SkMask& mask,
634                               const SkIRect& clip) {
635    if (mask.fFormat == SkMask::kBW_Format) {
636        SkRGB16_BlendBW(fDevice, mask, clip, 256 - fScale, fColor16);
637        return;
638    }
639
640    uint16_t* SK_RESTRICT device = fDevice.getAddr16(clip.fLeft, clip.fTop);
641    const uint8_t* SK_RESTRICT alpha = mask.getAddr8(clip.fLeft, clip.fTop);
642    int width = clip.width();
643    int height = clip.height();
644    size_t      deviceRB = fDevice.rowBytes() - (width << 1);
645    unsigned    maskRB = mask.fRowBytes - width;
646    uint32_t    color32 = fExpandedRaw16;
647
648    unsigned scale256 = fScale;
649    do {
650        int w = width;
651        do {
652            unsigned aa = *alpha++;
653            unsigned scale = SkAlpha255To256(aa) * scale256 >> (8 + 3);
654            uint32_t src32 = color32 * scale;
655            uint32_t dst32 = SkExpand_rgb_16(*device) * (32 - scale);
656            *device++ = SkCompact_rgb_16((src32 + dst32) >> 5);
657        } while (--w != 0);
658        device = (uint16_t*)((char*)device + deviceRB);
659        alpha += maskRB;
660    } while (--height != 0);
661}
662
663void SkRGB16_Blitter::blitV(int x, int y, int height, SkAlpha alpha) {
664    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
665    size_t    deviceRB = fDevice.rowBytes();
666
667    // TODO: respect fDoDither
668    unsigned scale5 = SkAlpha255To256(alpha) * fScale >> (8 + 3);
669    uint32_t src32 =  fExpandedRaw16 * scale5;
670    scale5 = 32 - scale5;
671    do {
672        uint32_t dst32 = SkExpand_rgb_16(*device) * scale5;
673        *device = SkCompact_rgb_16((src32 + dst32) >> 5);
674        device = (uint16_t*)((char*)device + deviceRB);
675    } while (--height != 0);
676}
677
678void SkRGB16_Blitter::blitRect(int x, int y, int width, int height) {
679    SkASSERT(x + width <= fDevice.width() && y + height <= fDevice.height());
680    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
681    size_t    deviceRB = fDevice.rowBytes();
682    SkPMColor src32 = fSrcColor32;
683
684    while (--height >= 0) {
685        blend32_16_row(src32, device, width);
686        device = (uint16_t*)((char*)device + deviceRB);
687    }
688}
689
690///////////////////////////////////////////////////////////////////////////////
691
692SkRGB16_Shader16_Blitter::SkRGB16_Shader16_Blitter(const SkBitmap& device,
693                                                   const SkPaint& paint)
694    : SkRGB16_Shader_Blitter(device, paint) {
695    SkASSERT(SkShader::CanCallShadeSpan16(fShaderFlags));
696}
697
698void SkRGB16_Shader16_Blitter::blitH(int x, int y, int width) {
699    SkASSERT(x + width <= fDevice.width());
700
701    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
702    SkShader*   shader = fShader;
703
704    int alpha = shader->getSpan16Alpha();
705    if (0xFF == alpha) {
706        shader->shadeSpan16(x, y, device, width);
707    } else {
708        uint16_t* span16 = (uint16_t*)fBuffer;
709        shader->shadeSpan16(x, y, span16, width);
710        SkBlendRGB16(span16, device, SkAlpha255To256(alpha), width);
711    }
712}
713
714void SkRGB16_Shader16_Blitter::blitRect(int x, int y, int width, int height) {
715    SkShader*   shader = fShader;
716    uint16_t*   dst = fDevice.getAddr16(x, y);
717    size_t      dstRB = fDevice.rowBytes();
718    int         alpha = shader->getSpan16Alpha();
719
720    if (0xFF == alpha) {
721        if (fShaderFlags & SkShader::kConstInY16_Flag) {
722            // have the shader blit directly into the device the first time
723            shader->shadeSpan16(x, y, dst, width);
724            // and now just memcpy that line on the subsequent lines
725            if (--height > 0) {
726                const uint16_t* orig = dst;
727                do {
728                    dst = (uint16_t*)((char*)dst + dstRB);
729                    memcpy(dst, orig, width << 1);
730                } while (--height);
731            }
732        } else {    // need to call shadeSpan16 for every line
733            do {
734                shader->shadeSpan16(x, y, dst, width);
735                y += 1;
736                dst = (uint16_t*)((char*)dst + dstRB);
737            } while (--height);
738        }
739    } else {
740        int scale = SkAlpha255To256(alpha);
741        uint16_t* span16 = (uint16_t*)fBuffer;
742        if (fShaderFlags & SkShader::kConstInY16_Flag) {
743            shader->shadeSpan16(x, y, span16, width);
744            do {
745                SkBlendRGB16(span16, dst, scale, width);
746                dst = (uint16_t*)((char*)dst + dstRB);
747            } while (--height);
748        } else {
749            do {
750                shader->shadeSpan16(x, y, span16, width);
751                SkBlendRGB16(span16, dst, scale, width);
752                y += 1;
753                dst = (uint16_t*)((char*)dst + dstRB);
754            } while (--height);
755        }
756    }
757}
758
759void SkRGB16_Shader16_Blitter::blitAntiH(int x, int y,
760                                         const SkAlpha* SK_RESTRICT antialias,
761                                         const int16_t* SK_RESTRICT runs) {
762    SkShader*   shader = fShader;
763    SkPMColor* SK_RESTRICT span = fBuffer;
764    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
765
766    int alpha = shader->getSpan16Alpha();
767    uint16_t* span16 = (uint16_t*)span;
768
769    if (0xFF == alpha) {
770        for (;;) {
771            int count = *runs;
772            if (count <= 0) {
773                break;
774            }
775            SkASSERT(count <= fDevice.width()); // don't overrun fBuffer
776
777            int aa = *antialias;
778            if (aa == 255) {
779                // go direct to the device!
780                shader->shadeSpan16(x, y, device, count);
781            } else if (aa) {
782                shader->shadeSpan16(x, y, span16, count);
783                SkBlendRGB16(span16, device, SkAlpha255To256(aa), count);
784            }
785            device += count;
786            runs += count;
787            antialias += count;
788            x += count;
789        }
790    } else {  // span alpha is < 255
791        alpha = SkAlpha255To256(alpha);
792        for (;;) {
793            int count = *runs;
794            if (count <= 0) {
795                break;
796            }
797            SkASSERT(count <= fDevice.width()); // don't overrun fBuffer
798
799            int aa = SkAlphaMul(*antialias, alpha);
800            if (aa) {
801                shader->shadeSpan16(x, y, span16, count);
802                SkBlendRGB16(span16, device, SkAlpha255To256(aa), count);
803            }
804
805            device += count;
806            runs += count;
807            antialias += count;
808            x += count;
809        }
810    }
811}
812
813///////////////////////////////////////////////////////////////////////////////
814
815SkRGB16_Shader_Blitter::SkRGB16_Shader_Blitter(const SkBitmap& device,
816                                               const SkPaint& paint)
817: INHERITED(device, paint) {
818    SkASSERT(paint.getXfermode() == NULL);
819
820    fBuffer = (SkPMColor*)sk_malloc_throw(device.width() * sizeof(SkPMColor));
821
822    // compute SkBlitRow::Procs
823    unsigned flags = 0;
824
825    uint32_t shaderFlags = fShaderFlags;
826    // shaders take care of global alpha, so we never set it in SkBlitRow
827    if (!(shaderFlags & SkShader::kOpaqueAlpha_Flag)) {
828        flags |= SkBlitRow::kSrcPixelAlpha_Flag;
829    }
830    // don't dither if the shader is really 16bit
831    if (paint.isDither() && !(shaderFlags & SkShader::kIntrinsicly16_Flag)) {
832        flags |= SkBlitRow::kDither_Flag;
833    }
834    // used when we know our global alpha is 0xFF
835    fOpaqueProc = SkBlitRow::Factory(flags, SkBitmap::kRGB_565_Config);
836    // used when we know our global alpha is < 0xFF
837    fAlphaProc  = SkBlitRow::Factory(flags | SkBlitRow::kGlobalAlpha_Flag,
838                                     SkBitmap::kRGB_565_Config);
839}
840
841SkRGB16_Shader_Blitter::~SkRGB16_Shader_Blitter() {
842    sk_free(fBuffer);
843}
844
845void SkRGB16_Shader_Blitter::blitH(int x, int y, int width) {
846    SkASSERT(x + width <= fDevice.width());
847
848    fShader->shadeSpan(x, y, fBuffer, width);
849    // shaders take care of global alpha, so we pass 0xFF (should be ignored)
850    fOpaqueProc(fDevice.getAddr16(x, y), fBuffer, width, 0xFF, x, y);
851}
852
853void SkRGB16_Shader_Blitter::blitRect(int x, int y, int width, int height) {
854    SkShader*       shader = fShader;
855    SkBlitRow::Proc proc = fOpaqueProc;
856    SkPMColor*      buffer = fBuffer;
857    uint16_t*       dst = fDevice.getAddr16(x, y);
858    size_t          dstRB = fDevice.rowBytes();
859
860    if (fShaderFlags & SkShader::kConstInY32_Flag) {
861        shader->shadeSpan(x, y, buffer, width);
862        do {
863            proc(dst, buffer, width, 0xFF, x, y);
864            y += 1;
865            dst = (uint16_t*)((char*)dst + dstRB);
866        } while (--height);
867    } else {
868        do {
869            shader->shadeSpan(x, y, buffer, width);
870            proc(dst, buffer, width, 0xFF, x, y);
871            y += 1;
872            dst = (uint16_t*)((char*)dst + dstRB);
873        } while (--height);
874    }
875}
876
877static inline int count_nonzero_span(const int16_t runs[], const SkAlpha aa[]) {
878    int count = 0;
879    for (;;) {
880        int n = *runs;
881        if (n == 0 || *aa == 0) {
882            break;
883        }
884        runs += n;
885        aa += n;
886        count += n;
887    }
888    return count;
889}
890
891void SkRGB16_Shader_Blitter::blitAntiH(int x, int y,
892                                       const SkAlpha* SK_RESTRICT antialias,
893                                       const int16_t* SK_RESTRICT runs) {
894    SkShader*   shader = fShader;
895    SkPMColor* SK_RESTRICT span = fBuffer;
896    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
897
898    for (;;) {
899        int count = *runs;
900        if (count <= 0) {
901            break;
902        }
903        int aa = *antialias;
904        if (0 == aa) {
905            device += count;
906            runs += count;
907            antialias += count;
908            x += count;
909            continue;
910        }
911
912        int nonZeroCount = count + count_nonzero_span(runs + count, antialias + count);
913
914        SkASSERT(nonZeroCount <= fDevice.width()); // don't overrun fBuffer
915        shader->shadeSpan(x, y, span, nonZeroCount);
916
917        SkPMColor* localSpan = span;
918        for (;;) {
919            SkBlitRow::Proc proc = (aa == 0xFF) ? fOpaqueProc : fAlphaProc;
920            proc(device, localSpan, count, aa, x, y);
921
922            x += count;
923            device += count;
924            runs += count;
925            antialias += count;
926            nonZeroCount -= count;
927            if (nonZeroCount == 0) {
928                break;
929            }
930            localSpan += count;
931            SkASSERT(nonZeroCount > 0);
932            count = *runs;
933            SkASSERT(count > 0);
934            aa = *antialias;
935        }
936    }
937}
938
939///////////////////////////////////////////////////////////////////////
940
941SkRGB16_Shader_Xfermode_Blitter::SkRGB16_Shader_Xfermode_Blitter(
942                                const SkBitmap& device, const SkPaint& paint)
943: INHERITED(device, paint) {
944    fXfermode = paint.getXfermode();
945    SkASSERT(fXfermode);
946    fXfermode->ref();
947
948    int width = device.width();
949    fBuffer = (SkPMColor*)sk_malloc_throw((width + (SkAlign4(width) >> 2)) * sizeof(SkPMColor));
950    fAAExpand = (uint8_t*)(fBuffer + width);
951}
952
953SkRGB16_Shader_Xfermode_Blitter::~SkRGB16_Shader_Xfermode_Blitter() {
954    fXfermode->unref();
955    sk_free(fBuffer);
956}
957
958void SkRGB16_Shader_Xfermode_Blitter::blitH(int x, int y, int width) {
959    SkASSERT(x + width <= fDevice.width());
960
961    uint16_t*   device = fDevice.getAddr16(x, y);
962    SkPMColor*  span = fBuffer;
963
964    fShader->shadeSpan(x, y, span, width);
965    fXfermode->xfer16(device, span, width, NULL);
966}
967
968void SkRGB16_Shader_Xfermode_Blitter::blitAntiH(int x, int y,
969                                const SkAlpha* SK_RESTRICT antialias,
970                                const int16_t* SK_RESTRICT runs) {
971    SkShader*   shader = fShader;
972    SkXfermode* mode = fXfermode;
973    SkPMColor* SK_RESTRICT span = fBuffer;
974    uint8_t* SK_RESTRICT aaExpand = fAAExpand;
975    uint16_t* SK_RESTRICT device = fDevice.getAddr16(x, y);
976
977    for (;;) {
978        int count = *runs;
979        if (count <= 0) {
980            break;
981        }
982        int aa = *antialias;
983        if (0 == aa) {
984            device += count;
985            runs += count;
986            antialias += count;
987            x += count;
988            continue;
989        }
990
991        int nonZeroCount = count + count_nonzero_span(runs + count,
992                                                      antialias + count);
993
994        SkASSERT(nonZeroCount <= fDevice.width()); // don't overrun fBuffer
995        shader->shadeSpan(x, y, span, nonZeroCount);
996
997        x += nonZeroCount;
998        SkPMColor* localSpan = span;
999        for (;;) {
1000            if (aa == 0xFF) {
1001                mode->xfer16(device, localSpan, count, NULL);
1002            } else {
1003                SkASSERT(aa);
1004                memset(aaExpand, aa, count);
1005                mode->xfer16(device, localSpan, count, aaExpand);
1006            }
1007            device += count;
1008            runs += count;
1009            antialias += count;
1010            nonZeroCount -= count;
1011            if (nonZeroCount == 0) {
1012                break;
1013            }
1014            localSpan += count;
1015            SkASSERT(nonZeroCount > 0);
1016            count = *runs;
1017            SkASSERT(count > 0);
1018            aa = *antialias;
1019        }
1020    }
1021}
1022
1023///////////////////////////////////////////////////////////////////////////////
1024
1025SkBlitter* SkBlitter_ChooseD565(const SkBitmap& device, const SkPaint& paint,
1026                                void* storage, size_t storageSize) {
1027    SkBlitter* blitter;
1028    SkShader* shader = paint.getShader();
1029    SkXfermode* mode = paint.getXfermode();
1030
1031    // we require a shader if there is an xfermode, handled by our caller
1032    SkASSERT(NULL == mode || NULL != shader);
1033
1034    if (shader) {
1035        if (mode) {
1036            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Shader_Xfermode_Blitter,
1037                                  storage, storageSize, (device, paint));
1038        } else if (shader->canCallShadeSpan16()) {
1039            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Shader16_Blitter,
1040                                  storage, storageSize, (device, paint));
1041        } else {
1042            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Shader_Blitter,
1043                                  storage, storageSize, (device, paint));
1044        }
1045    } else {
1046        // no shader, no xfermode, (and we always ignore colorfilter)
1047        SkColor color = paint.getColor();
1048        if (0 == SkColorGetA(color)) {
1049            SK_PLACEMENT_NEW(blitter, SkNullBlitter, storage, storageSize);
1050#ifdef USE_BLACK_BLITTER
1051        } else if (SK_ColorBLACK == color) {
1052            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Black_Blitter, storage,
1053                                  storageSize, (device, paint));
1054#endif
1055        } else if (0xFF == SkColorGetA(color)) {
1056            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Opaque_Blitter, storage,
1057                                  storageSize, (device, paint));
1058        } else {
1059            SK_PLACEMENT_NEW_ARGS(blitter, SkRGB16_Blitter, storage,
1060                                  storageSize, (device, paint));
1061        }
1062    }
1063
1064    return blitter;
1065}
1066