1/* libs/pixelflinger/scanline.cpp
2**
3** Copyright 2006-2011, 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
19#define LOG_TAG "pixelflinger"
20
21#include <assert.h>
22#include <stdlib.h>
23#include <stdio.h>
24#include <string.h>
25
26#include <cutils/memory.h>
27#include <cutils/log.h>
28
29#include "buffer.h"
30#include "scanline.h"
31
32#include "codeflinger/CodeCache.h"
33#include "codeflinger/GGLAssembler.h"
34#include "codeflinger/ARMAssembler.h"
35#if defined(__mips__)
36#include "codeflinger/MIPSAssembler.h"
37#endif
38//#include "codeflinger/ARMAssemblerOptimizer.h"
39
40// ----------------------------------------------------------------------------
41
42#define ANDROID_CODEGEN_GENERIC     0   // force generic pixel pipeline
43#define ANDROID_CODEGEN_C           1   // hand-written C, fallback generic
44#define ANDROID_CODEGEN_ASM         2   // hand-written asm, fallback generic
45#define ANDROID_CODEGEN_GENERATED   3   // hand-written asm, fallback codegen
46
47#ifdef NDEBUG
48#   define ANDROID_RELEASE
49#   define ANDROID_CODEGEN      ANDROID_CODEGEN_GENERATED
50#else
51#   define ANDROID_DEBUG
52#   define ANDROID_CODEGEN      ANDROID_CODEGEN_GENERATED
53#endif
54
55#if defined(__arm__) || defined(__mips__)
56#   define ANDROID_ARM_CODEGEN  1
57#else
58#   define ANDROID_ARM_CODEGEN  0
59#endif
60
61#define DEBUG__CODEGEN_ONLY     0
62
63/* Set to 1 to dump to the log the states that need a new
64 * code-generated scanline callback, i.e. those that don't
65 * have a corresponding shortcut function.
66 */
67#define DEBUG_NEEDS  0
68
69#ifdef __mips__
70#define ASSEMBLY_SCRATCH_SIZE   4096
71#else
72#define ASSEMBLY_SCRATCH_SIZE   2048
73#endif
74
75// ----------------------------------------------------------------------------
76namespace android {
77// ----------------------------------------------------------------------------
78
79static void init_y(context_t*, int32_t);
80static void init_y_noop(context_t*, int32_t);
81static void init_y_packed(context_t*, int32_t);
82static void init_y_error(context_t*, int32_t);
83
84static void step_y__generic(context_t* c);
85static void step_y__nop(context_t*);
86static void step_y__smooth(context_t* c);
87static void step_y__tmu(context_t* c);
88static void step_y__w(context_t* c);
89
90static void scanline(context_t* c);
91static void scanline_perspective(context_t* c);
92static void scanline_perspective_single(context_t* c);
93static void scanline_t32cb16blend(context_t* c);
94static void scanline_t32cb16blend_dither(context_t* c);
95static void scanline_t32cb16blend_srca(context_t* c);
96static void scanline_t32cb16blend_clamp(context_t* c);
97static void scanline_t32cb16blend_clamp_dither(context_t* c);
98static void scanline_t32cb16blend_clamp_mod(context_t* c);
99static void scanline_x32cb16blend_clamp_mod(context_t* c);
100static void scanline_t32cb16blend_clamp_mod_dither(context_t* c);
101static void scanline_x32cb16blend_clamp_mod_dither(context_t* c);
102static void scanline_t32cb16(context_t* c);
103static void scanline_t32cb16_dither(context_t* c);
104static void scanline_t32cb16_clamp(context_t* c);
105static void scanline_t32cb16_clamp_dither(context_t* c);
106static void scanline_col32cb16blend(context_t* c);
107static void scanline_t16cb16_clamp(context_t* c);
108static void scanline_t16cb16blend_clamp_mod(context_t* c);
109static void scanline_memcpy(context_t* c);
110static void scanline_memset8(context_t* c);
111static void scanline_memset16(context_t* c);
112static void scanline_memset32(context_t* c);
113static void scanline_noop(context_t* c);
114static void scanline_set(context_t* c);
115static void scanline_clear(context_t* c);
116
117static void rect_generic(context_t* c, size_t yc);
118static void rect_memcpy(context_t* c, size_t yc);
119
120#if defined( __arm__)
121extern "C" void scanline_t32cb16blend_arm(uint16_t*, uint32_t*, size_t);
122extern "C" void scanline_t32cb16_arm(uint16_t *dst, uint32_t *src, size_t ct);
123extern "C" void scanline_col32cb16blend_neon(uint16_t *dst, uint32_t *col, size_t ct);
124extern "C" void scanline_col32cb16blend_arm(uint16_t *dst, uint32_t col, size_t ct);
125#elif defined(__mips__)
126extern "C" void scanline_t32cb16blend_mips(uint16_t*, uint32_t*, size_t);
127#endif
128
129// ----------------------------------------------------------------------------
130
131static inline uint16_t  convertAbgr8888ToRgb565(uint32_t  pix)
132{
133    return uint16_t( ((pix << 8) & 0xf800) |
134                      ((pix >> 5) & 0x07e0) |
135                      ((pix >> 19) & 0x001f) );
136}
137
138struct shortcut_t {
139    needs_filter_t  filter;
140    const char*     desc;
141    void            (*scanline)(context_t*);
142    void            (*init_y)(context_t*, int32_t);
143};
144
145// Keep in sync with needs
146
147/* To understand the values here, have a look at:
148 *     system/core/include/private/pixelflinger/ggl_context.h
149 *
150 * Especially the lines defining and using GGL_RESERVE_NEEDS
151 *
152 * Quick reminders:
153 *   - the last nibble of the first value is the destination buffer format.
154 *   - the last nibble of the third value is the source texture format
155 *   - formats: 4=rgb565 1=abgr8888 2=xbgr8888
156 *
157 * In the descriptions below:
158 *
159 *   SRC      means we copy the source pixels to the destination
160 *
161 *   SRC_OVER means we blend the source pixels to the destination
162 *            with dstFactor = 1-srcA, srcFactor=1  (premultiplied source).
163 *            This mode is otherwise called 'blend'.
164 *
165 *   SRCA_OVER means we blend the source pixels to the destination
166 *             with dstFactor=srcA*(1-srcA) srcFactor=srcA (non-premul source).
167 *             This mode is otherwise called 'blend_srca'
168 *
169 *   clamp    means we fetch source pixels from a texture with u/v clamping
170 *
171 *   mod      means the source pixels are modulated (multiplied) by the
172 *            a/r/g/b of the current context's color. Typically used for
173 *            fade-in / fade-out.
174 *
175 *   dither   means we dither 32 bit values to 16 bits
176 */
177static shortcut_t shortcuts[] = {
178    { { { 0x03515104, 0x00000077, { 0x00000A01, 0x00000000 } },
179        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
180        "565 fb, 8888 tx, blend SRC_OVER", scanline_t32cb16blend, init_y_noop },
181    { { { 0x03010104, 0x00000077, { 0x00000A01, 0x00000000 } },
182        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
183        "565 fb, 8888 tx, SRC", scanline_t32cb16, init_y_noop  },
184    /* same as first entry, but with dithering */
185    { { { 0x03515104, 0x00000177, { 0x00000A01, 0x00000000 } },
186        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
187        "565 fb, 8888 tx, blend SRC_OVER dither", scanline_t32cb16blend_dither, init_y_noop },
188    /* same as second entry, but with dithering */
189    { { { 0x03010104, 0x00000177, { 0x00000A01, 0x00000000 } },
190        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
191        "565 fb, 8888 tx, SRC dither", scanline_t32cb16_dither, init_y_noop  },
192    /* this is used during the boot animation - CHEAT: ignore dithering */
193    { { { 0x03545404, 0x00000077, { 0x00000A01, 0x00000000 } },
194        { 0xFFFFFFFF, 0xFFFFFEFF, { 0xFFFFFFFF, 0x0000003F } } },
195        "565 fb, 8888 tx, blend dst:ONE_MINUS_SRCA src:SRCA", scanline_t32cb16blend_srca, init_y_noop },
196    /* special case for arbitrary texture coordinates (think scaling) */
197    { { { 0x03515104, 0x00000077, { 0x00000001, 0x00000000 } },
198        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
199        "565 fb, 8888 tx, SRC_OVER clamp", scanline_t32cb16blend_clamp, init_y },
200    { { { 0x03515104, 0x00000177, { 0x00000001, 0x00000000 } },
201        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
202        "565 fb, 8888 tx, SRC_OVER clamp dither", scanline_t32cb16blend_clamp_dither, init_y },
203    /* another case used during emulation */
204    { { { 0x03515104, 0x00000077, { 0x00001001, 0x00000000 } },
205        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
206        "565 fb, 8888 tx, SRC_OVER clamp modulate", scanline_t32cb16blend_clamp_mod, init_y },
207    /* and this */
208    { { { 0x03515104, 0x00000077, { 0x00001002, 0x00000000 } },
209        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
210        "565 fb, x888 tx, SRC_OVER clamp modulate", scanline_x32cb16blend_clamp_mod, init_y },
211    { { { 0x03515104, 0x00000177, { 0x00001001, 0x00000000 } },
212        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
213        "565 fb, 8888 tx, SRC_OVER clamp modulate dither", scanline_t32cb16blend_clamp_mod_dither, init_y },
214    { { { 0x03515104, 0x00000177, { 0x00001002, 0x00000000 } },
215        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
216        "565 fb, x888 tx, SRC_OVER clamp modulate dither", scanline_x32cb16blend_clamp_mod_dither, init_y },
217    { { { 0x03010104, 0x00000077, { 0x00000001, 0x00000000 } },
218        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
219        "565 fb, 8888 tx, SRC clamp", scanline_t32cb16_clamp, init_y  },
220    { { { 0x03010104, 0x00000077, { 0x00000002, 0x00000000 } },
221        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
222        "565 fb, x888 tx, SRC clamp", scanline_t32cb16_clamp, init_y  },
223    { { { 0x03010104, 0x00000177, { 0x00000001, 0x00000000 } },
224        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
225        "565 fb, 8888 tx, SRC clamp dither", scanline_t32cb16_clamp_dither, init_y  },
226    { { { 0x03010104, 0x00000177, { 0x00000002, 0x00000000 } },
227        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
228        "565 fb, x888 tx, SRC clamp dither", scanline_t32cb16_clamp_dither, init_y  },
229    { { { 0x03010104, 0x00000077, { 0x00000004, 0x00000000 } },
230        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
231        "565 fb, 565 tx, SRC clamp", scanline_t16cb16_clamp, init_y  },
232    { { { 0x03515104, 0x00000077, { 0x00001004, 0x00000000 } },
233        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
234        "565 fb, 565 tx, SRC_OVER clamp", scanline_t16cb16blend_clamp_mod, init_y  },
235    { { { 0x03515104, 0x00000077, { 0x00000000, 0x00000000 } },
236        { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0xFFFFFFFF } } },
237        "565 fb, 8888 fixed color", scanline_col32cb16blend, init_y_packed  },
238    { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
239        { 0x00000000, 0x00000007, { 0x00000000, 0x00000000 } } },
240        "(nop) alpha test", scanline_noop, init_y_noop },
241    { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
242        { 0x00000000, 0x00000070, { 0x00000000, 0x00000000 } } },
243        "(nop) depth test", scanline_noop, init_y_noop },
244    { { { 0x05000000, 0x00000000, { 0x00000000, 0x00000000 } },
245        { 0x0F000000, 0x00000080, { 0x00000000, 0x00000000 } } },
246        "(nop) logic_op", scanline_noop, init_y_noop },
247    { { { 0xF0000000, 0x00000000, { 0x00000000, 0x00000000 } },
248        { 0xF0000000, 0x00000080, { 0x00000000, 0x00000000 } } },
249        "(nop) color mask", scanline_noop, init_y_noop },
250    { { { 0x0F000000, 0x00000077, { 0x00000000, 0x00000000 } },
251        { 0xFF000000, 0x000000F7, { 0x00000000, 0x00000000 } } },
252        "(set) logic_op", scanline_set, init_y_noop },
253    { { { 0x00000000, 0x00000077, { 0x00000000, 0x00000000 } },
254        { 0xFF000000, 0x000000F7, { 0x00000000, 0x00000000 } } },
255        "(clear) logic_op", scanline_clear, init_y_noop },
256    { { { 0x03000000, 0x00000077, { 0x00000000, 0x00000000 } },
257        { 0xFFFFFF00, 0x000000F7, { 0x00000000, 0x00000000 } } },
258        "(clear) blending 0/0", scanline_clear, init_y_noop },
259    { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
260        { 0x0000003F, 0x00000000, { 0x00000000, 0x00000000 } } },
261        "(error) invalid color-buffer format", scanline_noop, init_y_error },
262};
263static const needs_filter_t noblend1to1 = {
264        // (disregard dithering, see below)
265        { 0x03010100, 0x00000077, { 0x00000A00, 0x00000000 } },
266        { 0xFFFFFFC0, 0xFFFFFEFF, { 0xFFFFFFC0, 0x0000003F } }
267};
268static  const needs_filter_t fill16noblend = {
269        { 0x03010100, 0x00000077, { 0x00000000, 0x00000000 } },
270        { 0xFFFFFFC0, 0xFFFFFFFF, { 0x0000003F, 0x0000003F } }
271};
272
273// ----------------------------------------------------------------------------
274
275#if ANDROID_ARM_CODEGEN
276
277#if defined(__mips__)
278static CodeCache gCodeCache(32 * 1024);
279#else
280static CodeCache gCodeCache(12 * 1024);
281#endif
282
283class ScanlineAssembly : public Assembly {
284    AssemblyKey<needs_t> mKey;
285public:
286    ScanlineAssembly(needs_t needs, size_t size)
287        : Assembly(size), mKey(needs) { }
288    const AssemblyKey<needs_t>& key() const { return mKey; }
289};
290#endif
291
292// ----------------------------------------------------------------------------
293
294void ggl_init_scanline(context_t* c)
295{
296    c->init_y = init_y;
297    c->step_y = step_y__generic;
298    c->scanline = scanline;
299}
300
301void ggl_uninit_scanline(context_t* c)
302{
303    if (c->state.buffers.coverage)
304        free(c->state.buffers.coverage);
305#if ANDROID_ARM_CODEGEN
306    if (c->scanline_as)
307        c->scanline_as->decStrong(c);
308#endif
309}
310
311// ----------------------------------------------------------------------------
312
313static void pick_scanline(context_t* c)
314{
315#if (!defined(DEBUG__CODEGEN_ONLY) || (DEBUG__CODEGEN_ONLY == 0))
316
317#if ANDROID_CODEGEN == ANDROID_CODEGEN_GENERIC
318    c->init_y = init_y;
319    c->step_y = step_y__generic;
320    c->scanline = scanline;
321    return;
322#endif
323
324    //printf("*** needs [%08lx:%08lx:%08lx:%08lx]\n",
325    //    c->state.needs.n, c->state.needs.p,
326    //    c->state.needs.t[0], c->state.needs.t[1]);
327
328    // first handle the special case that we cannot test with a filter
329    const uint32_t cb_format = GGL_READ_NEEDS(CB_FORMAT, c->state.needs.n);
330    if (GGL_READ_NEEDS(T_FORMAT, c->state.needs.t[0]) == cb_format) {
331        if (c->state.needs.match(noblend1to1)) {
332            // this will match regardless of dithering state, since both
333            // src and dest have the same format anyway, there is no dithering
334            // to be done.
335            const GGLFormat* f =
336                &(c->formats[GGL_READ_NEEDS(T_FORMAT, c->state.needs.t[0])]);
337            if ((f->components == GGL_RGB) ||
338                (f->components == GGL_RGBA) ||
339                (f->components == GGL_LUMINANCE) ||
340                (f->components == GGL_LUMINANCE_ALPHA))
341            {
342                // format must have all of RGB components
343                // (so the current color doesn't show through)
344                c->scanline = scanline_memcpy;
345                c->init_y = init_y_noop;
346                return;
347            }
348        }
349    }
350
351    if (c->state.needs.match(fill16noblend)) {
352        c->init_y = init_y_packed;
353        switch (c->formats[cb_format].size) {
354        case 1: c->scanline = scanline_memset8;  return;
355        case 2: c->scanline = scanline_memset16; return;
356        case 4: c->scanline = scanline_memset32; return;
357        }
358    }
359
360    const int numFilters = sizeof(shortcuts)/sizeof(shortcut_t);
361    for (int i=0 ; i<numFilters ; i++) {
362        if (c->state.needs.match(shortcuts[i].filter)) {
363            c->scanline = shortcuts[i].scanline;
364            c->init_y = shortcuts[i].init_y;
365            return;
366        }
367    }
368
369#if DEBUG_NEEDS
370    ALOGI("Needs: n=0x%08x p=0x%08x t0=0x%08x t1=0x%08x",
371         c->state.needs.n, c->state.needs.p,
372         c->state.needs.t[0], c->state.needs.t[1]);
373#endif
374
375#endif // DEBUG__CODEGEN_ONLY
376
377    c->init_y = init_y;
378    c->step_y = step_y__generic;
379
380#if ANDROID_ARM_CODEGEN
381    // we're going to have to generate some code...
382    // here, generate code for our pixel pipeline
383    const AssemblyKey<needs_t> key(c->state.needs);
384    sp<Assembly> assembly = gCodeCache.lookup(key);
385    if (assembly == 0) {
386        // create a new assembly region
387        sp<ScanlineAssembly> a = new ScanlineAssembly(c->state.needs,
388                ASSEMBLY_SCRATCH_SIZE);
389        // initialize our assembler
390#if defined(__arm__)
391        GGLAssembler assembler( new ARMAssembler(a) );
392        //GGLAssembler assembler(
393        //        new ARMAssemblerOptimizer(new ARMAssembler(a)) );
394#endif
395#if defined(__mips__)
396        GGLAssembler assembler( new ArmToMipsAssembler(a) );
397#endif
398        // generate the scanline code for the given needs
399        int err = assembler.scanline(c->state.needs, c);
400        if (ggl_likely(!err)) {
401            // finally, cache this assembly
402            err = gCodeCache.cache(a->key(), a);
403        }
404        if (ggl_unlikely(err)) {
405            ALOGE("error generating or caching assembly. Reverting to NOP.");
406            c->scanline = scanline_noop;
407            c->init_y = init_y_noop;
408            c->step_y = step_y__nop;
409            return;
410        }
411        assembly = a;
412    }
413
414    // release the previous assembly
415    if (c->scanline_as) {
416        c->scanline_as->decStrong(c);
417    }
418
419    //ALOGI("using generated pixel-pipeline");
420    c->scanline_as = assembly.get();
421    c->scanline_as->incStrong(c); //  hold on to assembly
422    c->scanline = (void(*)(context_t* c))assembly->base();
423#else
424//    ALOGW("using generic (slow) pixel-pipeline");
425    c->scanline = scanline;
426#endif
427}
428
429void ggl_pick_scanline(context_t* c)
430{
431    pick_scanline(c);
432    if ((c->state.enables & GGL_ENABLE_W) &&
433        (c->state.enables & GGL_ENABLE_TMUS))
434    {
435        c->span = c->scanline;
436        c->scanline = scanline_perspective;
437        if (!(c->state.enabled_tmu & (c->state.enabled_tmu - 1))) {
438            // only one TMU enabled
439            c->scanline = scanline_perspective_single;
440        }
441    }
442}
443
444// ----------------------------------------------------------------------------
445
446static void blending(context_t* c, pixel_t* fragment, pixel_t* fb);
447static void blend_factor(context_t* c, pixel_t* r, uint32_t factor,
448        const pixel_t* src, const pixel_t* dst);
449static void rescale(uint32_t& u, uint8_t& su, uint32_t& v, uint8_t& sv);
450
451#if ANDROID_ARM_CODEGEN && (ANDROID_CODEGEN == ANDROID_CODEGEN_GENERATED)
452
453// no need to compile the generic-pipeline, it can't be reached
454void scanline(context_t*)
455{
456}
457
458#else
459
460void rescale(uint32_t& u, uint8_t& su, uint32_t& v, uint8_t& sv)
461{
462    if (su && sv) {
463        if (su > sv) {
464            v = ggl_expand(v, sv, su);
465            sv = su;
466        } else if (su < sv) {
467            u = ggl_expand(u, su, sv);
468            su = sv;
469        }
470    }
471}
472
473void blending(context_t* c, pixel_t* fragment, pixel_t* fb)
474{
475    rescale(fragment->c[0], fragment->s[0], fb->c[0], fb->s[0]);
476    rescale(fragment->c[1], fragment->s[1], fb->c[1], fb->s[1]);
477    rescale(fragment->c[2], fragment->s[2], fb->c[2], fb->s[2]);
478    rescale(fragment->c[3], fragment->s[3], fb->c[3], fb->s[3]);
479
480    pixel_t sf, df;
481    blend_factor(c, &sf, c->state.blend.src, fragment, fb);
482    blend_factor(c, &df, c->state.blend.dst, fragment, fb);
483
484    fragment->c[1] =
485            gglMulAddx(fragment->c[1], sf.c[1], gglMulx(fb->c[1], df.c[1]));
486    fragment->c[2] =
487            gglMulAddx(fragment->c[2], sf.c[2], gglMulx(fb->c[2], df.c[2]));
488    fragment->c[3] =
489            gglMulAddx(fragment->c[3], sf.c[3], gglMulx(fb->c[3], df.c[3]));
490
491    if (c->state.blend.alpha_separate) {
492        blend_factor(c, &sf, c->state.blend.src_alpha, fragment, fb);
493        blend_factor(c, &df, c->state.blend.dst_alpha, fragment, fb);
494    }
495
496    fragment->c[0] =
497            gglMulAddx(fragment->c[0], sf.c[0], gglMulx(fb->c[0], df.c[0]));
498
499    // clamp to 1.0
500    if (fragment->c[0] >= (1LU<<fragment->s[0]))
501        fragment->c[0] = (1<<fragment->s[0])-1;
502    if (fragment->c[1] >= (1LU<<fragment->s[1]))
503        fragment->c[1] = (1<<fragment->s[1])-1;
504    if (fragment->c[2] >= (1LU<<fragment->s[2]))
505        fragment->c[2] = (1<<fragment->s[2])-1;
506    if (fragment->c[3] >= (1LU<<fragment->s[3]))
507        fragment->c[3] = (1<<fragment->s[3])-1;
508}
509
510static inline int blendfactor(uint32_t x, uint32_t size, uint32_t def = 0)
511{
512    if (!size)
513        return def;
514
515    // scale to 16 bits
516    if (size > 16) {
517        x >>= (size - 16);
518    } else if (size < 16) {
519        x = ggl_expand(x, size, 16);
520    }
521    x += x >> 15;
522    return x;
523}
524
525void blend_factor(context_t* c, pixel_t* r,
526        uint32_t factor, const pixel_t* src, const pixel_t* dst)
527{
528    switch (factor) {
529        case GGL_ZERO:
530            r->c[1] =
531            r->c[2] =
532            r->c[3] =
533            r->c[0] = 0;
534            break;
535        case GGL_ONE:
536            r->c[1] =
537            r->c[2] =
538            r->c[3] =
539            r->c[0] = FIXED_ONE;
540            break;
541        case GGL_DST_COLOR:
542            r->c[1] = blendfactor(dst->c[1], dst->s[1]);
543            r->c[2] = blendfactor(dst->c[2], dst->s[2]);
544            r->c[3] = blendfactor(dst->c[3], dst->s[3]);
545            r->c[0] = blendfactor(dst->c[0], dst->s[0]);
546            break;
547        case GGL_SRC_COLOR:
548            r->c[1] = blendfactor(src->c[1], src->s[1]);
549            r->c[2] = blendfactor(src->c[2], src->s[2]);
550            r->c[3] = blendfactor(src->c[3], src->s[3]);
551            r->c[0] = blendfactor(src->c[0], src->s[0]);
552            break;
553        case GGL_ONE_MINUS_DST_COLOR:
554            r->c[1] = FIXED_ONE - blendfactor(dst->c[1], dst->s[1]);
555            r->c[2] = FIXED_ONE - blendfactor(dst->c[2], dst->s[2]);
556            r->c[3] = FIXED_ONE - blendfactor(dst->c[3], dst->s[3]);
557            r->c[0] = FIXED_ONE - blendfactor(dst->c[0], dst->s[0]);
558            break;
559        case GGL_ONE_MINUS_SRC_COLOR:
560            r->c[1] = FIXED_ONE - blendfactor(src->c[1], src->s[1]);
561            r->c[2] = FIXED_ONE - blendfactor(src->c[2], src->s[2]);
562            r->c[3] = FIXED_ONE - blendfactor(src->c[3], src->s[3]);
563            r->c[0] = FIXED_ONE - blendfactor(src->c[0], src->s[0]);
564            break;
565        case GGL_SRC_ALPHA:
566            r->c[1] =
567            r->c[2] =
568            r->c[3] =
569            r->c[0] = blendfactor(src->c[0], src->s[0], FIXED_ONE);
570            break;
571        case GGL_ONE_MINUS_SRC_ALPHA:
572            r->c[1] =
573            r->c[2] =
574            r->c[3] =
575            r->c[0] = FIXED_ONE - blendfactor(src->c[0], src->s[0], FIXED_ONE);
576            break;
577        case GGL_DST_ALPHA:
578            r->c[1] =
579            r->c[2] =
580            r->c[3] =
581            r->c[0] = blendfactor(dst->c[0], dst->s[0], FIXED_ONE);
582            break;
583        case GGL_ONE_MINUS_DST_ALPHA:
584            r->c[1] =
585            r->c[2] =
586            r->c[3] =
587            r->c[0] = FIXED_ONE - blendfactor(dst->c[0], dst->s[0], FIXED_ONE);
588            break;
589        case GGL_SRC_ALPHA_SATURATE:
590            // XXX: GGL_SRC_ALPHA_SATURATE
591            break;
592    }
593}
594
595static GGLfixed wrapping(int32_t coord, uint32_t size, int tx_wrap)
596{
597    GGLfixed d;
598    if (tx_wrap == GGL_REPEAT) {
599        d = (uint32_t(coord)>>16) * size;
600    } else if (tx_wrap == GGL_CLAMP) { // CLAMP_TO_EDGE semantics
601        const GGLfixed clamp_min = FIXED_HALF;
602        const GGLfixed clamp_max = (size << 16) - FIXED_HALF;
603        if (coord < clamp_min)     coord = clamp_min;
604        if (coord > clamp_max)     coord = clamp_max;
605        d = coord;
606    } else { // 1:1
607        const GGLfixed clamp_min = 0;
608        const GGLfixed clamp_max = (size << 16);
609        if (coord < clamp_min)     coord = clamp_min;
610        if (coord > clamp_max)     coord = clamp_max;
611        d = coord;
612    }
613    return d;
614}
615
616static inline
617GGLcolor ADJUST_COLOR_ITERATOR(GGLcolor v, GGLcolor dvdx, int len)
618{
619    const int32_t end = dvdx * (len-1) + v;
620    if (end < 0)
621        v -= end;
622    v &= ~(v>>31);
623    return v;
624}
625
626void scanline(context_t* c)
627{
628    const uint32_t enables = c->state.enables;
629    const int xs = c->iterators.xl;
630    const int x1 = c->iterators.xr;
631	int xc = x1 - xs;
632    const int16_t* covPtr = c->state.buffers.coverage + xs;
633
634    // All iterated values are sampled at the pixel center
635
636    // reset iterators for that scanline...
637    GGLcolor r, g, b, a;
638    iterators_t& ci = c->iterators;
639    if (enables & GGL_ENABLE_SMOOTH) {
640        r = (xs * c->shade.drdx) + ci.ydrdy;
641        g = (xs * c->shade.dgdx) + ci.ydgdy;
642        b = (xs * c->shade.dbdx) + ci.ydbdy;
643        a = (xs * c->shade.dadx) + ci.ydady;
644        r = ADJUST_COLOR_ITERATOR(r, c->shade.drdx, xc);
645        g = ADJUST_COLOR_ITERATOR(g, c->shade.dgdx, xc);
646        b = ADJUST_COLOR_ITERATOR(b, c->shade.dbdx, xc);
647        a = ADJUST_COLOR_ITERATOR(a, c->shade.dadx, xc);
648    } else {
649        r = ci.ydrdy;
650        g = ci.ydgdy;
651        b = ci.ydbdy;
652        a = ci.ydady;
653    }
654
655    // z iterators are 1.31
656    GGLfixed z = (xs * c->shade.dzdx) + ci.ydzdy;
657    GGLfixed f = (xs * c->shade.dfdx) + ci.ydfdy;
658
659    struct {
660        GGLfixed s, t;
661    } tc[GGL_TEXTURE_UNIT_COUNT];
662    if (enables & GGL_ENABLE_TMUS) {
663        for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
664            if (c->state.texture[i].enable) {
665                texture_iterators_t& ti = c->state.texture[i].iterators;
666                if (enables & GGL_ENABLE_W) {
667                    tc[i].s = ti.ydsdy;
668                    tc[i].t = ti.ydtdy;
669                } else {
670                    tc[i].s = (xs * ti.dsdx) + ti.ydsdy;
671                    tc[i].t = (xs * ti.dtdx) + ti.ydtdy;
672                }
673            }
674        }
675    }
676
677    pixel_t fragment;
678    pixel_t texel;
679    pixel_t fb;
680
681	uint32_t x = xs;
682	uint32_t y = c->iterators.y;
683
684	while (xc--) {
685
686        { // just a scope
687
688		// read color (convert to 8 bits by keeping only the integer part)
689        fragment.s[1] = fragment.s[2] =
690        fragment.s[3] = fragment.s[0] = 8;
691        fragment.c[1] = r >> (GGL_COLOR_BITS-8);
692        fragment.c[2] = g >> (GGL_COLOR_BITS-8);
693        fragment.c[3] = b >> (GGL_COLOR_BITS-8);
694        fragment.c[0] = a >> (GGL_COLOR_BITS-8);
695
696		// texturing
697        if (enables & GGL_ENABLE_TMUS) {
698            for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
699                texture_t& tx = c->state.texture[i];
700                if (!tx.enable)
701                    continue;
702                texture_iterators_t& ti = tx.iterators;
703                int32_t u, v;
704
705                // s-coordinate
706                if (tx.s_coord != GGL_ONE_TO_ONE) {
707                    const int w = tx.surface.width;
708                    u = wrapping(tc[i].s, w, tx.s_wrap);
709                    tc[i].s += ti.dsdx;
710                } else {
711                    u = (((tx.shade.is0>>16) + x)<<16) + FIXED_HALF;
712                }
713
714                // t-coordinate
715                if (tx.t_coord != GGL_ONE_TO_ONE) {
716                    const int h = tx.surface.height;
717                    v = wrapping(tc[i].t, h, tx.t_wrap);
718                    tc[i].t += ti.dtdx;
719                } else {
720                    v = (((tx.shade.it0>>16) + y)<<16) + FIXED_HALF;
721                }
722
723                // read texture
724                if (tx.mag_filter == GGL_NEAREST &&
725                    tx.min_filter == GGL_NEAREST)
726                {
727                    u >>= 16;
728                    v >>= 16;
729                    tx.surface.read(&tx.surface, c, u, v, &texel);
730                } else {
731                    const int w = tx.surface.width;
732                    const int h = tx.surface.height;
733                    u -= FIXED_HALF;
734                    v -= FIXED_HALF;
735                    int u0 = u >> 16;
736                    int v0 = v >> 16;
737                    int u1 = u0 + 1;
738                    int v1 = v0 + 1;
739                    if (tx.s_wrap == GGL_REPEAT) {
740                        if (u0<0)  u0 += w;
741                        if (u1<0)  u1 += w;
742                        if (u0>=w) u0 -= w;
743                        if (u1>=w) u1 -= w;
744                    } else {
745                        if (u0<0)  u0 = 0;
746                        if (u1<0)  u1 = 0;
747                        if (u0>=w) u0 = w-1;
748                        if (u1>=w) u1 = w-1;
749                    }
750                    if (tx.t_wrap == GGL_REPEAT) {
751                        if (v0<0)  v0 += h;
752                        if (v1<0)  v1 += h;
753                        if (v0>=h) v0 -= h;
754                        if (v1>=h) v1 -= h;
755                    } else {
756                        if (v0<0)  v0 = 0;
757                        if (v1<0)  v1 = 0;
758                        if (v0>=h) v0 = h-1;
759                        if (v1>=h) v1 = h-1;
760                    }
761                    pixel_t texels[4];
762                    uint32_t mm[4];
763                    tx.surface.read(&tx.surface, c, u0, v0, &texels[0]);
764                    tx.surface.read(&tx.surface, c, u0, v1, &texels[1]);
765                    tx.surface.read(&tx.surface, c, u1, v0, &texels[2]);
766                    tx.surface.read(&tx.surface, c, u1, v1, &texels[3]);
767                    u = (u >> 12) & 0xF;
768                    v = (v >> 12) & 0xF;
769                    u += u>>3;
770                    v += v>>3;
771                    mm[0] = (0x10 - u) * (0x10 - v);
772                    mm[1] = (0x10 - u) * v;
773                    mm[2] = u * (0x10 - v);
774                    mm[3] = 0x100 - (mm[0] + mm[1] + mm[2]);
775                    for (int j=0 ; j<4 ; j++) {
776                        texel.s[j] = texels[0].s[j];
777                        if (!texel.s[j]) continue;
778                        texel.s[j] += 8;
779                        texel.c[j] =    texels[0].c[j]*mm[0] +
780                                        texels[1].c[j]*mm[1] +
781                                        texels[2].c[j]*mm[2] +
782                                        texels[3].c[j]*mm[3] ;
783                    }
784                }
785
786                // Texture environnement...
787                for (int j=0 ; j<4 ; j++) {
788                    uint32_t& Cf = fragment.c[j];
789                    uint32_t& Ct = texel.c[j];
790                    uint8_t& sf  = fragment.s[j];
791                    uint8_t& st  = texel.s[j];
792                    uint32_t At = texel.c[0];
793                    uint8_t sat = texel.s[0];
794                    switch (tx.env) {
795                    case GGL_REPLACE:
796                        if (st) {
797                            Cf = Ct;
798                            sf = st;
799                        }
800                        break;
801                    case GGL_MODULATE:
802                        if (st) {
803                            uint32_t factor = Ct + (Ct>>(st-1));
804                            Cf = (Cf * factor) >> st;
805                        }
806                        break;
807                    case GGL_DECAL:
808                        if (sat) {
809                            rescale(Cf, sf, Ct, st);
810                            Cf += ((Ct - Cf) * (At + (At>>(sat-1)))) >> sat;
811                        }
812                        break;
813                    case GGL_BLEND:
814                        if (st) {
815                            uint32_t Cc = tx.env_color[i];
816                            if (sf>8)       Cc = (Cc * ((1<<sf)-1))>>8;
817                            else if (sf<8)  Cc = (Cc - (Cc>>(8-sf)))>>(8-sf);
818                            uint32_t factor = Ct + (Ct>>(st-1));
819                            Cf = ((((1<<st) - factor) * Cf) + Ct*Cc)>>st;
820                        }
821                        break;
822                    case GGL_ADD:
823                        if (st) {
824                            rescale(Cf, sf, Ct, st);
825                            Cf += Ct;
826                        }
827                        break;
828                    }
829                }
830            }
831		}
832
833        // coverage application
834        if (enables & GGL_ENABLE_AA) {
835            int16_t cf = *covPtr++;
836            fragment.c[0] = (int64_t(fragment.c[0]) * cf) >> 15;
837        }
838
839        // alpha-test
840        if (enables & GGL_ENABLE_ALPHA_TEST) {
841            GGLcolor ref = c->state.alpha_test.ref;
842            GGLcolor alpha = (uint64_t(fragment.c[0]) *
843                    ((1<<GGL_COLOR_BITS)-1)) / ((1<<fragment.s[0])-1);
844            switch (c->state.alpha_test.func) {
845            case GGL_NEVER:     goto discard;
846            case GGL_LESS:      if (alpha<ref)  break; goto discard;
847            case GGL_EQUAL:     if (alpha==ref) break; goto discard;
848            case GGL_LEQUAL:    if (alpha<=ref) break; goto discard;
849            case GGL_GREATER:   if (alpha>ref)  break; goto discard;
850            case GGL_NOTEQUAL:  if (alpha!=ref) break; goto discard;
851            case GGL_GEQUAL:    if (alpha>=ref) break; goto discard;
852            }
853        }
854
855        // depth test
856        if (c->state.buffers.depth.format) {
857            if (enables & GGL_ENABLE_DEPTH_TEST) {
858                surface_t* cb = &(c->state.buffers.depth);
859                uint16_t* p = (uint16_t*)(cb->data)+(x+(cb->stride*y));
860                uint16_t zz = uint32_t(z)>>(16);
861                uint16_t depth = *p;
862                switch (c->state.depth_test.func) {
863                case GGL_NEVER:     goto discard;
864                case GGL_LESS:      if (zz<depth)    break; goto discard;
865                case GGL_EQUAL:     if (zz==depth)   break; goto discard;
866                case GGL_LEQUAL:    if (zz<=depth)   break; goto discard;
867                case GGL_GREATER:   if (zz>depth)    break; goto discard;
868                case GGL_NOTEQUAL:  if (zz!=depth)   break; goto discard;
869                case GGL_GEQUAL:    if (zz>=depth)   break; goto discard;
870                }
871                // depth buffer is not enabled, if depth-test is not enabled
872/*
873        fragment.s[1] = fragment.s[2] =
874        fragment.s[3] = fragment.s[0] = 8;
875        fragment.c[1] =
876        fragment.c[2] =
877        fragment.c[3] =
878        fragment.c[0] = 255 - (zz>>8);
879*/
880                if (c->state.mask.depth) {
881                    *p = zz;
882                }
883            }
884        }
885
886        // fog
887        if (enables & GGL_ENABLE_FOG) {
888            for (int i=1 ; i<=3 ; i++) {
889                GGLfixed fc = (c->state.fog.color[i] * 0x10000) / 0xFF;
890                uint32_t& c = fragment.c[i];
891                uint8_t& s  = fragment.s[i];
892                c = (c * 0x10000) / ((1<<s)-1);
893                c = gglMulAddx(c, f, gglMulx(fc, 0x10000 - f));
894                s = 16;
895            }
896        }
897
898        // blending
899        if (enables & GGL_ENABLE_BLENDING) {
900            fb.c[1] = fb.c[2] = fb.c[3] = fb.c[0] = 0; // placate valgrind
901            fb.s[1] = fb.s[2] = fb.s[3] = fb.s[0] = 0;
902            c->state.buffers.color.read(
903                    &(c->state.buffers.color), c, x, y, &fb);
904            blending( c, &fragment, &fb );
905        }
906
907		// write
908        c->state.buffers.color.write(
909                &(c->state.buffers.color), c, x, y, &fragment);
910        }
911
912discard:
913		// iterate...
914        x += 1;
915        if (enables & GGL_ENABLE_SMOOTH) {
916            r += c->shade.drdx;
917            g += c->shade.dgdx;
918            b += c->shade.dbdx;
919            a += c->shade.dadx;
920        }
921        z += c->shade.dzdx;
922        f += c->shade.dfdx;
923	}
924}
925
926#endif // ANDROID_ARM_CODEGEN && (ANDROID_CODEGEN == ANDROID_CODEGEN_GENERATED)
927
928// ----------------------------------------------------------------------------
929#if 0
930#pragma mark -
931#pragma mark Scanline
932#endif
933
934/* Used to parse a 32-bit source texture linearly. Usage is:
935 *
936 * horz_iterator32  hi(context);
937 * while (...) {
938 *    uint32_t  src_pixel = hi.get_pixel32();
939 *    ...
940 * }
941 *
942 * Use only for one-to-one texture mapping.
943 */
944struct horz_iterator32 {
945    horz_iterator32(context_t* c) {
946        const int x = c->iterators.xl;
947        const int y = c->iterators.y;
948        texture_t& tx = c->state.texture[0];
949        const int32_t u = (tx.shade.is0>>16) + x;
950        const int32_t v = (tx.shade.it0>>16) + y;
951        m_src = reinterpret_cast<uint32_t*>(tx.surface.data)+(u+(tx.surface.stride*v));
952    }
953    uint32_t  get_pixel32() {
954        return *m_src++;
955    }
956protected:
957    uint32_t* m_src;
958};
959
960/* A variant for 16-bit source textures. */
961struct horz_iterator16 {
962    horz_iterator16(context_t* c) {
963        const int x = c->iterators.xl;
964        const int y = c->iterators.y;
965        texture_t& tx = c->state.texture[0];
966        const int32_t u = (tx.shade.is0>>16) + x;
967        const int32_t v = (tx.shade.it0>>16) + y;
968        m_src = reinterpret_cast<uint16_t*>(tx.surface.data)+(u+(tx.surface.stride*v));
969    }
970    uint16_t  get_pixel16() {
971        return *m_src++;
972    }
973protected:
974    uint16_t* m_src;
975};
976
977/* A clamp iterator is used to iterate inside a texture with GGL_CLAMP.
978 * After initialization, call get_src16() or get_src32() to get the current
979 * texture pixel value.
980 */
981struct clamp_iterator {
982    clamp_iterator(context_t* c) {
983        const int xs = c->iterators.xl;
984        texture_t& tx = c->state.texture[0];
985        texture_iterators_t& ti = tx.iterators;
986        m_s = (xs * ti.dsdx) + ti.ydsdy;
987        m_t = (xs * ti.dtdx) + ti.ydtdy;
988        m_ds = ti.dsdx;
989        m_dt = ti.dtdx;
990        m_width_m1 = tx.surface.width - 1;
991        m_height_m1 = tx.surface.height - 1;
992        m_data = tx.surface.data;
993        m_stride = tx.surface.stride;
994    }
995    uint16_t get_pixel16() {
996        int  u, v;
997        get_uv(u, v);
998        uint16_t* src = reinterpret_cast<uint16_t*>(m_data) + (u + (m_stride*v));
999        return src[0];
1000    }
1001    uint32_t get_pixel32() {
1002        int  u, v;
1003        get_uv(u, v);
1004        uint32_t* src = reinterpret_cast<uint32_t*>(m_data) + (u + (m_stride*v));
1005        return src[0];
1006    }
1007private:
1008    void   get_uv(int& u, int& v) {
1009        int  uu = m_s >> 16;
1010        int  vv = m_t >> 16;
1011        if (uu < 0)
1012            uu = 0;
1013        if (uu > m_width_m1)
1014            uu = m_width_m1;
1015        if (vv < 0)
1016            vv = 0;
1017        if (vv > m_height_m1)
1018            vv = m_height_m1;
1019        u = uu;
1020        v = vv;
1021        m_s += m_ds;
1022        m_t += m_dt;
1023    }
1024
1025    GGLfixed  m_s, m_t;
1026    GGLfixed  m_ds, m_dt;
1027    int       m_width_m1, m_height_m1;
1028    uint8_t*  m_data;
1029    int       m_stride;
1030};
1031
1032/*
1033 * The 'horizontal clamp iterator' variant corresponds to the case where
1034 * the 'v' coordinate doesn't change. This is useful to avoid one mult and
1035 * extra adds / checks per pixels, if the blending/processing operation after
1036 * this is very fast.
1037 */
1038static int is_context_horizontal(const context_t* c) {
1039    return (c->state.texture[0].iterators.dtdx == 0);
1040}
1041
1042struct horz_clamp_iterator {
1043    uint16_t  get_pixel16() {
1044        int  u = m_s >> 16;
1045        m_s += m_ds;
1046        if (u < 0)
1047            u = 0;
1048        if (u > m_width_m1)
1049            u = m_width_m1;
1050        const uint16_t* src = reinterpret_cast<const uint16_t*>(m_data);
1051        return src[u];
1052    }
1053    uint32_t  get_pixel32() {
1054        int  u = m_s >> 16;
1055        m_s += m_ds;
1056        if (u < 0)
1057            u = 0;
1058        if (u > m_width_m1)
1059            u = m_width_m1;
1060        const uint32_t* src = reinterpret_cast<const uint32_t*>(m_data);
1061        return src[u];
1062    }
1063protected:
1064    void init(const context_t* c, int shift);
1065    GGLfixed       m_s;
1066    GGLfixed       m_ds;
1067    int            m_width_m1;
1068    const uint8_t* m_data;
1069};
1070
1071void horz_clamp_iterator::init(const context_t* c, int shift)
1072{
1073    const int xs = c->iterators.xl;
1074    const texture_t& tx = c->state.texture[0];
1075    const texture_iterators_t& ti = tx.iterators;
1076    m_s = (xs * ti.dsdx) + ti.ydsdy;
1077    m_ds = ti.dsdx;
1078    m_width_m1 = tx.surface.width-1;
1079    m_data = tx.surface.data;
1080
1081    GGLfixed t = (xs * ti.dtdx) + ti.ydtdy;
1082    int      v = t >> 16;
1083    if (v < 0)
1084        v = 0;
1085    else if (v >= (int)tx.surface.height)
1086        v = (int)tx.surface.height-1;
1087
1088    m_data += (tx.surface.stride*v) << shift;
1089}
1090
1091struct horz_clamp_iterator16 : horz_clamp_iterator {
1092    horz_clamp_iterator16(const context_t* c) {
1093        init(c,1);
1094    };
1095};
1096
1097struct horz_clamp_iterator32 : horz_clamp_iterator {
1098    horz_clamp_iterator32(context_t* c) {
1099        init(c,2);
1100    };
1101};
1102
1103/* This is used to perform dithering operations.
1104 */
1105struct ditherer {
1106    ditherer(const context_t* c) {
1107        const int x = c->iterators.xl;
1108        const int y = c->iterators.y;
1109        m_line = &c->ditherMatrix[ ((y & GGL_DITHER_MASK)<<GGL_DITHER_ORDER_SHIFT) ];
1110        m_index = x & GGL_DITHER_MASK;
1111    }
1112    void step(void) {
1113        m_index++;
1114    }
1115    int  get_value(void) {
1116        int ret = m_line[m_index & GGL_DITHER_MASK];
1117        m_index++;
1118        return ret;
1119    }
1120    uint16_t abgr8888ToRgb565(uint32_t s) {
1121        uint32_t r = s & 0xff;
1122        uint32_t g = (s >> 8) & 0xff;
1123        uint32_t b = (s >> 16) & 0xff;
1124        return rgb888ToRgb565(r,g,b);
1125    }
1126    /* The following assumes that r/g/b are in the 0..255 range each */
1127    uint16_t rgb888ToRgb565(uint32_t& r, uint32_t& g, uint32_t &b) {
1128        int threshold = get_value();
1129        /* dither in on GGL_DITHER_BITS, and each of r, g, b is on 8 bits */
1130        r += (threshold >> (GGL_DITHER_BITS-8 +5));
1131        g += (threshold >> (GGL_DITHER_BITS-8 +6));
1132        b += (threshold >> (GGL_DITHER_BITS-8 +5));
1133        if (r > 0xff)
1134            r = 0xff;
1135        if (g > 0xff)
1136            g = 0xff;
1137        if (b > 0xff)
1138            b = 0xff;
1139        return uint16_t(((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3));
1140    }
1141protected:
1142    const uint8_t* m_line;
1143    int            m_index;
1144};
1145
1146/* This structure is used to blend (SRC_OVER) 32-bit source pixels
1147 * onto 16-bit destination ones. Usage is simply:
1148 *
1149 *   blender.blend(<32-bit-src-pixel-value>,<ptr-to-16-bit-dest-pixel>)
1150 */
1151struct blender_32to16 {
1152    blender_32to16(context_t* c) { }
1153    void write(uint32_t s, uint16_t* dst) {
1154        if (s == 0)
1155            return;
1156        s = GGL_RGBA_TO_HOST(s);
1157        int sA = (s>>24);
1158        if (sA == 0xff) {
1159            *dst = convertAbgr8888ToRgb565(s);
1160        } else {
1161            int f = 0x100 - (sA + (sA>>7));
1162            int sR = (s >> (   3))&0x1F;
1163            int sG = (s >> ( 8+2))&0x3F;
1164            int sB = (s >> (16+3))&0x1F;
1165            uint16_t d = *dst;
1166            int dR = (d>>11)&0x1f;
1167            int dG = (d>>5)&0x3f;
1168            int dB = (d)&0x1f;
1169            sR += (f*dR)>>8;
1170            sG += (f*dG)>>8;
1171            sB += (f*dB)>>8;
1172            *dst = uint16_t((sR<<11)|(sG<<5)|sB);
1173        }
1174    }
1175    void write(uint32_t s, uint16_t* dst, ditherer& di) {
1176        if (s == 0) {
1177            di.step();
1178            return;
1179        }
1180        s = GGL_RGBA_TO_HOST(s);
1181        int sA = (s>>24);
1182        if (sA == 0xff) {
1183            *dst = di.abgr8888ToRgb565(s);
1184        } else {
1185            int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
1186            int f = 0x100 - (sA + (sA>>7));
1187            int sR = (s >> (   3))&0x1F;
1188            int sG = (s >> ( 8+2))&0x3F;
1189            int sB = (s >> (16+3))&0x1F;
1190            uint16_t d = *dst;
1191            int dR = (d>>11)&0x1f;
1192            int dG = (d>>5)&0x3f;
1193            int dB = (d)&0x1f;
1194            sR = ((sR << 8) + f*dR + threshold)>>8;
1195            sG = ((sG << 8) + f*dG + threshold)>>8;
1196            sB = ((sB << 8) + f*dB + threshold)>>8;
1197            if (sR > 0x1f) sR = 0x1f;
1198            if (sG > 0x3f) sG = 0x3f;
1199            if (sB > 0x1f) sB = 0x1f;
1200            *dst = uint16_t((sR<<11)|(sG<<5)|sB);
1201        }
1202    }
1203};
1204
1205/* This blender does the same for the 'blend_srca' operation.
1206 * where dstFactor=srcA*(1-srcA) srcFactor=srcA
1207 */
1208struct blender_32to16_srcA {
1209    blender_32to16_srcA(const context_t* c) { }
1210    void write(uint32_t s, uint16_t* dst) {
1211        if (!s) {
1212            return;
1213        }
1214        uint16_t d = *dst;
1215        s = GGL_RGBA_TO_HOST(s);
1216        int sR = (s >> (   3))&0x1F;
1217        int sG = (s >> ( 8+2))&0x3F;
1218        int sB = (s >> (16+3))&0x1F;
1219        int sA = (s>>24);
1220        int f1 = (sA + (sA>>7));
1221        int f2 = 0x100-f1;
1222        int dR = (d>>11)&0x1f;
1223        int dG = (d>>5)&0x3f;
1224        int dB = (d)&0x1f;
1225        sR = (f1*sR + f2*dR)>>8;
1226        sG = (f1*sG + f2*dG)>>8;
1227        sB = (f1*sB + f2*dB)>>8;
1228        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
1229    }
1230};
1231
1232/* Common init code the modulating blenders */
1233struct blender_modulate {
1234    void init(const context_t* c) {
1235        const int r = c->iterators.ydrdy >> (GGL_COLOR_BITS-8);
1236        const int g = c->iterators.ydgdy >> (GGL_COLOR_BITS-8);
1237        const int b = c->iterators.ydbdy >> (GGL_COLOR_BITS-8);
1238        const int a = c->iterators.ydady >> (GGL_COLOR_BITS-8);
1239        m_r = r + (r >> 7);
1240        m_g = g + (g >> 7);
1241        m_b = b + (b >> 7);
1242        m_a = a + (a >> 7);
1243    }
1244protected:
1245    int m_r, m_g, m_b, m_a;
1246};
1247
1248/* This blender does a normal blend after modulation.
1249 */
1250struct blender_32to16_modulate : blender_modulate {
1251    blender_32to16_modulate(const context_t* c) {
1252        init(c);
1253    }
1254    void write(uint32_t s, uint16_t* dst) {
1255        // blend source and destination
1256        if (!s) {
1257            return;
1258        }
1259        s = GGL_RGBA_TO_HOST(s);
1260
1261        /* We need to modulate s */
1262        uint32_t  sA = (s >> 24);
1263        uint32_t  sB = (s >> 16) & 0xff;
1264        uint32_t  sG = (s >> 8) & 0xff;
1265        uint32_t  sR = s & 0xff;
1266
1267        sA = (sA*m_a) >> 8;
1268        /* Keep R/G/B scaled to 5.8 or 6.8 fixed float format */
1269        sR = (sR*m_r) >> (8 - 5);
1270        sG = (sG*m_g) >> (8 - 6);
1271        sB = (sB*m_b) >> (8 - 5);
1272
1273        /* Now do a normal blend */
1274        int f = 0x100 - (sA + (sA>>7));
1275        uint16_t d = *dst;
1276        int dR = (d>>11)&0x1f;
1277        int dG = (d>>5)&0x3f;
1278        int dB = (d)&0x1f;
1279        sR = (sR + f*dR)>>8;
1280        sG = (sG + f*dG)>>8;
1281        sB = (sB + f*dB)>>8;
1282        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
1283    }
1284    void write(uint32_t s, uint16_t* dst, ditherer& di) {
1285        // blend source and destination
1286        if (!s) {
1287            di.step();
1288            return;
1289        }
1290        s = GGL_RGBA_TO_HOST(s);
1291
1292        /* We need to modulate s */
1293        uint32_t  sA = (s >> 24);
1294        uint32_t  sB = (s >> 16) & 0xff;
1295        uint32_t  sG = (s >> 8) & 0xff;
1296        uint32_t  sR = s & 0xff;
1297
1298        sA = (sA*m_a) >> 8;
1299        /* keep R/G/B scaled to 5.8 or 6.8 fixed float format */
1300        sR = (sR*m_r) >> (8 - 5);
1301        sG = (sG*m_g) >> (8 - 6);
1302        sB = (sB*m_b) >> (8 - 5);
1303
1304        /* Scale threshold to 0.8 fixed float format */
1305        int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
1306        int f = 0x100 - (sA + (sA>>7));
1307        uint16_t d = *dst;
1308        int dR = (d>>11)&0x1f;
1309        int dG = (d>>5)&0x3f;
1310        int dB = (d)&0x1f;
1311        sR = (sR + f*dR + threshold)>>8;
1312        sG = (sG + f*dG + threshold)>>8;
1313        sB = (sB + f*dB + threshold)>>8;
1314        if (sR > 0x1f) sR = 0x1f;
1315        if (sG > 0x3f) sG = 0x3f;
1316        if (sB > 0x1f) sB = 0x1f;
1317        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
1318    }
1319};
1320
1321/* same as 32to16_modulate, except that the input is xRGB, instead of ARGB */
1322struct blender_x32to16_modulate : blender_modulate {
1323    blender_x32to16_modulate(const context_t* c) {
1324        init(c);
1325    }
1326    void write(uint32_t s, uint16_t* dst) {
1327        s = GGL_RGBA_TO_HOST(s);
1328
1329        uint32_t  sB = (s >> 16) & 0xff;
1330        uint32_t  sG = (s >> 8) & 0xff;
1331        uint32_t  sR = s & 0xff;
1332
1333        /* Keep R/G/B in 5.8 or 6.8 format */
1334        sR = (sR*m_r) >> (8 - 5);
1335        sG = (sG*m_g) >> (8 - 6);
1336        sB = (sB*m_b) >> (8 - 5);
1337
1338        int f = 0x100 - m_a;
1339        uint16_t d = *dst;
1340        int dR = (d>>11)&0x1f;
1341        int dG = (d>>5)&0x3f;
1342        int dB = (d)&0x1f;
1343        sR = (sR + f*dR)>>8;
1344        sG = (sG + f*dG)>>8;
1345        sB = (sB + f*dB)>>8;
1346        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
1347    }
1348    void write(uint32_t s, uint16_t* dst, ditherer& di) {
1349        s = GGL_RGBA_TO_HOST(s);
1350
1351        uint32_t  sB = (s >> 16) & 0xff;
1352        uint32_t  sG = (s >> 8) & 0xff;
1353        uint32_t  sR = s & 0xff;
1354
1355        sR = (sR*m_r) >> (8 - 5);
1356        sG = (sG*m_g) >> (8 - 6);
1357        sB = (sB*m_b) >> (8 - 5);
1358
1359        /* Now do a normal blend */
1360        int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
1361        int f = 0x100 - m_a;
1362        uint16_t d = *dst;
1363        int dR = (d>>11)&0x1f;
1364        int dG = (d>>5)&0x3f;
1365        int dB = (d)&0x1f;
1366        sR = (sR + f*dR + threshold)>>8;
1367        sG = (sG + f*dG + threshold)>>8;
1368        sB = (sB + f*dB + threshold)>>8;
1369        if (sR > 0x1f) sR = 0x1f;
1370        if (sG > 0x3f) sG = 0x3f;
1371        if (sB > 0x1f) sB = 0x1f;
1372        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
1373    }
1374};
1375
1376/* Same as above, but source is 16bit rgb565 */
1377struct blender_16to16_modulate : blender_modulate {
1378    blender_16to16_modulate(const context_t* c) {
1379        init(c);
1380    }
1381    void write(uint16_t s16, uint16_t* dst) {
1382        uint32_t  s = s16;
1383
1384        uint32_t  sR = s >> 11;
1385        uint32_t  sG = (s >> 5) & 0x3f;
1386        uint32_t  sB = s & 0x1f;
1387
1388        sR = (sR*m_r);
1389        sG = (sG*m_g);
1390        sB = (sB*m_b);
1391
1392        int f = 0x100 - m_a;
1393        uint16_t d = *dst;
1394        int dR = (d>>11)&0x1f;
1395        int dG = (d>>5)&0x3f;
1396        int dB = (d)&0x1f;
1397        sR = (sR + f*dR)>>8;
1398        sG = (sG + f*dG)>>8;
1399        sB = (sB + f*dB)>>8;
1400        *dst = uint16_t((sR<<11)|(sG<<5)|sB);
1401    }
1402};
1403
1404/* This is used to iterate over a 16-bit destination color buffer.
1405 * Usage is:
1406 *
1407 *   dst_iterator16  di(context);
1408 *   while (di.count--) {
1409 *       <do stuff with dest pixel at di.dst>
1410 *       di.dst++;
1411 *   }
1412 */
1413struct dst_iterator16 {
1414    dst_iterator16(const context_t* c) {
1415        const int x = c->iterators.xl;
1416        const int width = c->iterators.xr - x;
1417        const int32_t y = c->iterators.y;
1418        const surface_t* cb = &(c->state.buffers.color);
1419        count = width;
1420        dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
1421    }
1422    int        count;
1423    uint16_t*  dst;
1424};
1425
1426
1427static void scanline_t32cb16_clamp(context_t* c)
1428{
1429    dst_iterator16  di(c);
1430
1431    if (is_context_horizontal(c)) {
1432        /* Special case for simple horizontal scaling */
1433        horz_clamp_iterator32 ci(c);
1434        while (di.count--) {
1435            uint32_t s = ci.get_pixel32();
1436            *di.dst++ = convertAbgr8888ToRgb565(s);
1437        }
1438    } else {
1439        /* General case */
1440        clamp_iterator ci(c);
1441        while (di.count--) {
1442            uint32_t s = ci.get_pixel32();
1443            *di.dst++ = convertAbgr8888ToRgb565(s);
1444        }
1445    }
1446}
1447
1448static void scanline_t32cb16_dither(context_t* c)
1449{
1450    horz_iterator32 si(c);
1451    dst_iterator16  di(c);
1452    ditherer        dither(c);
1453
1454    while (di.count--) {
1455        uint32_t s = si.get_pixel32();
1456        *di.dst++ = dither.abgr8888ToRgb565(s);
1457    }
1458}
1459
1460static void scanline_t32cb16_clamp_dither(context_t* c)
1461{
1462    dst_iterator16  di(c);
1463    ditherer        dither(c);
1464
1465    if (is_context_horizontal(c)) {
1466        /* Special case for simple horizontal scaling */
1467        horz_clamp_iterator32 ci(c);
1468        while (di.count--) {
1469            uint32_t s = ci.get_pixel32();
1470            *di.dst++ = dither.abgr8888ToRgb565(s);
1471        }
1472    } else {
1473        /* General case */
1474        clamp_iterator ci(c);
1475        while (di.count--) {
1476            uint32_t s = ci.get_pixel32();
1477            *di.dst++ = dither.abgr8888ToRgb565(s);
1478        }
1479    }
1480}
1481
1482static void scanline_t32cb16blend_dither(context_t* c)
1483{
1484    dst_iterator16 di(c);
1485    ditherer       dither(c);
1486    blender_32to16 bl(c);
1487    horz_iterator32  hi(c);
1488    while (di.count--) {
1489        uint32_t s = hi.get_pixel32();
1490        bl.write(s, di.dst, dither);
1491        di.dst++;
1492    }
1493}
1494
1495static void scanline_t32cb16blend_clamp(context_t* c)
1496{
1497    dst_iterator16  di(c);
1498    blender_32to16  bl(c);
1499
1500    if (is_context_horizontal(c)) {
1501        horz_clamp_iterator32 ci(c);
1502        while (di.count--) {
1503            uint32_t s = ci.get_pixel32();
1504            bl.write(s, di.dst);
1505            di.dst++;
1506        }
1507    } else {
1508        clamp_iterator ci(c);
1509        while (di.count--) {
1510            uint32_t s = ci.get_pixel32();
1511            bl.write(s, di.dst);
1512            di.dst++;
1513        }
1514    }
1515}
1516
1517static void scanline_t32cb16blend_clamp_dither(context_t* c)
1518{
1519    dst_iterator16 di(c);
1520    ditherer       dither(c);
1521    blender_32to16 bl(c);
1522
1523    clamp_iterator ci(c);
1524    while (di.count--) {
1525        uint32_t s = ci.get_pixel32();
1526        bl.write(s, di.dst, dither);
1527        di.dst++;
1528    }
1529}
1530
1531void scanline_t32cb16blend_clamp_mod(context_t* c)
1532{
1533    dst_iterator16 di(c);
1534    blender_32to16_modulate bl(c);
1535
1536    clamp_iterator ci(c);
1537    while (di.count--) {
1538        uint32_t s = ci.get_pixel32();
1539        bl.write(s, di.dst);
1540        di.dst++;
1541    }
1542}
1543
1544void scanline_t32cb16blend_clamp_mod_dither(context_t* c)
1545{
1546    dst_iterator16 di(c);
1547    blender_32to16_modulate bl(c);
1548    ditherer dither(c);
1549
1550    clamp_iterator ci(c);
1551    while (di.count--) {
1552        uint32_t s = ci.get_pixel32();
1553        bl.write(s, di.dst, dither);
1554        di.dst++;
1555    }
1556}
1557
1558/* Variant of scanline_t32cb16blend_clamp_mod with a xRGB texture */
1559void scanline_x32cb16blend_clamp_mod(context_t* c)
1560{
1561    dst_iterator16 di(c);
1562    blender_x32to16_modulate  bl(c);
1563
1564    clamp_iterator ci(c);
1565    while (di.count--) {
1566        uint32_t s = ci.get_pixel32();
1567        bl.write(s, di.dst);
1568        di.dst++;
1569    }
1570}
1571
1572void scanline_x32cb16blend_clamp_mod_dither(context_t* c)
1573{
1574    dst_iterator16 di(c);
1575    blender_x32to16_modulate  bl(c);
1576    ditherer dither(c);
1577
1578    clamp_iterator ci(c);
1579    while (di.count--) {
1580        uint32_t s = ci.get_pixel32();
1581        bl.write(s, di.dst, dither);
1582        di.dst++;
1583    }
1584}
1585
1586void scanline_t16cb16_clamp(context_t* c)
1587{
1588    dst_iterator16  di(c);
1589
1590    /* Special case for simple horizontal scaling */
1591    if (is_context_horizontal(c)) {
1592        horz_clamp_iterator16 ci(c);
1593        while (di.count--) {
1594            *di.dst++ = ci.get_pixel16();
1595        }
1596    } else {
1597        clamp_iterator ci(c);
1598        while (di.count--) {
1599            *di.dst++ = ci.get_pixel16();
1600        }
1601    }
1602}
1603
1604
1605
1606template <typename T, typename U>
1607static inline __attribute__((const))
1608T interpolate(int y, T v0, U dvdx, U dvdy) {
1609    // interpolates in pixel's centers
1610    // v = v0 + (y + 0.5) * dvdy + (0.5 * dvdx)
1611    return (y * dvdy) + (v0 + ((dvdy + dvdx) >> 1));
1612}
1613
1614// ----------------------------------------------------------------------------
1615#if 0
1616#pragma mark -
1617#endif
1618
1619void init_y(context_t* c, int32_t ys)
1620{
1621    const uint32_t enables = c->state.enables;
1622
1623    // compute iterators...
1624    iterators_t& ci = c->iterators;
1625
1626    // sample in the center
1627    ci.y = ys;
1628
1629    if (enables & (GGL_ENABLE_DEPTH_TEST|GGL_ENABLE_W|GGL_ENABLE_FOG)) {
1630        ci.ydzdy = interpolate(ys, c->shade.z0, c->shade.dzdx, c->shade.dzdy);
1631        ci.ydwdy = interpolate(ys, c->shade.w0, c->shade.dwdx, c->shade.dwdy);
1632        ci.ydfdy = interpolate(ys, c->shade.f0, c->shade.dfdx, c->shade.dfdy);
1633    }
1634
1635    if (ggl_unlikely(enables & GGL_ENABLE_SMOOTH)) {
1636        ci.ydrdy = interpolate(ys, c->shade.r0, c->shade.drdx, c->shade.drdy);
1637        ci.ydgdy = interpolate(ys, c->shade.g0, c->shade.dgdx, c->shade.dgdy);
1638        ci.ydbdy = interpolate(ys, c->shade.b0, c->shade.dbdx, c->shade.dbdy);
1639        ci.ydady = interpolate(ys, c->shade.a0, c->shade.dadx, c->shade.dady);
1640        c->step_y = step_y__smooth;
1641    } else {
1642        ci.ydrdy = c->shade.r0;
1643        ci.ydgdy = c->shade.g0;
1644        ci.ydbdy = c->shade.b0;
1645        ci.ydady = c->shade.a0;
1646        // XXX: do only if needed, or make sure this is fast
1647        c->packed = ggl_pack_color(c, c->state.buffers.color.format,
1648                ci.ydrdy, ci.ydgdy, ci.ydbdy, ci.ydady);
1649        c->packed8888 = ggl_pack_color(c, GGL_PIXEL_FORMAT_RGBA_8888,
1650                ci.ydrdy, ci.ydgdy, ci.ydbdy, ci.ydady);
1651    }
1652
1653    // initialize the variables we need in the shader
1654    generated_vars_t& gen = c->generated_vars;
1655    gen.argb[GGLFormat::ALPHA].c  = ci.ydady;
1656    gen.argb[GGLFormat::ALPHA].dx = c->shade.dadx;
1657    gen.argb[GGLFormat::RED  ].c  = ci.ydrdy;
1658    gen.argb[GGLFormat::RED  ].dx = c->shade.drdx;
1659    gen.argb[GGLFormat::GREEN].c  = ci.ydgdy;
1660    gen.argb[GGLFormat::GREEN].dx = c->shade.dgdx;
1661    gen.argb[GGLFormat::BLUE ].c  = ci.ydbdy;
1662    gen.argb[GGLFormat::BLUE ].dx = c->shade.dbdx;
1663    gen.dzdx = c->shade.dzdx;
1664    gen.f    = ci.ydfdy;
1665    gen.dfdx = c->shade.dfdx;
1666
1667    if (enables & GGL_ENABLE_TMUS) {
1668        for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1669            texture_t& t = c->state.texture[i];
1670            if (!t.enable) continue;
1671
1672            texture_iterators_t& ti = t.iterators;
1673            if (t.s_coord == GGL_ONE_TO_ONE && t.t_coord == GGL_ONE_TO_ONE) {
1674                // we need to set all of these to 0 because in some cases
1675                // step_y__generic() or step_y__tmu() will be used and
1676                // therefore will update dtdy, however, in 1:1 mode
1677                // this is always done by the scanline rasterizer.
1678                ti.dsdx = ti.dsdy = ti.dtdx = ti.dtdy = 0;
1679                ti.ydsdy = t.shade.is0;
1680                ti.ydtdy = t.shade.it0;
1681            } else {
1682                const int adjustSWrap = ((t.s_wrap==GGL_CLAMP)?0:16);
1683                const int adjustTWrap = ((t.t_wrap==GGL_CLAMP)?0:16);
1684                ti.sscale = t.shade.sscale + adjustSWrap;
1685                ti.tscale = t.shade.tscale + adjustTWrap;
1686                if (!(enables & GGL_ENABLE_W)) {
1687                    // S coordinate
1688                    const int32_t sscale = ti.sscale;
1689                    const int32_t sy = interpolate(ys,
1690                            t.shade.is0, t.shade.idsdx, t.shade.idsdy);
1691                    if (sscale>=0) {
1692                        ti.ydsdy= sy            << sscale;
1693                        ti.dsdx = t.shade.idsdx << sscale;
1694                        ti.dsdy = t.shade.idsdy << sscale;
1695                    } else {
1696                        ti.ydsdy= sy            >> -sscale;
1697                        ti.dsdx = t.shade.idsdx >> -sscale;
1698                        ti.dsdy = t.shade.idsdy >> -sscale;
1699                    }
1700                    // T coordinate
1701                    const int32_t tscale = ti.tscale;
1702                    const int32_t ty = interpolate(ys,
1703                            t.shade.it0, t.shade.idtdx, t.shade.idtdy);
1704                    if (tscale>=0) {
1705                        ti.ydtdy= ty            << tscale;
1706                        ti.dtdx = t.shade.idtdx << tscale;
1707                        ti.dtdy = t.shade.idtdy << tscale;
1708                    } else {
1709                        ti.ydtdy= ty            >> -tscale;
1710                        ti.dtdx = t.shade.idtdx >> -tscale;
1711                        ti.dtdy = t.shade.idtdy >> -tscale;
1712                    }
1713                }
1714            }
1715            // mirror for generated code...
1716            generated_tex_vars_t& gen = c->generated_vars.texture[i];
1717            gen.width   = t.surface.width;
1718            gen.height  = t.surface.height;
1719            gen.stride  = t.surface.stride;
1720            gen.data    = int32_t(t.surface.data);
1721            gen.dsdx = ti.dsdx;
1722            gen.dtdx = ti.dtdx;
1723        }
1724    }
1725
1726    // choose the y-stepper
1727    c->step_y = step_y__nop;
1728    if (enables & GGL_ENABLE_FOG) {
1729        c->step_y = step_y__generic;
1730    } else if (enables & GGL_ENABLE_TMUS) {
1731        if (enables & GGL_ENABLE_SMOOTH) {
1732            c->step_y = step_y__generic;
1733        } else if (enables & GGL_ENABLE_W) {
1734            c->step_y = step_y__w;
1735        } else {
1736            c->step_y = step_y__tmu;
1737        }
1738    } else {
1739        if (enables & GGL_ENABLE_SMOOTH) {
1740            c->step_y = step_y__smooth;
1741        }
1742    }
1743
1744    // choose the rectangle blitter
1745    c->rect = rect_generic;
1746    if ((c->step_y == step_y__nop) &&
1747        (c->scanline == scanline_memcpy))
1748    {
1749        c->rect = rect_memcpy;
1750    }
1751}
1752
1753void init_y_packed(context_t* c, int32_t y0)
1754{
1755    uint8_t f = c->state.buffers.color.format;
1756    c->packed = ggl_pack_color(c, f,
1757            c->shade.r0, c->shade.g0, c->shade.b0, c->shade.a0);
1758    c->packed8888 = ggl_pack_color(c, GGL_PIXEL_FORMAT_RGBA_8888,
1759            c->shade.r0, c->shade.g0, c->shade.b0, c->shade.a0);
1760    c->iterators.y = y0;
1761    c->step_y = step_y__nop;
1762    // choose the rectangle blitter
1763    c->rect = rect_generic;
1764    if (c->scanline == scanline_memcpy) {
1765        c->rect = rect_memcpy;
1766    }
1767}
1768
1769void init_y_noop(context_t* c, int32_t y0)
1770{
1771    c->iterators.y = y0;
1772    c->step_y = step_y__nop;
1773    // choose the rectangle blitter
1774    c->rect = rect_generic;
1775    if (c->scanline == scanline_memcpy) {
1776        c->rect = rect_memcpy;
1777    }
1778}
1779
1780void init_y_error(context_t* c, int32_t y0)
1781{
1782    // woooops, shoud never happen,
1783    // fail gracefully (don't display anything)
1784    init_y_noop(c, y0);
1785    ALOGE("color-buffer has an invalid format!");
1786}
1787
1788// ----------------------------------------------------------------------------
1789#if 0
1790#pragma mark -
1791#endif
1792
1793void step_y__generic(context_t* c)
1794{
1795    const uint32_t enables = c->state.enables;
1796
1797    // iterate...
1798    iterators_t& ci = c->iterators;
1799    ci.y += 1;
1800
1801    if (enables & GGL_ENABLE_SMOOTH) {
1802        ci.ydrdy += c->shade.drdy;
1803        ci.ydgdy += c->shade.dgdy;
1804        ci.ydbdy += c->shade.dbdy;
1805        ci.ydady += c->shade.dady;
1806    }
1807
1808    const uint32_t mask =
1809            GGL_ENABLE_DEPTH_TEST |
1810            GGL_ENABLE_W |
1811            GGL_ENABLE_FOG;
1812    if (enables & mask) {
1813        ci.ydzdy += c->shade.dzdy;
1814        ci.ydwdy += c->shade.dwdy;
1815        ci.ydfdy += c->shade.dfdy;
1816    }
1817
1818    if ((enables & GGL_ENABLE_TMUS) && (!(enables & GGL_ENABLE_W))) {
1819        for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1820            if (c->state.texture[i].enable) {
1821                texture_iterators_t& ti = c->state.texture[i].iterators;
1822                ti.ydsdy += ti.dsdy;
1823                ti.ydtdy += ti.dtdy;
1824            }
1825        }
1826    }
1827}
1828
1829void step_y__nop(context_t* c)
1830{
1831    c->iterators.y += 1;
1832    c->iterators.ydzdy += c->shade.dzdy;
1833}
1834
1835void step_y__smooth(context_t* c)
1836{
1837    iterators_t& ci = c->iterators;
1838    ci.y += 1;
1839    ci.ydrdy += c->shade.drdy;
1840    ci.ydgdy += c->shade.dgdy;
1841    ci.ydbdy += c->shade.dbdy;
1842    ci.ydady += c->shade.dady;
1843    ci.ydzdy += c->shade.dzdy;
1844}
1845
1846void step_y__w(context_t* c)
1847{
1848    iterators_t& ci = c->iterators;
1849    ci.y += 1;
1850    ci.ydzdy += c->shade.dzdy;
1851    ci.ydwdy += c->shade.dwdy;
1852}
1853
1854void step_y__tmu(context_t* c)
1855{
1856    iterators_t& ci = c->iterators;
1857    ci.y += 1;
1858    ci.ydzdy += c->shade.dzdy;
1859    for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1860        if (c->state.texture[i].enable) {
1861            texture_iterators_t& ti = c->state.texture[i].iterators;
1862            ti.ydsdy += ti.dsdy;
1863            ti.ydtdy += ti.dtdy;
1864        }
1865    }
1866}
1867
1868// ----------------------------------------------------------------------------
1869#if 0
1870#pragma mark -
1871#endif
1872
1873void scanline_perspective(context_t* c)
1874{
1875    struct {
1876        union {
1877            struct {
1878                int32_t s, sq;
1879                int32_t t, tq;
1880            };
1881            struct {
1882                int32_t v, q;
1883            } st[2];
1884        };
1885    } tc[GGL_TEXTURE_UNIT_COUNT] __attribute__((aligned(16)));
1886
1887    // XXX: we should have a special case when dwdx = 0
1888
1889    // 32 pixels spans works okay. 16 is a lot better,
1890    // but hey, it's a software renderer...
1891    const uint32_t SPAN_BITS = 5;
1892    const uint32_t ys = c->iterators.y;
1893    const uint32_t xs = c->iterators.xl;
1894    const uint32_t x1 = c->iterators.xr;
1895	const uint32_t xc = x1 - xs;
1896    uint32_t remainder = xc & ((1<<SPAN_BITS)-1);
1897    uint32_t numSpans = xc >> SPAN_BITS;
1898
1899    const iterators_t& ci = c->iterators;
1900    int32_t w0 = (xs * c->shade.dwdx) + ci.ydwdy;
1901    int32_t q0 = gglRecipQ(w0, 30);
1902    const int iwscale = 32 - gglClz(q0);
1903
1904    const int32_t dwdx = c->shade.dwdx << SPAN_BITS;
1905    int32_t xl = c->iterators.xl;
1906
1907    // We process s & t with a loop to reduce the code size
1908    // (and i-cache pressure).
1909
1910    for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1911        const texture_t& tmu = c->state.texture[i];
1912        if (!tmu.enable) continue;
1913        int32_t s =   tmu.shade.is0 +
1914                     (tmu.shade.idsdy * ys) + (tmu.shade.idsdx * xs) +
1915                     ((tmu.shade.idsdx + tmu.shade.idsdy)>>1);
1916        int32_t t =   tmu.shade.it0 +
1917                     (tmu.shade.idtdy * ys) + (tmu.shade.idtdx * xs) +
1918                     ((tmu.shade.idtdx + tmu.shade.idtdy)>>1);
1919        tc[i].s  = s;
1920        tc[i].t  = t;
1921        tc[i].sq = gglMulx(s, q0, iwscale);
1922        tc[i].tq = gglMulx(t, q0, iwscale);
1923    }
1924
1925    int32_t span = 0;
1926    do {
1927        int32_t w1;
1928        if (ggl_likely(numSpans)) {
1929            w1 = w0 + dwdx;
1930        } else {
1931            if (remainder) {
1932                // finish off the scanline...
1933                span = remainder;
1934                w1 = (c->shade.dwdx * span) + w0;
1935            } else {
1936                break;
1937            }
1938        }
1939        int32_t q1 = gglRecipQ(w1, 30);
1940        for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1941            texture_t& tmu = c->state.texture[i];
1942            if (!tmu.enable) continue;
1943            texture_iterators_t& ti = tmu.iterators;
1944
1945            for (int j=0 ; j<2 ; j++) {
1946                int32_t v = tc[i].st[j].v;
1947                if (span)   v += (tmu.shade.st[j].dx)*span;
1948                else        v += (tmu.shade.st[j].dx)<<SPAN_BITS;
1949                const int32_t v0 = tc[i].st[j].q;
1950                const int32_t v1 = gglMulx(v, q1, iwscale);
1951                int32_t dvdx = v1 - v0;
1952                if (span)   dvdx /= span;
1953                else        dvdx >>= SPAN_BITS;
1954                tc[i].st[j].v = v;
1955                tc[i].st[j].q = v1;
1956
1957                const int scale = ti.st[j].scale + (iwscale - 30);
1958                if (scale >= 0) {
1959                    ti.st[j].ydvdy = v0   << scale;
1960                    ti.st[j].dvdx  = dvdx << scale;
1961                } else {
1962                    ti.st[j].ydvdy = v0   >> -scale;
1963                    ti.st[j].dvdx  = dvdx >> -scale;
1964                }
1965            }
1966            generated_tex_vars_t& gen = c->generated_vars.texture[i];
1967            gen.dsdx = ti.st[0].dvdx;
1968            gen.dtdx = ti.st[1].dvdx;
1969        }
1970        c->iterators.xl = xl;
1971        c->iterators.xr = xl = xl + (span ? span : (1<<SPAN_BITS));
1972        w0 = w1;
1973        q0 = q1;
1974        c->span(c);
1975    } while(numSpans--);
1976}
1977
1978void scanline_perspective_single(context_t* c)
1979{
1980    // 32 pixels spans works okay. 16 is a lot better,
1981    // but hey, it's a software renderer...
1982    const uint32_t SPAN_BITS = 5;
1983    const uint32_t ys = c->iterators.y;
1984    const uint32_t xs = c->iterators.xl;
1985    const uint32_t x1 = c->iterators.xr;
1986	const uint32_t xc = x1 - xs;
1987
1988    const iterators_t& ci = c->iterators;
1989    int32_t w = (xs * c->shade.dwdx) + ci.ydwdy;
1990    int32_t iw = gglRecipQ(w, 30);
1991    const int iwscale = 32 - gglClz(iw);
1992
1993    const int i = 31 - gglClz(c->state.enabled_tmu);
1994    generated_tex_vars_t& gen = c->generated_vars.texture[i];
1995    texture_t& tmu = c->state.texture[i];
1996    texture_iterators_t& ti = tmu.iterators;
1997    const int sscale = ti.sscale + (iwscale - 30);
1998    const int tscale = ti.tscale + (iwscale - 30);
1999    int32_t s =   tmu.shade.is0 +
2000                 (tmu.shade.idsdy * ys) + (tmu.shade.idsdx * xs) +
2001                 ((tmu.shade.idsdx + tmu.shade.idsdy)>>1);
2002    int32_t t =   tmu.shade.it0 +
2003                 (tmu.shade.idtdy * ys) + (tmu.shade.idtdx * xs) +
2004                 ((tmu.shade.idtdx + tmu.shade.idtdy)>>1);
2005    int32_t s0 = gglMulx(s, iw, iwscale);
2006    int32_t t0 = gglMulx(t, iw, iwscale);
2007    int32_t xl = c->iterators.xl;
2008
2009    int32_t sq, tq, dsdx, dtdx;
2010    int32_t premainder = xc & ((1<<SPAN_BITS)-1);
2011    uint32_t numSpans = xc >> SPAN_BITS;
2012    if (c->shade.dwdx == 0) {
2013        // XXX: we could choose to do this if the error is small enough
2014        numSpans = 0;
2015        premainder = xc;
2016        goto no_perspective;
2017    }
2018
2019    if (premainder) {
2020        w += c->shade.dwdx   * premainder;
2021        iw = gglRecipQ(w, 30);
2022no_perspective:
2023        s += tmu.shade.idsdx * premainder;
2024        t += tmu.shade.idtdx * premainder;
2025        sq = gglMulx(s, iw, iwscale);
2026        tq = gglMulx(t, iw, iwscale);
2027        dsdx = (sq - s0) / premainder;
2028        dtdx = (tq - t0) / premainder;
2029        c->iterators.xl = xl;
2030        c->iterators.xr = xl = xl + premainder;
2031        goto finish;
2032    }
2033
2034    while (numSpans--) {
2035        w += c->shade.dwdx   << SPAN_BITS;
2036        s += tmu.shade.idsdx << SPAN_BITS;
2037        t += tmu.shade.idtdx << SPAN_BITS;
2038        iw = gglRecipQ(w, 30);
2039        sq = gglMulx(s, iw, iwscale);
2040        tq = gglMulx(t, iw, iwscale);
2041        dsdx = (sq - s0) >> SPAN_BITS;
2042        dtdx = (tq - t0) >> SPAN_BITS;
2043        c->iterators.xl = xl;
2044        c->iterators.xr = xl = xl + (1<<SPAN_BITS);
2045finish:
2046        if (sscale >= 0) {
2047            ti.ydsdy = s0   << sscale;
2048            ti.dsdx  = dsdx << sscale;
2049        } else {
2050            ti.ydsdy = s0   >>-sscale;
2051            ti.dsdx  = dsdx >>-sscale;
2052        }
2053        if (tscale >= 0) {
2054            ti.ydtdy = t0   << tscale;
2055            ti.dtdx  = dtdx << tscale;
2056        } else {
2057            ti.ydtdy = t0   >>-tscale;
2058            ti.dtdx  = dtdx >>-tscale;
2059        }
2060        s0 = sq;
2061        t0 = tq;
2062        gen.dsdx = ti.dsdx;
2063        gen.dtdx = ti.dtdx;
2064        c->span(c);
2065    }
2066}
2067
2068// ----------------------------------------------------------------------------
2069
2070void scanline_col32cb16blend(context_t* c)
2071{
2072    int32_t x = c->iterators.xl;
2073    size_t ct = c->iterators.xr - x;
2074    int32_t y = c->iterators.y;
2075    surface_t* cb = &(c->state.buffers.color);
2076    union {
2077        uint16_t* dst;
2078        uint32_t* dst32;
2079    };
2080    dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
2081
2082#if ((ANDROID_CODEGEN >= ANDROID_CODEGEN_ASM) && defined(__arm__))
2083#if defined(__ARM_HAVE_NEON) && BYTE_ORDER == LITTLE_ENDIAN
2084    scanline_col32cb16blend_neon(dst, &(c->packed8888), ct);
2085#else  // defined(__ARM_HAVE_NEON) && BYTE_ORDER == LITTLE_ENDIAN
2086    scanline_col32cb16blend_arm(dst, GGL_RGBA_TO_HOST(c->packed8888), ct);
2087#endif // defined(__ARM_HAVE_NEON) && BYTE_ORDER == LITTLE_ENDIAN
2088#else
2089    uint32_t s = GGL_RGBA_TO_HOST(c->packed8888);
2090    int sA = (s>>24);
2091    int f = 0x100 - (sA + (sA>>7));
2092    while (ct--) {
2093        uint16_t d = *dst;
2094        int dR = (d>>11)&0x1f;
2095        int dG = (d>>5)&0x3f;
2096        int dB = (d)&0x1f;
2097        int sR = (s >> (   3))&0x1F;
2098        int sG = (s >> ( 8+2))&0x3F;
2099        int sB = (s >> (16+3))&0x1F;
2100        sR += (f*dR)>>8;
2101        sG += (f*dG)>>8;
2102        sB += (f*dB)>>8;
2103        *dst++ = uint16_t((sR<<11)|(sG<<5)|sB);
2104    }
2105#endif
2106
2107}
2108
2109void scanline_t32cb16(context_t* c)
2110{
2111    int32_t x = c->iterators.xl;
2112    size_t ct = c->iterators.xr - x;
2113    int32_t y = c->iterators.y;
2114    surface_t* cb = &(c->state.buffers.color);
2115    union {
2116        uint16_t* dst;
2117        uint32_t* dst32;
2118    };
2119    dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
2120
2121    surface_t* tex = &(c->state.texture[0].surface);
2122    const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
2123    const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
2124    uint32_t *src = reinterpret_cast<uint32_t*>(tex->data)+(u+(tex->stride*v));
2125    int sR, sG, sB;
2126    uint32_t s, d;
2127
2128    if (ct==1 || uint32_t(dst)&2) {
2129last_one:
2130        s = GGL_RGBA_TO_HOST( *src++ );
2131        *dst++ = convertAbgr8888ToRgb565(s);
2132        ct--;
2133    }
2134
2135    while (ct >= 2) {
2136#if BYTE_ORDER == BIG_ENDIAN
2137        s = GGL_RGBA_TO_HOST( *src++ );
2138        d = convertAbgr8888ToRgb565_hi16(s);
2139
2140        s = GGL_RGBA_TO_HOST( *src++ );
2141        d |= convertAbgr8888ToRgb565(s);
2142#else
2143        s = GGL_RGBA_TO_HOST( *src++ );
2144        d = convertAbgr8888ToRgb565(s);
2145
2146        s = GGL_RGBA_TO_HOST( *src++ );
2147        d |= convertAbgr8888ToRgb565(s) << 16;
2148#endif
2149        *dst32++ = d;
2150        ct -= 2;
2151    }
2152
2153    if (ct > 0) {
2154        goto last_one;
2155    }
2156}
2157
2158void scanline_t32cb16blend(context_t* c)
2159{
2160#if ((ANDROID_CODEGEN >= ANDROID_CODEGEN_ASM) && (defined(__arm__) || defined(__mips)))
2161    int32_t x = c->iterators.xl;
2162    size_t ct = c->iterators.xr - x;
2163    int32_t y = c->iterators.y;
2164    surface_t* cb = &(c->state.buffers.color);
2165    uint16_t* dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
2166
2167    surface_t* tex = &(c->state.texture[0].surface);
2168    const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
2169    const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
2170    uint32_t *src = reinterpret_cast<uint32_t*>(tex->data)+(u+(tex->stride*v));
2171
2172#ifdef __arm__
2173    scanline_t32cb16blend_arm(dst, src, ct);
2174#else
2175    scanline_t32cb16blend_mips(dst, src, ct);
2176#endif
2177#else
2178    dst_iterator16  di(c);
2179    horz_iterator32  hi(c);
2180    blender_32to16  bl(c);
2181    while (di.count--) {
2182        uint32_t s = hi.get_pixel32();
2183        bl.write(s, di.dst);
2184        di.dst++;
2185    }
2186#endif
2187}
2188
2189void scanline_t32cb16blend_srca(context_t* c)
2190{
2191    dst_iterator16  di(c);
2192    horz_iterator32  hi(c);
2193    blender_32to16_srcA  blender(c);
2194
2195    while (di.count--) {
2196        uint32_t s = hi.get_pixel32();
2197        blender.write(s,di.dst);
2198        di.dst++;
2199    }
2200}
2201
2202void scanline_t16cb16blend_clamp_mod(context_t* c)
2203{
2204    const int a = c->iterators.ydady >> (GGL_COLOR_BITS-8);
2205    if (a == 0) {
2206        return;
2207    }
2208
2209    if (a == 255) {
2210        scanline_t16cb16_clamp(c);
2211        return;
2212    }
2213
2214    dst_iterator16  di(c);
2215    blender_16to16_modulate  blender(c);
2216    clamp_iterator  ci(c);
2217
2218    while (di.count--) {
2219        uint16_t s = ci.get_pixel16();
2220        blender.write(s, di.dst);
2221        di.dst++;
2222    }
2223}
2224
2225void scanline_memcpy(context_t* c)
2226{
2227    int32_t x = c->iterators.xl;
2228    size_t ct = c->iterators.xr - x;
2229    int32_t y = c->iterators.y;
2230    surface_t* cb = &(c->state.buffers.color);
2231    const GGLFormat* fp = &(c->formats[cb->format]);
2232    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
2233                            (x + (cb->stride * y)) * fp->size;
2234
2235    surface_t* tex = &(c->state.texture[0].surface);
2236    const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
2237    const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
2238    uint8_t *src = reinterpret_cast<uint8_t*>(tex->data) +
2239                            (u + (tex->stride * v)) * fp->size;
2240
2241    const size_t size = ct * fp->size;
2242    memcpy(dst, src, size);
2243}
2244
2245void scanline_memset8(context_t* c)
2246{
2247    int32_t x = c->iterators.xl;
2248    size_t ct = c->iterators.xr - x;
2249    int32_t y = c->iterators.y;
2250    surface_t* cb = &(c->state.buffers.color);
2251    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) + (x+(cb->stride*y));
2252    uint32_t packed = c->packed;
2253    memset(dst, packed, ct);
2254}
2255
2256void scanline_memset16(context_t* c)
2257{
2258    int32_t x = c->iterators.xl;
2259    size_t ct = c->iterators.xr - x;
2260    int32_t y = c->iterators.y;
2261    surface_t* cb = &(c->state.buffers.color);
2262    uint16_t* dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
2263    uint32_t packed = c->packed;
2264    android_memset16(dst, packed, ct*2);
2265}
2266
2267void scanline_memset32(context_t* c)
2268{
2269    int32_t x = c->iterators.xl;
2270    size_t ct = c->iterators.xr - x;
2271    int32_t y = c->iterators.y;
2272    surface_t* cb = &(c->state.buffers.color);
2273    uint32_t* dst = reinterpret_cast<uint32_t*>(cb->data) + (x+(cb->stride*y));
2274    uint32_t packed = GGL_HOST_TO_RGBA(c->packed);
2275    android_memset32(dst, packed, ct*4);
2276}
2277
2278void scanline_clear(context_t* c)
2279{
2280    int32_t x = c->iterators.xl;
2281    size_t ct = c->iterators.xr - x;
2282    int32_t y = c->iterators.y;
2283    surface_t* cb = &(c->state.buffers.color);
2284    const GGLFormat* fp = &(c->formats[cb->format]);
2285    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
2286                            (x + (cb->stride * y)) * fp->size;
2287    const size_t size = ct * fp->size;
2288    memset(dst, 0, size);
2289}
2290
2291void scanline_set(context_t* c)
2292{
2293    int32_t x = c->iterators.xl;
2294    size_t ct = c->iterators.xr - x;
2295    int32_t y = c->iterators.y;
2296    surface_t* cb = &(c->state.buffers.color);
2297    const GGLFormat* fp = &(c->formats[cb->format]);
2298    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
2299                            (x + (cb->stride * y)) * fp->size;
2300    const size_t size = ct * fp->size;
2301    memset(dst, 0xFF, size);
2302}
2303
2304void scanline_noop(context_t* c)
2305{
2306}
2307
2308void rect_generic(context_t* c, size_t yc)
2309{
2310    do {
2311        c->scanline(c);
2312        c->step_y(c);
2313    } while (--yc);
2314}
2315
2316void rect_memcpy(context_t* c, size_t yc)
2317{
2318    int32_t x = c->iterators.xl;
2319    size_t ct = c->iterators.xr - x;
2320    int32_t y = c->iterators.y;
2321    surface_t* cb = &(c->state.buffers.color);
2322    const GGLFormat* fp = &(c->formats[cb->format]);
2323    uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
2324                            (x + (cb->stride * y)) * fp->size;
2325
2326    surface_t* tex = &(c->state.texture[0].surface);
2327    const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
2328    const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
2329    uint8_t *src = reinterpret_cast<uint8_t*>(tex->data) +
2330                            (u + (tex->stride * v)) * fp->size;
2331
2332    if (cb->stride == tex->stride && ct == size_t(cb->stride)) {
2333        memcpy(dst, src, ct * fp->size * yc);
2334    } else {
2335        const size_t size = ct * fp->size;
2336        const size_t dbpr = cb->stride  * fp->size;
2337        const size_t sbpr = tex->stride * fp->size;
2338        do {
2339            memcpy(dst, src, size);
2340            dst += dbpr;
2341            src += sbpr;
2342        } while (--yc);
2343    }
2344}
2345// ----------------------------------------------------------------------------
2346}; // namespace android
2347
2348