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