1 /* libs/pixelflinger/scanline.cpp
2 **
3 ** Copyright 2006, The Android Open Source Project
4 **
5 ** Licensed under the Apache License, Version 2.0 (the "License");
6 ** you may not use this file except in compliance with the License.
7 ** You may obtain a copy of the License at
8 **
9 ** http://www.apache.org/licenses/LICENSE-2.0
10 **
11 ** Unless required by applicable law or agreed to in writing, software
12 ** distributed under the License is distributed on an "AS IS" BASIS,
13 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 ** See the License for the specific language governing permissions and
15 ** limitations under the License.
16 */
17
18
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
61 #define ASSEMBLY_SCRATCH_SIZE 2048
62
63 // ----------------------------------------------------------------------------
64 namespace android {
65 // ----------------------------------------------------------------------------
66
67 static void init_y(context_t*, int32_t);
68 static void init_y_noop(context_t*, int32_t);
69 static void init_y_packed(context_t*, int32_t);
70 static void init_y_error(context_t*, int32_t);
71
72 static void step_y__generic(context_t* c);
73 static void step_y__nop(context_t*);
74 static void step_y__smooth(context_t* c);
75 static void step_y__tmu(context_t* c);
76 static void step_y__w(context_t* c);
77
78 static void scanline(context_t* c);
79 static void scanline_perspective(context_t* c);
80 static void scanline_perspective_single(context_t* c);
81 static void scanline_t32cb16blend(context_t* c);
82 static void scanline_t32cb16(context_t* c);
83 static void scanline_memcpy(context_t* c);
84 static void scanline_memset8(context_t* c);
85 static void scanline_memset16(context_t* c);
86 static void scanline_memset32(context_t* c);
87 static void scanline_noop(context_t* c);
88 static void scanline_set(context_t* c);
89 static void scanline_clear(context_t* c);
90
91 static void rect_generic(context_t* c, size_t yc);
92 static void rect_memcpy(context_t* c, size_t yc);
93
94 extern "C" void scanline_t32cb16blend_arm(uint16_t*, uint32_t*, size_t);
95 extern "C" void scanline_t32cb16_arm(uint16_t *dst, uint32_t *src, size_t ct);
96
97 // ----------------------------------------------------------------------------
98
99 struct shortcut_t {
100 needs_filter_t filter;
101 const char* desc;
102 void (*scanline)(context_t*);
103 void (*init_y)(context_t*, int32_t);
104 };
105
106 // Keep in sync with needs
107 static shortcut_t shortcuts[] = {
108 { { { 0x03515104, 0x00000077, { 0x00000A01, 0x00000000 } },
109 { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
110 "565 fb, 8888 tx, blend", scanline_t32cb16blend, init_y_noop },
111 { { { 0x03010104, 0x00000077, { 0x00000A01, 0x00000000 } },
112 { 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
113 "565 fb, 8888 tx", scanline_t32cb16, init_y_noop },
114 { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
115 { 0x00000000, 0x00000007, { 0x00000000, 0x00000000 } } },
116 "(nop) alpha test", scanline_noop, init_y_noop },
117 { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
118 { 0x00000000, 0x00000070, { 0x00000000, 0x00000000 } } },
119 "(nop) depth test", scanline_noop, init_y_noop },
120 { { { 0x05000000, 0x00000000, { 0x00000000, 0x00000000 } },
121 { 0x0F000000, 0x00000080, { 0x00000000, 0x00000000 } } },
122 "(nop) logic_op", scanline_noop, init_y_noop },
123 { { { 0xF0000000, 0x00000000, { 0x00000000, 0x00000000 } },
124 { 0xF0000000, 0x00000080, { 0x00000000, 0x00000000 } } },
125 "(nop) color mask", scanline_noop, init_y_noop },
126 { { { 0x0F000000, 0x00000077, { 0x00000000, 0x00000000 } },
127 { 0xFF000000, 0x000000F7, { 0x00000000, 0x00000000 } } },
128 "(set) logic_op", scanline_set, init_y_noop },
129 { { { 0x00000000, 0x00000077, { 0x00000000, 0x00000000 } },
130 { 0xFF000000, 0x000000F7, { 0x00000000, 0x00000000 } } },
131 "(clear) logic_op", scanline_clear, init_y_noop },
132 { { { 0x03000000, 0x00000077, { 0x00000000, 0x00000000 } },
133 { 0xFFFFFF00, 0x000000F7, { 0x00000000, 0x00000000 } } },
134 "(clear) blending 0/0", scanline_clear, init_y_noop },
135 { { { 0x00000000, 0x00000000, { 0x00000000, 0x00000000 } },
136 { 0x0000003F, 0x00000000, { 0x00000000, 0x00000000 } } },
137 "(error) invalid color-buffer format", scanline_noop, init_y_error },
138 };
139 static const needs_filter_t noblend1to1 = {
140 // (disregard dithering, see below)
141 { 0x03010100, 0x00000077, { 0x00000A00, 0x00000000 } },
142 { 0xFFFFFFC0, 0xFFFFFEFF, { 0xFFFFFFC0, 0x0000003F } }
143 };
144 static const needs_filter_t fill16noblend = {
145 { 0x03010100, 0x00000077, { 0x00000000, 0x00000000 } },
146 { 0xFFFFFFC0, 0xFFFFFFFF, { 0x0000003F, 0x0000003F } }
147 };
148
149 // ----------------------------------------------------------------------------
150
151 #if ANDROID_ARM_CODEGEN
152 static CodeCache gCodeCache(12 * 1024);
153
154 class ScanlineAssembly : public Assembly {
155 AssemblyKey<needs_t> mKey;
156 public:
ScanlineAssembly(needs_t needs,size_t size)157 ScanlineAssembly(needs_t needs, size_t size)
158 : Assembly(size), mKey(needs) { }
key() const159 const AssemblyKey<needs_t>& key() const { return mKey; }
160 };
161 #endif
162
163 // ----------------------------------------------------------------------------
164
ggl_init_scanline(context_t * c)165 void ggl_init_scanline(context_t* c)
166 {
167 c->init_y = init_y;
168 c->step_y = step_y__generic;
169 c->scanline = scanline;
170 }
171
ggl_uninit_scanline(context_t * c)172 void ggl_uninit_scanline(context_t* c)
173 {
174 if (c->state.buffers.coverage)
175 free(c->state.buffers.coverage);
176 #if ANDROID_ARM_CODEGEN
177 if (c->scanline_as)
178 c->scanline_as->decStrong(c);
179 #endif
180 }
181
182 // ----------------------------------------------------------------------------
183
pick_scanline(context_t * c)184 static void pick_scanline(context_t* c)
185 {
186 #if (!defined(DEBUG__CODEGEN_ONLY) || (DEBUG__CODEGEN_ONLY == 0))
187
188 #if ANDROID_CODEGEN == ANDROID_CODEGEN_GENERIC
189 c->init_y = init_y;
190 c->step_y = step_y__generic;
191 c->scanline = scanline;
192 return;
193 #endif
194
195 //printf("*** needs [%08lx:%08lx:%08lx:%08lx]\n",
196 // c->state.needs.n, c->state.needs.p,
197 // c->state.needs.t[0], c->state.needs.t[1]);
198
199 // first handle the special case that we cannot test with a filter
200 const uint32_t cb_format = GGL_READ_NEEDS(CB_FORMAT, c->state.needs.n);
201 if (GGL_READ_NEEDS(T_FORMAT, c->state.needs.t[0]) == cb_format) {
202 if (c->state.needs.match(noblend1to1)) {
203 // this will match regardless of dithering state, since both
204 // src and dest have the same format anyway, there is no dithering
205 // to be done.
206 const GGLFormat* f =
207 &(c->formats[GGL_READ_NEEDS(T_FORMAT, c->state.needs.t[0])]);
208 if ((f->components == GGL_RGB) ||
209 (f->components == GGL_RGBA) ||
210 (f->components == GGL_LUMINANCE) ||
211 (f->components == GGL_LUMINANCE_ALPHA))
212 {
213 // format must have all of RGB components
214 // (so the current color doesn't show through)
215 c->scanline = scanline_memcpy;
216 c->init_y = init_y_noop;
217 return;
218 }
219 }
220 }
221
222 if (c->state.needs.match(fill16noblend)) {
223 c->init_y = init_y_packed;
224 switch (c->formats[cb_format].size) {
225 case 1: c->scanline = scanline_memset8; return;
226 case 2: c->scanline = scanline_memset16; return;
227 case 4: c->scanline = scanline_memset32; return;
228 }
229 }
230
231 const int numFilters = sizeof(shortcuts)/sizeof(shortcut_t);
232 for (int i=0 ; i<numFilters ; i++) {
233 if (c->state.needs.match(shortcuts[i].filter)) {
234 c->scanline = shortcuts[i].scanline;
235 c->init_y = shortcuts[i].init_y;
236 return;
237 }
238 }
239
240 #endif // DEBUG__CODEGEN_ONLY
241
242 c->init_y = init_y;
243 c->step_y = step_y__generic;
244
245 #if ANDROID_ARM_CODEGEN
246 // we're going to have to generate some code...
247 // here, generate code for our pixel pipeline
248 const AssemblyKey<needs_t> key(c->state.needs);
249 sp<Assembly> assembly = gCodeCache.lookup(key);
250 if (assembly == 0) {
251 // create a new assembly region
252 sp<ScanlineAssembly> a = new ScanlineAssembly(c->state.needs,
253 ASSEMBLY_SCRATCH_SIZE);
254 // initialize our assembler
255 GGLAssembler assembler( new ARMAssembler(a) );
256 //GGLAssembler assembler(
257 // new ARMAssemblerOptimizer(new ARMAssembler(a)) );
258 // generate the scanline code for the given needs
259 int err = assembler.scanline(c->state.needs, c);
260 if (ggl_likely(!err)) {
261 // finally, cache this assembly
262 err = gCodeCache.cache(a->key(), a);
263 }
264 if (ggl_unlikely(err)) {
265 LOGE("error generating or caching assembly. Reverting to NOP.");
266 c->scanline = scanline_noop;
267 c->init_y = init_y_noop;
268 c->step_y = step_y__nop;
269 return;
270 }
271 assembly = a;
272 }
273
274 // release the previous assembly
275 if (c->scanline_as) {
276 c->scanline_as->decStrong(c);
277 }
278
279 //LOGI("using generated pixel-pipeline");
280 c->scanline_as = assembly.get();
281 c->scanline_as->incStrong(c); // hold on to assembly
282 c->scanline = (void(*)(context_t* c))assembly->base();
283 #else
284 // LOGW("using generic (slow) pixel-pipeline");
285 c->scanline = scanline;
286 #endif
287 }
288
ggl_pick_scanline(context_t * c)289 void ggl_pick_scanline(context_t* c)
290 {
291 pick_scanline(c);
292 if ((c->state.enables & GGL_ENABLE_W) &&
293 (c->state.enables & GGL_ENABLE_TMUS))
294 {
295 c->span = c->scanline;
296 c->scanline = scanline_perspective;
297 if (!(c->state.enabled_tmu & (c->state.enabled_tmu - 1))) {
298 // only one TMU enabled
299 c->scanline = scanline_perspective_single;
300 }
301 }
302 }
303
304 // ----------------------------------------------------------------------------
305
306 static void blending(context_t* c, pixel_t* fragment, pixel_t* fb);
307 static void blend_factor(context_t* c, pixel_t* r, uint32_t factor,
308 const pixel_t* src, const pixel_t* dst);
309 static void rescale(uint32_t& u, uint8_t& su, uint32_t& v, uint8_t& sv);
310
311 #if ANDROID_ARM_CODEGEN && (ANDROID_CODEGEN == ANDROID_CODEGEN_GENERATED)
312
313 // no need to compile the generic-pipeline, it can't be reached
scanline(context_t *)314 void scanline(context_t*)
315 {
316 }
317
318 #else
319
rescale(uint32_t & u,uint8_t & su,uint32_t & v,uint8_t & sv)320 void rescale(uint32_t& u, uint8_t& su, uint32_t& v, uint8_t& sv)
321 {
322 if (su && sv) {
323 if (su > sv) {
324 v = ggl_expand(v, sv, su);
325 sv = su;
326 } else if (su < sv) {
327 u = ggl_expand(u, su, sv);
328 su = sv;
329 }
330 }
331 }
332
blending(context_t * c,pixel_t * fragment,pixel_t * fb)333 void blending(context_t* c, pixel_t* fragment, pixel_t* fb)
334 {
335 rescale(fragment->c[0], fragment->s[0], fb->c[0], fb->s[0]);
336 rescale(fragment->c[1], fragment->s[1], fb->c[1], fb->s[1]);
337 rescale(fragment->c[2], fragment->s[2], fb->c[2], fb->s[2]);
338 rescale(fragment->c[3], fragment->s[3], fb->c[3], fb->s[3]);
339
340 pixel_t sf, df;
341 blend_factor(c, &sf, c->state.blend.src, fragment, fb);
342 blend_factor(c, &df, c->state.blend.dst, fragment, fb);
343
344 fragment->c[1] =
345 gglMulAddx(fragment->c[1], sf.c[1], gglMulx(fb->c[1], df.c[1]));
346 fragment->c[2] =
347 gglMulAddx(fragment->c[2], sf.c[2], gglMulx(fb->c[2], df.c[2]));
348 fragment->c[3] =
349 gglMulAddx(fragment->c[3], sf.c[3], gglMulx(fb->c[3], df.c[3]));
350
351 if (c->state.blend.alpha_separate) {
352 blend_factor(c, &sf, c->state.blend.src_alpha, fragment, fb);
353 blend_factor(c, &df, c->state.blend.dst_alpha, fragment, fb);
354 }
355
356 fragment->c[0] =
357 gglMulAddx(fragment->c[0], sf.c[0], gglMulx(fb->c[0], df.c[0]));
358
359 // clamp to 1.0
360 if (fragment->c[0] >= (1LU<<fragment->s[0]))
361 fragment->c[0] = (1<<fragment->s[0])-1;
362 if (fragment->c[1] >= (1LU<<fragment->s[1]))
363 fragment->c[1] = (1<<fragment->s[1])-1;
364 if (fragment->c[2] >= (1LU<<fragment->s[2]))
365 fragment->c[2] = (1<<fragment->s[2])-1;
366 if (fragment->c[3] >= (1LU<<fragment->s[3]))
367 fragment->c[3] = (1<<fragment->s[3])-1;
368 }
369
blendfactor(uint32_t x,uint32_t size,uint32_t def=0)370 static inline int blendfactor(uint32_t x, uint32_t size, uint32_t def = 0)
371 {
372 if (!size)
373 return def;
374
375 // scale to 16 bits
376 if (size > 16) {
377 x >>= (size - 16);
378 } else if (size < 16) {
379 x = ggl_expand(x, size, 16);
380 }
381 x += x >> 15;
382 return x;
383 }
384
blend_factor(context_t * c,pixel_t * r,uint32_t factor,const pixel_t * src,const pixel_t * dst)385 void blend_factor(context_t* c, pixel_t* r,
386 uint32_t factor, const pixel_t* src, const pixel_t* dst)
387 {
388 switch (factor) {
389 case GGL_ZERO:
390 r->c[1] =
391 r->c[2] =
392 r->c[3] =
393 r->c[0] = 0;
394 break;
395 case GGL_ONE:
396 r->c[1] =
397 r->c[2] =
398 r->c[3] =
399 r->c[0] = FIXED_ONE;
400 break;
401 case GGL_DST_COLOR:
402 r->c[1] = blendfactor(dst->c[1], dst->s[1]);
403 r->c[2] = blendfactor(dst->c[2], dst->s[2]);
404 r->c[3] = blendfactor(dst->c[3], dst->s[3]);
405 r->c[0] = blendfactor(dst->c[0], dst->s[0]);
406 break;
407 case GGL_SRC_COLOR:
408 r->c[1] = blendfactor(src->c[1], src->s[1]);
409 r->c[2] = blendfactor(src->c[2], src->s[2]);
410 r->c[3] = blendfactor(src->c[3], src->s[3]);
411 r->c[0] = blendfactor(src->c[0], src->s[0]);
412 break;
413 case GGL_ONE_MINUS_DST_COLOR:
414 r->c[1] = FIXED_ONE - blendfactor(dst->c[1], dst->s[1]);
415 r->c[2] = FIXED_ONE - blendfactor(dst->c[2], dst->s[2]);
416 r->c[3] = FIXED_ONE - blendfactor(dst->c[3], dst->s[3]);
417 r->c[0] = FIXED_ONE - blendfactor(dst->c[0], dst->s[0]);
418 break;
419 case GGL_ONE_MINUS_SRC_COLOR:
420 r->c[1] = FIXED_ONE - blendfactor(src->c[1], src->s[1]);
421 r->c[2] = FIXED_ONE - blendfactor(src->c[2], src->s[2]);
422 r->c[3] = FIXED_ONE - blendfactor(src->c[3], src->s[3]);
423 r->c[0] = FIXED_ONE - blendfactor(src->c[0], src->s[0]);
424 break;
425 case GGL_SRC_ALPHA:
426 r->c[1] =
427 r->c[2] =
428 r->c[3] =
429 r->c[0] = blendfactor(src->c[0], src->s[0], FIXED_ONE);
430 break;
431 case GGL_ONE_MINUS_SRC_ALPHA:
432 r->c[1] =
433 r->c[2] =
434 r->c[3] =
435 r->c[0] = FIXED_ONE - blendfactor(src->c[0], src->s[0], FIXED_ONE);
436 break;
437 case GGL_DST_ALPHA:
438 r->c[1] =
439 r->c[2] =
440 r->c[3] =
441 r->c[0] = blendfactor(dst->c[0], dst->s[0], FIXED_ONE);
442 break;
443 case GGL_ONE_MINUS_DST_ALPHA:
444 r->c[1] =
445 r->c[2] =
446 r->c[3] =
447 r->c[0] = FIXED_ONE - blendfactor(dst->c[0], dst->s[0], FIXED_ONE);
448 break;
449 case GGL_SRC_ALPHA_SATURATE:
450 // XXX: GGL_SRC_ALPHA_SATURATE
451 break;
452 }
453 }
454
wrapping(int32_t coord,uint32_t size,int tx_wrap)455 static GGLfixed wrapping(int32_t coord, uint32_t size, int tx_wrap)
456 {
457 GGLfixed d;
458 if (tx_wrap == GGL_REPEAT) {
459 d = (uint32_t(coord)>>16) * size;
460 } else if (tx_wrap == GGL_CLAMP) { // CLAMP_TO_EDGE semantics
461 const GGLfixed clamp_min = FIXED_HALF;
462 const GGLfixed clamp_max = (size << 16) - FIXED_HALF;
463 if (coord < clamp_min) coord = clamp_min;
464 if (coord > clamp_max) coord = clamp_max;
465 d = coord;
466 } else { // 1:1
467 const GGLfixed clamp_min = 0;
468 const GGLfixed clamp_max = (size << 16);
469 if (coord < clamp_min) coord = clamp_min;
470 if (coord > clamp_max) coord = clamp_max;
471 d = coord;
472 }
473 return d;
474 }
475
476 static inline
ADJUST_COLOR_ITERATOR(GGLcolor v,GGLcolor dvdx,int len)477 GGLcolor ADJUST_COLOR_ITERATOR(GGLcolor v, GGLcolor dvdx, int len)
478 {
479 const int32_t end = dvdx * (len-1) + v;
480 if (end < 0)
481 v -= end;
482 v &= ~(v>>31);
483 return v;
484 }
485
scanline(context_t * c)486 void scanline(context_t* c)
487 {
488 const uint32_t enables = c->state.enables;
489 const int xs = c->iterators.xl;
490 const int x1 = c->iterators.xr;
491 int xc = x1 - xs;
492 const int16_t* covPtr = c->state.buffers.coverage + xs;
493
494 // All iterated values are sampled at the pixel center
495
496 // reset iterators for that scanline...
497 GGLcolor r, g, b, a;
498 iterators_t& ci = c->iterators;
499 if (enables & GGL_ENABLE_SMOOTH) {
500 r = (xs * c->shade.drdx) + ci.ydrdy;
501 g = (xs * c->shade.dgdx) + ci.ydgdy;
502 b = (xs * c->shade.dbdx) + ci.ydbdy;
503 a = (xs * c->shade.dadx) + ci.ydady;
504 r = ADJUST_COLOR_ITERATOR(r, c->shade.drdx, xc);
505 g = ADJUST_COLOR_ITERATOR(g, c->shade.dgdx, xc);
506 b = ADJUST_COLOR_ITERATOR(b, c->shade.dbdx, xc);
507 a = ADJUST_COLOR_ITERATOR(a, c->shade.dadx, xc);
508 } else {
509 r = ci.ydrdy;
510 g = ci.ydgdy;
511 b = ci.ydbdy;
512 a = ci.ydady;
513 }
514
515 // z iterators are 1.31
516 GGLfixed z = (xs * c->shade.dzdx) + ci.ydzdy;
517 GGLfixed f = (xs * c->shade.dfdx) + ci.ydfdy;
518
519 struct {
520 GGLfixed s, t;
521 } tc[GGL_TEXTURE_UNIT_COUNT];
522 if (enables & GGL_ENABLE_TMUS) {
523 for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
524 if (c->state.texture[i].enable) {
525 texture_iterators_t& ti = c->state.texture[i].iterators;
526 if (enables & GGL_ENABLE_W) {
527 tc[i].s = ti.ydsdy;
528 tc[i].t = ti.ydtdy;
529 } else {
530 tc[i].s = (xs * ti.dsdx) + ti.ydsdy;
531 tc[i].t = (xs * ti.dtdx) + ti.ydtdy;
532 }
533 }
534 }
535 }
536
537 pixel_t fragment;
538 pixel_t texel;
539 pixel_t fb;
540
541 uint32_t x = xs;
542 uint32_t y = c->iterators.y;
543
544 while (xc--) {
545
546 { // just a scope
547
548 // read color (convert to 8 bits by keeping only the integer part)
549 fragment.s[1] = fragment.s[2] =
550 fragment.s[3] = fragment.s[0] = 8;
551 fragment.c[1] = r >> (GGL_COLOR_BITS-8);
552 fragment.c[2] = g >> (GGL_COLOR_BITS-8);
553 fragment.c[3] = b >> (GGL_COLOR_BITS-8);
554 fragment.c[0] = a >> (GGL_COLOR_BITS-8);
555
556 // texturing
557 if (enables & GGL_ENABLE_TMUS) {
558 for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
559 texture_t& tx = c->state.texture[i];
560 if (!tx.enable)
561 continue;
562 texture_iterators_t& ti = tx.iterators;
563 int32_t u, v;
564
565 // s-coordinate
566 if (tx.s_coord != GGL_ONE_TO_ONE) {
567 const int w = tx.surface.width;
568 u = wrapping(tc[i].s, w, tx.s_wrap);
569 tc[i].s += ti.dsdx;
570 } else {
571 u = (((tx.shade.is0>>16) + x)<<16) + FIXED_HALF;
572 }
573
574 // t-coordinate
575 if (tx.t_coord != GGL_ONE_TO_ONE) {
576 const int h = tx.surface.height;
577 v = wrapping(tc[i].t, h, tx.t_wrap);
578 tc[i].t += ti.dtdx;
579 } else {
580 v = (((tx.shade.it0>>16) + y)<<16) + FIXED_HALF;
581 }
582
583 // read texture
584 if (tx.mag_filter == GGL_NEAREST &&
585 tx.min_filter == GGL_NEAREST)
586 {
587 u >>= 16;
588 v >>= 16;
589 tx.surface.read(&tx.surface, c, u, v, &texel);
590 } else {
591 const int w = tx.surface.width;
592 const int h = tx.surface.height;
593 u -= FIXED_HALF;
594 v -= FIXED_HALF;
595 int u0 = u >> 16;
596 int v0 = v >> 16;
597 int u1 = u0 + 1;
598 int v1 = v0 + 1;
599 if (tx.s_wrap == GGL_REPEAT) {
600 if (u0<0) u0 += w;
601 if (u1<0) u1 += w;
602 if (u0>=w) u0 -= w;
603 if (u1>=w) u1 -= w;
604 } else {
605 if (u0<0) u0 = 0;
606 if (u1<0) u1 = 0;
607 if (u0>=w) u0 = w-1;
608 if (u1>=w) u1 = w-1;
609 }
610 if (tx.t_wrap == GGL_REPEAT) {
611 if (v0<0) v0 += h;
612 if (v1<0) v1 += h;
613 if (v0>=h) v0 -= h;
614 if (v1>=h) v1 -= h;
615 } else {
616 if (v0<0) v0 = 0;
617 if (v1<0) v1 = 0;
618 if (v0>=h) v0 = h-1;
619 if (v1>=h) v1 = h-1;
620 }
621 pixel_t texels[4];
622 uint32_t mm[4];
623 tx.surface.read(&tx.surface, c, u0, v0, &texels[0]);
624 tx.surface.read(&tx.surface, c, u0, v1, &texels[1]);
625 tx.surface.read(&tx.surface, c, u1, v0, &texels[2]);
626 tx.surface.read(&tx.surface, c, u1, v1, &texels[3]);
627 u = (u >> 12) & 0xF;
628 v = (v >> 12) & 0xF;
629 u += u>>3;
630 v += v>>3;
631 mm[0] = (0x10 - u) * (0x10 - v);
632 mm[1] = (0x10 - u) * v;
633 mm[2] = u * (0x10 - v);
634 mm[3] = 0x100 - (mm[0] + mm[1] + mm[2]);
635 for (int j=0 ; j<4 ; j++) {
636 texel.s[j] = texels[0].s[j];
637 if (!texel.s[j]) continue;
638 texel.s[j] += 8;
639 texel.c[j] = texels[0].c[j]*mm[0] +
640 texels[1].c[j]*mm[1] +
641 texels[2].c[j]*mm[2] +
642 texels[3].c[j]*mm[3] ;
643 }
644 }
645
646 // Texture environnement...
647 for (int j=0 ; j<4 ; j++) {
648 uint32_t& Cf = fragment.c[j];
649 uint32_t& Ct = texel.c[j];
650 uint8_t& sf = fragment.s[j];
651 uint8_t& st = texel.s[j];
652 uint32_t At = texel.c[0];
653 uint8_t sat = texel.s[0];
654 switch (tx.env) {
655 case GGL_REPLACE:
656 if (st) {
657 Cf = Ct;
658 sf = st;
659 }
660 break;
661 case GGL_MODULATE:
662 if (st) {
663 uint32_t factor = Ct + (Ct>>(st-1));
664 Cf = (Cf * factor) >> st;
665 }
666 break;
667 case GGL_DECAL:
668 if (sat) {
669 rescale(Cf, sf, Ct, st);
670 Cf += ((Ct - Cf) * (At + (At>>(sat-1)))) >> sat;
671 }
672 break;
673 case GGL_BLEND:
674 if (st) {
675 uint32_t Cc = tx.env_color[i];
676 if (sf>8) Cc = (Cc * ((1<<sf)-1))>>8;
677 else if (sf<8) Cc = (Cc - (Cc>>(8-sf)))>>(8-sf);
678 uint32_t factor = Ct + (Ct>>(st-1));
679 Cf = ((((1<<st) - factor) * Cf) + Ct*Cc)>>st;
680 }
681 break;
682 case GGL_ADD:
683 if (st) {
684 rescale(Cf, sf, Ct, st);
685 Cf += Ct;
686 }
687 break;
688 }
689 }
690 }
691 }
692
693 // coverage application
694 if (enables & GGL_ENABLE_AA) {
695 int16_t cf = *covPtr++;
696 fragment.c[0] = (int64_t(fragment.c[0]) * cf) >> 15;
697 }
698
699 // alpha-test
700 if (enables & GGL_ENABLE_ALPHA_TEST) {
701 GGLcolor ref = c->state.alpha_test.ref;
702 GGLcolor alpha = (uint64_t(fragment.c[0]) *
703 ((1<<GGL_COLOR_BITS)-1)) / ((1<<fragment.s[0])-1);
704 switch (c->state.alpha_test.func) {
705 case GGL_NEVER: goto discard;
706 case GGL_LESS: if (alpha<ref) break; goto discard;
707 case GGL_EQUAL: if (alpha==ref) break; goto discard;
708 case GGL_LEQUAL: if (alpha<=ref) break; goto discard;
709 case GGL_GREATER: if (alpha>ref) break; goto discard;
710 case GGL_NOTEQUAL: if (alpha!=ref) break; goto discard;
711 case GGL_GEQUAL: if (alpha>=ref) break; goto discard;
712 }
713 }
714
715 // depth test
716 if (c->state.buffers.depth.format) {
717 if (enables & GGL_ENABLE_DEPTH_TEST) {
718 surface_t* cb = &(c->state.buffers.depth);
719 uint16_t* p = (uint16_t*)(cb->data)+(x+(cb->stride*y));
720 uint16_t zz = uint32_t(z)>>(16);
721 uint16_t depth = *p;
722 switch (c->state.depth_test.func) {
723 case GGL_NEVER: goto discard;
724 case GGL_LESS: if (zz<depth) break; goto discard;
725 case GGL_EQUAL: if (zz==depth) break; goto discard;
726 case GGL_LEQUAL: if (zz<=depth) break; goto discard;
727 case GGL_GREATER: if (zz>depth) break; goto discard;
728 case GGL_NOTEQUAL: if (zz!=depth) break; goto discard;
729 case GGL_GEQUAL: if (zz>=depth) break; goto discard;
730 }
731 // depth buffer is not enabled, if depth-test is not enabled
732 /*
733 fragment.s[1] = fragment.s[2] =
734 fragment.s[3] = fragment.s[0] = 8;
735 fragment.c[1] =
736 fragment.c[2] =
737 fragment.c[3] =
738 fragment.c[0] = 255 - (zz>>8);
739 */
740 if (c->state.mask.depth) {
741 *p = zz;
742 }
743 }
744 }
745
746 // fog
747 if (enables & GGL_ENABLE_FOG) {
748 for (int i=1 ; i<=3 ; i++) {
749 GGLfixed fc = (c->state.fog.color[i] * 0x10000) / 0xFF;
750 uint32_t& c = fragment.c[i];
751 uint8_t& s = fragment.s[i];
752 c = (c * 0x10000) / ((1<<s)-1);
753 c = gglMulAddx(c, f, gglMulx(fc, 0x10000 - f));
754 s = 16;
755 }
756 }
757
758 // blending
759 if (enables & GGL_ENABLE_BLENDING) {
760 fb.c[1] = fb.c[2] = fb.c[3] = fb.c[0] = 0; // placate valgrind
761 fb.s[1] = fb.s[2] = fb.s[3] = fb.s[0] = 0;
762 c->state.buffers.color.read(
763 &(c->state.buffers.color), c, x, y, &fb);
764 blending( c, &fragment, &fb );
765 }
766
767 // write
768 c->state.buffers.color.write(
769 &(c->state.buffers.color), c, x, y, &fragment);
770 }
771
772 discard:
773 // iterate...
774 x += 1;
775 if (enables & GGL_ENABLE_SMOOTH) {
776 r += c->shade.drdx;
777 g += c->shade.dgdx;
778 b += c->shade.dbdx;
779 a += c->shade.dadx;
780 }
781 z += c->shade.dzdx;
782 f += c->shade.dfdx;
783 }
784 }
785
786 #endif // ANDROID_ARM_CODEGEN && (ANDROID_CODEGEN == ANDROID_CODEGEN_GENERATED)
787
788 // ----------------------------------------------------------------------------
789 #if 0
790 #pragma mark -
791 #pragma mark Scanline
792 #endif
793
794 template <typename T, typename U>
795 static inline __attribute__((const))
interpolate(int y,T v0,U dvdx,U dvdy)796 T interpolate(int y, T v0, U dvdx, U dvdy) {
797 // interpolates in pixel's centers
798 // v = v0 + (y + 0.5) * dvdy + (0.5 * dvdx)
799 return (y * dvdy) + (v0 + ((dvdy + dvdx) >> 1));
800 }
801
802 // ----------------------------------------------------------------------------
803 #if 0
804 #pragma mark -
805 #endif
806
init_y(context_t * c,int32_t ys)807 void init_y(context_t* c, int32_t ys)
808 {
809 const uint32_t enables = c->state.enables;
810
811 // compute iterators...
812 iterators_t& ci = c->iterators;
813
814 // sample in the center
815 ci.y = ys;
816
817 if (enables & (GGL_ENABLE_DEPTH_TEST|GGL_ENABLE_W|GGL_ENABLE_FOG)) {
818 ci.ydzdy = interpolate(ys, c->shade.z0, c->shade.dzdx, c->shade.dzdy);
819 ci.ydwdy = interpolate(ys, c->shade.w0, c->shade.dwdx, c->shade.dwdy);
820 ci.ydfdy = interpolate(ys, c->shade.f0, c->shade.dfdx, c->shade.dfdy);
821 }
822
823 if (ggl_unlikely(enables & GGL_ENABLE_SMOOTH)) {
824 ci.ydrdy = interpolate(ys, c->shade.r0, c->shade.drdx, c->shade.drdy);
825 ci.ydgdy = interpolate(ys, c->shade.g0, c->shade.dgdx, c->shade.dgdy);
826 ci.ydbdy = interpolate(ys, c->shade.b0, c->shade.dbdx, c->shade.dbdy);
827 ci.ydady = interpolate(ys, c->shade.a0, c->shade.dadx, c->shade.dady);
828 c->step_y = step_y__smooth;
829 } else {
830 ci.ydrdy = c->shade.r0;
831 ci.ydgdy = c->shade.g0;
832 ci.ydbdy = c->shade.b0;
833 ci.ydady = c->shade.a0;
834 // XXX: do only if needed, or make sure this is fast
835 c->packed = ggl_pack_color(c, c->state.buffers.color.format,
836 ci.ydrdy, ci.ydgdy, ci.ydbdy, ci.ydady);
837 c->packed8888 = ggl_pack_color(c, GGL_PIXEL_FORMAT_RGBA_8888,
838 ci.ydrdy, ci.ydgdy, ci.ydbdy, ci.ydady);
839 }
840
841 // initialize the variables we need in the shader
842 generated_vars_t& gen = c->generated_vars;
843 gen.argb[GGLFormat::ALPHA].c = ci.ydady;
844 gen.argb[GGLFormat::ALPHA].dx = c->shade.dadx;
845 gen.argb[GGLFormat::RED ].c = ci.ydrdy;
846 gen.argb[GGLFormat::RED ].dx = c->shade.drdx;
847 gen.argb[GGLFormat::GREEN].c = ci.ydgdy;
848 gen.argb[GGLFormat::GREEN].dx = c->shade.dgdx;
849 gen.argb[GGLFormat::BLUE ].c = ci.ydbdy;
850 gen.argb[GGLFormat::BLUE ].dx = c->shade.dbdx;
851 gen.dzdx = c->shade.dzdx;
852 gen.f = ci.ydfdy;
853 gen.dfdx = c->shade.dfdx;
854
855 if (enables & GGL_ENABLE_TMUS) {
856 for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
857 texture_t& t = c->state.texture[i];
858 if (!t.enable) continue;
859
860 texture_iterators_t& ti = t.iterators;
861 if (t.s_coord == GGL_ONE_TO_ONE && t.t_coord == GGL_ONE_TO_ONE) {
862 // we need to set all of these to 0 because in some cases
863 // step_y__generic() or step_y__tmu() will be used and
864 // therefore will update dtdy, however, in 1:1 mode
865 // this is always done by the scanline rasterizer.
866 ti.dsdx = ti.dsdy = ti.dtdx = ti.dtdy = 0;
867 ti.ydsdy = t.shade.is0;
868 ti.ydtdy = t.shade.it0;
869 } else {
870 const int adjustSWrap = ((t.s_wrap==GGL_CLAMP)?0:16);
871 const int adjustTWrap = ((t.t_wrap==GGL_CLAMP)?0:16);
872 ti.sscale = t.shade.sscale + adjustSWrap;
873 ti.tscale = t.shade.tscale + adjustTWrap;
874 if (!(enables & GGL_ENABLE_W)) {
875 // S coordinate
876 const int32_t sscale = ti.sscale;
877 const int32_t sy = interpolate(ys,
878 t.shade.is0, t.shade.idsdx, t.shade.idsdy);
879 if (sscale>=0) {
880 ti.ydsdy= sy << sscale;
881 ti.dsdx = t.shade.idsdx << sscale;
882 ti.dsdy = t.shade.idsdy << sscale;
883 } else {
884 ti.ydsdy= sy >> -sscale;
885 ti.dsdx = t.shade.idsdx >> -sscale;
886 ti.dsdy = t.shade.idsdy >> -sscale;
887 }
888 // T coordinate
889 const int32_t tscale = ti.tscale;
890 const int32_t ty = interpolate(ys,
891 t.shade.it0, t.shade.idtdx, t.shade.idtdy);
892 if (tscale>=0) {
893 ti.ydtdy= ty << tscale;
894 ti.dtdx = t.shade.idtdx << tscale;
895 ti.dtdy = t.shade.idtdy << tscale;
896 } else {
897 ti.ydtdy= ty >> -tscale;
898 ti.dtdx = t.shade.idtdx >> -tscale;
899 ti.dtdy = t.shade.idtdy >> -tscale;
900 }
901 }
902 }
903 // mirror for generated code...
904 generated_tex_vars_t& gen = c->generated_vars.texture[i];
905 gen.width = t.surface.width;
906 gen.height = t.surface.height;
907 gen.stride = t.surface.stride;
908 gen.data = int32_t(t.surface.data);
909 gen.dsdx = ti.dsdx;
910 gen.dtdx = ti.dtdx;
911 }
912 }
913
914 // choose the y-stepper
915 c->step_y = step_y__nop;
916 if (enables & GGL_ENABLE_FOG) {
917 c->step_y = step_y__generic;
918 } else if (enables & GGL_ENABLE_TMUS) {
919 if (enables & GGL_ENABLE_SMOOTH) {
920 c->step_y = step_y__generic;
921 } else if (enables & GGL_ENABLE_W) {
922 c->step_y = step_y__w;
923 } else {
924 c->step_y = step_y__tmu;
925 }
926 } else {
927 if (enables & GGL_ENABLE_SMOOTH) {
928 c->step_y = step_y__smooth;
929 }
930 }
931
932 // choose the rectangle blitter
933 c->rect = rect_generic;
934 if ((c->step_y == step_y__nop) &&
935 (c->scanline == scanline_memcpy))
936 {
937 c->rect = rect_memcpy;
938 }
939 }
940
init_y_packed(context_t * c,int32_t y0)941 void init_y_packed(context_t* c, int32_t y0)
942 {
943 uint8_t f = c->state.buffers.color.format;
944 c->packed = ggl_pack_color(c, f,
945 c->shade.r0, c->shade.g0, c->shade.b0, c->shade.a0);
946 c->iterators.y = y0;
947 c->step_y = step_y__nop;
948 // choose the rectangle blitter
949 c->rect = rect_generic;
950 if (c->scanline == scanline_memcpy) {
951 c->rect = rect_memcpy;
952 }
953 }
954
init_y_noop(context_t * c,int32_t y0)955 void init_y_noop(context_t* c, int32_t y0)
956 {
957 c->iterators.y = y0;
958 c->step_y = step_y__nop;
959 // choose the rectangle blitter
960 c->rect = rect_generic;
961 if (c->scanline == scanline_memcpy) {
962 c->rect = rect_memcpy;
963 }
964 }
965
init_y_error(context_t * c,int32_t y0)966 void init_y_error(context_t* c, int32_t y0)
967 {
968 // woooops, shoud never happen,
969 // fail gracefully (don't display anything)
970 init_y_noop(c, y0);
971 LOGE("color-buffer has an invalid format!");
972 }
973
974 // ----------------------------------------------------------------------------
975 #if 0
976 #pragma mark -
977 #endif
978
step_y__generic(context_t * c)979 void step_y__generic(context_t* c)
980 {
981 const uint32_t enables = c->state.enables;
982
983 // iterate...
984 iterators_t& ci = c->iterators;
985 ci.y += 1;
986
987 if (enables & GGL_ENABLE_SMOOTH) {
988 ci.ydrdy += c->shade.drdy;
989 ci.ydgdy += c->shade.dgdy;
990 ci.ydbdy += c->shade.dbdy;
991 ci.ydady += c->shade.dady;
992 }
993
994 const uint32_t mask =
995 GGL_ENABLE_DEPTH_TEST |
996 GGL_ENABLE_W |
997 GGL_ENABLE_FOG;
998 if (enables & mask) {
999 ci.ydzdy += c->shade.dzdy;
1000 ci.ydwdy += c->shade.dwdy;
1001 ci.ydfdy += c->shade.dfdy;
1002 }
1003
1004 if ((enables & GGL_ENABLE_TMUS) && (!(enables & GGL_ENABLE_W))) {
1005 for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1006 if (c->state.texture[i].enable) {
1007 texture_iterators_t& ti = c->state.texture[i].iterators;
1008 ti.ydsdy += ti.dsdy;
1009 ti.ydtdy += ti.dtdy;
1010 }
1011 }
1012 }
1013 }
1014
step_y__nop(context_t * c)1015 void step_y__nop(context_t* c)
1016 {
1017 c->iterators.y += 1;
1018 c->iterators.ydzdy += c->shade.dzdy;
1019 }
1020
step_y__smooth(context_t * c)1021 void step_y__smooth(context_t* c)
1022 {
1023 iterators_t& ci = c->iterators;
1024 ci.y += 1;
1025 ci.ydrdy += c->shade.drdy;
1026 ci.ydgdy += c->shade.dgdy;
1027 ci.ydbdy += c->shade.dbdy;
1028 ci.ydady += c->shade.dady;
1029 ci.ydzdy += c->shade.dzdy;
1030 }
1031
step_y__w(context_t * c)1032 void step_y__w(context_t* c)
1033 {
1034 iterators_t& ci = c->iterators;
1035 ci.y += 1;
1036 ci.ydzdy += c->shade.dzdy;
1037 ci.ydwdy += c->shade.dwdy;
1038 }
1039
step_y__tmu(context_t * c)1040 void step_y__tmu(context_t* c)
1041 {
1042 iterators_t& ci = c->iterators;
1043 ci.y += 1;
1044 ci.ydzdy += c->shade.dzdy;
1045 for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1046 if (c->state.texture[i].enable) {
1047 texture_iterators_t& ti = c->state.texture[i].iterators;
1048 ti.ydsdy += ti.dsdy;
1049 ti.ydtdy += ti.dtdy;
1050 }
1051 }
1052 }
1053
1054 // ----------------------------------------------------------------------------
1055 #if 0
1056 #pragma mark -
1057 #endif
1058
scanline_perspective(context_t * c)1059 void scanline_perspective(context_t* c)
1060 {
1061 struct {
1062 union {
1063 struct {
1064 int32_t s, sq;
1065 int32_t t, tq;
1066 };
1067 struct {
1068 int32_t v, q;
1069 } st[2];
1070 };
1071 } tc[GGL_TEXTURE_UNIT_COUNT] __attribute__((aligned(16)));
1072
1073 // XXX: we should have a special case when dwdx = 0
1074
1075 // 32 pixels spans works okay. 16 is a lot better,
1076 // but hey, it's a software renderer...
1077 const uint32_t SPAN_BITS = 5;
1078 const uint32_t ys = c->iterators.y;
1079 const uint32_t xs = c->iterators.xl;
1080 const uint32_t x1 = c->iterators.xr;
1081 const uint32_t xc = x1 - xs;
1082 uint32_t remainder = xc & ((1<<SPAN_BITS)-1);
1083 uint32_t numSpans = xc >> SPAN_BITS;
1084
1085 const iterators_t& ci = c->iterators;
1086 int32_t w0 = (xs * c->shade.dwdx) + ci.ydwdy;
1087 int32_t q0 = gglRecipQ(w0, 30);
1088 const int iwscale = 32 - gglClz(q0);
1089
1090 const int32_t dwdx = c->shade.dwdx << SPAN_BITS;
1091 int32_t xl = c->iterators.xl;
1092
1093 // We process s & t with a loop to reduce the code size
1094 // (and i-cache pressure).
1095
1096 for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1097 const texture_t& tmu = c->state.texture[i];
1098 if (!tmu.enable) continue;
1099 int32_t s = tmu.shade.is0 +
1100 (tmu.shade.idsdy * ys) + (tmu.shade.idsdx * xs) +
1101 ((tmu.shade.idsdx + tmu.shade.idsdy)>>1);
1102 int32_t t = tmu.shade.it0 +
1103 (tmu.shade.idtdy * ys) + (tmu.shade.idtdx * xs) +
1104 ((tmu.shade.idtdx + tmu.shade.idtdy)>>1);
1105 tc[i].s = s;
1106 tc[i].t = t;
1107 tc[i].sq = gglMulx(s, q0, iwscale);
1108 tc[i].tq = gglMulx(t, q0, iwscale);
1109 }
1110
1111 int32_t span = 0;
1112 do {
1113 int32_t w1;
1114 if (ggl_likely(numSpans)) {
1115 w1 = w0 + dwdx;
1116 } else {
1117 if (remainder) {
1118 // finish off the scanline...
1119 span = remainder;
1120 w1 = (c->shade.dwdx * span) + w0;
1121 } else {
1122 break;
1123 }
1124 }
1125 int32_t q1 = gglRecipQ(w1, 30);
1126 for (int i=0 ; i<GGL_TEXTURE_UNIT_COUNT ; ++i) {
1127 texture_t& tmu = c->state.texture[i];
1128 if (!tmu.enable) continue;
1129 texture_iterators_t& ti = tmu.iterators;
1130
1131 for (int j=0 ; j<2 ; j++) {
1132 int32_t v = tc[i].st[j].v;
1133 if (span) v += (tmu.shade.st[j].dx)*span;
1134 else v += (tmu.shade.st[j].dx)<<SPAN_BITS;
1135 const int32_t v0 = tc[i].st[j].q;
1136 const int32_t v1 = gglMulx(v, q1, iwscale);
1137 int32_t dvdx = v1 - v0;
1138 if (span) dvdx /= span;
1139 else dvdx >>= SPAN_BITS;
1140 tc[i].st[j].v = v;
1141 tc[i].st[j].q = v1;
1142
1143 const int scale = ti.st[j].scale + (iwscale - 30);
1144 if (scale >= 0) {
1145 ti.st[j].ydvdy = v0 << scale;
1146 ti.st[j].dvdx = dvdx << scale;
1147 } else {
1148 ti.st[j].ydvdy = v0 >> -scale;
1149 ti.st[j].dvdx = dvdx >> -scale;
1150 }
1151 }
1152 generated_tex_vars_t& gen = c->generated_vars.texture[i];
1153 gen.dsdx = ti.st[0].dvdx;
1154 gen.dtdx = ti.st[1].dvdx;
1155 }
1156 c->iterators.xl = xl;
1157 c->iterators.xr = xl = xl + (span ? span : (1<<SPAN_BITS));
1158 w0 = w1;
1159 q0 = q1;
1160 c->span(c);
1161 } while(numSpans--);
1162 }
1163
scanline_perspective_single(context_t * c)1164 void scanline_perspective_single(context_t* c)
1165 {
1166 // 32 pixels spans works okay. 16 is a lot better,
1167 // but hey, it's a software renderer...
1168 const uint32_t SPAN_BITS = 5;
1169 const uint32_t ys = c->iterators.y;
1170 const uint32_t xs = c->iterators.xl;
1171 const uint32_t x1 = c->iterators.xr;
1172 const uint32_t xc = x1 - xs;
1173
1174 const iterators_t& ci = c->iterators;
1175 int32_t w = (xs * c->shade.dwdx) + ci.ydwdy;
1176 int32_t iw = gglRecipQ(w, 30);
1177 const int iwscale = 32 - gglClz(iw);
1178
1179 const int i = 31 - gglClz(c->state.enabled_tmu);
1180 generated_tex_vars_t& gen = c->generated_vars.texture[i];
1181 texture_t& tmu = c->state.texture[i];
1182 texture_iterators_t& ti = tmu.iterators;
1183 const int sscale = ti.sscale + (iwscale - 30);
1184 const int tscale = ti.tscale + (iwscale - 30);
1185 int32_t s = tmu.shade.is0 +
1186 (tmu.shade.idsdy * ys) + (tmu.shade.idsdx * xs) +
1187 ((tmu.shade.idsdx + tmu.shade.idsdy)>>1);
1188 int32_t t = tmu.shade.it0 +
1189 (tmu.shade.idtdy * ys) + (tmu.shade.idtdx * xs) +
1190 ((tmu.shade.idtdx + tmu.shade.idtdy)>>1);
1191 int32_t s0 = gglMulx(s, iw, iwscale);
1192 int32_t t0 = gglMulx(t, iw, iwscale);
1193 int32_t xl = c->iterators.xl;
1194
1195 int32_t sq, tq, dsdx, dtdx;
1196 int32_t premainder = xc & ((1<<SPAN_BITS)-1);
1197 uint32_t numSpans = xc >> SPAN_BITS;
1198 if (c->shade.dwdx == 0) {
1199 // XXX: we could choose to do this if the error is small enough
1200 numSpans = 0;
1201 premainder = xc;
1202 goto no_perspective;
1203 }
1204
1205 if (premainder) {
1206 w += c->shade.dwdx * premainder;
1207 iw = gglRecipQ(w, 30);
1208 no_perspective:
1209 s += tmu.shade.idsdx * premainder;
1210 t += tmu.shade.idtdx * premainder;
1211 sq = gglMulx(s, iw, iwscale);
1212 tq = gglMulx(t, iw, iwscale);
1213 dsdx = (sq - s0) / premainder;
1214 dtdx = (tq - t0) / premainder;
1215 c->iterators.xl = xl;
1216 c->iterators.xr = xl = xl + premainder;
1217 goto finish;
1218 }
1219
1220 while (numSpans--) {
1221 w += c->shade.dwdx << SPAN_BITS;
1222 s += tmu.shade.idsdx << SPAN_BITS;
1223 t += tmu.shade.idtdx << SPAN_BITS;
1224 iw = gglRecipQ(w, 30);
1225 sq = gglMulx(s, iw, iwscale);
1226 tq = gglMulx(t, iw, iwscale);
1227 dsdx = (sq - s0) >> SPAN_BITS;
1228 dtdx = (tq - t0) >> SPAN_BITS;
1229 c->iterators.xl = xl;
1230 c->iterators.xr = xl = xl + (1<<SPAN_BITS);
1231 finish:
1232 if (sscale >= 0) {
1233 ti.ydsdy = s0 << sscale;
1234 ti.dsdx = dsdx << sscale;
1235 } else {
1236 ti.ydsdy = s0 >>-sscale;
1237 ti.dsdx = dsdx >>-sscale;
1238 }
1239 if (tscale >= 0) {
1240 ti.ydtdy = t0 << tscale;
1241 ti.dtdx = dtdx << tscale;
1242 } else {
1243 ti.ydtdy = t0 >>-tscale;
1244 ti.dtdx = dtdx >>-tscale;
1245 }
1246 s0 = sq;
1247 t0 = tq;
1248 gen.dsdx = ti.dsdx;
1249 gen.dtdx = ti.dtdx;
1250 c->span(c);
1251 }
1252 }
1253
1254 // ----------------------------------------------------------------------------
1255
scanline_t32cb16(context_t * c)1256 void scanline_t32cb16(context_t* c)
1257 {
1258 int32_t x = c->iterators.xl;
1259 size_t ct = c->iterators.xr - x;
1260 int32_t y = c->iterators.y;
1261 surface_t* cb = &(c->state.buffers.color);
1262 union {
1263 uint16_t* dst;
1264 uint32_t* dst32;
1265 };
1266 dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
1267
1268 surface_t* tex = &(c->state.texture[0].surface);
1269 const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
1270 const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
1271 uint32_t *src = reinterpret_cast<uint32_t*>(tex->data)+(u+(tex->stride*v));
1272 int sR, sG, sB;
1273 uint32_t s, d;
1274
1275 if (ct==1 || uint32_t(dst)&2) {
1276 last_one:
1277 s = GGL_RGBA_TO_HOST( *src++ );
1278 sR = (s >> ( 3))&0x1F;
1279 sG = (s >> ( 8+2))&0x3F;
1280 sB = (s >> (16+3))&0x1F;
1281 *dst++ = uint16_t((sR<<11)|(sG<<5)|sB);
1282 ct--;
1283 }
1284
1285 while (ct >= 2) {
1286 s = GGL_RGBA_TO_HOST( *src++ );
1287 sR = (s >> ( 3))&0x1F;
1288 sG = (s >> ( 8+2))&0x3F;
1289 sB = (s >> (16+3))&0x1F;
1290 d = (sR<<11)|(sG<<5)|sB;
1291
1292 s = GGL_RGBA_TO_HOST( *src++ );
1293 sR = (s >> ( 3))&0x1F;
1294 sG = (s >> ( 8+2))&0x3F;
1295 sB = (s >> (16+3))&0x1F;
1296 d |= ((sR<<11)|(sG<<5)|sB)<<16;
1297
1298 #if BYTE_ORDER == BIG_ENDIAN
1299 d = (d>>16) | (d<<16);
1300 #endif
1301
1302 *dst32++ = d;
1303 ct -= 2;
1304 }
1305
1306 if (ct > 0) {
1307 goto last_one;
1308 }
1309 }
1310
scanline_t32cb16blend(context_t * c)1311 void scanline_t32cb16blend(context_t* c)
1312 {
1313 int32_t x = c->iterators.xl;
1314 size_t ct = c->iterators.xr - x;
1315 int32_t y = c->iterators.y;
1316 surface_t* cb = &(c->state.buffers.color);
1317 uint16_t* dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
1318
1319 surface_t* tex = &(c->state.texture[0].surface);
1320 const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
1321 const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
1322 uint32_t *src = reinterpret_cast<uint32_t*>(tex->data)+(u+(tex->stride*v));
1323
1324 #if ((ANDROID_CODEGEN >= ANDROID_CODEGEN_ASM) && defined(__arm__))
1325 scanline_t32cb16blend_arm(dst, src, ct);
1326 #else
1327 while (ct--) {
1328 uint32_t s = *src++;
1329 if (!s) {
1330 dst++;
1331 continue;
1332 }
1333 uint16_t d = *dst;
1334 s = GGL_RGBA_TO_HOST(s);
1335 int sR = (s >> ( 3))&0x1F;
1336 int sG = (s >> ( 8+2))&0x3F;
1337 int sB = (s >> (16+3))&0x1F;
1338 int sA = (s>>24);
1339 int f = 0x100 - (sA + (sA>>7));
1340 int dR = (d>>11)&0x1f;
1341 int dG = (d>>5)&0x3f;
1342 int dB = (d)&0x1f;
1343 sR += (f*dR)>>8;
1344 sG += (f*dG)>>8;
1345 sB += (f*dB)>>8;
1346 *dst++ = uint16_t((sR<<11)|(sG<<5)|sB);
1347 }
1348 #endif
1349 }
1350
scanline_memcpy(context_t * c)1351 void scanline_memcpy(context_t* c)
1352 {
1353 int32_t x = c->iterators.xl;
1354 size_t ct = c->iterators.xr - x;
1355 int32_t y = c->iterators.y;
1356 surface_t* cb = &(c->state.buffers.color);
1357 const GGLFormat* fp = &(c->formats[cb->format]);
1358 uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
1359 (x + (cb->stride * y)) * fp->size;
1360
1361 surface_t* tex = &(c->state.texture[0].surface);
1362 const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
1363 const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
1364 uint8_t *src = reinterpret_cast<uint8_t*>(tex->data) +
1365 (u + (tex->stride * v)) * fp->size;
1366
1367 const size_t size = ct * fp->size;
1368 memcpy(dst, src, size);
1369 }
1370
scanline_memset8(context_t * c)1371 void scanline_memset8(context_t* c)
1372 {
1373 int32_t x = c->iterators.xl;
1374 size_t ct = c->iterators.xr - x;
1375 int32_t y = c->iterators.y;
1376 surface_t* cb = &(c->state.buffers.color);
1377 uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) + (x+(cb->stride*y));
1378 uint32_t packed = c->packed;
1379 memset(dst, packed, ct);
1380 }
1381
scanline_memset16(context_t * c)1382 void scanline_memset16(context_t* c)
1383 {
1384 int32_t x = c->iterators.xl;
1385 size_t ct = c->iterators.xr - x;
1386 int32_t y = c->iterators.y;
1387 surface_t* cb = &(c->state.buffers.color);
1388 uint16_t* dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
1389 uint32_t packed = c->packed;
1390 android_memset16(dst, packed, ct*2);
1391 }
1392
scanline_memset32(context_t * c)1393 void scanline_memset32(context_t* c)
1394 {
1395 int32_t x = c->iterators.xl;
1396 size_t ct = c->iterators.xr - x;
1397 int32_t y = c->iterators.y;
1398 surface_t* cb = &(c->state.buffers.color);
1399 uint32_t* dst = reinterpret_cast<uint32_t*>(cb->data) + (x+(cb->stride*y));
1400 uint32_t packed = GGL_HOST_TO_RGBA(c->packed);
1401 android_memset32(dst, packed, ct*4);
1402 }
1403
scanline_clear(context_t * c)1404 void scanline_clear(context_t* c)
1405 {
1406 int32_t x = c->iterators.xl;
1407 size_t ct = c->iterators.xr - x;
1408 int32_t y = c->iterators.y;
1409 surface_t* cb = &(c->state.buffers.color);
1410 const GGLFormat* fp = &(c->formats[cb->format]);
1411 uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
1412 (x + (cb->stride * y)) * fp->size;
1413 const size_t size = ct * fp->size;
1414 memset(dst, 0, size);
1415 }
1416
scanline_set(context_t * c)1417 void scanline_set(context_t* c)
1418 {
1419 int32_t x = c->iterators.xl;
1420 size_t ct = c->iterators.xr - x;
1421 int32_t y = c->iterators.y;
1422 surface_t* cb = &(c->state.buffers.color);
1423 const GGLFormat* fp = &(c->formats[cb->format]);
1424 uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
1425 (x + (cb->stride * y)) * fp->size;
1426 const size_t size = ct * fp->size;
1427 memset(dst, 0xFF, size);
1428 }
1429
scanline_noop(context_t * c)1430 void scanline_noop(context_t* c)
1431 {
1432 }
1433
rect_generic(context_t * c,size_t yc)1434 void rect_generic(context_t* c, size_t yc)
1435 {
1436 do {
1437 c->scanline(c);
1438 c->step_y(c);
1439 } while (--yc);
1440 }
1441
rect_memcpy(context_t * c,size_t yc)1442 void rect_memcpy(context_t* c, size_t yc)
1443 {
1444 int32_t x = c->iterators.xl;
1445 size_t ct = c->iterators.xr - x;
1446 int32_t y = c->iterators.y;
1447 surface_t* cb = &(c->state.buffers.color);
1448 const GGLFormat* fp = &(c->formats[cb->format]);
1449 uint8_t* dst = reinterpret_cast<uint8_t*>(cb->data) +
1450 (x + (cb->stride * y)) * fp->size;
1451
1452 surface_t* tex = &(c->state.texture[0].surface);
1453 const int32_t u = (c->state.texture[0].shade.is0>>16) + x;
1454 const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
1455 uint8_t *src = reinterpret_cast<uint8_t*>(tex->data) +
1456 (u + (tex->stride * v)) * fp->size;
1457
1458 if (cb->stride == tex->stride && ct == size_t(cb->stride)) {
1459 memcpy(dst, src, ct * fp->size * yc);
1460 } else {
1461 const size_t size = ct * fp->size;
1462 const size_t dbpr = cb->stride * fp->size;
1463 const size_t sbpr = tex->stride * fp->size;
1464 do {
1465 memcpy(dst, src, size);
1466 dst += dbpr;
1467 src += sbpr;
1468 } while (--yc);
1469 }
1470 }
1471 // ----------------------------------------------------------------------------
1472 }; // namespace android
1473
1474 using namespace android;
ggl_test_codegen(uint32_t n,uint32_t p,uint32_t t0,uint32_t t1)1475 extern "C" void ggl_test_codegen(uint32_t n, uint32_t p, uint32_t t0, uint32_t t1)
1476 {
1477 #if ANDROID_ARM_CODEGEN
1478 GGLContext* c;
1479 gglInit(&c);
1480 needs_t needs;
1481 needs.n = n;
1482 needs.p = p;
1483 needs.t[0] = t0;
1484 needs.t[1] = t1;
1485 sp<ScanlineAssembly> a(new ScanlineAssembly(needs, ASSEMBLY_SCRATCH_SIZE));
1486 GGLAssembler assembler( new ARMAssembler(a) );
1487 int err = assembler.scanline(needs, (context_t*)c);
1488 if (err != 0) {
1489 printf("error %08x (%s)\n", err, strerror(-err));
1490 }
1491 gglUninit(c);
1492 #else
1493 printf("This test runs only on ARM\n");
1494 #endif
1495 }
1496
1497