projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge remote branch 'origin/nvc0'
[mesa.git] / src / gallium / auxiliary / tgsi / tgsi_exec.c
1 /**************************************************************************
2 *
3 * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 * Copyright 2009-2010 VMware, Inc. All rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28
29 /**
30 * TGSI interpreter/executor.
31 *
32 * Flow control information:
33 *
34 * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
35 * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
36 * care since a condition may be true for some quad components but false
37 * for other components.
38 *
39 * We basically execute all statements (even if they're in the part of
40 * an IF/ELSE clause that's "not taken") and use a special mask to
41 * control writing to destination registers. This is the ExecMask.
42 * See store_dest().
43 *
44 * The ExecMask is computed from three other masks (CondMask, LoopMask and
45 * ContMask) which are controlled by the flow control instructions (namely:
46 * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
47 *
48 *
49 * Authors:
50 * Michal Krol
51 * Brian Paul
52 */
53
54 #include "pipe/p_compiler.h"
55 #include "pipe/p_state.h"
56 #include "pipe/p_shader_tokens.h"
57 #include "tgsi/tgsi_dump.h"
58 #include "tgsi/tgsi_parse.h"
59 #include "tgsi/tgsi_util.h"
60 #include "tgsi_exec.h"
61 #include "util/u_memory.h"
62 #include "util/u_math.h"
63
64
65 #define FAST_MATH 1
66
67 #define TILE_TOP_LEFT 0
68 #define TILE_TOP_RIGHT 1
69 #define TILE_BOTTOM_LEFT 2
70 #define TILE_BOTTOM_RIGHT 3
71
72 static void
73 micro_abs(union tgsi_exec_channel *dst,
74 const union tgsi_exec_channel *src)
75 {
76 dst->f[0] = fabsf(src->f[0]);
77 dst->f[1] = fabsf(src->f[1]);
78 dst->f[2] = fabsf(src->f[2]);
79 dst->f[3] = fabsf(src->f[3]);
80 }
81
82 static void
83 micro_arl(union tgsi_exec_channel *dst,
84 const union tgsi_exec_channel *src)
85 {
86 dst->i[0] = (int)floorf(src->f[0]);
87 dst->i[1] = (int)floorf(src->f[1]);
88 dst->i[2] = (int)floorf(src->f[2]);
89 dst->i[3] = (int)floorf(src->f[3]);
90 }
91
92 static void
93 micro_arr(union tgsi_exec_channel *dst,
94 const union tgsi_exec_channel *src)
95 {
96 dst->i[0] = (int)floorf(src->f[0] + 0.5f);
97 dst->i[1] = (int)floorf(src->f[1] + 0.5f);
98 dst->i[2] = (int)floorf(src->f[2] + 0.5f);
99 dst->i[3] = (int)floorf(src->f[3] + 0.5f);
100 }
101
102 static void
103 micro_ceil(union tgsi_exec_channel *dst,
104 const union tgsi_exec_channel *src)
105 {
106 dst->f[0] = ceilf(src->f[0]);
107 dst->f[1] = ceilf(src->f[1]);
108 dst->f[2] = ceilf(src->f[2]);
109 dst->f[3] = ceilf(src->f[3]);
110 }
111
112 static void
113 micro_clamp(union tgsi_exec_channel *dst,
114 const union tgsi_exec_channel *src0,
115 const union tgsi_exec_channel *src1,
116 const union tgsi_exec_channel *src2)
117 {
118 dst->f[0] = src0->f[0] < src1->f[0] ? src1->f[0] : src0->f[0] > src2->f[0] ? src2->f[0] : src0->f[0];
119 dst->f[1] = src0->f[1] < src1->f[1] ? src1->f[1] : src0->f[1] > src2->f[1] ? src2->f[1] : src0->f[1];
120 dst->f[2] = src0->f[2] < src1->f[2] ? src1->f[2] : src0->f[2] > src2->f[2] ? src2->f[2] : src0->f[2];
121 dst->f[3] = src0->f[3] < src1->f[3] ? src1->f[3] : src0->f[3] > src2->f[3] ? src2->f[3] : src0->f[3];
122 }
123
124 static void
125 micro_cmp(union tgsi_exec_channel *dst,
126 const union tgsi_exec_channel *src0,
127 const union tgsi_exec_channel *src1,
128 const union tgsi_exec_channel *src2)
129 {
130 dst->f[0] = src0->f[0] < 0.0f ? src1->f[0] : src2->f[0];
131 dst->f[1] = src0->f[1] < 0.0f ? src1->f[1] : src2->f[1];
132 dst->f[2] = src0->f[2] < 0.0f ? src1->f[2] : src2->f[2];
133 dst->f[3] = src0->f[3] < 0.0f ? src1->f[3] : src2->f[3];
134 }
135
136 static void
137 micro_cnd(union tgsi_exec_channel *dst,
138 const union tgsi_exec_channel *src0,
139 const union tgsi_exec_channel *src1,
140 const union tgsi_exec_channel *src2)
141 {
142 dst->f[0] = src2->f[0] > 0.5f ? src0->f[0] : src1->f[0];
143 dst->f[1] = src2->f[1] > 0.5f ? src0->f[1] : src1->f[1];
144 dst->f[2] = src2->f[2] > 0.5f ? src0->f[2] : src1->f[2];
145 dst->f[3] = src2->f[3] > 0.5f ? src0->f[3] : src1->f[3];
146 }
147
148 static void
149 micro_cos(union tgsi_exec_channel *dst,
150 const union tgsi_exec_channel *src)
151 {
152 dst->f[0] = cosf(src->f[0]);
153 dst->f[1] = cosf(src->f[1]);
154 dst->f[2] = cosf(src->f[2]);
155 dst->f[3] = cosf(src->f[3]);
156 }
157
158 static void
159 micro_ddx(union tgsi_exec_channel *dst,
160 const union tgsi_exec_channel *src)
161 {
162 dst->f[0] =
163 dst->f[1] =
164 dst->f[2] =
165 dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_BOTTOM_LEFT];
166 }
167
168 static void
169 micro_ddy(union tgsi_exec_channel *dst,
170 const union tgsi_exec_channel *src)
171 {
172 dst->f[0] =
173 dst->f[1] =
174 dst->f[2] =
175 dst->f[3] = src->f[TILE_BOTTOM_LEFT] - src->f[TILE_TOP_LEFT];
176 }
177
178 static void
179 micro_exp2(union tgsi_exec_channel *dst,
180 const union tgsi_exec_channel *src)
181 {
182 #if FAST_MATH
183 dst->f[0] = util_fast_exp2(src->f[0]);
184 dst->f[1] = util_fast_exp2(src->f[1]);
185 dst->f[2] = util_fast_exp2(src->f[2]);
186 dst->f[3] = util_fast_exp2(src->f[3]);
187 #else
188 #if DEBUG
189 /* Inf is okay for this instruction, so clamp it to silence assertions. */
190 uint i;
191 union tgsi_exec_channel clamped;
192
193 for (i = 0; i < 4; i++) {
194 if (src->f[i] > 127.99999f) {
195 clamped.f[i] = 127.99999f;
196 } else if (src->f[i] < -126.99999f) {
197 clamped.f[i] = -126.99999f;
198 } else {
199 clamped.f[i] = src->f[i];
200 }
201 }
202 src = &clamped;
203 #endif /* DEBUG */
204
205 dst->f[0] = powf(2.0f, src->f[0]);
206 dst->f[1] = powf(2.0f, src->f[1]);
207 dst->f[2] = powf(2.0f, src->f[2]);
208 dst->f[3] = powf(2.0f, src->f[3]);
209 #endif /* FAST_MATH */
210 }
211
212 static void
213 micro_flr(union tgsi_exec_channel *dst,
214 const union tgsi_exec_channel *src)
215 {
216 dst->f[0] = floorf(src->f[0]);
217 dst->f[1] = floorf(src->f[1]);
218 dst->f[2] = floorf(src->f[2]);
219 dst->f[3] = floorf(src->f[3]);
220 }
221
222 static void
223 micro_frc(union tgsi_exec_channel *dst,
224 const union tgsi_exec_channel *src)
225 {
226 dst->f[0] = src->f[0] - floorf(src->f[0]);
227 dst->f[1] = src->f[1] - floorf(src->f[1]);
228 dst->f[2] = src->f[2] - floorf(src->f[2]);
229 dst->f[3] = src->f[3] - floorf(src->f[3]);
230 }
231
232 static void
233 micro_iabs(union tgsi_exec_channel *dst,
234 const union tgsi_exec_channel *src)
235 {
236 dst->i[0] = src->i[0] >= 0 ? src->i[0] : -src->i[0];
237 dst->i[1] = src->i[1] >= 0 ? src->i[1] : -src->i[1];
238 dst->i[2] = src->i[2] >= 0 ? src->i[2] : -src->i[2];
239 dst->i[3] = src->i[3] >= 0 ? src->i[3] : -src->i[3];
240 }
241
242 static void
243 micro_ineg(union tgsi_exec_channel *dst,
244 const union tgsi_exec_channel *src)
245 {
246 dst->i[0] = -src->i[0];
247 dst->i[1] = -src->i[1];
248 dst->i[2] = -src->i[2];
249 dst->i[3] = -src->i[3];
250 }
251
252 static void
253 micro_lg2(union tgsi_exec_channel *dst,
254 const union tgsi_exec_channel *src)
255 {
256 #if FAST_MATH
257 dst->f[0] = util_fast_log2(src->f[0]);
258 dst->f[1] = util_fast_log2(src->f[1]);
259 dst->f[2] = util_fast_log2(src->f[2]);
260 dst->f[3] = util_fast_log2(src->f[3]);
261 #else
262 dst->f[0] = logf(src->f[0]) * 1.442695f;
263 dst->f[1] = logf(src->f[1]) * 1.442695f;
264 dst->f[2] = logf(src->f[2]) * 1.442695f;
265 dst->f[3] = logf(src->f[3]) * 1.442695f;
266 #endif
267 }
268
269 static void
270 micro_lrp(union tgsi_exec_channel *dst,
271 const union tgsi_exec_channel *src0,
272 const union tgsi_exec_channel *src1,
273 const union tgsi_exec_channel *src2)
274 {
275 dst->f[0] = src0->f[0] * (src1->f[0] - src2->f[0]) + src2->f[0];
276 dst->f[1] = src0->f[1] * (src1->f[1] - src2->f[1]) + src2->f[1];
277 dst->f[2] = src0->f[2] * (src1->f[2] - src2->f[2]) + src2->f[2];
278 dst->f[3] = src0->f[3] * (src1->f[3] - src2->f[3]) + src2->f[3];
279 }
280
281 static void
282 micro_mad(union tgsi_exec_channel *dst,
283 const union tgsi_exec_channel *src0,
284 const union tgsi_exec_channel *src1,
285 const union tgsi_exec_channel *src2)
286 {
287 dst->f[0] = src0->f[0] * src1->f[0] + src2->f[0];
288 dst->f[1] = src0->f[1] * src1->f[1] + src2->f[1];
289 dst->f[2] = src0->f[2] * src1->f[2] + src2->f[2];
290 dst->f[3] = src0->f[3] * src1->f[3] + src2->f[3];
291 }
292
293 static void
294 micro_mov(union tgsi_exec_channel *dst,
295 const union tgsi_exec_channel *src)
296 {
297 dst->u[0] = src->u[0];
298 dst->u[1] = src->u[1];
299 dst->u[2] = src->u[2];
300 dst->u[3] = src->u[3];
301 }
302
303 static void
304 micro_rcp(union tgsi_exec_channel *dst,
305 const union tgsi_exec_channel *src)
306 {
307 #if 0 /* for debugging */
308 assert(src->f[0] != 0.0f);
309 assert(src->f[1] != 0.0f);
310 assert(src->f[2] != 0.0f);
311 assert(src->f[3] != 0.0f);
312 #endif
313 dst->f[0] = 1.0f / src->f[0];
314 dst->f[1] = 1.0f / src->f[1];
315 dst->f[2] = 1.0f / src->f[2];
316 dst->f[3] = 1.0f / src->f[3];
317 }
318
319 static void
320 micro_rnd(union tgsi_exec_channel *dst,
321 const union tgsi_exec_channel *src)
322 {
323 dst->f[0] = floorf(src->f[0] + 0.5f);
324 dst->f[1] = floorf(src->f[1] + 0.5f);
325 dst->f[2] = floorf(src->f[2] + 0.5f);
326 dst->f[3] = floorf(src->f[3] + 0.5f);
327 }
328
329 static void
330 micro_rsq(union tgsi_exec_channel *dst,
331 const union tgsi_exec_channel *src)
332 {
333 #if 0 /* for debugging */
334 assert(src->f[0] != 0.0f);
335 assert(src->f[1] != 0.0f);
336 assert(src->f[2] != 0.0f);
337 assert(src->f[3] != 0.0f);
338 #endif
339 dst->f[0] = 1.0f / sqrtf(fabsf(src->f[0]));
340 dst->f[1] = 1.0f / sqrtf(fabsf(src->f[1]));
341 dst->f[2] = 1.0f / sqrtf(fabsf(src->f[2]));
342 dst->f[3] = 1.0f / sqrtf(fabsf(src->f[3]));
343 }
344
345 static void
346 micro_seq(union tgsi_exec_channel *dst,
347 const union tgsi_exec_channel *src0,
348 const union tgsi_exec_channel *src1)
349 {
350 dst->f[0] = src0->f[0] == src1->f[0] ? 1.0f : 0.0f;
351 dst->f[1] = src0->f[1] == src1->f[1] ? 1.0f : 0.0f;
352 dst->f[2] = src0->f[2] == src1->f[2] ? 1.0f : 0.0f;
353 dst->f[3] = src0->f[3] == src1->f[3] ? 1.0f : 0.0f;
354 }
355
356 static void
357 micro_sge(union tgsi_exec_channel *dst,
358 const union tgsi_exec_channel *src0,
359 const union tgsi_exec_channel *src1)
360 {
361 dst->f[0] = src0->f[0] >= src1->f[0] ? 1.0f : 0.0f;
362 dst->f[1] = src0->f[1] >= src1->f[1] ? 1.0f : 0.0f;
363 dst->f[2] = src0->f[2] >= src1->f[2] ? 1.0f : 0.0f;
364 dst->f[3] = src0->f[3] >= src1->f[3] ? 1.0f : 0.0f;
365 }
366
367 static void
368 micro_sgn(union tgsi_exec_channel *dst,
369 const union tgsi_exec_channel *src)
370 {
371 dst->f[0] = src->f[0] < 0.0f ? -1.0f : src->f[0] > 0.0f ? 1.0f : 0.0f;
372 dst->f[1] = src->f[1] < 0.0f ? -1.0f : src->f[1] > 0.0f ? 1.0f : 0.0f;
373 dst->f[2] = src->f[2] < 0.0f ? -1.0f : src->f[2] > 0.0f ? 1.0f : 0.0f;
374 dst->f[3] = src->f[3] < 0.0f ? -1.0f : src->f[3] > 0.0f ? 1.0f : 0.0f;
375 }
376
377 static void
378 micro_sgt(union tgsi_exec_channel *dst,
379 const union tgsi_exec_channel *src0,
380 const union tgsi_exec_channel *src1)
381 {
382 dst->f[0] = src0->f[0] > src1->f[0] ? 1.0f : 0.0f;
383 dst->f[1] = src0->f[1] > src1->f[1] ? 1.0f : 0.0f;
384 dst->f[2] = src0->f[2] > src1->f[2] ? 1.0f : 0.0f;
385 dst->f[3] = src0->f[3] > src1->f[3] ? 1.0f : 0.0f;
386 }
387
388 static void
389 micro_sin(union tgsi_exec_channel *dst,
390 const union tgsi_exec_channel *src)
391 {
392 dst->f[0] = sinf(src->f[0]);
393 dst->f[1] = sinf(src->f[1]);
394 dst->f[2] = sinf(src->f[2]);
395 dst->f[3] = sinf(src->f[3]);
396 }
397
398 static void
399 micro_sle(union tgsi_exec_channel *dst,
400 const union tgsi_exec_channel *src0,
401 const union tgsi_exec_channel *src1)
402 {
403 dst->f[0] = src0->f[0] <= src1->f[0] ? 1.0f : 0.0f;
404 dst->f[1] = src0->f[1] <= src1->f[1] ? 1.0f : 0.0f;
405 dst->f[2] = src0->f[2] <= src1->f[2] ? 1.0f : 0.0f;
406 dst->f[3] = src0->f[3] <= src1->f[3] ? 1.0f : 0.0f;
407 }
408
409 static void
410 micro_slt(union tgsi_exec_channel *dst,
411 const union tgsi_exec_channel *src0,
412 const union tgsi_exec_channel *src1)
413 {
414 dst->f[0] = src0->f[0] < src1->f[0] ? 1.0f : 0.0f;
415 dst->f[1] = src0->f[1] < src1->f[1] ? 1.0f : 0.0f;
416 dst->f[2] = src0->f[2] < src1->f[2] ? 1.0f : 0.0f;
417 dst->f[3] = src0->f[3] < src1->f[3] ? 1.0f : 0.0f;
418 }
419
420 static void
421 micro_sne(union tgsi_exec_channel *dst,
422 const union tgsi_exec_channel *src0,
423 const union tgsi_exec_channel *src1)
424 {
425 dst->f[0] = src0->f[0] != src1->f[0] ? 1.0f : 0.0f;
426 dst->f[1] = src0->f[1] != src1->f[1] ? 1.0f : 0.0f;
427 dst->f[2] = src0->f[2] != src1->f[2] ? 1.0f : 0.0f;
428 dst->f[3] = src0->f[3] != src1->f[3] ? 1.0f : 0.0f;
429 }
430
431 static void
432 micro_sfl(union tgsi_exec_channel *dst)
433 {
434 dst->f[0] = 0.0f;
435 dst->f[1] = 0.0f;
436 dst->f[2] = 0.0f;
437 dst->f[3] = 0.0f;
438 }
439
440 static void
441 micro_str(union tgsi_exec_channel *dst)
442 {
443 dst->f[0] = 1.0f;
444 dst->f[1] = 1.0f;
445 dst->f[2] = 1.0f;
446 dst->f[3] = 1.0f;
447 }
448
449 static void
450 micro_trunc(union tgsi_exec_channel *dst,
451 const union tgsi_exec_channel *src)
452 {
453 dst->f[0] = (float)(int)src->f[0];
454 dst->f[1] = (float)(int)src->f[1];
455 dst->f[2] = (float)(int)src->f[2];
456 dst->f[3] = (float)(int)src->f[3];
457 }
458
459
460 #define CHAN_X 0
461 #define CHAN_Y 1
462 #define CHAN_Z 2
463 #define CHAN_W 3
464
465 enum tgsi_exec_datatype {
466 TGSI_EXEC_DATA_FLOAT,
467 TGSI_EXEC_DATA_INT,
468 TGSI_EXEC_DATA_UINT
469 };
470
471 /*
472 * Shorthand locations of various utility registers (_I = Index, _C = Channel)
473 */
474 #define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I
475 #define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C
476 #define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I
477 #define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C
478 #define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I
479 #define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C
480
481
482 /** The execution mask depends on the conditional mask and the loop mask */
483 #define UPDATE_EXEC_MASK(MACH) \
484 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask
485
486
487 static const union tgsi_exec_channel ZeroVec =
488 { { 0.0, 0.0, 0.0, 0.0 } };
489
490 static const union tgsi_exec_channel OneVec = {
491 {1.0f, 1.0f, 1.0f, 1.0f}
492 };
493
494 static const union tgsi_exec_channel P128Vec = {
495 {128.0f, 128.0f, 128.0f, 128.0f}
496 };
497
498 static const union tgsi_exec_channel M128Vec = {
499 {-128.0f, -128.0f, -128.0f, -128.0f}
500 };
501
502
503 /**
504 * Assert that none of the float values in 'chan' are infinite or NaN.
505 * NaN and Inf may occur normally during program execution and should
506 * not lead to crashes, etc. But when debugging, it's helpful to catch
507 * them.
508 */
509 static INLINE void
510 check_inf_or_nan(const union tgsi_exec_channel *chan)
511 {
512 assert(!util_is_inf_or_nan((chan)->f[0]));
513 assert(!util_is_inf_or_nan((chan)->f[1]));
514 assert(!util_is_inf_or_nan((chan)->f[2]));
515 assert(!util_is_inf_or_nan((chan)->f[3]));
516 }
517
518
519 #ifdef DEBUG
520 static void
521 print_chan(const char *msg, const union tgsi_exec_channel *chan)
522 {
523 debug_printf("%s = {%f, %f, %f, %f}\n",
524 msg, chan->f[0], chan->f[1], chan->f[2], chan->f[3]);
525 }
526 #endif
527
528
529 #ifdef DEBUG
530 static void
531 print_temp(const struct tgsi_exec_machine *mach, uint index)
532 {
533 const struct tgsi_exec_vector *tmp = &mach->Temps[index];
534 int i;
535 debug_printf("Temp[%u] =\n", index);
536 for (i = 0; i < 4; i++) {
537 debug_printf(" %c: { %f, %f, %f, %f }\n",
538 "XYZW"[i],
539 tmp->xyzw[i].f[0],
540 tmp->xyzw[i].f[1],
541 tmp->xyzw[i].f[2],
542 tmp->xyzw[i].f[3]);
543 }
544 }
545 #endif
546
547
548 void
549 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
550 unsigned num_bufs,
551 const void **bufs,
552 const unsigned *buf_sizes)
553 {
554 unsigned i;
555
556 for (i = 0; i < num_bufs; i++) {
557 mach->Consts[i] = bufs[i];
558 mach->ConstsSize[i] = buf_sizes[i];
559 }
560 }
561
562
563 /**
564 * Check if there's a potential src/dst register data dependency when
565 * using SOA execution.
566 * Example:
567 * MOV T, T.yxwz;
568 * This would expand into:
569 * MOV t0, t1;
570 * MOV t1, t0;
571 * MOV t2, t3;
572 * MOV t3, t2;
573 * The second instruction will have the wrong value for t0 if executed as-is.
574 */
575 boolean
576 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst)
577 {
578 uint i, chan;
579
580 uint writemask = inst->Dst[0].Register.WriteMask;
581 if (writemask == TGSI_WRITEMASK_X ||
582 writemask == TGSI_WRITEMASK_Y ||
583 writemask == TGSI_WRITEMASK_Z ||
584 writemask == TGSI_WRITEMASK_W ||
585 writemask == TGSI_WRITEMASK_NONE) {
586 /* no chance of data dependency */
587 return FALSE;
588 }
589
590 /* loop over src regs */
591 for (i = 0; i < inst->Instruction.NumSrcRegs; i++) {
592 if ((inst->Src[i].Register.File ==
593 inst->Dst[0].Register.File) &&
594 ((inst->Src[i].Register.Index ==
595 inst->Dst[0].Register.Index) ||
596 inst->Src[i].Register.Indirect ||
597 inst->Dst[0].Register.Indirect)) {
598 /* loop over dest channels */
599 uint channelsWritten = 0x0;
600 for (chan = 0; chan < NUM_CHANNELS; chan++) {
601 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
602 /* check if we're reading a channel that's been written */
603 uint swizzle = tgsi_util_get_full_src_register_swizzle(&inst->Src[i], chan);
604 if (channelsWritten & (1 << swizzle)) {
605 return TRUE;
606 }
607
608 channelsWritten |= (1 << chan);
609 }
610 }
611 }
612 }
613 return FALSE;
614 }
615
616
617 /**
618 * Initialize machine state by expanding tokens to full instructions,
619 * allocating temporary storage, setting up constants, etc.
620 * After this, we can call tgsi_exec_machine_run() many times.
621 */
622 void
623 tgsi_exec_machine_bind_shader(
624 struct tgsi_exec_machine *mach,
625 const struct tgsi_token *tokens,
626 uint numSamplers,
627 struct tgsi_sampler **samplers)
628 {
629 uint k;
630 struct tgsi_parse_context parse;
631 struct tgsi_full_instruction *instructions;
632 struct tgsi_full_declaration *declarations;
633 uint maxInstructions = 10, numInstructions = 0;
634 uint maxDeclarations = 10, numDeclarations = 0;
635
636 #if 0
637 tgsi_dump(tokens, 0);
638 #endif
639
640 util_init_math();
641
642 if (numSamplers) {
643 assert(samplers);
644 }
645
646 mach->Tokens = tokens;
647 mach->Samplers = samplers;
648
649 if (!tokens) {
650 /* unbind and free all */
651 if (mach->Declarations) {
652 FREE( mach->Declarations );
653 }
654 mach->Declarations = NULL;
655 mach->NumDeclarations = 0;
656
657 if (mach->Instructions) {
658 FREE( mach->Instructions );
659 }
660 mach->Instructions = NULL;
661 mach->NumInstructions = 0;
662
663 return;
664 }
665
666 k = tgsi_parse_init (&parse, mach->Tokens);
667 if (k != TGSI_PARSE_OK) {
668 debug_printf( "Problem parsing!\n" );
669 return;
670 }
671
672 mach->Processor = parse.FullHeader.Processor.Processor;
673 mach->ImmLimit = 0;
674
675 if (mach->Processor == TGSI_PROCESSOR_GEOMETRY &&
676 !mach->UsedGeometryShader) {
677 struct tgsi_exec_vector *inputs =
678 align_malloc(sizeof(struct tgsi_exec_vector) *
679 TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS,
680 16);
681 struct tgsi_exec_vector *outputs =
682 align_malloc(sizeof(struct tgsi_exec_vector) *
683 TGSI_MAX_TOTAL_VERTICES, 16);
684
685 if (!inputs)
686 return;
687 if (!outputs) {
688 align_free(inputs);
689 return;
690 }
691
692 align_free(mach->Inputs);
693 align_free(mach->Outputs);
694
695 mach->Inputs = inputs;
696 mach->Outputs = outputs;
697 mach->UsedGeometryShader = TRUE;
698 }
699
700 declarations = (struct tgsi_full_declaration *)
701 MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) );
702
703 if (!declarations) {
704 return;
705 }
706
707 instructions = (struct tgsi_full_instruction *)
708 MALLOC( maxInstructions * sizeof(struct tgsi_full_instruction) );
709
710 if (!instructions) {
711 FREE( declarations );
712 return;
713 }
714
715 while( !tgsi_parse_end_of_tokens( &parse ) ) {
716 uint i;
717
718 tgsi_parse_token( &parse );
719 switch( parse.FullToken.Token.Type ) {
720 case TGSI_TOKEN_TYPE_DECLARATION:
721 /* save expanded declaration */
722 if (numDeclarations == maxDeclarations) {
723 declarations = REALLOC(declarations,
724 maxDeclarations
725 * sizeof(struct tgsi_full_declaration),
726 (maxDeclarations + 10)
727 * sizeof(struct tgsi_full_declaration));
728 maxDeclarations += 10;
729 }
730 if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_OUTPUT) {
731 unsigned reg;
732 for (reg = parse.FullToken.FullDeclaration.Range.First;
733 reg <= parse.FullToken.FullDeclaration.Range.Last;
734 ++reg) {
735 ++mach->NumOutputs;
736 }
737 }
738 if (parse.FullToken.FullDeclaration.Declaration.File ==
739 TGSI_FILE_IMMEDIATE_ARRAY) {
740 unsigned reg;
741 struct tgsi_full_declaration *decl =
742 &parse.FullToken.FullDeclaration;
743 debug_assert(decl->Range.Last < TGSI_EXEC_NUM_IMMEDIATES);
744 for (reg = decl->Range.First; reg <= decl->Range.Last; ++reg) {
745 for( i = 0; i < 4; i++ ) {
746 int idx = reg * 4 + i;
747 mach->ImmArray[reg][i] = decl->ImmediateData.u[idx].Float;
748 }
749 }
750 }
751 memcpy(declarations + numDeclarations,
752 &parse.FullToken.FullDeclaration,
753 sizeof(declarations[0]));
754 numDeclarations++;
755 break;
756
757 case TGSI_TOKEN_TYPE_IMMEDIATE:
758 {
759 uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
760 assert( size <= 4 );
761 assert( mach->ImmLimit + 1 <= TGSI_EXEC_NUM_IMMEDIATES );
762
763 for( i = 0; i < size; i++ ) {
764 mach->Imms[mach->ImmLimit][i] =
765 parse.FullToken.FullImmediate.u[i].Float;
766 }
767 mach->ImmLimit += 1;
768 }
769 break;
770
771 case TGSI_TOKEN_TYPE_INSTRUCTION:
772
773 /* save expanded instruction */
774 if (numInstructions == maxInstructions) {
775 instructions = REALLOC(instructions,
776 maxInstructions
777 * sizeof(struct tgsi_full_instruction),
778 (maxInstructions + 10)
779 * sizeof(struct tgsi_full_instruction));
780 maxInstructions += 10;
781 }
782
783 memcpy(instructions + numInstructions,
784 &parse.FullToken.FullInstruction,
785 sizeof(instructions[0]));
786
787 numInstructions++;
788 break;
789
790 case TGSI_TOKEN_TYPE_PROPERTY:
791 break;
792
793 default:
794 assert( 0 );
795 }
796 }
797 tgsi_parse_free (&parse);
798
799 if (mach->Declarations) {
800 FREE( mach->Declarations );
801 }
802 mach->Declarations = declarations;
803 mach->NumDeclarations = numDeclarations;
804
805 if (mach->Instructions) {
806 FREE( mach->Instructions );
807 }
808 mach->Instructions = instructions;
809 mach->NumInstructions = numInstructions;
810 }
811
812
813 struct tgsi_exec_machine *
814 tgsi_exec_machine_create( void )
815 {
816 struct tgsi_exec_machine *mach;
817 uint i;
818
819 mach = align_malloc( sizeof *mach, 16 );
820 if (!mach)
821 goto fail;
822
823 memset(mach, 0, sizeof(*mach));
824
825 mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR];
826 mach->MaxGeometryShaderOutputs = TGSI_MAX_TOTAL_VERTICES;
827 mach->Predicates = &mach->Temps[TGSI_EXEC_TEMP_P0];
828
829 mach->Inputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16);
830 mach->Outputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16);
831 if (!mach->Inputs || !mach->Outputs)
832 goto fail;
833
834 /* Setup constants needed by the SSE2 executor. */
835 for( i = 0; i < 4; i++ ) {
836 mach->Temps[TGSI_EXEC_TEMP_00000000_I].xyzw[TGSI_EXEC_TEMP_00000000_C].u[i] = 0x00000000;
837 mach->Temps[TGSI_EXEC_TEMP_7FFFFFFF_I].xyzw[TGSI_EXEC_TEMP_7FFFFFFF_C].u[i] = 0x7FFFFFFF;
838 mach->Temps[TGSI_EXEC_TEMP_80000000_I].xyzw[TGSI_EXEC_TEMP_80000000_C].u[i] = 0x80000000;
839 mach->Temps[TGSI_EXEC_TEMP_FFFFFFFF_I].xyzw[TGSI_EXEC_TEMP_FFFFFFFF_C].u[i] = 0xFFFFFFFF; /* not used */
840 mach->Temps[TGSI_EXEC_TEMP_ONE_I].xyzw[TGSI_EXEC_TEMP_ONE_C].f[i] = 1.0f;
841 mach->Temps[TGSI_EXEC_TEMP_TWO_I].xyzw[TGSI_EXEC_TEMP_TWO_C].f[i] = 2.0f; /* not used */
842 mach->Temps[TGSI_EXEC_TEMP_128_I].xyzw[TGSI_EXEC_TEMP_128_C].f[i] = 128.0f;
843 mach->Temps[TGSI_EXEC_TEMP_MINUS_128_I].xyzw[TGSI_EXEC_TEMP_MINUS_128_C].f[i] = -128.0f;
844 mach->Temps[TGSI_EXEC_TEMP_THREE_I].xyzw[TGSI_EXEC_TEMP_THREE_C].f[i] = 3.0f;
845 mach->Temps[TGSI_EXEC_TEMP_HALF_I].xyzw[TGSI_EXEC_TEMP_HALF_C].f[i] = 0.5f;
846 }
847
848 #ifdef DEBUG
849 /* silence warnings */
850 (void) print_chan;
851 (void) print_temp;
852 #endif
853
854 return mach;
855
856 fail:
857 if (mach) {
858 align_free(mach->Inputs);
859 align_free(mach->Outputs);
860 align_free(mach);
861 }
862 return NULL;
863 }
864
865
866 void
867 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach)
868 {
869 if (mach) {
870 if (mach->Instructions)
871 FREE(mach->Instructions);
872 if (mach->Declarations)
873 FREE(mach->Declarations);
874
875 align_free(mach->Inputs);
876 align_free(mach->Outputs);
877
878 align_free(mach);
879 }
880 }
881
882 static void
883 micro_add(union tgsi_exec_channel *dst,
884 const union tgsi_exec_channel *src0,
885 const union tgsi_exec_channel *src1)
886 {
887 dst->f[0] = src0->f[0] + src1->f[0];
888 dst->f[1] = src0->f[1] + src1->f[1];
889 dst->f[2] = src0->f[2] + src1->f[2];
890 dst->f[3] = src0->f[3] + src1->f[3];
891 }
892
893 static void
894 micro_div(
895 union tgsi_exec_channel *dst,
896 const union tgsi_exec_channel *src0,
897 const union tgsi_exec_channel *src1 )
898 {
899 if (src1->f[0] != 0) {
900 dst->f[0] = src0->f[0] / src1->f[0];
901 }
902 if (src1->f[1] != 0) {
903 dst->f[1] = src0->f[1] / src1->f[1];
904 }
905 if (src1->f[2] != 0) {
906 dst->f[2] = src0->f[2] / src1->f[2];
907 }
908 if (src1->f[3] != 0) {
909 dst->f[3] = src0->f[3] / src1->f[3];
910 }
911 }
912
913 static void
914 micro_rcc(union tgsi_exec_channel *dst,
915 const union tgsi_exec_channel *src)
916 {
917 uint i;
918
919 for (i = 0; i < 4; i++) {
920 float recip = 1.0f / src->f[i];
921
922 if (recip > 0.0f) {
923 if (recip > 1.884467e+019f) {
924 dst->f[i] = 1.884467e+019f;
925 }
926 else if (recip < 5.42101e-020f) {
927 dst->f[i] = 5.42101e-020f;
928 }
929 else {
930 dst->f[i] = recip;
931 }
932 }
933 else {
934 if (recip < -1.884467e+019f) {
935 dst->f[i] = -1.884467e+019f;
936 }
937 else if (recip > -5.42101e-020f) {
938 dst->f[i] = -5.42101e-020f;
939 }
940 else {
941 dst->f[i] = recip;
942 }
943 }
944 }
945 }
946
947 static void
948 micro_lt(
949 union tgsi_exec_channel *dst,
950 const union tgsi_exec_channel *src0,
951 const union tgsi_exec_channel *src1,
952 const union tgsi_exec_channel *src2,
953 const union tgsi_exec_channel *src3 )
954 {
955 dst->f[0] = src0->f[0] < src1->f[0] ? src2->f[0] : src3->f[0];
956 dst->f[1] = src0->f[1] < src1->f[1] ? src2->f[1] : src3->f[1];
957 dst->f[2] = src0->f[2] < src1->f[2] ? src2->f[2] : src3->f[2];
958 dst->f[3] = src0->f[3] < src1->f[3] ? src2->f[3] : src3->f[3];
959 }
960
961 static void
962 micro_max(union tgsi_exec_channel *dst,
963 const union tgsi_exec_channel *src0,
964 const union tgsi_exec_channel *src1)
965 {
966 dst->f[0] = src0->f[0] > src1->f[0] ? src0->f[0] : src1->f[0];
967 dst->f[1] = src0->f[1] > src1->f[1] ? src0->f[1] : src1->f[1];
968 dst->f[2] = src0->f[2] > src1->f[2] ? src0->f[2] : src1->f[2];
969 dst->f[3] = src0->f[3] > src1->f[3] ? src0->f[3] : src1->f[3];
970 }
971
972 static void
973 micro_min(union tgsi_exec_channel *dst,
974 const union tgsi_exec_channel *src0,
975 const union tgsi_exec_channel *src1)
976 {
977 dst->f[0] = src0->f[0] < src1->f[0] ? src0->f[0] : src1->f[0];
978 dst->f[1] = src0->f[1] < src1->f[1] ? src0->f[1] : src1->f[1];
979 dst->f[2] = src0->f[2] < src1->f[2] ? src0->f[2] : src1->f[2];
980 dst->f[3] = src0->f[3] < src1->f[3] ? src0->f[3] : src1->f[3];
981 }
982
983 static void
984 micro_mul(union tgsi_exec_channel *dst,
985 const union tgsi_exec_channel *src0,
986 const union tgsi_exec_channel *src1)
987 {
988 dst->f[0] = src0->f[0] * src1->f[0];
989 dst->f[1] = src0->f[1] * src1->f[1];
990 dst->f[2] = src0->f[2] * src1->f[2];
991 dst->f[3] = src0->f[3] * src1->f[3];
992 }
993
994 static void
995 micro_neg(
996 union tgsi_exec_channel *dst,
997 const union tgsi_exec_channel *src )
998 {
999 dst->f[0] = -src->f[0];
1000 dst->f[1] = -src->f[1];
1001 dst->f[2] = -src->f[2];
1002 dst->f[3] = -src->f[3];
1003 }
1004
1005 static void
1006 micro_pow(
1007 union tgsi_exec_channel *dst,
1008 const union tgsi_exec_channel *src0,
1009 const union tgsi_exec_channel *src1 )
1010 {
1011 #if FAST_MATH
1012 dst->f[0] = util_fast_pow( src0->f[0], src1->f[0] );
1013 dst->f[1] = util_fast_pow( src0->f[1], src1->f[1] );
1014 dst->f[2] = util_fast_pow( src0->f[2], src1->f[2] );
1015 dst->f[3] = util_fast_pow( src0->f[3], src1->f[3] );
1016 #else
1017 dst->f[0] = powf( src0->f[0], src1->f[0] );
1018 dst->f[1] = powf( src0->f[1], src1->f[1] );
1019 dst->f[2] = powf( src0->f[2], src1->f[2] );
1020 dst->f[3] = powf( src0->f[3], src1->f[3] );
1021 #endif
1022 }
1023
1024 static void
1025 micro_sub(union tgsi_exec_channel *dst,
1026 const union tgsi_exec_channel *src0,
1027 const union tgsi_exec_channel *src1)
1028 {
1029 dst->f[0] = src0->f[0] - src1->f[0];
1030 dst->f[1] = src0->f[1] - src1->f[1];
1031 dst->f[2] = src0->f[2] - src1->f[2];
1032 dst->f[3] = src0->f[3] - src1->f[3];
1033 }
1034
1035 static void
1036 fetch_src_file_channel(const struct tgsi_exec_machine *mach,
1037 const uint file,
1038 const uint swizzle,
1039 const union tgsi_exec_channel *index,
1040 const union tgsi_exec_channel *index2D,
1041 union tgsi_exec_channel *chan)
1042 {
1043 uint i;
1044
1045 assert(swizzle < 4);
1046
1047 switch (file) {
1048 case TGSI_FILE_CONSTANT:
1049 for (i = 0; i < QUAD_SIZE; i++) {
1050 assert(index2D->i[i] >= 0 && index2D->i[i] < PIPE_MAX_CONSTANT_BUFFERS);
1051 assert(mach->Consts[index2D->i[i]]);
1052
1053 if (index->i[i] < 0) {
1054 chan->u[i] = 0;
1055 } else {
1056 /* NOTE: copying the const value as a uint instead of float */
1057 const uint constbuf = index2D->i[i];
1058 const uint *buf = (const uint *)mach->Consts[constbuf];
1059 const int pos = index->i[i] * 4 + swizzle;
1060 /* const buffer bounds check */
1061 if (pos < 0 || pos >= mach->ConstsSize[constbuf]) {
1062 if (0) {
1063 /* Debug: print warning */
1064 static int count = 0;
1065 if (count++ < 100)
1066 debug_printf("TGSI Exec: const buffer index %d"
1067 " out of bounds\n", pos);
1068 }
1069 chan->u[i] = 0;
1070 }
1071 else
1072 chan->u[i] = buf[pos];
1073 }
1074 }
1075 break;
1076
1077 case TGSI_FILE_INPUT:
1078 for (i = 0; i < QUAD_SIZE; i++) {
1079 /*
1080 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
1081 debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n",
1082 index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i],
1083 index2D->i[i], index->i[i]);
1084 }*/
1085 int pos = index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i];
1086 assert(pos >= 0);
1087 assert(pos < TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS);
1088 chan->u[i] = mach->Inputs[pos].xyzw[swizzle].u[i];
1089 }
1090 break;
1091
1092 case TGSI_FILE_SYSTEM_VALUE:
1093 /* XXX no swizzling at this point. Will be needed if we put
1094 * gl_FragCoord, for example, in a sys value register.
1095 */
1096 for (i = 0; i < QUAD_SIZE; i++) {
1097 chan->f[i] = mach->SystemValue[index->i[i]][0];
1098 }
1099 break;
1100
1101 case TGSI_FILE_TEMPORARY:
1102 for (i = 0; i < QUAD_SIZE; i++) {
1103 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS);
1104 assert(index2D->i[i] == 0);
1105
1106 chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i];
1107 }
1108 break;
1109
1110 case TGSI_FILE_TEMPORARY_ARRAY:
1111 for (i = 0; i < QUAD_SIZE; i++) {
1112 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS);
1113 assert(index2D->i[i] < TGSI_EXEC_NUM_TEMP_ARRAYS);
1114
1115 chan->u[i] =
1116 mach->TempArray[index2D->i[i]][index->i[i]].xyzw[swizzle].u[i];
1117 }
1118 break;
1119
1120 case TGSI_FILE_IMMEDIATE:
1121 for (i = 0; i < QUAD_SIZE; i++) {
1122 assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit);
1123 assert(index2D->i[i] == 0);
1124
1125 chan->f[i] = mach->Imms[index->i[i]][swizzle];
1126 }
1127 break;
1128
1129 case TGSI_FILE_IMMEDIATE_ARRAY:
1130 for (i = 0; i < QUAD_SIZE; i++) {
1131 assert(index2D->i[i] == 0);
1132
1133 chan->f[i] = mach->ImmArray[index->i[i]][swizzle];
1134 }
1135 break;
1136
1137 case TGSI_FILE_ADDRESS:
1138 for (i = 0; i < QUAD_SIZE; i++) {
1139 assert(index->i[i] >= 0);
1140 assert(index2D->i[i] == 0);
1141
1142 chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i];
1143 }
1144 break;
1145
1146 case TGSI_FILE_PREDICATE:
1147 for (i = 0; i < QUAD_SIZE; i++) {
1148 assert(index->i[i] >= 0 && index->i[i] < TGSI_EXEC_NUM_PREDS);
1149 assert(index2D->i[i] == 0);
1150
1151 chan->u[i] = mach->Predicates[0].xyzw[swizzle].u[i];
1152 }
1153 break;
1154
1155 case TGSI_FILE_OUTPUT:
1156 /* vertex/fragment output vars can be read too */
1157 for (i = 0; i < QUAD_SIZE; i++) {
1158 assert(index->i[i] >= 0);
1159 assert(index2D->i[i] == 0);
1160
1161 chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i];
1162 }
1163 break;
1164
1165 default:
1166 assert(0);
1167 for (i = 0; i < QUAD_SIZE; i++) {
1168 chan->u[i] = 0;
1169 }
1170 }
1171 }
1172
1173 static void
1174 fetch_source(const struct tgsi_exec_machine *mach,
1175 union tgsi_exec_channel *chan,
1176 const struct tgsi_full_src_register *reg,
1177 const uint chan_index,
1178 enum tgsi_exec_datatype src_datatype)
1179 {
1180 union tgsi_exec_channel index;
1181 union tgsi_exec_channel index2D;
1182 uint swizzle;
1183
1184 /* We start with a direct index into a register file.
1185 *
1186 * file[1],
1187 * where:
1188 * file = Register.File
1189 * [1] = Register.Index
1190 */
1191 index.i[0] =
1192 index.i[1] =
1193 index.i[2] =
1194 index.i[3] = reg->Register.Index;
1195
1196 /* There is an extra source register that indirectly subscripts
1197 * a register file. The direct index now becomes an offset
1198 * that is being added to the indirect register.
1199 *
1200 * file[ind[2].x+1],
1201 * where:
1202 * ind = Indirect.File
1203 * [2] = Indirect.Index
1204 * .x = Indirect.SwizzleX
1205 */
1206 if (reg->Register.Indirect) {
1207 union tgsi_exec_channel index2;
1208 union tgsi_exec_channel indir_index;
1209 const uint execmask = mach->ExecMask;
1210 uint i;
1211
1212 /* which address register (always zero now) */
1213 index2.i[0] =
1214 index2.i[1] =
1215 index2.i[2] =
1216 index2.i[3] = reg->Indirect.Index;
1217 assert(reg->Indirect.File == TGSI_FILE_ADDRESS);
1218 /* get current value of address register[swizzle] */
1219 swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
1220 fetch_src_file_channel(mach,
1221 reg->Indirect.File,
1222 swizzle,
1223 &index2,
1224 &ZeroVec,
1225 &indir_index);
1226
1227 /* add value of address register to the offset */
1228 index.i[0] += indir_index.i[0];
1229 index.i[1] += indir_index.i[1];
1230 index.i[2] += indir_index.i[2];
1231 index.i[3] += indir_index.i[3];
1232
1233 /* for disabled execution channels, zero-out the index to
1234 * avoid using a potential garbage value.
1235 */
1236 for (i = 0; i < QUAD_SIZE; i++) {
1237 if ((execmask & (1 << i)) == 0)
1238 index.i[i] = 0;
1239 }
1240 }
1241
1242 /* There is an extra source register that is a second
1243 * subscript to a register file. Effectively it means that
1244 * the register file is actually a 2D array of registers.
1245 *
1246 * file[3][1],
1247 * where:
1248 * [3] = Dimension.Index
1249 */
1250 if (reg->Register.Dimension) {
1251 index2D.i[0] =
1252 index2D.i[1] =
1253 index2D.i[2] =
1254 index2D.i[3] = reg->Dimension.Index;
1255
1256 /* Again, the second subscript index can be addressed indirectly
1257 * identically to the first one.
1258 * Nothing stops us from indirectly addressing the indirect register,
1259 * but there is no need for that, so we won't exercise it.
1260 *
1261 * file[ind[4].y+3][1],
1262 * where:
1263 * ind = DimIndirect.File
1264 * [4] = DimIndirect.Index
1265 * .y = DimIndirect.SwizzleX
1266 */
1267 if (reg->Dimension.Indirect) {
1268 union tgsi_exec_channel index2;
1269 union tgsi_exec_channel indir_index;
1270 const uint execmask = mach->ExecMask;
1271 uint i;
1272
1273 index2.i[0] =
1274 index2.i[1] =
1275 index2.i[2] =
1276 index2.i[3] = reg->DimIndirect.Index;
1277
1278 swizzle = tgsi_util_get_src_register_swizzle( &reg->DimIndirect, CHAN_X );
1279 fetch_src_file_channel(mach,
1280 reg->DimIndirect.File,
1281 swizzle,
1282 &index2,
1283 &ZeroVec,
1284 &indir_index);
1285
1286 index2D.i[0] += indir_index.i[0];
1287 index2D.i[1] += indir_index.i[1];
1288 index2D.i[2] += indir_index.i[2];
1289 index2D.i[3] += indir_index.i[3];
1290
1291 /* for disabled execution channels, zero-out the index to
1292 * avoid using a potential garbage value.
1293 */
1294 for (i = 0; i < QUAD_SIZE; i++) {
1295 if ((execmask & (1 << i)) == 0) {
1296 index2D.i[i] = 0;
1297 }
1298 }
1299 }
1300
1301 /* If by any chance there was a need for a 3D array of register
1302 * files, we would have to check whether Dimension is followed
1303 * by a dimension register and continue the saga.
1304 */
1305 } else {
1306 index2D.i[0] =
1307 index2D.i[1] =
1308 index2D.i[2] =
1309 index2D.i[3] = 0;
1310 }
1311
1312 swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index );
1313 fetch_src_file_channel(mach,
1314 reg->Register.File,
1315 swizzle,
1316 &index,
1317 &index2D,
1318 chan);
1319
1320 if (reg->Register.Absolute) {
1321 if (src_datatype == TGSI_EXEC_DATA_FLOAT) {
1322 micro_abs(chan, chan);
1323 } else {
1324 micro_iabs(chan, chan);
1325 }
1326 }
1327
1328 if (reg->Register.Negate) {
1329 if (src_datatype == TGSI_EXEC_DATA_FLOAT) {
1330 micro_neg(chan, chan);
1331 } else {
1332 micro_ineg(chan, chan);
1333 }
1334 }
1335 }
1336
1337 static void
1338 store_dest(struct tgsi_exec_machine *mach,
1339 const union tgsi_exec_channel *chan,
1340 const struct tgsi_full_dst_register *reg,
1341 const struct tgsi_full_instruction *inst,
1342 uint chan_index,
1343 enum tgsi_exec_datatype dst_datatype)
1344 {
1345 uint i;
1346 union tgsi_exec_channel null;
1347 union tgsi_exec_channel *dst;
1348 union tgsi_exec_channel index2D;
1349 uint execmask = mach->ExecMask;
1350 int offset = 0; /* indirection offset */
1351 int index;
1352
1353 /* for debugging */
1354 if (0 && dst_datatype == TGSI_EXEC_DATA_FLOAT) {
1355 check_inf_or_nan(chan);
1356 }
1357
1358 /* There is an extra source register that indirectly subscripts
1359 * a register file. The direct index now becomes an offset
1360 * that is being added to the indirect register.
1361 *
1362 * file[ind[2].x+1],
1363 * where:
1364 * ind = Indirect.File
1365 * [2] = Indirect.Index
1366 * .x = Indirect.SwizzleX
1367 */
1368 if (reg->Register.Indirect) {
1369 union tgsi_exec_channel index;
1370 union tgsi_exec_channel indir_index;
1371 uint swizzle;
1372
1373 /* which address register (always zero for now) */
1374 index.i[0] =
1375 index.i[1] =
1376 index.i[2] =
1377 index.i[3] = reg->Indirect.Index;
1378
1379 /* get current value of address register[swizzle] */
1380 swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
1381
1382 /* fetch values from the address/indirection register */
1383 fetch_src_file_channel(mach,
1384 reg->Indirect.File,
1385 swizzle,
1386 &index,
1387 &ZeroVec,
1388 &indir_index);
1389
1390 /* save indirection offset */
1391 offset = indir_index.i[0];
1392 }
1393
1394 /* There is an extra source register that is a second
1395 * subscript to a register file. Effectively it means that
1396 * the register file is actually a 2D array of registers.
1397 *
1398 * file[3][1],
1399 * where:
1400 * [3] = Dimension.Index
1401 */
1402 if (reg->Register.Dimension) {
1403 index2D.i[0] =
1404 index2D.i[1] =
1405 index2D.i[2] =
1406 index2D.i[3] = reg->Dimension.Index;
1407
1408 /* Again, the second subscript index can be addressed indirectly
1409 * identically to the first one.
1410 * Nothing stops us from indirectly addressing the indirect register,
1411 * but there is no need for that, so we won't exercise it.
1412 *
1413 * file[ind[4].y+3][1],
1414 * where:
1415 * ind = DimIndirect.File
1416 * [4] = DimIndirect.Index
1417 * .y = DimIndirect.SwizzleX
1418 */
1419 if (reg->Dimension.Indirect) {
1420 union tgsi_exec_channel index2;
1421 union tgsi_exec_channel indir_index;
1422 const uint execmask = mach->ExecMask;
1423 unsigned swizzle;
1424 uint i;
1425
1426 index2.i[0] =
1427 index2.i[1] =
1428 index2.i[2] =
1429 index2.i[3] = reg->DimIndirect.Index;
1430
1431 swizzle = tgsi_util_get_src_register_swizzle( &reg->DimIndirect, CHAN_X );
1432 fetch_src_file_channel(mach,
1433 reg->DimIndirect.File,
1434 swizzle,
1435 &index2,
1436 &ZeroVec,
1437 &indir_index);
1438
1439 index2D.i[0] += indir_index.i[0];
1440 index2D.i[1] += indir_index.i[1];
1441 index2D.i[2] += indir_index.i[2];
1442 index2D.i[3] += indir_index.i[3];
1443
1444 /* for disabled execution channels, zero-out the index to
1445 * avoid using a potential garbage value.
1446 */
1447 for (i = 0; i < QUAD_SIZE; i++) {
1448 if ((execmask & (1 << i)) == 0) {
1449 index2D.i[i] = 0;
1450 }
1451 }
1452 }
1453
1454 /* If by any chance there was a need for a 3D array of register
1455 * files, we would have to check whether Dimension is followed
1456 * by a dimension register and continue the saga.
1457 */
1458 } else {
1459 index2D.i[0] =
1460 index2D.i[1] =
1461 index2D.i[2] =
1462 index2D.i[3] = 0;
1463 }
1464
1465 switch (reg->Register.File) {
1466 case TGSI_FILE_NULL:
1467 dst = &null;
1468 break;
1469
1470 case TGSI_FILE_OUTPUT:
1471 index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0]
1472 + reg->Register.Index;
1473 dst = &mach->Outputs[offset + index].xyzw[chan_index];
1474 #if 0
1475 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
1476 fprintf(stderr, "STORING OUT[%d] mask(%d), = (", offset + index, execmask);
1477 for (i = 0; i < QUAD_SIZE; i++)
1478 if (execmask & (1 << i))
1479 fprintf(stderr, "%f, ", chan->f[i]);
1480 fprintf(stderr, ")\n");
1481 }
1482 #endif
1483 break;
1484
1485 case TGSI_FILE_TEMPORARY:
1486 index = reg->Register.Index;
1487 assert( index < TGSI_EXEC_NUM_TEMPS );
1488 dst = &mach->Temps[offset + index].xyzw[chan_index];
1489 break;
1490
1491 case TGSI_FILE_TEMPORARY_ARRAY:
1492 index = reg->Register.Index;
1493 assert( index < TGSI_EXEC_NUM_TEMPS );
1494 assert( index2D.i[0] < TGSI_EXEC_NUM_TEMP_ARRAYS );
1495 /* XXX we use index2D.i[0] here but somehow we might
1496 * end up with someone trying to store indirectly in
1497 * different buffers */
1498 dst = &mach->TempArray[index2D.i[0]][offset + index].xyzw[chan_index];
1499 break;
1500
1501 case TGSI_FILE_ADDRESS:
1502 index = reg->Register.Index;
1503 dst = &mach->Addrs[index].xyzw[chan_index];
1504 break;
1505
1506 case TGSI_FILE_PREDICATE:
1507 index = reg->Register.Index;
1508 assert(index < TGSI_EXEC_NUM_PREDS);
1509 dst = &mach->Predicates[index].xyzw[chan_index];
1510 break;
1511
1512 default:
1513 assert( 0 );
1514 return;
1515 }
1516
1517 if (inst->Instruction.Predicate) {
1518 uint swizzle;
1519 union tgsi_exec_channel *pred;
1520
1521 switch (chan_index) {
1522 case CHAN_X:
1523 swizzle = inst->Predicate.SwizzleX;
1524 break;
1525 case CHAN_Y:
1526 swizzle = inst->Predicate.SwizzleY;
1527 break;
1528 case CHAN_Z:
1529 swizzle = inst->Predicate.SwizzleZ;
1530 break;
1531 case CHAN_W:
1532 swizzle = inst->Predicate.SwizzleW;
1533 break;
1534 default:
1535 assert(0);
1536 return;
1537 }
1538
1539 assert(inst->Predicate.Index == 0);
1540
1541 pred = &mach->Predicates[inst->Predicate.Index].xyzw[swizzle];
1542
1543 if (inst->Predicate.Negate) {
1544 for (i = 0; i < QUAD_SIZE; i++) {
1545 if (pred->u[i]) {
1546 execmask &= ~(1 << i);
1547 }
1548 }
1549 } else {
1550 for (i = 0; i < QUAD_SIZE; i++) {
1551 if (!pred->u[i]) {
1552 execmask &= ~(1 << i);
1553 }
1554 }
1555 }
1556 }
1557
1558 switch (inst->Instruction.Saturate) {
1559 case TGSI_SAT_NONE:
1560 for (i = 0; i < QUAD_SIZE; i++)
1561 if (execmask & (1 << i))
1562 dst->i[i] = chan->i[i];
1563 break;
1564
1565 case TGSI_SAT_ZERO_ONE:
1566 for (i = 0; i < QUAD_SIZE; i++)
1567 if (execmask & (1 << i)) {
1568 if (chan->f[i] < 0.0f)
1569 dst->f[i] = 0.0f;
1570 else if (chan->f[i] > 1.0f)
1571 dst->f[i] = 1.0f;
1572 else
1573 dst->i[i] = chan->i[i];
1574 }
1575 break;
1576
1577 case TGSI_SAT_MINUS_PLUS_ONE:
1578 for (i = 0; i < QUAD_SIZE; i++)
1579 if (execmask & (1 << i)) {
1580 if (chan->f[i] < -1.0f)
1581 dst->f[i] = -1.0f;
1582 else if (chan->f[i] > 1.0f)
1583 dst->f[i] = 1.0f;
1584 else
1585 dst->i[i] = chan->i[i];
1586 }
1587 break;
1588
1589 default:
1590 assert( 0 );
1591 }
1592 }
1593
1594 #define FETCH(VAL,INDEX,CHAN)\
1595 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT)
1596
1597
1598 /**
1599 * Execute ARB-style KIL which is predicated by a src register.
1600 * Kill fragment if any of the four values is less than zero.
1601 */
1602 static void
1603 exec_kil(struct tgsi_exec_machine *mach,
1604 const struct tgsi_full_instruction *inst)
1605 {
1606 uint uniquemask;
1607 uint chan_index;
1608 uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1609 union tgsi_exec_channel r[1];
1610
1611 /* This mask stores component bits that were already tested. */
1612 uniquemask = 0;
1613
1614 for (chan_index = 0; chan_index < 4; chan_index++)
1615 {
1616 uint swizzle;
1617 uint i;
1618
1619 /* unswizzle channel */
1620 swizzle = tgsi_util_get_full_src_register_swizzle (
1621 &inst->Src[0],
1622 chan_index);
1623
1624 /* check if the component has not been already tested */
1625 if (uniquemask & (1 << swizzle))
1626 continue;
1627 uniquemask |= 1 << swizzle;
1628
1629 FETCH(&r[0], 0, chan_index);
1630 for (i = 0; i < 4; i++)
1631 if (r[0].f[i] < 0.0f)
1632 kilmask |= 1 << i;
1633 }
1634
1635 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
1636 }
1637
1638 /**
1639 * Execute NVIDIA-style KIL which is predicated by a condition code.
1640 * Kill fragment if the condition code is TRUE.
1641 */
1642 static void
1643 exec_kilp(struct tgsi_exec_machine *mach,
1644 const struct tgsi_full_instruction *inst)
1645 {
1646 uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1647
1648 /* "unconditional" kil */
1649 kilmask = mach->ExecMask;
1650 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
1651 }
1652
1653 static void
1654 emit_vertex(struct tgsi_exec_machine *mach)
1655 {
1656 /* FIXME: check for exec mask correctly
1657 unsigned i;
1658 for (i = 0; i < QUAD_SIZE; ++i) {
1659 if ((mach->ExecMask & (1 << i)))
1660 */
1661 if (mach->ExecMask) {
1662 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += mach->NumOutputs;
1663 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++;
1664 }
1665 }
1666
1667 static void
1668 emit_primitive(struct tgsi_exec_machine *mach)
1669 {
1670 unsigned *prim_count = &mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0];
1671 /* FIXME: check for exec mask correctly
1672 unsigned i;
1673 for (i = 0; i < QUAD_SIZE; ++i) {
1674 if ((mach->ExecMask & (1 << i)))
1675 */
1676 if (mach->ExecMask) {
1677 ++(*prim_count);
1678 debug_assert((*prim_count * mach->NumOutputs) < mach->MaxGeometryShaderOutputs);
1679 mach->Primitives[*prim_count] = 0;
1680 }
1681 }
1682
1683 static void
1684 conditional_emit_primitive(struct tgsi_exec_machine *mach)
1685 {
1686 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
1687 int emitted_verts =
1688 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]];
1689 if (emitted_verts) {
1690 emit_primitive(mach);
1691 }
1692 }
1693 }
1694
1695
1696 /*
1697 * Fetch four texture samples using STR texture coordinates.
1698 */
1699 static void
1700 fetch_texel( struct tgsi_sampler *sampler,
1701 const union tgsi_exec_channel *s,
1702 const union tgsi_exec_channel *t,
1703 const union tgsi_exec_channel *p,
1704 const union tgsi_exec_channel *c0,
1705 enum tgsi_sampler_control control,
1706 union tgsi_exec_channel *r,
1707 union tgsi_exec_channel *g,
1708 union tgsi_exec_channel *b,
1709 union tgsi_exec_channel *a )
1710 {
1711 uint j;
1712 float rgba[NUM_CHANNELS][QUAD_SIZE];
1713
1714 sampler->get_samples(sampler, s->f, t->f, p->f, c0->f, control, rgba);
1715
1716 for (j = 0; j < 4; j++) {
1717 r->f[j] = rgba[0][j];
1718 g->f[j] = rgba[1][j];
1719 b->f[j] = rgba[2][j];
1720 a->f[j] = rgba[3][j];
1721 }
1722 }
1723
1724
1725 #define TEX_MODIFIER_NONE 0
1726 #define TEX_MODIFIER_PROJECTED 1
1727 #define TEX_MODIFIER_LOD_BIAS 2
1728 #define TEX_MODIFIER_EXPLICIT_LOD 3
1729
1730
1731 static void
1732 exec_tex(struct tgsi_exec_machine *mach,
1733 const struct tgsi_full_instruction *inst,
1734 uint modifier)
1735 {
1736 const uint unit = inst->Src[1].Register.Index;
1737 union tgsi_exec_channel r[4];
1738 const union tgsi_exec_channel *lod = &ZeroVec;
1739 enum tgsi_sampler_control control;
1740 uint chan;
1741
1742 if (modifier != TEX_MODIFIER_NONE) {
1743 FETCH(&r[3], 0, CHAN_W);
1744 if (modifier != TEX_MODIFIER_PROJECTED) {
1745 lod = &r[3];
1746 }
1747 }
1748
1749 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) {
1750 control = tgsi_sampler_lod_explicit;
1751 } else {
1752 control = tgsi_sampler_lod_bias;
1753 }
1754
1755 switch (inst->Texture.Texture) {
1756 case TGSI_TEXTURE_1D:
1757 case TGSI_TEXTURE_SHADOW1D:
1758 FETCH(&r[0], 0, CHAN_X);
1759
1760 if (modifier == TEX_MODIFIER_PROJECTED) {
1761 micro_div(&r[0], &r[0], &r[3]);
1762 }
1763
1764 fetch_texel(mach->Samplers[unit],
1765 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */
1766 control,
1767 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
1768 break;
1769
1770 case TGSI_TEXTURE_2D:
1771 case TGSI_TEXTURE_RECT:
1772 case TGSI_TEXTURE_SHADOW2D:
1773 case TGSI_TEXTURE_SHADOWRECT:
1774 FETCH(&r[0], 0, CHAN_X);
1775 FETCH(&r[1], 0, CHAN_Y);
1776 FETCH(&r[2], 0, CHAN_Z);
1777
1778 if (modifier == TEX_MODIFIER_PROJECTED) {
1779 micro_div(&r[0], &r[0], &r[3]);
1780 micro_div(&r[1], &r[1], &r[3]);
1781 micro_div(&r[2], &r[2], &r[3]);
1782 }
1783
1784 fetch_texel(mach->Samplers[unit],
1785 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */
1786 control,
1787 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1788 break;
1789
1790 case TGSI_TEXTURE_1D_ARRAY:
1791 FETCH(&r[0], 0, CHAN_X);
1792 FETCH(&r[1], 0, CHAN_Y);
1793
1794 if (modifier == TEX_MODIFIER_PROJECTED) {
1795 micro_div(&r[0], &r[0], &r[3]);
1796 }
1797
1798 fetch_texel(mach->Samplers[unit],
1799 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */
1800 control,
1801 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1802 break;
1803
1804 case TGSI_TEXTURE_2D_ARRAY:
1805 FETCH(&r[0], 0, CHAN_X);
1806 FETCH(&r[1], 0, CHAN_Y);
1807 FETCH(&r[2], 0, CHAN_Z);
1808
1809 if (modifier == TEX_MODIFIER_PROJECTED) {
1810 micro_div(&r[0], &r[0], &r[3]);
1811 micro_div(&r[1], &r[1], &r[3]);
1812 }
1813
1814 fetch_texel(mach->Samplers[unit],
1815 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */
1816 control,
1817 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1818 break;
1819
1820 case TGSI_TEXTURE_3D:
1821 case TGSI_TEXTURE_CUBE:
1822 FETCH(&r[0], 0, CHAN_X);
1823 FETCH(&r[1], 0, CHAN_Y);
1824 FETCH(&r[2], 0, CHAN_Z);
1825
1826 if (modifier == TEX_MODIFIER_PROJECTED) {
1827 micro_div(&r[0], &r[0], &r[3]);
1828 micro_div(&r[1], &r[1], &r[3]);
1829 micro_div(&r[2], &r[2], &r[3]);
1830 }
1831
1832 fetch_texel(mach->Samplers[unit],
1833 &r[0], &r[1], &r[2], lod,
1834 control,
1835 &r[0], &r[1], &r[2], &r[3]);
1836 break;
1837
1838 default:
1839 assert(0);
1840 }
1841
1842 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1843 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1844 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
1845 }
1846 }
1847 }
1848
1849 static void
1850 exec_txd(struct tgsi_exec_machine *mach,
1851 const struct tgsi_full_instruction *inst)
1852 {
1853 const uint unit = inst->Src[3].Register.Index;
1854 union tgsi_exec_channel r[4];
1855 uint chan;
1856
1857 /*
1858 * XXX: This is fake TXD -- the derivatives are not taken into account, yet.
1859 */
1860
1861 switch (inst->Texture.Texture) {
1862 case TGSI_TEXTURE_1D:
1863 case TGSI_TEXTURE_SHADOW1D:
1864
1865 FETCH(&r[0], 0, CHAN_X);
1866
1867 fetch_texel(mach->Samplers[unit],
1868 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */
1869 tgsi_sampler_lod_bias,
1870 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
1871 break;
1872
1873 case TGSI_TEXTURE_2D:
1874 case TGSI_TEXTURE_RECT:
1875 case TGSI_TEXTURE_SHADOW2D:
1876 case TGSI_TEXTURE_SHADOWRECT:
1877
1878 FETCH(&r[0], 0, CHAN_X);
1879 FETCH(&r[1], 0, CHAN_Y);
1880 FETCH(&r[2], 0, CHAN_Z);
1881
1882 fetch_texel(mach->Samplers[unit],
1883 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */
1884 tgsi_sampler_lod_bias,
1885 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1886 break;
1887
1888 case TGSI_TEXTURE_3D:
1889 case TGSI_TEXTURE_CUBE:
1890
1891 FETCH(&r[0], 0, CHAN_X);
1892 FETCH(&r[1], 0, CHAN_Y);
1893 FETCH(&r[2], 0, CHAN_Z);
1894
1895 fetch_texel(mach->Samplers[unit],
1896 &r[0], &r[1], &r[2], &ZeroVec,
1897 tgsi_sampler_lod_bias,
1898 &r[0], &r[1], &r[2], &r[3]);
1899 break;
1900
1901 default:
1902 assert(0);
1903 }
1904
1905 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1906 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1907 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
1908 }
1909 }
1910 }
1911
1912
1913
1914 static void
1915 exec_sample(struct tgsi_exec_machine *mach,
1916 const struct tgsi_full_instruction *inst,
1917 uint modifier)
1918 {
1919 const uint resource_unit = inst->Src[1].Register.Index;
1920 const uint sampler_unit = inst->Src[2].Register.Index;
1921 union tgsi_exec_channel r[4];
1922 const union tgsi_exec_channel *lod = &ZeroVec;
1923 enum tgsi_sampler_control control;
1924 uint chan;
1925
1926 if (modifier != TEX_MODIFIER_NONE) {
1927 if (modifier == TEX_MODIFIER_LOD_BIAS)
1928 FETCH(&r[3], 3, CHAN_X);
1929 else /*TEX_MODIFIER_LOD*/
1930 FETCH(&r[3], 0, CHAN_W);
1931
1932 if (modifier != TEX_MODIFIER_PROJECTED) {
1933 lod = &r[3];
1934 }
1935 }
1936
1937 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) {
1938 control = tgsi_sampler_lod_explicit;
1939 } else {
1940 control = tgsi_sampler_lod_bias;
1941 }
1942
1943 switch (mach->Resources[resource_unit].Resource) {
1944 case TGSI_TEXTURE_1D:
1945 case TGSI_TEXTURE_SHADOW1D:
1946 FETCH(&r[0], 0, CHAN_X);
1947
1948 if (modifier == TEX_MODIFIER_PROJECTED) {
1949 micro_div(&r[0], &r[0], &r[3]);
1950 }
1951
1952 fetch_texel(mach->Samplers[sampler_unit],
1953 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */
1954 control,
1955 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
1956 break;
1957
1958 case TGSI_TEXTURE_2D:
1959 case TGSI_TEXTURE_RECT:
1960 case TGSI_TEXTURE_SHADOW2D:
1961 case TGSI_TEXTURE_SHADOWRECT:
1962 FETCH(&r[0], 0, CHAN_X);
1963 FETCH(&r[1], 0, CHAN_Y);
1964 FETCH(&r[2], 0, CHAN_Z);
1965
1966 if (modifier == TEX_MODIFIER_PROJECTED) {
1967 micro_div(&r[0], &r[0], &r[3]);
1968 micro_div(&r[1], &r[1], &r[3]);
1969 micro_div(&r[2], &r[2], &r[3]);
1970 }
1971
1972 fetch_texel(mach->Samplers[sampler_unit],
1973 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */
1974 control,
1975 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1976 break;
1977
1978 case TGSI_TEXTURE_3D:
1979 case TGSI_TEXTURE_CUBE:
1980 FETCH(&r[0], 0, CHAN_X);
1981 FETCH(&r[1], 0, CHAN_Y);
1982 FETCH(&r[2], 0, CHAN_Z);
1983
1984 if (modifier == TEX_MODIFIER_PROJECTED) {
1985 micro_div(&r[0], &r[0], &r[3]);
1986 micro_div(&r[1], &r[1], &r[3]);
1987 micro_div(&r[2], &r[2], &r[3]);
1988 }
1989
1990 fetch_texel(mach->Samplers[sampler_unit],
1991 &r[0], &r[1], &r[2], lod,
1992 control,
1993 &r[0], &r[1], &r[2], &r[3]);
1994 break;
1995
1996 default:
1997 assert(0);
1998 }
1999
2000 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2001 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2002 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2003 }
2004 }
2005 }
2006
2007 static void
2008 exec_sample_d(struct tgsi_exec_machine *mach,
2009 const struct tgsi_full_instruction *inst)
2010 {
2011 const uint resource_unit = inst->Src[1].Register.Index;
2012 const uint sampler_unit = inst->Src[2].Register.Index;
2013 union tgsi_exec_channel r[4];
2014 uint chan;
2015 /*
2016 * XXX: This is fake SAMPLE_D -- the derivatives are not taken into account, yet.
2017 */
2018
2019 switch (mach->Resources[resource_unit].Resource) {
2020 case TGSI_TEXTURE_1D:
2021 case TGSI_TEXTURE_SHADOW1D:
2022
2023 FETCH(&r[0], 0, CHAN_X);
2024
2025 fetch_texel(mach->Samplers[sampler_unit],
2026 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */
2027 tgsi_sampler_lod_bias,
2028 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
2029 break;
2030
2031 case TGSI_TEXTURE_2D:
2032 case TGSI_TEXTURE_RECT:
2033 case TGSI_TEXTURE_SHADOW2D:
2034 case TGSI_TEXTURE_SHADOWRECT:
2035
2036 FETCH(&r[0], 0, CHAN_X);
2037 FETCH(&r[1], 0, CHAN_Y);
2038 FETCH(&r[2], 0, CHAN_Z);
2039
2040 fetch_texel(mach->Samplers[sampler_unit],
2041 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */
2042 tgsi_sampler_lod_bias,
2043 &r[0], &r[1], &r[2], &r[3]); /* outputs */
2044 break;
2045
2046 case TGSI_TEXTURE_3D:
2047 case TGSI_TEXTURE_CUBE:
2048
2049 FETCH(&r[0], 0, CHAN_X);
2050 FETCH(&r[1], 0, CHAN_Y);
2051 FETCH(&r[2], 0, CHAN_Z);
2052
2053 fetch_texel(mach->Samplers[sampler_unit],
2054 &r[0], &r[1], &r[2], &ZeroVec,
2055 tgsi_sampler_lod_bias,
2056 &r[0], &r[1], &r[2], &r[3]);
2057 break;
2058
2059 default:
2060 assert(0);
2061 }
2062
2063 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2064 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2065 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2066 }
2067 }
2068 }
2069
2070
2071 /**
2072 * Evaluate a constant-valued coefficient at the position of the
2073 * current quad.
2074 */
2075 static void
2076 eval_constant_coef(
2077 struct tgsi_exec_machine *mach,
2078 unsigned attrib,
2079 unsigned chan )
2080 {
2081 unsigned i;
2082
2083 for( i = 0; i < QUAD_SIZE; i++ ) {
2084 mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan];
2085 }
2086 }
2087
2088 /**
2089 * Evaluate a linear-valued coefficient at the position of the
2090 * current quad.
2091 */
2092 static void
2093 eval_linear_coef(
2094 struct tgsi_exec_machine *mach,
2095 unsigned attrib,
2096 unsigned chan )
2097 {
2098 const float x = mach->QuadPos.xyzw[0].f[0];
2099 const float y = mach->QuadPos.xyzw[1].f[0];
2100 const float dadx = mach->InterpCoefs[attrib].dadx[chan];
2101 const float dady = mach->InterpCoefs[attrib].dady[chan];
2102 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
2103 mach->Inputs[attrib].xyzw[chan].f[0] = a0;
2104 mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx;
2105 mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady;
2106 mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady;
2107 }
2108
2109 /**
2110 * Evaluate a perspective-valued coefficient at the position of the
2111 * current quad.
2112 */
2113 static void
2114 eval_perspective_coef(
2115 struct tgsi_exec_machine *mach,
2116 unsigned attrib,
2117 unsigned chan )
2118 {
2119 const float x = mach->QuadPos.xyzw[0].f[0];
2120 const float y = mach->QuadPos.xyzw[1].f[0];
2121 const float dadx = mach->InterpCoefs[attrib].dadx[chan];
2122 const float dady = mach->InterpCoefs[attrib].dady[chan];
2123 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
2124 const float *w = mach->QuadPos.xyzw[3].f;
2125 /* divide by W here */
2126 mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0];
2127 mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1];
2128 mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2];
2129 mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3];
2130 }
2131
2132
2133 typedef void (* eval_coef_func)(
2134 struct tgsi_exec_machine *mach,
2135 unsigned attrib,
2136 unsigned chan );
2137
2138 static void
2139 exec_declaration(struct tgsi_exec_machine *mach,
2140 const struct tgsi_full_declaration *decl)
2141 {
2142 if (decl->Declaration.File == TGSI_FILE_RESOURCE) {
2143 mach->Resources[decl->Range.First] = decl->Resource;
2144 return;
2145 }
2146
2147 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
2148 if (decl->Declaration.File == TGSI_FILE_INPUT) {
2149 uint first, last, mask;
2150
2151 first = decl->Range.First;
2152 last = decl->Range.Last;
2153 mask = decl->Declaration.UsageMask;
2154
2155 /* XXX we could remove this special-case code since
2156 * mach->InterpCoefs[first].a0 should already have the
2157 * front/back-face value. But we should first update the
2158 * ureg code to emit the right UsageMask value (WRITEMASK_X).
2159 * Then, we could remove the tgsi_exec_machine::Face field.
2160 */
2161 /* XXX make FACE a system value */
2162 if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) {
2163 uint i;
2164
2165 assert(decl->Semantic.Index == 0);
2166 assert(first == last);
2167
2168 for (i = 0; i < QUAD_SIZE; i++) {
2169 mach->Inputs[first].xyzw[0].f[i] = mach->Face;
2170 }
2171 } else {
2172 eval_coef_func eval;
2173 uint i, j;
2174
2175 switch (decl->Declaration.Interpolate) {
2176 case TGSI_INTERPOLATE_CONSTANT:
2177 eval = eval_constant_coef;
2178 break;
2179
2180 case TGSI_INTERPOLATE_LINEAR:
2181 eval = eval_linear_coef;
2182 break;
2183
2184 case TGSI_INTERPOLATE_PERSPECTIVE:
2185 eval = eval_perspective_coef;
2186 break;
2187
2188 default:
2189 assert(0);
2190 return;
2191 }
2192
2193 for (j = 0; j < NUM_CHANNELS; j++) {
2194 if (mask & (1 << j)) {
2195 for (i = first; i <= last; i++) {
2196 eval(mach, i, j);
2197 }
2198 }
2199 }
2200 }
2201 }
2202 }
2203
2204 if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) {
2205 mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First;
2206 }
2207 }
2208
2209
2210 typedef void (* micro_op)(union tgsi_exec_channel *dst);
2211
2212 static void
2213 exec_vector(struct tgsi_exec_machine *mach,
2214 const struct tgsi_full_instruction *inst,
2215 micro_op op,
2216 enum tgsi_exec_datatype dst_datatype)
2217 {
2218 unsigned int chan;
2219
2220 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2221 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2222 union tgsi_exec_channel dst;
2223
2224 op(&dst);
2225 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype);
2226 }
2227 }
2228 }
2229
2230 typedef void (* micro_unary_op)(union tgsi_exec_channel *dst,
2231 const union tgsi_exec_channel *src);
2232
2233 static void
2234 exec_scalar_unary(struct tgsi_exec_machine *mach,
2235 const struct tgsi_full_instruction *inst,
2236 micro_unary_op op,
2237 enum tgsi_exec_datatype dst_datatype,
2238 enum tgsi_exec_datatype src_datatype)
2239 {
2240 unsigned int chan;
2241 union tgsi_exec_channel src;
2242 union tgsi_exec_channel dst;
2243
2244 fetch_source(mach, &src, &inst->Src[0], CHAN_X, src_datatype);
2245 op(&dst, &src);
2246 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2247 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2248 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype);
2249 }
2250 }
2251 }
2252
2253 static void
2254 exec_vector_unary(struct tgsi_exec_machine *mach,
2255 const struct tgsi_full_instruction *inst,
2256 micro_unary_op op,
2257 enum tgsi_exec_datatype dst_datatype,
2258 enum tgsi_exec_datatype src_datatype)
2259 {
2260 unsigned int chan;
2261 struct tgsi_exec_vector dst;
2262
2263 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2264 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2265 union tgsi_exec_channel src;
2266
2267 fetch_source(mach, &src, &inst->Src[0], chan, src_datatype);
2268 op(&dst.xyzw[chan], &src);
2269 }
2270 }
2271 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2272 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2273 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype);
2274 }
2275 }
2276 }
2277
2278 typedef void (* micro_binary_op)(union tgsi_exec_channel *dst,
2279 const union tgsi_exec_channel *src0,
2280 const union tgsi_exec_channel *src1);
2281
2282 static void
2283 exec_scalar_binary(struct tgsi_exec_machine *mach,
2284 const struct tgsi_full_instruction *inst,
2285 micro_binary_op op,
2286 enum tgsi_exec_datatype dst_datatype,
2287 enum tgsi_exec_datatype src_datatype)
2288 {
2289 unsigned int chan;
2290 union tgsi_exec_channel src[2];
2291 union tgsi_exec_channel dst;
2292
2293 fetch_source(mach, &src[0], &inst->Src[0], CHAN_X, src_datatype);
2294 fetch_source(mach, &src[1], &inst->Src[1], CHAN_Y, src_datatype);
2295 op(&dst, &src[0], &src[1]);
2296 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2297 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2298 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype);
2299 }
2300 }
2301 }
2302
2303 static void
2304 exec_vector_binary(struct tgsi_exec_machine *mach,
2305 const struct tgsi_full_instruction *inst,
2306 micro_binary_op op,
2307 enum tgsi_exec_datatype dst_datatype,
2308 enum tgsi_exec_datatype src_datatype)
2309 {
2310 unsigned int chan;
2311 struct tgsi_exec_vector dst;
2312
2313 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2314 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2315 union tgsi_exec_channel src[2];
2316
2317 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype);
2318 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype);
2319 op(&dst.xyzw[chan], &src[0], &src[1]);
2320 }
2321 }
2322 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2323 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2324 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype);
2325 }
2326 }
2327 }
2328
2329 typedef void (* micro_trinary_op)(union tgsi_exec_channel *dst,
2330 const union tgsi_exec_channel *src0,
2331 const union tgsi_exec_channel *src1,
2332 const union tgsi_exec_channel *src2);
2333
2334 static void
2335 exec_vector_trinary(struct tgsi_exec_machine *mach,
2336 const struct tgsi_full_instruction *inst,
2337 micro_trinary_op op,
2338 enum tgsi_exec_datatype dst_datatype,
2339 enum tgsi_exec_datatype src_datatype)
2340 {
2341 unsigned int chan;
2342 struct tgsi_exec_vector dst;
2343
2344 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2345 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2346 union tgsi_exec_channel src[3];
2347
2348 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype);
2349 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype);
2350 fetch_source(mach, &src[2], &inst->Src[2], chan, src_datatype);
2351 op(&dst.xyzw[chan], &src[0], &src[1], &src[2]);
2352 }
2353 }
2354 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2355 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2356 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype);
2357 }
2358 }
2359 }
2360
2361 static void
2362 exec_dp3(struct tgsi_exec_machine *mach,
2363 const struct tgsi_full_instruction *inst)
2364 {
2365 unsigned int chan;
2366 union tgsi_exec_channel arg[3];
2367
2368 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2369 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2370 micro_mul(&arg[2], &arg[0], &arg[1]);
2371
2372 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) {
2373 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
2374 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT);
2375 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
2376 }
2377
2378 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2379 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2380 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2381 }
2382 }
2383 }
2384
2385 static void
2386 exec_dp4(struct tgsi_exec_machine *mach,
2387 const struct tgsi_full_instruction *inst)
2388 {
2389 unsigned int chan;
2390 union tgsi_exec_channel arg[3];
2391
2392 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2393 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2394 micro_mul(&arg[2], &arg[0], &arg[1]);
2395
2396 for (chan = CHAN_Y; chan <= CHAN_W; chan++) {
2397 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
2398 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT);
2399 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
2400 }
2401
2402 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2403 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2404 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2405 }
2406 }
2407 }
2408
2409 static void
2410 exec_dp2a(struct tgsi_exec_machine *mach,
2411 const struct tgsi_full_instruction *inst)
2412 {
2413 unsigned int chan;
2414 union tgsi_exec_channel arg[3];
2415
2416 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2417 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2418 micro_mul(&arg[2], &arg[0], &arg[1]);
2419
2420 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2421 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2422 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]);
2423
2424 fetch_source(mach, &arg[1], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2425 micro_add(&arg[0], &arg[0], &arg[1]);
2426
2427 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2428 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2429 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2430 }
2431 }
2432 }
2433
2434 static void
2435 exec_dph(struct tgsi_exec_machine *mach,
2436 const struct tgsi_full_instruction *inst)
2437 {
2438 unsigned int chan;
2439 union tgsi_exec_channel arg[3];
2440
2441 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2442 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2443 micro_mul(&arg[2], &arg[0], &arg[1]);
2444
2445 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2446 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2447 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
2448
2449 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2450 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2451 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]);
2452
2453 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT);
2454 micro_add(&arg[0], &arg[0], &arg[1]);
2455
2456 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2457 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2458 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2459 }
2460 }
2461 }
2462
2463 static void
2464 exec_dp2(struct tgsi_exec_machine *mach,
2465 const struct tgsi_full_instruction *inst)
2466 {
2467 unsigned int chan;
2468 union tgsi_exec_channel arg[3];
2469
2470 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2471 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2472 micro_mul(&arg[2], &arg[0], &arg[1]);
2473
2474 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2475 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2476 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
2477
2478 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2479 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2480 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2481 }
2482 }
2483 }
2484
2485 static void
2486 exec_nrm4(struct tgsi_exec_machine *mach,
2487 const struct tgsi_full_instruction *inst)
2488 {
2489 unsigned int chan;
2490 union tgsi_exec_channel arg[4];
2491 union tgsi_exec_channel scale;
2492
2493 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2494 micro_mul(&scale, &arg[0], &arg[0]);
2495
2496 for (chan = CHAN_Y; chan <= CHAN_W; chan++) {
2497 union tgsi_exec_channel product;
2498
2499 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
2500 micro_mul(&product, &arg[chan], &arg[chan]);
2501 micro_add(&scale, &scale, &product);
2502 }
2503
2504 micro_rsq(&scale, &scale);
2505
2506 for (chan = CHAN_X; chan <= CHAN_W; chan++) {
2507 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2508 micro_mul(&arg[chan], &arg[chan], &scale);
2509 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2510 }
2511 }
2512 }
2513
2514 static void
2515 exec_nrm3(struct tgsi_exec_machine *mach,
2516 const struct tgsi_full_instruction *inst)
2517 {
2518 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) {
2519 unsigned int chan;
2520 union tgsi_exec_channel arg[3];
2521 union tgsi_exec_channel scale;
2522
2523 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2524 micro_mul(&scale, &arg[0], &arg[0]);
2525
2526 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) {
2527 union tgsi_exec_channel product;
2528
2529 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
2530 micro_mul(&product, &arg[chan], &arg[chan]);
2531 micro_add(&scale, &scale, &product);
2532 }
2533
2534 micro_rsq(&scale, &scale);
2535
2536 for (chan = CHAN_X; chan <= CHAN_Z; chan++) {
2537 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2538 micro_mul(&arg[chan], &arg[chan], &scale);
2539 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2540 }
2541 }
2542 }
2543
2544 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2545 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2546 }
2547 }
2548
2549 static void
2550 exec_scs(struct tgsi_exec_machine *mach,
2551 const struct tgsi_full_instruction *inst)
2552 {
2553 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XY) {
2554 union tgsi_exec_channel arg;
2555 union tgsi_exec_channel result;
2556
2557 fetch_source(mach, &arg, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2558
2559 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2560 micro_cos(&result, &arg);
2561 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2562 }
2563 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2564 micro_sin(&result, &arg);
2565 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2566 }
2567 }
2568 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2569 store_dest(mach, &ZeroVec, &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2570 }
2571 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2572 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2573 }
2574 }
2575
2576 static void
2577 exec_x2d(struct tgsi_exec_machine *mach,
2578 const struct tgsi_full_instruction *inst)
2579 {
2580 union tgsi_exec_channel r[4];
2581 union tgsi_exec_channel d[2];
2582
2583 fetch_source(mach, &r[0], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2584 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2585 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XZ) {
2586 fetch_source(mach, &r[2], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2587 micro_mul(&r[2], &r[2], &r[0]);
2588 fetch_source(mach, &r[3], &inst->Src[2], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2589 micro_mul(&r[3], &r[3], &r[1]);
2590 micro_add(&r[2], &r[2], &r[3]);
2591 fetch_source(mach, &r[3], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2592 micro_add(&d[0], &r[2], &r[3]);
2593 }
2594 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YW) {
2595 fetch_source(mach, &r[2], &inst->Src[2], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2596 micro_mul(&r[2], &r[2], &r[0]);
2597 fetch_source(mach, &r[3], &inst->Src[2], CHAN_W, TGSI_EXEC_DATA_FLOAT);
2598 micro_mul(&r[3], &r[3], &r[1]);
2599 micro_add(&r[2], &r[2], &r[3]);
2600 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2601 micro_add(&d[1], &r[2], &r[3]);
2602 }
2603 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2604 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2605 }
2606 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2607 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2608 }
2609 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2610 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2611 }
2612 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2613 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2614 }
2615 }
2616
2617 static void
2618 exec_rfl(struct tgsi_exec_machine *mach,
2619 const struct tgsi_full_instruction *inst)
2620 {
2621 union tgsi_exec_channel r[9];
2622
2623 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) {
2624 /* r0 = dp3(src0, src0) */
2625 fetch_source(mach, &r[2], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2626 micro_mul(&r[0], &r[2], &r[2]);
2627 fetch_source(mach, &r[4], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2628 micro_mul(&r[8], &r[4], &r[4]);
2629 micro_add(&r[0], &r[0], &r[8]);
2630 fetch_source(mach, &r[6], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2631 micro_mul(&r[8], &r[6], &r[6]);
2632 micro_add(&r[0], &r[0], &r[8]);
2633
2634 /* r1 = dp3(src0, src1) */
2635 fetch_source(mach, &r[3], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2636 micro_mul(&r[1], &r[2], &r[3]);
2637 fetch_source(mach, &r[5], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2638 micro_mul(&r[8], &r[4], &r[5]);
2639 micro_add(&r[1], &r[1], &r[8]);
2640 fetch_source(mach, &r[7], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2641 micro_mul(&r[8], &r[6], &r[7]);
2642 micro_add(&r[1], &r[1], &r[8]);
2643
2644 /* r1 = 2 * r1 / r0 */
2645 micro_add(&r[1], &r[1], &r[1]);
2646 micro_div(&r[1], &r[1], &r[0]);
2647
2648 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2649 micro_mul(&r[2], &r[2], &r[1]);
2650 micro_sub(&r[2], &r[2], &r[3]);
2651 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2652 }
2653 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2654 micro_mul(&r[4], &r[4], &r[1]);
2655 micro_sub(&r[4], &r[4], &r[5]);
2656 store_dest(mach, &r[4], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2657 }
2658 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2659 micro_mul(&r[6], &r[6], &r[1]);
2660 micro_sub(&r[6], &r[6], &r[7]);
2661 store_dest(mach, &r[6], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2662 }
2663 }
2664 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2665 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2666 }
2667 }
2668
2669 static void
2670 exec_xpd(struct tgsi_exec_machine *mach,
2671 const struct tgsi_full_instruction *inst)
2672 {
2673 union tgsi_exec_channel r[6];
2674 union tgsi_exec_channel d[3];
2675
2676 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2677 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2678
2679 micro_mul(&r[2], &r[0], &r[1]);
2680
2681 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2682 fetch_source(mach, &r[4], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2683
2684 micro_mul(&r[5], &r[3], &r[4] );
2685 micro_sub(&d[CHAN_X], &r[2], &r[5]);
2686
2687 fetch_source(mach, &r[2], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2688
2689 micro_mul(&r[3], &r[3], &r[2]);
2690
2691 fetch_source(mach, &r[5], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2692
2693 micro_mul(&r[1], &r[1], &r[5]);
2694 micro_sub(&d[CHAN_Y], &r[3], &r[1]);
2695
2696 micro_mul(&r[5], &r[5], &r[4]);
2697 micro_mul(&r[0], &r[0], &r[2]);
2698 micro_sub(&d[CHAN_Z], &r[5], &r[0]);
2699
2700 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2701 store_dest(mach, &d[CHAN_X], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2702 }
2703 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2704 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2705 }
2706 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2707 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2708 }
2709 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2710 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2711 }
2712 }
2713
2714 static void
2715 exec_dst(struct tgsi_exec_machine *mach,
2716 const struct tgsi_full_instruction *inst)
2717 {
2718 union tgsi_exec_channel r[2];
2719 union tgsi_exec_channel d[4];
2720
2721 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2722 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2723 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2724 micro_mul(&d[CHAN_Y], &r[0], &r[1]);
2725 }
2726 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2727 fetch_source(mach, &d[CHAN_Z], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2728 }
2729 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2730 fetch_source(mach, &d[CHAN_W], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT);
2731 }
2732
2733 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2734 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2735 }
2736 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2737 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2738 }
2739 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2740 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2741 }
2742 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2743 store_dest(mach, &d[CHAN_W], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2744 }
2745 }
2746
2747 static void
2748 exec_log(struct tgsi_exec_machine *mach,
2749 const struct tgsi_full_instruction *inst)
2750 {
2751 union tgsi_exec_channel r[3];
2752
2753 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2754 micro_abs(&r[2], &r[0]); /* r2 = abs(r0) */
2755 micro_lg2(&r[1], &r[2]); /* r1 = lg2(r2) */
2756 micro_flr(&r[0], &r[1]); /* r0 = floor(r1) */
2757 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2758 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2759 }
2760 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2761 micro_exp2(&r[0], &r[0]); /* r0 = 2 ^ r0 */
2762 micro_div(&r[0], &r[2], &r[0]); /* r0 = r2 / r0 */
2763 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2764 }
2765 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2766 store_dest(mach, &r[1], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2767 }
2768 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2769 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2770 }
2771 }
2772
2773 static void
2774 exec_exp(struct tgsi_exec_machine *mach,
2775 const struct tgsi_full_instruction *inst)
2776 {
2777 union tgsi_exec_channel r[3];
2778
2779 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2780 micro_flr(&r[1], &r[0]); /* r1 = floor(r0) */
2781 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2782 micro_exp2(&r[2], &r[1]); /* r2 = 2 ^ r1 */
2783 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2784 }
2785 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2786 micro_sub(&r[2], &r[0], &r[1]); /* r2 = r0 - r1 */
2787 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2788 }
2789 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2790 micro_exp2(&r[2], &r[0]); /* r2 = 2 ^ r0 */
2791 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2792 }
2793 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2794 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2795 }
2796 }
2797
2798 static void
2799 exec_lit(struct tgsi_exec_machine *mach,
2800 const struct tgsi_full_instruction *inst)
2801 {
2802 union tgsi_exec_channel r[3];
2803 union tgsi_exec_channel d[3];
2804
2805 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2806 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2807 }
2808 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YZ) {
2809 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2810 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2811 micro_max(&d[CHAN_Y], &r[0], &ZeroVec);
2812 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2813 }
2814
2815 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2816 fetch_source(mach, &r[1], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2817 micro_max(&r[1], &r[1], &ZeroVec);
2818
2819 fetch_source(mach, &r[2], &inst->Src[0], CHAN_W, TGSI_EXEC_DATA_FLOAT);
2820 micro_min(&r[2], &r[2], &P128Vec);
2821 micro_max(&r[2], &r[2], &M128Vec);
2822 micro_pow(&r[1], &r[1], &r[2]);
2823 micro_lt(&d[CHAN_Z], &ZeroVec, &r[0], &r[1], &ZeroVec);
2824 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2825 }
2826 }
2827 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2828 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2829 }
2830 }
2831
2832 static void
2833 exec_break(struct tgsi_exec_machine *mach)
2834 {
2835 if (mach->BreakType == TGSI_EXEC_BREAK_INSIDE_LOOP) {
2836 /* turn off loop channels for each enabled exec channel */
2837 mach->LoopMask &= ~mach->ExecMask;
2838 /* Todo: if mach->LoopMask == 0, jump to end of loop */
2839 UPDATE_EXEC_MASK(mach);
2840 } else {
2841 assert(mach->BreakType == TGSI_EXEC_BREAK_INSIDE_SWITCH);
2842
2843 mach->Switch.mask = 0x0;
2844
2845 UPDATE_EXEC_MASK(mach);
2846 }
2847 }
2848
2849 static void
2850 exec_switch(struct tgsi_exec_machine *mach,
2851 const struct tgsi_full_instruction *inst)
2852 {
2853 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING);
2854 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
2855
2856 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch;
2857 fetch_source(mach, &mach->Switch.selector, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT);
2858 mach->Switch.mask = 0x0;
2859 mach->Switch.defaultMask = 0x0;
2860
2861 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
2862 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_SWITCH;
2863
2864 UPDATE_EXEC_MASK(mach);
2865 }
2866
2867 static void
2868 exec_case(struct tgsi_exec_machine *mach,
2869 const struct tgsi_full_instruction *inst)
2870 {
2871 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask;
2872 union tgsi_exec_channel src;
2873 uint mask = 0;
2874
2875 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT);
2876
2877 if (mach->Switch.selector.u[0] == src.u[0]) {
2878 mask |= 0x1;
2879 }
2880 if (mach->Switch.selector.u[1] == src.u[1]) {
2881 mask |= 0x2;
2882 }
2883 if (mach->Switch.selector.u[2] == src.u[2]) {
2884 mask |= 0x4;
2885 }
2886 if (mach->Switch.selector.u[3] == src.u[3]) {
2887 mask |= 0x8;
2888 }
2889
2890 mach->Switch.defaultMask |= mask;
2891
2892 mach->Switch.mask |= mask & prevMask;
2893
2894 UPDATE_EXEC_MASK(mach);
2895 }
2896
2897 static void
2898 exec_default(struct tgsi_exec_machine *mach)
2899 {
2900 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask;
2901
2902 mach->Switch.mask |= ~mach->Switch.defaultMask & prevMask;
2903
2904 UPDATE_EXEC_MASK(mach);
2905 }
2906
2907 static void
2908 exec_endswitch(struct tgsi_exec_machine *mach)
2909 {
2910 mach->Switch = mach->SwitchStack[--mach->SwitchStackTop];
2911 mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
2912
2913 UPDATE_EXEC_MASK(mach);
2914 }
2915
2916 static void
2917 micro_i2f(union tgsi_exec_channel *dst,
2918 const union tgsi_exec_channel *src)
2919 {
2920 dst->f[0] = (float)src->i[0];
2921 dst->f[1] = (float)src->i[1];
2922 dst->f[2] = (float)src->i[2];
2923 dst->f[3] = (float)src->i[3];
2924 }
2925
2926 static void
2927 micro_not(union tgsi_exec_channel *dst,
2928 const union tgsi_exec_channel *src)
2929 {
2930 dst->u[0] = ~src->u[0];
2931 dst->u[1] = ~src->u[1];
2932 dst->u[2] = ~src->u[2];
2933 dst->u[3] = ~src->u[3];
2934 }
2935
2936 static void
2937 micro_shl(union tgsi_exec_channel *dst,
2938 const union tgsi_exec_channel *src0,
2939 const union tgsi_exec_channel *src1)
2940 {
2941 dst->u[0] = src0->u[0] << src1->u[0];
2942 dst->u[1] = src0->u[1] << src1->u[1];
2943 dst->u[2] = src0->u[2] << src1->u[2];
2944 dst->u[3] = src0->u[3] << src1->u[3];
2945 }
2946
2947 static void
2948 micro_and(union tgsi_exec_channel *dst,
2949 const union tgsi_exec_channel *src0,
2950 const union tgsi_exec_channel *src1)
2951 {
2952 dst->u[0] = src0->u[0] & src1->u[0];
2953 dst->u[1] = src0->u[1] & src1->u[1];
2954 dst->u[2] = src0->u[2] & src1->u[2];
2955 dst->u[3] = src0->u[3] & src1->u[3];
2956 }
2957
2958 static void
2959 micro_or(union tgsi_exec_channel *dst,
2960 const union tgsi_exec_channel *src0,
2961 const union tgsi_exec_channel *src1)
2962 {
2963 dst->u[0] = src0->u[0] | src1->u[0];
2964 dst->u[1] = src0->u[1] | src1->u[1];
2965 dst->u[2] = src0->u[2] | src1->u[2];
2966 dst->u[3] = src0->u[3] | src1->u[3];
2967 }
2968
2969 static void
2970 micro_xor(union tgsi_exec_channel *dst,
2971 const union tgsi_exec_channel *src0,
2972 const union tgsi_exec_channel *src1)
2973 {
2974 dst->u[0] = src0->u[0] ^ src1->u[0];
2975 dst->u[1] = src0->u[1] ^ src1->u[1];
2976 dst->u[2] = src0->u[2] ^ src1->u[2];
2977 dst->u[3] = src0->u[3] ^ src1->u[3];
2978 }
2979
2980 static void
2981 micro_f2i(union tgsi_exec_channel *dst,
2982 const union tgsi_exec_channel *src)
2983 {
2984 dst->i[0] = (int)src->f[0];
2985 dst->i[1] = (int)src->f[1];
2986 dst->i[2] = (int)src->f[2];
2987 dst->i[3] = (int)src->f[3];
2988 }
2989
2990 static void
2991 micro_idiv(union tgsi_exec_channel *dst,
2992 const union tgsi_exec_channel *src0,
2993 const union tgsi_exec_channel *src1)
2994 {
2995 dst->i[0] = src0->i[0] / src1->i[0];
2996 dst->i[1] = src0->i[1] / src1->i[1];
2997 dst->i[2] = src0->i[2] / src1->i[2];
2998 dst->i[3] = src0->i[3] / src1->i[3];
2999 }
3000
3001 static void
3002 micro_imax(union tgsi_exec_channel *dst,
3003 const union tgsi_exec_channel *src0,
3004 const union tgsi_exec_channel *src1)
3005 {
3006 dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0];
3007 dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1];
3008 dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2];
3009 dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3];
3010 }
3011
3012 static void
3013 micro_imin(union tgsi_exec_channel *dst,
3014 const union tgsi_exec_channel *src0,
3015 const union tgsi_exec_channel *src1)
3016 {
3017 dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0];
3018 dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1];
3019 dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2];
3020 dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3];
3021 }
3022
3023 static void
3024 micro_isge(union tgsi_exec_channel *dst,
3025 const union tgsi_exec_channel *src0,
3026 const union tgsi_exec_channel *src1)
3027 {
3028 dst->i[0] = src0->i[0] >= src1->i[0] ? -1 : 0;
3029 dst->i[1] = src0->i[1] >= src1->i[1] ? -1 : 0;
3030 dst->i[2] = src0->i[2] >= src1->i[2] ? -1 : 0;
3031 dst->i[3] = src0->i[3] >= src1->i[3] ? -1 : 0;
3032 }
3033
3034 static void
3035 micro_ishr(union tgsi_exec_channel *dst,
3036 const union tgsi_exec_channel *src0,
3037 const union tgsi_exec_channel *src1)
3038 {
3039 dst->i[0] = src0->i[0] >> src1->i[0];
3040 dst->i[1] = src0->i[1] >> src1->i[1];
3041 dst->i[2] = src0->i[2] >> src1->i[2];
3042 dst->i[3] = src0->i[3] >> src1->i[3];
3043 }
3044
3045 static void
3046 micro_islt(union tgsi_exec_channel *dst,
3047 const union tgsi_exec_channel *src0,
3048 const union tgsi_exec_channel *src1)
3049 {
3050 dst->i[0] = src0->i[0] < src1->i[0] ? -1 : 0;
3051 dst->i[1] = src0->i[1] < src1->i[1] ? -1 : 0;
3052 dst->i[2] = src0->i[2] < src1->i[2] ? -1 : 0;
3053 dst->i[3] = src0->i[3] < src1->i[3] ? -1 : 0;
3054 }
3055
3056 static void
3057 micro_f2u(union tgsi_exec_channel *dst,
3058 const union tgsi_exec_channel *src)
3059 {
3060 dst->u[0] = (uint)src->f[0];
3061 dst->u[1] = (uint)src->f[1];
3062 dst->u[2] = (uint)src->f[2];
3063 dst->u[3] = (uint)src->f[3];
3064 }
3065
3066 static void
3067 micro_u2f(union tgsi_exec_channel *dst,
3068 const union tgsi_exec_channel *src)
3069 {
3070 dst->f[0] = (float)src->u[0];
3071 dst->f[1] = (float)src->u[1];
3072 dst->f[2] = (float)src->u[2];
3073 dst->f[3] = (float)src->u[3];
3074 }
3075
3076 static void
3077 micro_uadd(union tgsi_exec_channel *dst,
3078 const union tgsi_exec_channel *src0,
3079 const union tgsi_exec_channel *src1)
3080 {
3081 dst->u[0] = src0->u[0] + src1->u[0];
3082 dst->u[1] = src0->u[1] + src1->u[1];
3083 dst->u[2] = src0->u[2] + src1->u[2];
3084 dst->u[3] = src0->u[3] + src1->u[3];
3085 }
3086
3087 static void
3088 micro_udiv(union tgsi_exec_channel *dst,
3089 const union tgsi_exec_channel *src0,
3090 const union tgsi_exec_channel *src1)
3091 {
3092 dst->u[0] = src0->u[0] / src1->u[0];
3093 dst->u[1] = src0->u[1] / src1->u[1];
3094 dst->u[2] = src0->u[2] / src1->u[2];
3095 dst->u[3] = src0->u[3] / src1->u[3];
3096 }
3097
3098 static void
3099 micro_umad(union tgsi_exec_channel *dst,
3100 const union tgsi_exec_channel *src0,
3101 const union tgsi_exec_channel *src1,
3102 const union tgsi_exec_channel *src2)
3103 {
3104 dst->u[0] = src0->u[0] * src1->u[0] + src2->u[0];
3105 dst->u[1] = src0->u[1] * src1->u[1] + src2->u[1];
3106 dst->u[2] = src0->u[2] * src1->u[2] + src2->u[2];
3107 dst->u[3] = src0->u[3] * src1->u[3] + src2->u[3];
3108 }
3109
3110 static void
3111 micro_umax(union tgsi_exec_channel *dst,
3112 const union tgsi_exec_channel *src0,
3113 const union tgsi_exec_channel *src1)
3114 {
3115 dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0];
3116 dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1];
3117 dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2];
3118 dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3];
3119 }
3120
3121 static void
3122 micro_umin(union tgsi_exec_channel *dst,
3123 const union tgsi_exec_channel *src0,
3124 const union tgsi_exec_channel *src1)
3125 {
3126 dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0];
3127 dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1];
3128 dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2];
3129 dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3];
3130 }
3131
3132 static void
3133 micro_umod(union tgsi_exec_channel *dst,
3134 const union tgsi_exec_channel *src0,
3135 const union tgsi_exec_channel *src1)
3136 {
3137 dst->u[0] = src0->u[0] % src1->u[0];
3138 dst->u[1] = src0->u[1] % src1->u[1];
3139 dst->u[2] = src0->u[2] % src1->u[2];
3140 dst->u[3] = src0->u[3] % src1->u[3];
3141 }
3142
3143 static void
3144 micro_umul(union tgsi_exec_channel *dst,
3145 const union tgsi_exec_channel *src0,
3146 const union tgsi_exec_channel *src1)
3147 {
3148 dst->u[0] = src0->u[0] * src1->u[0];
3149 dst->u[1] = src0->u[1] * src1->u[1];
3150 dst->u[2] = src0->u[2] * src1->u[2];
3151 dst->u[3] = src0->u[3] * src1->u[3];
3152 }
3153
3154 static void
3155 micro_useq(union tgsi_exec_channel *dst,
3156 const union tgsi_exec_channel *src0,
3157 const union tgsi_exec_channel *src1)
3158 {
3159 dst->u[0] = src0->u[0] == src1->u[0] ? ~0 : 0;
3160 dst->u[1] = src0->u[1] == src1->u[1] ? ~0 : 0;
3161 dst->u[2] = src0->u[2] == src1->u[2] ? ~0 : 0;
3162 dst->u[3] = src0->u[3] == src1->u[3] ? ~0 : 0;
3163 }
3164
3165 static void
3166 micro_usge(union tgsi_exec_channel *dst,
3167 const union tgsi_exec_channel *src0,
3168 const union tgsi_exec_channel *src1)
3169 {
3170 dst->u[0] = src0->u[0] >= src1->u[0] ? ~0 : 0;
3171 dst->u[1] = src0->u[1] >= src1->u[1] ? ~0 : 0;
3172 dst->u[2] = src0->u[2] >= src1->u[2] ? ~0 : 0;
3173 dst->u[3] = src0->u[3] >= src1->u[3] ? ~0 : 0;
3174 }
3175
3176 static void
3177 micro_ushr(union tgsi_exec_channel *dst,
3178 const union tgsi_exec_channel *src0,
3179 const union tgsi_exec_channel *src1)
3180 {
3181 dst->u[0] = src0->u[0] >> src1->u[0];
3182 dst->u[1] = src0->u[1] >> src1->u[1];
3183 dst->u[2] = src0->u[2] >> src1->u[2];
3184 dst->u[3] = src0->u[3] >> src1->u[3];
3185 }
3186
3187 static void
3188 micro_uslt(union tgsi_exec_channel *dst,
3189 const union tgsi_exec_channel *src0,
3190 const union tgsi_exec_channel *src1)
3191 {
3192 dst->u[0] = src0->u[0] < src1->u[0] ? ~0 : 0;
3193 dst->u[1] = src0->u[1] < src1->u[1] ? ~0 : 0;
3194 dst->u[2] = src0->u[2] < src1->u[2] ? ~0 : 0;
3195 dst->u[3] = src0->u[3] < src1->u[3] ? ~0 : 0;
3196 }
3197
3198 static void
3199 micro_usne(union tgsi_exec_channel *dst,
3200 const union tgsi_exec_channel *src0,
3201 const union tgsi_exec_channel *src1)
3202 {
3203 dst->u[0] = src0->u[0] != src1->u[0] ? ~0 : 0;
3204 dst->u[1] = src0->u[1] != src1->u[1] ? ~0 : 0;
3205 dst->u[2] = src0->u[2] != src1->u[2] ? ~0 : 0;
3206 dst->u[3] = src0->u[3] != src1->u[3] ? ~0 : 0;
3207 }
3208
3209 static void
3210 exec_instruction(
3211 struct tgsi_exec_machine *mach,
3212 const struct tgsi_full_instruction *inst,
3213 int *pc )
3214 {
3215 union tgsi_exec_channel r[10];
3216
3217 (*pc)++;
3218
3219 switch (inst->Instruction.Opcode) {
3220 case TGSI_OPCODE_ARL:
3221 exec_vector_unary(mach, inst, micro_arl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT);
3222 break;
3223
3224 case TGSI_OPCODE_MOV:
3225 exec_vector_unary(mach, inst, micro_mov, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT);
3226 break;
3227
3228 case TGSI_OPCODE_LIT:
3229 exec_lit(mach, inst);
3230 break;
3231
3232 case TGSI_OPCODE_RCP:
3233 exec_scalar_unary(mach, inst, micro_rcp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3234 break;
3235
3236 case TGSI_OPCODE_RSQ:
3237 exec_scalar_unary(mach, inst, micro_rsq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3238 break;
3239
3240 case TGSI_OPCODE_EXP:
3241 exec_exp(mach, inst);
3242 break;
3243
3244 case TGSI_OPCODE_LOG:
3245 exec_log(mach, inst);
3246 break;
3247
3248 case TGSI_OPCODE_MUL:
3249 exec_vector_binary(mach, inst, micro_mul, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3250 break;
3251
3252 case TGSI_OPCODE_ADD:
3253 exec_vector_binary(mach, inst, micro_add, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3254 break;
3255
3256 case TGSI_OPCODE_DP3:
3257 exec_dp3(mach, inst);
3258 break;
3259
3260 case TGSI_OPCODE_DP4:
3261 exec_dp4(mach, inst);
3262 break;
3263
3264 case TGSI_OPCODE_DST:
3265 exec_dst(mach, inst);
3266 break;
3267
3268 case TGSI_OPCODE_MIN:
3269 exec_vector_binary(mach, inst, micro_min, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3270 break;
3271
3272 case TGSI_OPCODE_MAX:
3273 exec_vector_binary(mach, inst, micro_max, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3274 break;
3275
3276 case TGSI_OPCODE_SLT:
3277 exec_vector_binary(mach, inst, micro_slt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3278 break;
3279
3280 case TGSI_OPCODE_SGE:
3281 exec_vector_binary(mach, inst, micro_sge, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3282 break;
3283
3284 case TGSI_OPCODE_MAD:
3285 exec_vector_trinary(mach, inst, micro_mad, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3286 break;
3287
3288 case TGSI_OPCODE_SUB:
3289 exec_vector_binary(mach, inst, micro_sub, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3290 break;
3291
3292 case TGSI_OPCODE_LRP:
3293 exec_vector_trinary(mach, inst, micro_lrp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3294 break;
3295
3296 case TGSI_OPCODE_CND:
3297 exec_vector_trinary(mach, inst, micro_cnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3298 break;
3299
3300 case TGSI_OPCODE_DP2A:
3301 exec_dp2a(mach, inst);
3302 break;
3303
3304 case TGSI_OPCODE_FRC:
3305 exec_vector_unary(mach, inst, micro_frc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3306 break;
3307
3308 case TGSI_OPCODE_CLAMP:
3309 exec_vector_trinary(mach, inst, micro_clamp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3310 break;
3311
3312 case TGSI_OPCODE_FLR:
3313 exec_vector_unary(mach, inst, micro_flr, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3314 break;
3315
3316 case TGSI_OPCODE_ROUND:
3317 exec_vector_unary(mach, inst, micro_rnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3318 break;
3319
3320 case TGSI_OPCODE_EX2:
3321 exec_scalar_unary(mach, inst, micro_exp2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3322 break;
3323
3324 case TGSI_OPCODE_LG2:
3325 exec_scalar_unary(mach, inst, micro_lg2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3326 break;
3327
3328 case TGSI_OPCODE_POW:
3329 exec_scalar_binary(mach, inst, micro_pow, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3330 break;
3331
3332 case TGSI_OPCODE_XPD:
3333 exec_xpd(mach, inst);
3334 break;
3335
3336 case TGSI_OPCODE_ABS:
3337 exec_vector_unary(mach, inst, micro_abs, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3338 break;
3339
3340 case TGSI_OPCODE_RCC:
3341 exec_scalar_unary(mach, inst, micro_rcc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3342 break;
3343
3344 case TGSI_OPCODE_DPH:
3345 exec_dph(mach, inst);
3346 break;
3347
3348 case TGSI_OPCODE_COS:
3349 exec_scalar_unary(mach, inst, micro_cos, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3350 break;
3351
3352 case TGSI_OPCODE_DDX:
3353 exec_vector_unary(mach, inst, micro_ddx, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3354 break;
3355
3356 case TGSI_OPCODE_DDY:
3357 exec_vector_unary(mach, inst, micro_ddy, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3358 break;
3359
3360 case TGSI_OPCODE_KILP:
3361 exec_kilp (mach, inst);
3362 break;
3363
3364 case TGSI_OPCODE_KIL:
3365 exec_kil (mach, inst);
3366 break;
3367
3368 case TGSI_OPCODE_PK2H:
3369 assert (0);
3370 break;
3371
3372 case TGSI_OPCODE_PK2US:
3373 assert (0);
3374 break;
3375
3376 case TGSI_OPCODE_PK4B:
3377 assert (0);
3378 break;
3379
3380 case TGSI_OPCODE_PK4UB:
3381 assert (0);
3382 break;
3383
3384 case TGSI_OPCODE_RFL:
3385 exec_rfl(mach, inst);
3386 break;
3387
3388 case TGSI_OPCODE_SEQ:
3389 exec_vector_binary(mach, inst, micro_seq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3390 break;
3391
3392 case TGSI_OPCODE_SFL:
3393 exec_vector(mach, inst, micro_sfl, TGSI_EXEC_DATA_FLOAT);
3394 break;
3395
3396 case TGSI_OPCODE_SGT:
3397 exec_vector_binary(mach, inst, micro_sgt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3398 break;
3399
3400 case TGSI_OPCODE_SIN:
3401 exec_scalar_unary(mach, inst, micro_sin, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3402 break;
3403
3404 case TGSI_OPCODE_SLE:
3405 exec_vector_binary(mach, inst, micro_sle, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3406 break;
3407
3408 case TGSI_OPCODE_SNE:
3409 exec_vector_binary(mach, inst, micro_sne, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3410 break;
3411
3412 case TGSI_OPCODE_STR:
3413 exec_vector(mach, inst, micro_str, TGSI_EXEC_DATA_FLOAT);
3414 break;
3415
3416 case TGSI_OPCODE_TEX:
3417 /* simple texture lookup */
3418 /* src[0] = texcoord */
3419 /* src[1] = sampler unit */
3420 exec_tex(mach, inst, TEX_MODIFIER_NONE);
3421 break;
3422
3423 case TGSI_OPCODE_TXB:
3424 /* Texture lookup with lod bias */
3425 /* src[0] = texcoord (src[0].w = LOD bias) */
3426 /* src[1] = sampler unit */
3427 exec_tex(mach, inst, TEX_MODIFIER_LOD_BIAS);
3428 break;
3429
3430 case TGSI_OPCODE_TXD:
3431 /* Texture lookup with explict partial derivatives */
3432 /* src[0] = texcoord */
3433 /* src[1] = d[strq]/dx */
3434 /* src[2] = d[strq]/dy */
3435 /* src[3] = sampler unit */
3436 exec_txd(mach, inst);
3437 break;
3438
3439 case TGSI_OPCODE_TXL:
3440 /* Texture lookup with explit LOD */
3441 /* src[0] = texcoord (src[0].w = LOD) */
3442 /* src[1] = sampler unit */
3443 exec_tex(mach, inst, TEX_MODIFIER_EXPLICIT_LOD);
3444 break;
3445
3446 case TGSI_OPCODE_TXP:
3447 /* Texture lookup with projection */
3448 /* src[0] = texcoord (src[0].w = projection) */
3449 /* src[1] = sampler unit */
3450 exec_tex(mach, inst, TEX_MODIFIER_PROJECTED);
3451 break;
3452
3453 case TGSI_OPCODE_UP2H:
3454 assert (0);
3455 break;
3456
3457 case TGSI_OPCODE_UP2US:
3458 assert (0);
3459 break;
3460
3461 case TGSI_OPCODE_UP4B:
3462 assert (0);
3463 break;
3464
3465 case TGSI_OPCODE_UP4UB:
3466 assert (0);
3467 break;
3468
3469 case TGSI_OPCODE_X2D:
3470 exec_x2d(mach, inst);
3471 break;
3472
3473 case TGSI_OPCODE_ARA:
3474 assert (0);
3475 break;
3476
3477 case TGSI_OPCODE_ARR:
3478 exec_vector_unary(mach, inst, micro_arr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT);
3479 break;
3480
3481 case TGSI_OPCODE_BRA:
3482 assert (0);
3483 break;
3484
3485 case TGSI_OPCODE_CAL:
3486 /* skip the call if no execution channels are enabled */
3487 if (mach->ExecMask) {
3488 /* do the call */
3489
3490 /* First, record the depths of the execution stacks.
3491 * This is important for deeply nested/looped return statements.
3492 * We have to unwind the stacks by the correct amount. For a
3493 * real code generator, we could determine the number of entries
3494 * to pop off each stack with simple static analysis and avoid
3495 * implementing this data structure at run time.
3496 */
3497 mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop;
3498 mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop;
3499 mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop;
3500 mach->CallStack[mach->CallStackTop].SwitchStackTop = mach->SwitchStackTop;
3501 mach->CallStack[mach->CallStackTop].BreakStackTop = mach->BreakStackTop;
3502 /* note that PC was already incremented above */
3503 mach->CallStack[mach->CallStackTop].ReturnAddr = *pc;
3504
3505 mach->CallStackTop++;
3506
3507 /* Second, push the Cond, Loop, Cont, Func stacks */
3508 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
3509 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3510 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3511 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING);
3512 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
3513 assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
3514
3515 mach->CondStack[mach->CondStackTop++] = mach->CondMask;
3516 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
3517 mach->ContStack[mach->ContStackTop++] = mach->ContMask;
3518 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch;
3519 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
3520 mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
3521
3522 /* Finally, jump to the subroutine */
3523 *pc = inst->Label.Label;
3524 }
3525 break;
3526
3527 case TGSI_OPCODE_RET:
3528 mach->FuncMask &= ~mach->ExecMask;
3529 UPDATE_EXEC_MASK(mach);
3530
3531 if (mach->FuncMask == 0x0) {
3532 /* really return now (otherwise, keep executing */
3533
3534 if (mach->CallStackTop == 0) {
3535 /* returning from main() */
3536 mach->CondStackTop = 0;
3537 mach->LoopStackTop = 0;
3538 *pc = -1;
3539 return;
3540 }
3541
3542 assert(mach->CallStackTop > 0);
3543 mach->CallStackTop--;
3544
3545 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
3546 mach->CondMask = mach->CondStack[mach->CondStackTop];
3547
3548 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
3549 mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
3550
3551 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
3552 mach->ContMask = mach->ContStack[mach->ContStackTop];
3553
3554 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop;
3555 mach->Switch = mach->SwitchStack[mach->SwitchStackTop];
3556
3557 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop;
3558 mach->BreakType = mach->BreakStack[mach->BreakStackTop];
3559
3560 assert(mach->FuncStackTop > 0);
3561 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
3562
3563 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
3564
3565 UPDATE_EXEC_MASK(mach);
3566 }
3567 break;
3568
3569 case TGSI_OPCODE_SSG:
3570 exec_vector_unary(mach, inst, micro_sgn, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3571 break;
3572
3573 case TGSI_OPCODE_CMP:
3574 exec_vector_trinary(mach, inst, micro_cmp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3575 break;
3576
3577 case TGSI_OPCODE_SCS:
3578 exec_scs(mach, inst);
3579 break;
3580
3581 case TGSI_OPCODE_NRM:
3582 exec_nrm3(mach, inst);
3583 break;
3584
3585 case TGSI_OPCODE_NRM4:
3586 exec_nrm4(mach, inst);
3587 break;
3588
3589 case TGSI_OPCODE_DIV:
3590 exec_vector_binary(mach, inst, micro_div, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3591 break;
3592
3593 case TGSI_OPCODE_DP2:
3594 exec_dp2(mach, inst);
3595 break;
3596
3597 case TGSI_OPCODE_IF:
3598 /* push CondMask */
3599 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
3600 mach->CondStack[mach->CondStackTop++] = mach->CondMask;
3601 FETCH( &r[0], 0, CHAN_X );
3602 /* update CondMask */
3603 if( ! r[0].u[0] ) {
3604 mach->CondMask &= ~0x1;
3605 }
3606 if( ! r[0].u[1] ) {
3607 mach->CondMask &= ~0x2;
3608 }
3609 if( ! r[0].u[2] ) {
3610 mach->CondMask &= ~0x4;
3611 }
3612 if( ! r[0].u[3] ) {
3613 mach->CondMask &= ~0x8;
3614 }
3615 UPDATE_EXEC_MASK(mach);
3616 /* Todo: If CondMask==0, jump to ELSE */
3617 break;
3618
3619 case TGSI_OPCODE_ELSE:
3620 /* invert CondMask wrt previous mask */
3621 {
3622 uint prevMask;
3623 assert(mach->CondStackTop > 0);
3624 prevMask = mach->CondStack[mach->CondStackTop - 1];
3625 mach->CondMask = ~mach->CondMask & prevMask;
3626 UPDATE_EXEC_MASK(mach);
3627 /* Todo: If CondMask==0, jump to ENDIF */
3628 }
3629 break;
3630
3631 case TGSI_OPCODE_ENDIF:
3632 /* pop CondMask */
3633 assert(mach->CondStackTop > 0);
3634 mach->CondMask = mach->CondStack[--mach->CondStackTop];
3635 UPDATE_EXEC_MASK(mach);
3636 break;
3637
3638 case TGSI_OPCODE_END:
3639 /* make sure we end primitives which haven't
3640 * been explicitly emitted */
3641 conditional_emit_primitive(mach);
3642 /* halt execution */
3643 *pc = -1;
3644 break;
3645
3646 case TGSI_OPCODE_PUSHA:
3647 assert (0);
3648 break;
3649
3650 case TGSI_OPCODE_POPA:
3651 assert (0);
3652 break;
3653
3654 case TGSI_OPCODE_CEIL:
3655 exec_vector_unary(mach, inst, micro_ceil, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3656 break;
3657
3658 case TGSI_OPCODE_I2F:
3659 exec_vector_unary(mach, inst, micro_i2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_INT);
3660 break;
3661
3662 case TGSI_OPCODE_NOT:
3663 exec_vector_unary(mach, inst, micro_not, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3664 break;
3665
3666 case TGSI_OPCODE_TRUNC:
3667 exec_vector_unary(mach, inst, micro_trunc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3668 break;
3669
3670 case TGSI_OPCODE_SHL:
3671 exec_vector_binary(mach, inst, micro_shl, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3672 break;
3673
3674 case TGSI_OPCODE_AND:
3675 exec_vector_binary(mach, inst, micro_and, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3676 break;
3677
3678 case TGSI_OPCODE_OR:
3679 exec_vector_binary(mach, inst, micro_or, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3680 break;
3681
3682 case TGSI_OPCODE_MOD:
3683 assert (0);
3684 break;
3685
3686 case TGSI_OPCODE_XOR:
3687 exec_vector_binary(mach, inst, micro_xor, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3688 break;
3689
3690 case TGSI_OPCODE_SAD:
3691 assert (0);
3692 break;
3693
3694 case TGSI_OPCODE_TXF:
3695 assert (0);
3696 break;
3697
3698 case TGSI_OPCODE_TXQ:
3699 assert (0);
3700 break;
3701
3702 case TGSI_OPCODE_EMIT:
3703 emit_vertex(mach);
3704 break;
3705
3706 case TGSI_OPCODE_ENDPRIM:
3707 emit_primitive(mach);
3708 break;
3709
3710 case TGSI_OPCODE_BGNLOOP:
3711 /* push LoopMask and ContMasks */
3712 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3713 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3714 assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3715 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
3716
3717 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
3718 mach->ContStack[mach->ContStackTop++] = mach->ContMask;
3719 mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1;
3720 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
3721 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP;
3722 break;
3723
3724 case TGSI_OPCODE_ENDLOOP:
3725 /* Restore ContMask, but don't pop */
3726 assert(mach->ContStackTop > 0);
3727 mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
3728 UPDATE_EXEC_MASK(mach);
3729 if (mach->ExecMask) {
3730 /* repeat loop: jump to instruction just past BGNLOOP */
3731 assert(mach->LoopLabelStackTop > 0);
3732 *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1;
3733 }
3734 else {
3735 /* exit loop: pop LoopMask */
3736 assert(mach->LoopStackTop > 0);
3737 mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
3738 /* pop ContMask */
3739 assert(mach->ContStackTop > 0);
3740 mach->ContMask = mach->ContStack[--mach->ContStackTop];
3741 assert(mach->LoopLabelStackTop > 0);
3742 --mach->LoopLabelStackTop;
3743
3744 mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
3745 }
3746 UPDATE_EXEC_MASK(mach);
3747 break;
3748
3749 case TGSI_OPCODE_BRK:
3750 exec_break(mach);
3751 break;
3752
3753 case TGSI_OPCODE_CONT:
3754 /* turn off cont channels for each enabled exec channel */
3755 mach->ContMask &= ~mach->ExecMask;
3756 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3757 UPDATE_EXEC_MASK(mach);
3758 break;
3759
3760 case TGSI_OPCODE_BGNSUB:
3761 /* no-op */
3762 break;
3763
3764 case TGSI_OPCODE_ENDSUB:
3765 /*
3766 * XXX: This really should be a no-op. We should never reach this opcode.
3767 */
3768
3769 assert(mach->CallStackTop > 0);
3770 mach->CallStackTop--;
3771
3772 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
3773 mach->CondMask = mach->CondStack[mach->CondStackTop];
3774
3775 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
3776 mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
3777
3778 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
3779 mach->ContMask = mach->ContStack[mach->ContStackTop];
3780
3781 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop;
3782 mach->Switch = mach->SwitchStack[mach->SwitchStackTop];
3783
3784 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop;
3785 mach->BreakType = mach->BreakStack[mach->BreakStackTop];
3786
3787 assert(mach->FuncStackTop > 0);
3788 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
3789
3790 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
3791
3792 UPDATE_EXEC_MASK(mach);
3793 break;
3794
3795 case TGSI_OPCODE_NOP:
3796 break;
3797
3798 case TGSI_OPCODE_BREAKC:
3799 FETCH(&r[0], 0, CHAN_X);
3800 /* update CondMask */
3801 if (r[0].u[0] && (mach->ExecMask & 0x1)) {
3802 mach->LoopMask &= ~0x1;
3803 }
3804 if (r[0].u[1] && (mach->ExecMask & 0x2)) {
3805 mach->LoopMask &= ~0x2;
3806 }
3807 if (r[0].u[2] && (mach->ExecMask & 0x4)) {
3808 mach->LoopMask &= ~0x4;
3809 }
3810 if (r[0].u[3] && (mach->ExecMask & 0x8)) {
3811 mach->LoopMask &= ~0x8;
3812 }
3813 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3814 UPDATE_EXEC_MASK(mach);
3815 break;
3816
3817 case TGSI_OPCODE_F2I:
3818 exec_vector_unary(mach, inst, micro_f2i, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT);
3819 break;
3820
3821 case TGSI_OPCODE_IDIV:
3822 exec_vector_binary(mach, inst, micro_idiv, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3823 break;
3824
3825 case TGSI_OPCODE_IMAX:
3826 exec_vector_binary(mach, inst, micro_imax, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3827 break;
3828
3829 case TGSI_OPCODE_IMIN:
3830 exec_vector_binary(mach, inst, micro_imin, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3831 break;
3832
3833 case TGSI_OPCODE_INEG:
3834 exec_vector_unary(mach, inst, micro_ineg, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3835 break;
3836
3837 case TGSI_OPCODE_ISGE:
3838 exec_vector_binary(mach, inst, micro_isge, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3839 break;
3840
3841 case TGSI_OPCODE_ISHR:
3842 exec_vector_binary(mach, inst, micro_ishr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3843 break;
3844
3845 case TGSI_OPCODE_ISLT:
3846 exec_vector_binary(mach, inst, micro_islt, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3847 break;
3848
3849 case TGSI_OPCODE_F2U:
3850 exec_vector_unary(mach, inst, micro_f2u, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT);
3851 break;
3852
3853 case TGSI_OPCODE_U2F:
3854 exec_vector_unary(mach, inst, micro_u2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_UINT);
3855 break;
3856
3857 case TGSI_OPCODE_UADD:
3858 exec_vector_binary(mach, inst, micro_uadd, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3859 break;
3860
3861 case TGSI_OPCODE_UDIV:
3862 exec_vector_binary(mach, inst, micro_udiv, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3863 break;
3864
3865 case TGSI_OPCODE_UMAD:
3866 exec_vector_trinary(mach, inst, micro_umad, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3867 break;
3868
3869 case TGSI_OPCODE_UMAX:
3870 exec_vector_binary(mach, inst, micro_umax, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3871 break;
3872
3873 case TGSI_OPCODE_UMIN:
3874 exec_vector_binary(mach, inst, micro_umin, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3875 break;
3876
3877 case TGSI_OPCODE_UMOD:
3878 exec_vector_binary(mach, inst, micro_umod, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3879 break;
3880
3881 case TGSI_OPCODE_UMUL:
3882 exec_vector_binary(mach, inst, micro_umul, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3883 break;
3884
3885 case TGSI_OPCODE_USEQ:
3886 exec_vector_binary(mach, inst, micro_useq, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3887 break;
3888
3889 case TGSI_OPCODE_USGE:
3890 exec_vector_binary(mach, inst, micro_usge, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3891 break;
3892
3893 case TGSI_OPCODE_USHR:
3894 exec_vector_binary(mach, inst, micro_ushr, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3895 break;
3896
3897 case TGSI_OPCODE_USLT:
3898 exec_vector_binary(mach, inst, micro_uslt, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3899 break;
3900
3901 case TGSI_OPCODE_USNE:
3902 exec_vector_binary(mach, inst, micro_usne, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3903 break;
3904
3905 case TGSI_OPCODE_SWITCH:
3906 exec_switch(mach, inst);
3907 break;
3908
3909 case TGSI_OPCODE_CASE:
3910 exec_case(mach, inst);
3911 break;
3912
3913 case TGSI_OPCODE_DEFAULT:
3914 exec_default(mach);
3915 break;
3916
3917 case TGSI_OPCODE_ENDSWITCH:
3918 exec_endswitch(mach);
3919 break;
3920
3921 case TGSI_OPCODE_LOAD:
3922 assert(0);
3923 break;
3924
3925 case TGSI_OPCODE_LOAD_MS:
3926 assert(0);
3927 break;
3928
3929 case TGSI_OPCODE_SAMPLE:
3930 exec_sample(mach, inst, TEX_MODIFIER_NONE);
3931 break;
3932
3933 case TGSI_OPCODE_SAMPLE_B:
3934 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS);
3935 break;
3936
3937 case TGSI_OPCODE_SAMPLE_C:
3938 exec_sample(mach, inst, TEX_MODIFIER_NONE);
3939 break;
3940
3941 case TGSI_OPCODE_SAMPLE_C_LZ:
3942 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS);
3943 break;
3944
3945 case TGSI_OPCODE_SAMPLE_D:
3946 exec_sample_d(mach, inst);
3947 break;
3948
3949 case TGSI_OPCODE_SAMPLE_L:
3950 exec_sample(mach, inst, TEX_MODIFIER_EXPLICIT_LOD);
3951 break;
3952
3953 case TGSI_OPCODE_GATHER4:
3954 assert(0);
3955 break;
3956
3957 case TGSI_OPCODE_RESINFO:
3958 assert(0);
3959 break;
3960
3961 case TGSI_OPCODE_SAMPLE_POS:
3962 assert(0);
3963 break;
3964
3965 case TGSI_OPCODE_SAMPLE_INFO:
3966 assert(0);
3967 break;
3968
3969 default:
3970 assert( 0 );
3971 }
3972 }
3973
3974
3975 #define DEBUG_EXECUTION 0
3976
3977
3978 /**
3979 * Run TGSI interpreter.
3980 * \return bitmask of "alive" quad components
3981 */
3982 uint
3983 tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
3984 {
3985 uint i;
3986 int pc = 0;
3987
3988 mach->CondMask = 0xf;
3989 mach->LoopMask = 0xf;
3990 mach->ContMask = 0xf;
3991 mach->FuncMask = 0xf;
3992 mach->ExecMask = 0xf;
3993
3994 mach->Switch.mask = 0xf;
3995
3996 assert(mach->CondStackTop == 0);
3997 assert(mach->LoopStackTop == 0);
3998 assert(mach->ContStackTop == 0);
3999 assert(mach->SwitchStackTop == 0);
4000 assert(mach->BreakStackTop == 0);
4001 assert(mach->CallStackTop == 0);
4002
4003 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0;
4004 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0;
4005
4006 if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) {
4007 mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0;
4008 mach->Primitives[0] = 0;
4009 }
4010
4011 /* execute declarations (interpolants) */
4012 for (i = 0; i < mach->NumDeclarations; i++) {
4013 exec_declaration( mach, mach->Declarations+i );
4014 }
4015
4016 {
4017 #if DEBUG_EXECUTION
4018 struct tgsi_exec_vector temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS];
4019 struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS];
4020 uint inst = 1;
4021
4022 memcpy(temps, mach->Temps, sizeof(temps));
4023 memcpy(outputs, mach->Outputs, sizeof(outputs));
4024 #endif
4025
4026 /* execute instructions, until pc is set to -1 */
4027 while (pc != -1) {
4028
4029 #if DEBUG_EXECUTION
4030 uint i;
4031
4032 tgsi_dump_instruction(&mach->Instructions[pc], inst++);
4033 #endif
4034
4035 assert(pc < (int) mach->NumInstructions);
4036 exec_instruction(mach, mach->Instructions + pc, &pc);
4037
4038 #if DEBUG_EXECUTION
4039 for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) {
4040 if (memcmp(&temps[i], &mach->Temps[i], sizeof(temps[i]))) {
4041 uint j;
4042
4043 memcpy(&temps[i], &mach->Temps[i], sizeof(temps[i]));
4044 debug_printf("TEMP[%2u] = ", i);
4045 for (j = 0; j < 4; j++) {
4046 if (j > 0) {
4047 debug_printf(" ");
4048 }
4049 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
4050 temps[i].xyzw[0].f[j], temps[i].xyzw[0].u[j],
4051 temps[i].xyzw[1].f[j], temps[i].xyzw[1].u[j],
4052 temps[i].xyzw[2].f[j], temps[i].xyzw[2].u[j],
4053 temps[i].xyzw[3].f[j], temps[i].xyzw[3].u[j]);
4054 }
4055 }
4056 }
4057 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {
4058 if (memcmp(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]))) {
4059 uint j;
4060
4061 memcpy(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]));
4062 debug_printf("OUT[%2u] = ", i);
4063 for (j = 0; j < 4; j++) {
4064 if (j > 0) {
4065 debug_printf(" ");
4066 }
4067 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
4068 outputs[i].xyzw[0].f[j], outputs[i].xyzw[0].u[j],
4069 outputs[i].xyzw[1].f[j], outputs[i].xyzw[1].u[j],
4070 outputs[i].xyzw[2].f[j], outputs[i].xyzw[2].u[j],
4071 outputs[i].xyzw[3].f[j], outputs[i].xyzw[3].u[j]);
4072 }
4073 }
4074 }
4075 #endif
4076 }
4077 }
4078
4079 #if 0
4080 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
4081 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
4082 /*
4083 * Scale back depth component.
4084 */
4085 for (i = 0; i < 4; i++)
4086 mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF;
4087 }
4088 #endif
4089
4090 /* Strictly speaking, these assertions aren't really needed but they
4091 * can potentially catch some bugs in the control flow code.
4092 */
4093 assert(mach->CondStackTop == 0);
4094 assert(mach->LoopStackTop == 0);
4095 assert(mach->ContStackTop == 0);
4096 assert(mach->SwitchStackTop == 0);
4097 assert(mach->BreakStackTop == 0);
4098 assert(mach->CallStackTop == 0);
4099
4100 return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
4101 }