Merge branch 'master' of git://anongit.freedesktop.org/mesa/mesa
[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 #define IFETCH(VAL,INDEX,CHAN)\
1598 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT)
1599
1600
1601 /**
1602 * Execute ARB-style KIL which is predicated by a src register.
1603 * Kill fragment if any of the four values is less than zero.
1604 */
1605 static void
1606 exec_kil(struct tgsi_exec_machine *mach,
1607 const struct tgsi_full_instruction *inst)
1608 {
1609 uint uniquemask;
1610 uint chan_index;
1611 uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1612 union tgsi_exec_channel r[1];
1613
1614 /* This mask stores component bits that were already tested. */
1615 uniquemask = 0;
1616
1617 for (chan_index = 0; chan_index < 4; chan_index++)
1618 {
1619 uint swizzle;
1620 uint i;
1621
1622 /* unswizzle channel */
1623 swizzle = tgsi_util_get_full_src_register_swizzle (
1624 &inst->Src[0],
1625 chan_index);
1626
1627 /* check if the component has not been already tested */
1628 if (uniquemask & (1 << swizzle))
1629 continue;
1630 uniquemask |= 1 << swizzle;
1631
1632 FETCH(&r[0], 0, chan_index);
1633 for (i = 0; i < 4; i++)
1634 if (r[0].f[i] < 0.0f)
1635 kilmask |= 1 << i;
1636 }
1637
1638 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
1639 }
1640
1641 /**
1642 * Execute NVIDIA-style KIL which is predicated by a condition code.
1643 * Kill fragment if the condition code is TRUE.
1644 */
1645 static void
1646 exec_kilp(struct tgsi_exec_machine *mach,
1647 const struct tgsi_full_instruction *inst)
1648 {
1649 uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1650
1651 /* "unconditional" kil */
1652 kilmask = mach->ExecMask;
1653 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
1654 }
1655
1656 static void
1657 emit_vertex(struct tgsi_exec_machine *mach)
1658 {
1659 /* FIXME: check for exec mask correctly
1660 unsigned i;
1661 for (i = 0; i < QUAD_SIZE; ++i) {
1662 if ((mach->ExecMask & (1 << i)))
1663 */
1664 if (mach->ExecMask) {
1665 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += mach->NumOutputs;
1666 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++;
1667 }
1668 }
1669
1670 static void
1671 emit_primitive(struct tgsi_exec_machine *mach)
1672 {
1673 unsigned *prim_count = &mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0];
1674 /* FIXME: check for exec mask correctly
1675 unsigned i;
1676 for (i = 0; i < QUAD_SIZE; ++i) {
1677 if ((mach->ExecMask & (1 << i)))
1678 */
1679 if (mach->ExecMask) {
1680 ++(*prim_count);
1681 debug_assert((*prim_count * mach->NumOutputs) < mach->MaxGeometryShaderOutputs);
1682 mach->Primitives[*prim_count] = 0;
1683 }
1684 }
1685
1686 static void
1687 conditional_emit_primitive(struct tgsi_exec_machine *mach)
1688 {
1689 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
1690 int emitted_verts =
1691 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]];
1692 if (emitted_verts) {
1693 emit_primitive(mach);
1694 }
1695 }
1696 }
1697
1698
1699 /*
1700 * Fetch four texture samples using STR texture coordinates.
1701 */
1702 static void
1703 fetch_texel( struct tgsi_sampler *sampler,
1704 const union tgsi_exec_channel *s,
1705 const union tgsi_exec_channel *t,
1706 const union tgsi_exec_channel *p,
1707 const union tgsi_exec_channel *c0,
1708 enum tgsi_sampler_control control,
1709 union tgsi_exec_channel *r,
1710 union tgsi_exec_channel *g,
1711 union tgsi_exec_channel *b,
1712 union tgsi_exec_channel *a )
1713 {
1714 uint j;
1715 float rgba[NUM_CHANNELS][QUAD_SIZE];
1716
1717 sampler->get_samples(sampler, s->f, t->f, p->f, c0->f, control, rgba);
1718
1719 for (j = 0; j < 4; j++) {
1720 r->f[j] = rgba[0][j];
1721 g->f[j] = rgba[1][j];
1722 b->f[j] = rgba[2][j];
1723 a->f[j] = rgba[3][j];
1724 }
1725 }
1726
1727
1728 #define TEX_MODIFIER_NONE 0
1729 #define TEX_MODIFIER_PROJECTED 1
1730 #define TEX_MODIFIER_LOD_BIAS 2
1731 #define TEX_MODIFIER_EXPLICIT_LOD 3
1732
1733
1734 static void
1735 exec_tex(struct tgsi_exec_machine *mach,
1736 const struct tgsi_full_instruction *inst,
1737 uint modifier)
1738 {
1739 const uint unit = inst->Src[1].Register.Index;
1740 union tgsi_exec_channel r[4];
1741 const union tgsi_exec_channel *lod = &ZeroVec;
1742 enum tgsi_sampler_control control;
1743 uint chan;
1744
1745 if (modifier != TEX_MODIFIER_NONE) {
1746 FETCH(&r[3], 0, CHAN_W);
1747 if (modifier != TEX_MODIFIER_PROJECTED) {
1748 lod = &r[3];
1749 }
1750 }
1751
1752 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) {
1753 control = tgsi_sampler_lod_explicit;
1754 } else {
1755 control = tgsi_sampler_lod_bias;
1756 }
1757
1758 switch (inst->Texture.Texture) {
1759 case TGSI_TEXTURE_1D:
1760 case TGSI_TEXTURE_SHADOW1D:
1761 FETCH(&r[0], 0, CHAN_X);
1762
1763 if (modifier == TEX_MODIFIER_PROJECTED) {
1764 micro_div(&r[0], &r[0], &r[3]);
1765 }
1766
1767 fetch_texel(mach->Samplers[unit],
1768 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */
1769 control,
1770 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
1771 break;
1772
1773 case TGSI_TEXTURE_2D:
1774 case TGSI_TEXTURE_RECT:
1775 case TGSI_TEXTURE_SHADOW2D:
1776 case TGSI_TEXTURE_SHADOWRECT:
1777 FETCH(&r[0], 0, CHAN_X);
1778 FETCH(&r[1], 0, CHAN_Y);
1779 FETCH(&r[2], 0, CHAN_Z);
1780
1781 if (modifier == TEX_MODIFIER_PROJECTED) {
1782 micro_div(&r[0], &r[0], &r[3]);
1783 micro_div(&r[1], &r[1], &r[3]);
1784 micro_div(&r[2], &r[2], &r[3]);
1785 }
1786
1787 fetch_texel(mach->Samplers[unit],
1788 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */
1789 control,
1790 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1791 break;
1792
1793 case TGSI_TEXTURE_1D_ARRAY:
1794 FETCH(&r[0], 0, CHAN_X);
1795 FETCH(&r[1], 0, CHAN_Y);
1796
1797 if (modifier == TEX_MODIFIER_PROJECTED) {
1798 micro_div(&r[0], &r[0], &r[3]);
1799 }
1800
1801 fetch_texel(mach->Samplers[unit],
1802 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */
1803 control,
1804 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1805 break;
1806
1807 case TGSI_TEXTURE_2D_ARRAY:
1808 FETCH(&r[0], 0, CHAN_X);
1809 FETCH(&r[1], 0, CHAN_Y);
1810 FETCH(&r[2], 0, CHAN_Z);
1811
1812 if (modifier == TEX_MODIFIER_PROJECTED) {
1813 micro_div(&r[0], &r[0], &r[3]);
1814 micro_div(&r[1], &r[1], &r[3]);
1815 }
1816
1817 fetch_texel(mach->Samplers[unit],
1818 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */
1819 control,
1820 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1821 break;
1822
1823 case TGSI_TEXTURE_3D:
1824 case TGSI_TEXTURE_CUBE:
1825 FETCH(&r[0], 0, CHAN_X);
1826 FETCH(&r[1], 0, CHAN_Y);
1827 FETCH(&r[2], 0, CHAN_Z);
1828
1829 if (modifier == TEX_MODIFIER_PROJECTED) {
1830 micro_div(&r[0], &r[0], &r[3]);
1831 micro_div(&r[1], &r[1], &r[3]);
1832 micro_div(&r[2], &r[2], &r[3]);
1833 }
1834
1835 fetch_texel(mach->Samplers[unit],
1836 &r[0], &r[1], &r[2], lod,
1837 control,
1838 &r[0], &r[1], &r[2], &r[3]);
1839 break;
1840
1841 default:
1842 assert(0);
1843 }
1844
1845 #if 0
1846 debug_printf("fetch r: %g %g %g %g\n",
1847 r[0].f[0], r[0].f[1], r[0].f[2], r[0].f[3]);
1848 debug_printf("fetch g: %g %g %g %g\n",
1849 r[1].f[0], r[1].f[1], r[1].f[2], r[1].f[3]);
1850 debug_printf("fetch b: %g %g %g %g\n",
1851 r[2].f[0], r[2].f[1], r[2].f[2], r[2].f[3]);
1852 debug_printf("fetch a: %g %g %g %g\n",
1853 r[3].f[0], r[3].f[1], r[3].f[2], r[3].f[3]);
1854 #endif
1855
1856 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1857 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1858 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
1859 }
1860 }
1861 }
1862
1863 static void
1864 exec_txd(struct tgsi_exec_machine *mach,
1865 const struct tgsi_full_instruction *inst)
1866 {
1867 const uint unit = inst->Src[3].Register.Index;
1868 union tgsi_exec_channel r[4];
1869 uint chan;
1870
1871 /*
1872 * XXX: This is fake TXD -- the derivatives are not taken into account, yet.
1873 */
1874
1875 switch (inst->Texture.Texture) {
1876 case TGSI_TEXTURE_1D:
1877 case TGSI_TEXTURE_SHADOW1D:
1878
1879 FETCH(&r[0], 0, CHAN_X);
1880
1881 fetch_texel(mach->Samplers[unit],
1882 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */
1883 tgsi_sampler_lod_bias,
1884 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
1885 break;
1886
1887 case TGSI_TEXTURE_2D:
1888 case TGSI_TEXTURE_RECT:
1889 case TGSI_TEXTURE_SHADOW2D:
1890 case TGSI_TEXTURE_SHADOWRECT:
1891
1892 FETCH(&r[0], 0, CHAN_X);
1893 FETCH(&r[1], 0, CHAN_Y);
1894 FETCH(&r[2], 0, CHAN_Z);
1895
1896 fetch_texel(mach->Samplers[unit],
1897 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */
1898 tgsi_sampler_lod_bias,
1899 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1900 break;
1901
1902 case TGSI_TEXTURE_3D:
1903 case TGSI_TEXTURE_CUBE:
1904
1905 FETCH(&r[0], 0, CHAN_X);
1906 FETCH(&r[1], 0, CHAN_Y);
1907 FETCH(&r[2], 0, CHAN_Z);
1908
1909 fetch_texel(mach->Samplers[unit],
1910 &r[0], &r[1], &r[2], &ZeroVec,
1911 tgsi_sampler_lod_bias,
1912 &r[0], &r[1], &r[2], &r[3]);
1913 break;
1914
1915 default:
1916 assert(0);
1917 }
1918
1919 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1920 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1921 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
1922 }
1923 }
1924 }
1925
1926
1927 static void
1928 exec_txf(struct tgsi_exec_machine *mach,
1929 const struct tgsi_full_instruction *inst)
1930 {
1931 struct tgsi_sampler *sampler;
1932 const uint unit = inst->Src[1].Register.Index;
1933 union tgsi_exec_channel r[4];
1934 uint chan;
1935 float rgba[NUM_CHANNELS][QUAD_SIZE];
1936 int j;
1937
1938 IFETCH(&r[3], 0, CHAN_W);
1939
1940 switch(inst->Texture.Texture) {
1941 case TGSI_TEXTURE_3D:
1942 case TGSI_TEXTURE_2D_ARRAY:
1943 IFETCH(&r[2], 0, CHAN_Z);
1944 /* fallthrough */
1945 case TGSI_TEXTURE_2D:
1946 case TGSI_TEXTURE_RECT:
1947 case TGSI_TEXTURE_SHADOW2D:
1948 case TGSI_TEXTURE_SHADOWRECT:
1949 case TGSI_TEXTURE_1D_ARRAY:
1950 IFETCH(&r[1], 0, CHAN_Y);
1951 /* fallthrough */
1952 case TGSI_TEXTURE_1D:
1953 case TGSI_TEXTURE_SHADOW1D:
1954 IFETCH(&r[0], 0, CHAN_X);
1955 break;
1956 default:
1957 assert(0);
1958 break;
1959 }
1960
1961 sampler = mach->Samplers[unit];
1962 sampler->get_texel(sampler, r[0].i, r[1].i, r[2].i, r[3].i, rgba);
1963
1964 for (j = 0; j < QUAD_SIZE; j++) {
1965 r[0].f[j] = rgba[0][j];
1966 r[1].f[j] = rgba[1][j];
1967 r[2].f[j] = rgba[2][j];
1968 r[3].f[j] = rgba[3][j];
1969 }
1970
1971 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1972 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1973 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
1974 }
1975 }
1976 }
1977
1978 static void
1979 exec_txq(struct tgsi_exec_machine *mach,
1980 const struct tgsi_full_instruction *inst)
1981 {
1982 struct tgsi_sampler *sampler;
1983 const uint unit = inst->Src[1].Register.Index;
1984 int result[4];
1985 union tgsi_exec_channel r[4], src;
1986 uint chan;
1987 int i,j;
1988
1989 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_INT);
1990 sampler = mach->Samplers[unit];
1991
1992 sampler->get_dims(sampler, src.i[0], result);
1993
1994 for (i = 0; i < QUAD_SIZE; i++) {
1995 for (j = 0; j < 4; j++) {
1996 r[j].i[i] = result[j];
1997 }
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,
2003 TGSI_EXEC_DATA_INT);
2004 }
2005 }
2006 }
2007
2008 static void
2009 exec_sample(struct tgsi_exec_machine *mach,
2010 const struct tgsi_full_instruction *inst,
2011 uint modifier)
2012 {
2013 const uint resource_unit = inst->Src[1].Register.Index;
2014 const uint sampler_unit = inst->Src[2].Register.Index;
2015 union tgsi_exec_channel r[4];
2016 const union tgsi_exec_channel *lod = &ZeroVec;
2017 enum tgsi_sampler_control control;
2018 uint chan;
2019
2020 if (modifier != TEX_MODIFIER_NONE) {
2021 if (modifier == TEX_MODIFIER_LOD_BIAS)
2022 FETCH(&r[3], 3, CHAN_X);
2023 else /*TEX_MODIFIER_LOD*/
2024 FETCH(&r[3], 0, CHAN_W);
2025
2026 if (modifier != TEX_MODIFIER_PROJECTED) {
2027 lod = &r[3];
2028 }
2029 }
2030
2031 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) {
2032 control = tgsi_sampler_lod_explicit;
2033 } else {
2034 control = tgsi_sampler_lod_bias;
2035 }
2036
2037 switch (mach->Resources[resource_unit].Resource) {
2038 case TGSI_TEXTURE_1D:
2039 case TGSI_TEXTURE_SHADOW1D:
2040 FETCH(&r[0], 0, CHAN_X);
2041
2042 if (modifier == TEX_MODIFIER_PROJECTED) {
2043 micro_div(&r[0], &r[0], &r[3]);
2044 }
2045
2046 fetch_texel(mach->Samplers[sampler_unit],
2047 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */
2048 control,
2049 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
2050 break;
2051
2052 case TGSI_TEXTURE_2D:
2053 case TGSI_TEXTURE_RECT:
2054 case TGSI_TEXTURE_SHADOW2D:
2055 case TGSI_TEXTURE_SHADOWRECT:
2056 FETCH(&r[0], 0, CHAN_X);
2057 FETCH(&r[1], 0, CHAN_Y);
2058 FETCH(&r[2], 0, CHAN_Z);
2059
2060 if (modifier == TEX_MODIFIER_PROJECTED) {
2061 micro_div(&r[0], &r[0], &r[3]);
2062 micro_div(&r[1], &r[1], &r[3]);
2063 micro_div(&r[2], &r[2], &r[3]);
2064 }
2065
2066 fetch_texel(mach->Samplers[sampler_unit],
2067 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */
2068 control,
2069 &r[0], &r[1], &r[2], &r[3]); /* outputs */
2070 break;
2071
2072 case TGSI_TEXTURE_3D:
2073 case TGSI_TEXTURE_CUBE:
2074 FETCH(&r[0], 0, CHAN_X);
2075 FETCH(&r[1], 0, CHAN_Y);
2076 FETCH(&r[2], 0, CHAN_Z);
2077
2078 if (modifier == TEX_MODIFIER_PROJECTED) {
2079 micro_div(&r[0], &r[0], &r[3]);
2080 micro_div(&r[1], &r[1], &r[3]);
2081 micro_div(&r[2], &r[2], &r[3]);
2082 }
2083
2084 fetch_texel(mach->Samplers[sampler_unit],
2085 &r[0], &r[1], &r[2], lod,
2086 control,
2087 &r[0], &r[1], &r[2], &r[3]);
2088 break;
2089
2090 default:
2091 assert(0);
2092 }
2093
2094 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2095 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2096 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2097 }
2098 }
2099 }
2100
2101 static void
2102 exec_sample_d(struct tgsi_exec_machine *mach,
2103 const struct tgsi_full_instruction *inst)
2104 {
2105 const uint resource_unit = inst->Src[1].Register.Index;
2106 const uint sampler_unit = inst->Src[2].Register.Index;
2107 union tgsi_exec_channel r[4];
2108 uint chan;
2109 /*
2110 * XXX: This is fake SAMPLE_D -- the derivatives are not taken into account, yet.
2111 */
2112
2113 switch (mach->Resources[resource_unit].Resource) {
2114 case TGSI_TEXTURE_1D:
2115 case TGSI_TEXTURE_SHADOW1D:
2116
2117 FETCH(&r[0], 0, CHAN_X);
2118
2119 fetch_texel(mach->Samplers[sampler_unit],
2120 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */
2121 tgsi_sampler_lod_bias,
2122 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
2123 break;
2124
2125 case TGSI_TEXTURE_2D:
2126 case TGSI_TEXTURE_RECT:
2127 case TGSI_TEXTURE_SHADOW2D:
2128 case TGSI_TEXTURE_SHADOWRECT:
2129
2130 FETCH(&r[0], 0, CHAN_X);
2131 FETCH(&r[1], 0, CHAN_Y);
2132 FETCH(&r[2], 0, CHAN_Z);
2133
2134 fetch_texel(mach->Samplers[sampler_unit],
2135 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */
2136 tgsi_sampler_lod_bias,
2137 &r[0], &r[1], &r[2], &r[3]); /* outputs */
2138 break;
2139
2140 case TGSI_TEXTURE_3D:
2141 case TGSI_TEXTURE_CUBE:
2142
2143 FETCH(&r[0], 0, CHAN_X);
2144 FETCH(&r[1], 0, CHAN_Y);
2145 FETCH(&r[2], 0, CHAN_Z);
2146
2147 fetch_texel(mach->Samplers[sampler_unit],
2148 &r[0], &r[1], &r[2], &ZeroVec,
2149 tgsi_sampler_lod_bias,
2150 &r[0], &r[1], &r[2], &r[3]);
2151 break;
2152
2153 default:
2154 assert(0);
2155 }
2156
2157 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2158 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2159 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2160 }
2161 }
2162 }
2163
2164
2165 /**
2166 * Evaluate a constant-valued coefficient at the position of the
2167 * current quad.
2168 */
2169 static void
2170 eval_constant_coef(
2171 struct tgsi_exec_machine *mach,
2172 unsigned attrib,
2173 unsigned chan )
2174 {
2175 unsigned i;
2176
2177 for( i = 0; i < QUAD_SIZE; i++ ) {
2178 mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan];
2179 }
2180 }
2181
2182 /**
2183 * Evaluate a linear-valued coefficient at the position of the
2184 * current quad.
2185 */
2186 static void
2187 eval_linear_coef(
2188 struct tgsi_exec_machine *mach,
2189 unsigned attrib,
2190 unsigned chan )
2191 {
2192 const float x = mach->QuadPos.xyzw[0].f[0];
2193 const float y = mach->QuadPos.xyzw[1].f[0];
2194 const float dadx = mach->InterpCoefs[attrib].dadx[chan];
2195 const float dady = mach->InterpCoefs[attrib].dady[chan];
2196 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
2197 mach->Inputs[attrib].xyzw[chan].f[0] = a0;
2198 mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx;
2199 mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady;
2200 mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady;
2201 }
2202
2203 /**
2204 * Evaluate a perspective-valued coefficient at the position of the
2205 * current quad.
2206 */
2207 static void
2208 eval_perspective_coef(
2209 struct tgsi_exec_machine *mach,
2210 unsigned attrib,
2211 unsigned chan )
2212 {
2213 const float x = mach->QuadPos.xyzw[0].f[0];
2214 const float y = mach->QuadPos.xyzw[1].f[0];
2215 const float dadx = mach->InterpCoefs[attrib].dadx[chan];
2216 const float dady = mach->InterpCoefs[attrib].dady[chan];
2217 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
2218 const float *w = mach->QuadPos.xyzw[3].f;
2219 /* divide by W here */
2220 mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0];
2221 mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1];
2222 mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2];
2223 mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3];
2224 }
2225
2226
2227 typedef void (* eval_coef_func)(
2228 struct tgsi_exec_machine *mach,
2229 unsigned attrib,
2230 unsigned chan );
2231
2232 static void
2233 exec_declaration(struct tgsi_exec_machine *mach,
2234 const struct tgsi_full_declaration *decl)
2235 {
2236 if (decl->Declaration.File == TGSI_FILE_RESOURCE) {
2237 mach->Resources[decl->Range.First] = decl->Resource;
2238 return;
2239 }
2240
2241 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
2242 if (decl->Declaration.File == TGSI_FILE_INPUT) {
2243 uint first, last, mask;
2244
2245 first = decl->Range.First;
2246 last = decl->Range.Last;
2247 mask = decl->Declaration.UsageMask;
2248
2249 /* XXX we could remove this special-case code since
2250 * mach->InterpCoefs[first].a0 should already have the
2251 * front/back-face value. But we should first update the
2252 * ureg code to emit the right UsageMask value (WRITEMASK_X).
2253 * Then, we could remove the tgsi_exec_machine::Face field.
2254 */
2255 /* XXX make FACE a system value */
2256 if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) {
2257 uint i;
2258
2259 assert(decl->Semantic.Index == 0);
2260 assert(first == last);
2261
2262 for (i = 0; i < QUAD_SIZE; i++) {
2263 mach->Inputs[first].xyzw[0].f[i] = mach->Face;
2264 }
2265 } else {
2266 eval_coef_func eval;
2267 uint i, j;
2268
2269 switch (decl->Declaration.Interpolate) {
2270 case TGSI_INTERPOLATE_CONSTANT:
2271 eval = eval_constant_coef;
2272 break;
2273
2274 case TGSI_INTERPOLATE_LINEAR:
2275 eval = eval_linear_coef;
2276 break;
2277
2278 case TGSI_INTERPOLATE_PERSPECTIVE:
2279 eval = eval_perspective_coef;
2280 break;
2281
2282 default:
2283 assert(0);
2284 return;
2285 }
2286
2287 for (j = 0; j < NUM_CHANNELS; j++) {
2288 if (mask & (1 << j)) {
2289 for (i = first; i <= last; i++) {
2290 eval(mach, i, j);
2291 }
2292 }
2293 }
2294 }
2295 }
2296 }
2297
2298 if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) {
2299 mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First;
2300 }
2301 }
2302
2303
2304 typedef void (* micro_op)(union tgsi_exec_channel *dst);
2305
2306 static void
2307 exec_vector(struct tgsi_exec_machine *mach,
2308 const struct tgsi_full_instruction *inst,
2309 micro_op op,
2310 enum tgsi_exec_datatype dst_datatype)
2311 {
2312 unsigned int chan;
2313
2314 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2315 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2316 union tgsi_exec_channel dst;
2317
2318 op(&dst);
2319 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype);
2320 }
2321 }
2322 }
2323
2324 typedef void (* micro_unary_op)(union tgsi_exec_channel *dst,
2325 const union tgsi_exec_channel *src);
2326
2327 static void
2328 exec_scalar_unary(struct tgsi_exec_machine *mach,
2329 const struct tgsi_full_instruction *inst,
2330 micro_unary_op op,
2331 enum tgsi_exec_datatype dst_datatype,
2332 enum tgsi_exec_datatype src_datatype)
2333 {
2334 unsigned int chan;
2335 union tgsi_exec_channel src;
2336 union tgsi_exec_channel dst;
2337
2338 fetch_source(mach, &src, &inst->Src[0], CHAN_X, src_datatype);
2339 op(&dst, &src);
2340 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2341 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2342 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype);
2343 }
2344 }
2345 }
2346
2347 static void
2348 exec_vector_unary(struct tgsi_exec_machine *mach,
2349 const struct tgsi_full_instruction *inst,
2350 micro_unary_op op,
2351 enum tgsi_exec_datatype dst_datatype,
2352 enum tgsi_exec_datatype src_datatype)
2353 {
2354 unsigned int chan;
2355 struct tgsi_exec_vector dst;
2356
2357 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2358 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2359 union tgsi_exec_channel src;
2360
2361 fetch_source(mach, &src, &inst->Src[0], chan, src_datatype);
2362 op(&dst.xyzw[chan], &src);
2363 }
2364 }
2365 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2366 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2367 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype);
2368 }
2369 }
2370 }
2371
2372 typedef void (* micro_binary_op)(union tgsi_exec_channel *dst,
2373 const union tgsi_exec_channel *src0,
2374 const union tgsi_exec_channel *src1);
2375
2376 static void
2377 exec_scalar_binary(struct tgsi_exec_machine *mach,
2378 const struct tgsi_full_instruction *inst,
2379 micro_binary_op op,
2380 enum tgsi_exec_datatype dst_datatype,
2381 enum tgsi_exec_datatype src_datatype)
2382 {
2383 unsigned int chan;
2384 union tgsi_exec_channel src[2];
2385 union tgsi_exec_channel dst;
2386
2387 fetch_source(mach, &src[0], &inst->Src[0], CHAN_X, src_datatype);
2388 fetch_source(mach, &src[1], &inst->Src[1], CHAN_Y, src_datatype);
2389 op(&dst, &src[0], &src[1]);
2390 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2391 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2392 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype);
2393 }
2394 }
2395 }
2396
2397 static void
2398 exec_vector_binary(struct tgsi_exec_machine *mach,
2399 const struct tgsi_full_instruction *inst,
2400 micro_binary_op op,
2401 enum tgsi_exec_datatype dst_datatype,
2402 enum tgsi_exec_datatype src_datatype)
2403 {
2404 unsigned int chan;
2405 struct tgsi_exec_vector dst;
2406
2407 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2408 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2409 union tgsi_exec_channel src[2];
2410
2411 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype);
2412 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype);
2413 op(&dst.xyzw[chan], &src[0], &src[1]);
2414 }
2415 }
2416 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2417 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2418 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype);
2419 }
2420 }
2421 }
2422
2423 typedef void (* micro_trinary_op)(union tgsi_exec_channel *dst,
2424 const union tgsi_exec_channel *src0,
2425 const union tgsi_exec_channel *src1,
2426 const union tgsi_exec_channel *src2);
2427
2428 static void
2429 exec_vector_trinary(struct tgsi_exec_machine *mach,
2430 const struct tgsi_full_instruction *inst,
2431 micro_trinary_op op,
2432 enum tgsi_exec_datatype dst_datatype,
2433 enum tgsi_exec_datatype src_datatype)
2434 {
2435 unsigned int chan;
2436 struct tgsi_exec_vector dst;
2437
2438 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2439 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2440 union tgsi_exec_channel src[3];
2441
2442 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype);
2443 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype);
2444 fetch_source(mach, &src[2], &inst->Src[2], chan, src_datatype);
2445 op(&dst.xyzw[chan], &src[0], &src[1], &src[2]);
2446 }
2447 }
2448 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2449 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2450 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype);
2451 }
2452 }
2453 }
2454
2455 static void
2456 exec_dp3(struct tgsi_exec_machine *mach,
2457 const struct tgsi_full_instruction *inst)
2458 {
2459 unsigned int chan;
2460 union tgsi_exec_channel arg[3];
2461
2462 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2463 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2464 micro_mul(&arg[2], &arg[0], &arg[1]);
2465
2466 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) {
2467 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
2468 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT);
2469 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
2470 }
2471
2472 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2473 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2474 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2475 }
2476 }
2477 }
2478
2479 static void
2480 exec_dp4(struct tgsi_exec_machine *mach,
2481 const struct tgsi_full_instruction *inst)
2482 {
2483 unsigned int chan;
2484 union tgsi_exec_channel arg[3];
2485
2486 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2487 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2488 micro_mul(&arg[2], &arg[0], &arg[1]);
2489
2490 for (chan = CHAN_Y; chan <= CHAN_W; chan++) {
2491 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
2492 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT);
2493 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
2494 }
2495
2496 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2497 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2498 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2499 }
2500 }
2501 }
2502
2503 static void
2504 exec_dp2a(struct tgsi_exec_machine *mach,
2505 const struct tgsi_full_instruction *inst)
2506 {
2507 unsigned int chan;
2508 union tgsi_exec_channel arg[3];
2509
2510 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2511 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2512 micro_mul(&arg[2], &arg[0], &arg[1]);
2513
2514 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2515 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2516 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]);
2517
2518 fetch_source(mach, &arg[1], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2519 micro_add(&arg[0], &arg[0], &arg[1]);
2520
2521 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2522 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2523 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2524 }
2525 }
2526 }
2527
2528 static void
2529 exec_dph(struct tgsi_exec_machine *mach,
2530 const struct tgsi_full_instruction *inst)
2531 {
2532 unsigned int chan;
2533 union tgsi_exec_channel arg[3];
2534
2535 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2536 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2537 micro_mul(&arg[2], &arg[0], &arg[1]);
2538
2539 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2540 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2541 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
2542
2543 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2544 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2545 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]);
2546
2547 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT);
2548 micro_add(&arg[0], &arg[0], &arg[1]);
2549
2550 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2551 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2552 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2553 }
2554 }
2555 }
2556
2557 static void
2558 exec_dp2(struct tgsi_exec_machine *mach,
2559 const struct tgsi_full_instruction *inst)
2560 {
2561 unsigned int chan;
2562 union tgsi_exec_channel arg[3];
2563
2564 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2565 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2566 micro_mul(&arg[2], &arg[0], &arg[1]);
2567
2568 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2569 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2570 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]);
2571
2572 for (chan = 0; chan < NUM_CHANNELS; chan++) {
2573 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2574 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2575 }
2576 }
2577 }
2578
2579 static void
2580 exec_nrm4(struct tgsi_exec_machine *mach,
2581 const struct tgsi_full_instruction *inst)
2582 {
2583 unsigned int chan;
2584 union tgsi_exec_channel arg[4];
2585 union tgsi_exec_channel scale;
2586
2587 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2588 micro_mul(&scale, &arg[0], &arg[0]);
2589
2590 for (chan = CHAN_Y; chan <= CHAN_W; chan++) {
2591 union tgsi_exec_channel product;
2592
2593 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
2594 micro_mul(&product, &arg[chan], &arg[chan]);
2595 micro_add(&scale, &scale, &product);
2596 }
2597
2598 micro_rsq(&scale, &scale);
2599
2600 for (chan = CHAN_X; chan <= CHAN_W; chan++) {
2601 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2602 micro_mul(&arg[chan], &arg[chan], &scale);
2603 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2604 }
2605 }
2606 }
2607
2608 static void
2609 exec_nrm3(struct tgsi_exec_machine *mach,
2610 const struct tgsi_full_instruction *inst)
2611 {
2612 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) {
2613 unsigned int chan;
2614 union tgsi_exec_channel arg[3];
2615 union tgsi_exec_channel scale;
2616
2617 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2618 micro_mul(&scale, &arg[0], &arg[0]);
2619
2620 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) {
2621 union tgsi_exec_channel product;
2622
2623 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT);
2624 micro_mul(&product, &arg[chan], &arg[chan]);
2625 micro_add(&scale, &scale, &product);
2626 }
2627
2628 micro_rsq(&scale, &scale);
2629
2630 for (chan = CHAN_X; chan <= CHAN_Z; chan++) {
2631 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
2632 micro_mul(&arg[chan], &arg[chan], &scale);
2633 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
2634 }
2635 }
2636 }
2637
2638 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2639 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2640 }
2641 }
2642
2643 static void
2644 exec_scs(struct tgsi_exec_machine *mach,
2645 const struct tgsi_full_instruction *inst)
2646 {
2647 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XY) {
2648 union tgsi_exec_channel arg;
2649 union tgsi_exec_channel result;
2650
2651 fetch_source(mach, &arg, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2652
2653 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2654 micro_cos(&result, &arg);
2655 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2656 }
2657 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2658 micro_sin(&result, &arg);
2659 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2660 }
2661 }
2662 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2663 store_dest(mach, &ZeroVec, &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2664 }
2665 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2666 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2667 }
2668 }
2669
2670 static void
2671 exec_x2d(struct tgsi_exec_machine *mach,
2672 const struct tgsi_full_instruction *inst)
2673 {
2674 union tgsi_exec_channel r[4];
2675 union tgsi_exec_channel d[2];
2676
2677 fetch_source(mach, &r[0], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2678 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2679 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XZ) {
2680 fetch_source(mach, &r[2], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2681 micro_mul(&r[2], &r[2], &r[0]);
2682 fetch_source(mach, &r[3], &inst->Src[2], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2683 micro_mul(&r[3], &r[3], &r[1]);
2684 micro_add(&r[2], &r[2], &r[3]);
2685 fetch_source(mach, &r[3], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2686 micro_add(&d[0], &r[2], &r[3]);
2687 }
2688 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YW) {
2689 fetch_source(mach, &r[2], &inst->Src[2], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2690 micro_mul(&r[2], &r[2], &r[0]);
2691 fetch_source(mach, &r[3], &inst->Src[2], CHAN_W, TGSI_EXEC_DATA_FLOAT);
2692 micro_mul(&r[3], &r[3], &r[1]);
2693 micro_add(&r[2], &r[2], &r[3]);
2694 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2695 micro_add(&d[1], &r[2], &r[3]);
2696 }
2697 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2698 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2699 }
2700 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2701 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2702 }
2703 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2704 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2705 }
2706 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2707 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2708 }
2709 }
2710
2711 static void
2712 exec_rfl(struct tgsi_exec_machine *mach,
2713 const struct tgsi_full_instruction *inst)
2714 {
2715 union tgsi_exec_channel r[9];
2716
2717 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) {
2718 /* r0 = dp3(src0, src0) */
2719 fetch_source(mach, &r[2], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2720 micro_mul(&r[0], &r[2], &r[2]);
2721 fetch_source(mach, &r[4], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2722 micro_mul(&r[8], &r[4], &r[4]);
2723 micro_add(&r[0], &r[0], &r[8]);
2724 fetch_source(mach, &r[6], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2725 micro_mul(&r[8], &r[6], &r[6]);
2726 micro_add(&r[0], &r[0], &r[8]);
2727
2728 /* r1 = dp3(src0, src1) */
2729 fetch_source(mach, &r[3], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2730 micro_mul(&r[1], &r[2], &r[3]);
2731 fetch_source(mach, &r[5], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2732 micro_mul(&r[8], &r[4], &r[5]);
2733 micro_add(&r[1], &r[1], &r[8]);
2734 fetch_source(mach, &r[7], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2735 micro_mul(&r[8], &r[6], &r[7]);
2736 micro_add(&r[1], &r[1], &r[8]);
2737
2738 /* r1 = 2 * r1 / r0 */
2739 micro_add(&r[1], &r[1], &r[1]);
2740 micro_div(&r[1], &r[1], &r[0]);
2741
2742 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2743 micro_mul(&r[2], &r[2], &r[1]);
2744 micro_sub(&r[2], &r[2], &r[3]);
2745 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2746 }
2747 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2748 micro_mul(&r[4], &r[4], &r[1]);
2749 micro_sub(&r[4], &r[4], &r[5]);
2750 store_dest(mach, &r[4], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2751 }
2752 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2753 micro_mul(&r[6], &r[6], &r[1]);
2754 micro_sub(&r[6], &r[6], &r[7]);
2755 store_dest(mach, &r[6], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2756 }
2757 }
2758 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2759 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2760 }
2761 }
2762
2763 static void
2764 exec_xpd(struct tgsi_exec_machine *mach,
2765 const struct tgsi_full_instruction *inst)
2766 {
2767 union tgsi_exec_channel r[6];
2768 union tgsi_exec_channel d[3];
2769
2770 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2771 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2772
2773 micro_mul(&r[2], &r[0], &r[1]);
2774
2775 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2776 fetch_source(mach, &r[4], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2777
2778 micro_mul(&r[5], &r[3], &r[4] );
2779 micro_sub(&d[CHAN_X], &r[2], &r[5]);
2780
2781 fetch_source(mach, &r[2], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2782
2783 micro_mul(&r[3], &r[3], &r[2]);
2784
2785 fetch_source(mach, &r[5], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2786
2787 micro_mul(&r[1], &r[1], &r[5]);
2788 micro_sub(&d[CHAN_Y], &r[3], &r[1]);
2789
2790 micro_mul(&r[5], &r[5], &r[4]);
2791 micro_mul(&r[0], &r[0], &r[2]);
2792 micro_sub(&d[CHAN_Z], &r[5], &r[0]);
2793
2794 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2795 store_dest(mach, &d[CHAN_X], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2796 }
2797 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2798 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2799 }
2800 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2801 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2802 }
2803 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2804 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2805 }
2806 }
2807
2808 static void
2809 exec_dst(struct tgsi_exec_machine *mach,
2810 const struct tgsi_full_instruction *inst)
2811 {
2812 union tgsi_exec_channel r[2];
2813 union tgsi_exec_channel d[4];
2814
2815 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2816 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2817 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2818 micro_mul(&d[CHAN_Y], &r[0], &r[1]);
2819 }
2820 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2821 fetch_source(mach, &d[CHAN_Z], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2822 }
2823 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2824 fetch_source(mach, &d[CHAN_W], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT);
2825 }
2826
2827 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2828 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2829 }
2830 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2831 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2832 }
2833 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2834 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2835 }
2836 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2837 store_dest(mach, &d[CHAN_W], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2838 }
2839 }
2840
2841 static void
2842 exec_log(struct tgsi_exec_machine *mach,
2843 const struct tgsi_full_instruction *inst)
2844 {
2845 union tgsi_exec_channel r[3];
2846
2847 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2848 micro_abs(&r[2], &r[0]); /* r2 = abs(r0) */
2849 micro_lg2(&r[1], &r[2]); /* r1 = lg2(r2) */
2850 micro_flr(&r[0], &r[1]); /* r0 = floor(r1) */
2851 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2852 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2853 }
2854 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2855 micro_exp2(&r[0], &r[0]); /* r0 = 2 ^ r0 */
2856 micro_div(&r[0], &r[2], &r[0]); /* r0 = r2 / r0 */
2857 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2858 }
2859 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2860 store_dest(mach, &r[1], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2861 }
2862 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2863 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2864 }
2865 }
2866
2867 static void
2868 exec_exp(struct tgsi_exec_machine *mach,
2869 const struct tgsi_full_instruction *inst)
2870 {
2871 union tgsi_exec_channel r[3];
2872
2873 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2874 micro_flr(&r[1], &r[0]); /* r1 = floor(r0) */
2875 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2876 micro_exp2(&r[2], &r[1]); /* r2 = 2 ^ r1 */
2877 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2878 }
2879 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2880 micro_sub(&r[2], &r[0], &r[1]); /* r2 = r0 - r1 */
2881 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2882 }
2883 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2884 micro_exp2(&r[2], &r[0]); /* r2 = 2 ^ r0 */
2885 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2886 }
2887 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2888 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2889 }
2890 }
2891
2892 static void
2893 exec_lit(struct tgsi_exec_machine *mach,
2894 const struct tgsi_full_instruction *inst)
2895 {
2896 union tgsi_exec_channel r[3];
2897 union tgsi_exec_channel d[3];
2898
2899 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
2900 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT);
2901 }
2902 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YZ) {
2903 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT);
2904 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
2905 micro_max(&d[CHAN_Y], &r[0], &ZeroVec);
2906 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2907 }
2908
2909 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
2910 fetch_source(mach, &r[1], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT);
2911 micro_max(&r[1], &r[1], &ZeroVec);
2912
2913 fetch_source(mach, &r[2], &inst->Src[0], CHAN_W, TGSI_EXEC_DATA_FLOAT);
2914 micro_min(&r[2], &r[2], &P128Vec);
2915 micro_max(&r[2], &r[2], &M128Vec);
2916 micro_pow(&r[1], &r[1], &r[2]);
2917 micro_lt(&d[CHAN_Z], &ZeroVec, &r[0], &r[1], &ZeroVec);
2918 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT);
2919 }
2920 }
2921 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
2922 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT);
2923 }
2924 }
2925
2926 static void
2927 exec_break(struct tgsi_exec_machine *mach)
2928 {
2929 if (mach->BreakType == TGSI_EXEC_BREAK_INSIDE_LOOP) {
2930 /* turn off loop channels for each enabled exec channel */
2931 mach->LoopMask &= ~mach->ExecMask;
2932 /* Todo: if mach->LoopMask == 0, jump to end of loop */
2933 UPDATE_EXEC_MASK(mach);
2934 } else {
2935 assert(mach->BreakType == TGSI_EXEC_BREAK_INSIDE_SWITCH);
2936
2937 mach->Switch.mask = 0x0;
2938
2939 UPDATE_EXEC_MASK(mach);
2940 }
2941 }
2942
2943 static void
2944 exec_switch(struct tgsi_exec_machine *mach,
2945 const struct tgsi_full_instruction *inst)
2946 {
2947 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING);
2948 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
2949
2950 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch;
2951 fetch_source(mach, &mach->Switch.selector, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT);
2952 mach->Switch.mask = 0x0;
2953 mach->Switch.defaultMask = 0x0;
2954
2955 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
2956 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_SWITCH;
2957
2958 UPDATE_EXEC_MASK(mach);
2959 }
2960
2961 static void
2962 exec_case(struct tgsi_exec_machine *mach,
2963 const struct tgsi_full_instruction *inst)
2964 {
2965 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask;
2966 union tgsi_exec_channel src;
2967 uint mask = 0;
2968
2969 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT);
2970
2971 if (mach->Switch.selector.u[0] == src.u[0]) {
2972 mask |= 0x1;
2973 }
2974 if (mach->Switch.selector.u[1] == src.u[1]) {
2975 mask |= 0x2;
2976 }
2977 if (mach->Switch.selector.u[2] == src.u[2]) {
2978 mask |= 0x4;
2979 }
2980 if (mach->Switch.selector.u[3] == src.u[3]) {
2981 mask |= 0x8;
2982 }
2983
2984 mach->Switch.defaultMask |= mask;
2985
2986 mach->Switch.mask |= mask & prevMask;
2987
2988 UPDATE_EXEC_MASK(mach);
2989 }
2990
2991 static void
2992 exec_default(struct tgsi_exec_machine *mach)
2993 {
2994 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask;
2995
2996 mach->Switch.mask |= ~mach->Switch.defaultMask & prevMask;
2997
2998 UPDATE_EXEC_MASK(mach);
2999 }
3000
3001 static void
3002 exec_endswitch(struct tgsi_exec_machine *mach)
3003 {
3004 mach->Switch = mach->SwitchStack[--mach->SwitchStackTop];
3005 mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
3006
3007 UPDATE_EXEC_MASK(mach);
3008 }
3009
3010 static void
3011 micro_i2f(union tgsi_exec_channel *dst,
3012 const union tgsi_exec_channel *src)
3013 {
3014 dst->f[0] = (float)src->i[0];
3015 dst->f[1] = (float)src->i[1];
3016 dst->f[2] = (float)src->i[2];
3017 dst->f[3] = (float)src->i[3];
3018 }
3019
3020 static void
3021 micro_not(union tgsi_exec_channel *dst,
3022 const union tgsi_exec_channel *src)
3023 {
3024 dst->u[0] = ~src->u[0];
3025 dst->u[1] = ~src->u[1];
3026 dst->u[2] = ~src->u[2];
3027 dst->u[3] = ~src->u[3];
3028 }
3029
3030 static void
3031 micro_shl(union tgsi_exec_channel *dst,
3032 const union tgsi_exec_channel *src0,
3033 const union tgsi_exec_channel *src1)
3034 {
3035 dst->u[0] = src0->u[0] << src1->u[0];
3036 dst->u[1] = src0->u[1] << src1->u[1];
3037 dst->u[2] = src0->u[2] << src1->u[2];
3038 dst->u[3] = src0->u[3] << src1->u[3];
3039 }
3040
3041 static void
3042 micro_and(union tgsi_exec_channel *dst,
3043 const union tgsi_exec_channel *src0,
3044 const union tgsi_exec_channel *src1)
3045 {
3046 dst->u[0] = src0->u[0] & src1->u[0];
3047 dst->u[1] = src0->u[1] & src1->u[1];
3048 dst->u[2] = src0->u[2] & src1->u[2];
3049 dst->u[3] = src0->u[3] & src1->u[3];
3050 }
3051
3052 static void
3053 micro_or(union tgsi_exec_channel *dst,
3054 const union tgsi_exec_channel *src0,
3055 const union tgsi_exec_channel *src1)
3056 {
3057 dst->u[0] = src0->u[0] | src1->u[0];
3058 dst->u[1] = src0->u[1] | src1->u[1];
3059 dst->u[2] = src0->u[2] | src1->u[2];
3060 dst->u[3] = src0->u[3] | src1->u[3];
3061 }
3062
3063 static void
3064 micro_xor(union tgsi_exec_channel *dst,
3065 const union tgsi_exec_channel *src0,
3066 const union tgsi_exec_channel *src1)
3067 {
3068 dst->u[0] = src0->u[0] ^ src1->u[0];
3069 dst->u[1] = src0->u[1] ^ src1->u[1];
3070 dst->u[2] = src0->u[2] ^ src1->u[2];
3071 dst->u[3] = src0->u[3] ^ src1->u[3];
3072 }
3073
3074 static void
3075 micro_mod(union tgsi_exec_channel *dst,
3076 const union tgsi_exec_channel *src0,
3077 const union tgsi_exec_channel *src1)
3078 {
3079 dst->i[0] = src0->i[0] % src1->i[0];
3080 dst->i[1] = src0->i[1] % src1->i[1];
3081 dst->i[2] = src0->i[2] % src1->i[2];
3082 dst->i[3] = src0->i[3] % src1->i[3];
3083 }
3084
3085 static void
3086 micro_f2i(union tgsi_exec_channel *dst,
3087 const union tgsi_exec_channel *src)
3088 {
3089 dst->i[0] = (int)src->f[0];
3090 dst->i[1] = (int)src->f[1];
3091 dst->i[2] = (int)src->f[2];
3092 dst->i[3] = (int)src->f[3];
3093 }
3094
3095 static void
3096 micro_idiv(union tgsi_exec_channel *dst,
3097 const union tgsi_exec_channel *src0,
3098 const union tgsi_exec_channel *src1)
3099 {
3100 dst->i[0] = src0->i[0] / src1->i[0];
3101 dst->i[1] = src0->i[1] / src1->i[1];
3102 dst->i[2] = src0->i[2] / src1->i[2];
3103 dst->i[3] = src0->i[3] / src1->i[3];
3104 }
3105
3106 static void
3107 micro_imax(union tgsi_exec_channel *dst,
3108 const union tgsi_exec_channel *src0,
3109 const union tgsi_exec_channel *src1)
3110 {
3111 dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0];
3112 dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1];
3113 dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2];
3114 dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3];
3115 }
3116
3117 static void
3118 micro_imin(union tgsi_exec_channel *dst,
3119 const union tgsi_exec_channel *src0,
3120 const union tgsi_exec_channel *src1)
3121 {
3122 dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0];
3123 dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1];
3124 dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2];
3125 dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3];
3126 }
3127
3128 static void
3129 micro_isge(union tgsi_exec_channel *dst,
3130 const union tgsi_exec_channel *src0,
3131 const union tgsi_exec_channel *src1)
3132 {
3133 dst->i[0] = src0->i[0] >= src1->i[0] ? -1 : 0;
3134 dst->i[1] = src0->i[1] >= src1->i[1] ? -1 : 0;
3135 dst->i[2] = src0->i[2] >= src1->i[2] ? -1 : 0;
3136 dst->i[3] = src0->i[3] >= src1->i[3] ? -1 : 0;
3137 }
3138
3139 static void
3140 micro_ishr(union tgsi_exec_channel *dst,
3141 const union tgsi_exec_channel *src0,
3142 const union tgsi_exec_channel *src1)
3143 {
3144 dst->i[0] = src0->i[0] >> src1->i[0];
3145 dst->i[1] = src0->i[1] >> src1->i[1];
3146 dst->i[2] = src0->i[2] >> src1->i[2];
3147 dst->i[3] = src0->i[3] >> src1->i[3];
3148 }
3149
3150 static void
3151 micro_islt(union tgsi_exec_channel *dst,
3152 const union tgsi_exec_channel *src0,
3153 const union tgsi_exec_channel *src1)
3154 {
3155 dst->i[0] = src0->i[0] < src1->i[0] ? -1 : 0;
3156 dst->i[1] = src0->i[1] < src1->i[1] ? -1 : 0;
3157 dst->i[2] = src0->i[2] < src1->i[2] ? -1 : 0;
3158 dst->i[3] = src0->i[3] < src1->i[3] ? -1 : 0;
3159 }
3160
3161 static void
3162 micro_f2u(union tgsi_exec_channel *dst,
3163 const union tgsi_exec_channel *src)
3164 {
3165 dst->u[0] = (uint)src->f[0];
3166 dst->u[1] = (uint)src->f[1];
3167 dst->u[2] = (uint)src->f[2];
3168 dst->u[3] = (uint)src->f[3];
3169 }
3170
3171 static void
3172 micro_u2f(union tgsi_exec_channel *dst,
3173 const union tgsi_exec_channel *src)
3174 {
3175 dst->f[0] = (float)src->u[0];
3176 dst->f[1] = (float)src->u[1];
3177 dst->f[2] = (float)src->u[2];
3178 dst->f[3] = (float)src->u[3];
3179 }
3180
3181 static void
3182 micro_uadd(union tgsi_exec_channel *dst,
3183 const union tgsi_exec_channel *src0,
3184 const union tgsi_exec_channel *src1)
3185 {
3186 dst->u[0] = src0->u[0] + src1->u[0];
3187 dst->u[1] = src0->u[1] + src1->u[1];
3188 dst->u[2] = src0->u[2] + src1->u[2];
3189 dst->u[3] = src0->u[3] + src1->u[3];
3190 }
3191
3192 static void
3193 micro_udiv(union tgsi_exec_channel *dst,
3194 const union tgsi_exec_channel *src0,
3195 const union tgsi_exec_channel *src1)
3196 {
3197 dst->u[0] = src0->u[0] / src1->u[0];
3198 dst->u[1] = src0->u[1] / src1->u[1];
3199 dst->u[2] = src0->u[2] / src1->u[2];
3200 dst->u[3] = src0->u[3] / src1->u[3];
3201 }
3202
3203 static void
3204 micro_umad(union tgsi_exec_channel *dst,
3205 const union tgsi_exec_channel *src0,
3206 const union tgsi_exec_channel *src1,
3207 const union tgsi_exec_channel *src2)
3208 {
3209 dst->u[0] = src0->u[0] * src1->u[0] + src2->u[0];
3210 dst->u[1] = src0->u[1] * src1->u[1] + src2->u[1];
3211 dst->u[2] = src0->u[2] * src1->u[2] + src2->u[2];
3212 dst->u[3] = src0->u[3] * src1->u[3] + src2->u[3];
3213 }
3214
3215 static void
3216 micro_umax(union tgsi_exec_channel *dst,
3217 const union tgsi_exec_channel *src0,
3218 const union tgsi_exec_channel *src1)
3219 {
3220 dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0];
3221 dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1];
3222 dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2];
3223 dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3];
3224 }
3225
3226 static void
3227 micro_umin(union tgsi_exec_channel *dst,
3228 const union tgsi_exec_channel *src0,
3229 const union tgsi_exec_channel *src1)
3230 {
3231 dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0];
3232 dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1];
3233 dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2];
3234 dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3];
3235 }
3236
3237 static void
3238 micro_umod(union tgsi_exec_channel *dst,
3239 const union tgsi_exec_channel *src0,
3240 const union tgsi_exec_channel *src1)
3241 {
3242 dst->u[0] = src0->u[0] % src1->u[0];
3243 dst->u[1] = src0->u[1] % src1->u[1];
3244 dst->u[2] = src0->u[2] % src1->u[2];
3245 dst->u[3] = src0->u[3] % src1->u[3];
3246 }
3247
3248 static void
3249 micro_umul(union tgsi_exec_channel *dst,
3250 const union tgsi_exec_channel *src0,
3251 const union tgsi_exec_channel *src1)
3252 {
3253 dst->u[0] = src0->u[0] * src1->u[0];
3254 dst->u[1] = src0->u[1] * src1->u[1];
3255 dst->u[2] = src0->u[2] * src1->u[2];
3256 dst->u[3] = src0->u[3] * src1->u[3];
3257 }
3258
3259 static void
3260 micro_useq(union tgsi_exec_channel *dst,
3261 const union tgsi_exec_channel *src0,
3262 const union tgsi_exec_channel *src1)
3263 {
3264 dst->u[0] = src0->u[0] == src1->u[0] ? ~0 : 0;
3265 dst->u[1] = src0->u[1] == src1->u[1] ? ~0 : 0;
3266 dst->u[2] = src0->u[2] == src1->u[2] ? ~0 : 0;
3267 dst->u[3] = src0->u[3] == src1->u[3] ? ~0 : 0;
3268 }
3269
3270 static void
3271 micro_usge(union tgsi_exec_channel *dst,
3272 const union tgsi_exec_channel *src0,
3273 const union tgsi_exec_channel *src1)
3274 {
3275 dst->u[0] = src0->u[0] >= src1->u[0] ? ~0 : 0;
3276 dst->u[1] = src0->u[1] >= src1->u[1] ? ~0 : 0;
3277 dst->u[2] = src0->u[2] >= src1->u[2] ? ~0 : 0;
3278 dst->u[3] = src0->u[3] >= src1->u[3] ? ~0 : 0;
3279 }
3280
3281 static void
3282 micro_ushr(union tgsi_exec_channel *dst,
3283 const union tgsi_exec_channel *src0,
3284 const union tgsi_exec_channel *src1)
3285 {
3286 dst->u[0] = src0->u[0] >> src1->u[0];
3287 dst->u[1] = src0->u[1] >> src1->u[1];
3288 dst->u[2] = src0->u[2] >> src1->u[2];
3289 dst->u[3] = src0->u[3] >> src1->u[3];
3290 }
3291
3292 static void
3293 micro_uslt(union tgsi_exec_channel *dst,
3294 const union tgsi_exec_channel *src0,
3295 const union tgsi_exec_channel *src1)
3296 {
3297 dst->u[0] = src0->u[0] < src1->u[0] ? ~0 : 0;
3298 dst->u[1] = src0->u[1] < src1->u[1] ? ~0 : 0;
3299 dst->u[2] = src0->u[2] < src1->u[2] ? ~0 : 0;
3300 dst->u[3] = src0->u[3] < src1->u[3] ? ~0 : 0;
3301 }
3302
3303 static void
3304 micro_usne(union tgsi_exec_channel *dst,
3305 const union tgsi_exec_channel *src0,
3306 const union tgsi_exec_channel *src1)
3307 {
3308 dst->u[0] = src0->u[0] != src1->u[0] ? ~0 : 0;
3309 dst->u[1] = src0->u[1] != src1->u[1] ? ~0 : 0;
3310 dst->u[2] = src0->u[2] != src1->u[2] ? ~0 : 0;
3311 dst->u[3] = src0->u[3] != src1->u[3] ? ~0 : 0;
3312 }
3313
3314 static void
3315 exec_instruction(
3316 struct tgsi_exec_machine *mach,
3317 const struct tgsi_full_instruction *inst,
3318 int *pc )
3319 {
3320 union tgsi_exec_channel r[10];
3321
3322 (*pc)++;
3323
3324 switch (inst->Instruction.Opcode) {
3325 case TGSI_OPCODE_ARL:
3326 exec_vector_unary(mach, inst, micro_arl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT);
3327 break;
3328
3329 case TGSI_OPCODE_MOV:
3330 exec_vector_unary(mach, inst, micro_mov, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT);
3331 break;
3332
3333 case TGSI_OPCODE_LIT:
3334 exec_lit(mach, inst);
3335 break;
3336
3337 case TGSI_OPCODE_RCP:
3338 exec_scalar_unary(mach, inst, micro_rcp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3339 break;
3340
3341 case TGSI_OPCODE_RSQ:
3342 exec_scalar_unary(mach, inst, micro_rsq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3343 break;
3344
3345 case TGSI_OPCODE_EXP:
3346 exec_exp(mach, inst);
3347 break;
3348
3349 case TGSI_OPCODE_LOG:
3350 exec_log(mach, inst);
3351 break;
3352
3353 case TGSI_OPCODE_MUL:
3354 exec_vector_binary(mach, inst, micro_mul, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3355 break;
3356
3357 case TGSI_OPCODE_ADD:
3358 exec_vector_binary(mach, inst, micro_add, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3359 break;
3360
3361 case TGSI_OPCODE_DP3:
3362 exec_dp3(mach, inst);
3363 break;
3364
3365 case TGSI_OPCODE_DP4:
3366 exec_dp4(mach, inst);
3367 break;
3368
3369 case TGSI_OPCODE_DST:
3370 exec_dst(mach, inst);
3371 break;
3372
3373 case TGSI_OPCODE_MIN:
3374 exec_vector_binary(mach, inst, micro_min, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3375 break;
3376
3377 case TGSI_OPCODE_MAX:
3378 exec_vector_binary(mach, inst, micro_max, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3379 break;
3380
3381 case TGSI_OPCODE_SLT:
3382 exec_vector_binary(mach, inst, micro_slt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3383 break;
3384
3385 case TGSI_OPCODE_SGE:
3386 exec_vector_binary(mach, inst, micro_sge, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3387 break;
3388
3389 case TGSI_OPCODE_MAD:
3390 exec_vector_trinary(mach, inst, micro_mad, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3391 break;
3392
3393 case TGSI_OPCODE_SUB:
3394 exec_vector_binary(mach, inst, micro_sub, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3395 break;
3396
3397 case TGSI_OPCODE_LRP:
3398 exec_vector_trinary(mach, inst, micro_lrp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3399 break;
3400
3401 case TGSI_OPCODE_CND:
3402 exec_vector_trinary(mach, inst, micro_cnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3403 break;
3404
3405 case TGSI_OPCODE_DP2A:
3406 exec_dp2a(mach, inst);
3407 break;
3408
3409 case TGSI_OPCODE_FRC:
3410 exec_vector_unary(mach, inst, micro_frc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3411 break;
3412
3413 case TGSI_OPCODE_CLAMP:
3414 exec_vector_trinary(mach, inst, micro_clamp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3415 break;
3416
3417 case TGSI_OPCODE_FLR:
3418 exec_vector_unary(mach, inst, micro_flr, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3419 break;
3420
3421 case TGSI_OPCODE_ROUND:
3422 exec_vector_unary(mach, inst, micro_rnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3423 break;
3424
3425 case TGSI_OPCODE_EX2:
3426 exec_scalar_unary(mach, inst, micro_exp2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3427 break;
3428
3429 case TGSI_OPCODE_LG2:
3430 exec_scalar_unary(mach, inst, micro_lg2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3431 break;
3432
3433 case TGSI_OPCODE_POW:
3434 exec_scalar_binary(mach, inst, micro_pow, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3435 break;
3436
3437 case TGSI_OPCODE_XPD:
3438 exec_xpd(mach, inst);
3439 break;
3440
3441 case TGSI_OPCODE_ABS:
3442 exec_vector_unary(mach, inst, micro_abs, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3443 break;
3444
3445 case TGSI_OPCODE_RCC:
3446 exec_scalar_unary(mach, inst, micro_rcc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3447 break;
3448
3449 case TGSI_OPCODE_DPH:
3450 exec_dph(mach, inst);
3451 break;
3452
3453 case TGSI_OPCODE_COS:
3454 exec_scalar_unary(mach, inst, micro_cos, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3455 break;
3456
3457 case TGSI_OPCODE_DDX:
3458 exec_vector_unary(mach, inst, micro_ddx, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3459 break;
3460
3461 case TGSI_OPCODE_DDY:
3462 exec_vector_unary(mach, inst, micro_ddy, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3463 break;
3464
3465 case TGSI_OPCODE_KILP:
3466 exec_kilp (mach, inst);
3467 break;
3468
3469 case TGSI_OPCODE_KIL:
3470 exec_kil (mach, inst);
3471 break;
3472
3473 case TGSI_OPCODE_PK2H:
3474 assert (0);
3475 break;
3476
3477 case TGSI_OPCODE_PK2US:
3478 assert (0);
3479 break;
3480
3481 case TGSI_OPCODE_PK4B:
3482 assert (0);
3483 break;
3484
3485 case TGSI_OPCODE_PK4UB:
3486 assert (0);
3487 break;
3488
3489 case TGSI_OPCODE_RFL:
3490 exec_rfl(mach, inst);
3491 break;
3492
3493 case TGSI_OPCODE_SEQ:
3494 exec_vector_binary(mach, inst, micro_seq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3495 break;
3496
3497 case TGSI_OPCODE_SFL:
3498 exec_vector(mach, inst, micro_sfl, TGSI_EXEC_DATA_FLOAT);
3499 break;
3500
3501 case TGSI_OPCODE_SGT:
3502 exec_vector_binary(mach, inst, micro_sgt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3503 break;
3504
3505 case TGSI_OPCODE_SIN:
3506 exec_scalar_unary(mach, inst, micro_sin, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3507 break;
3508
3509 case TGSI_OPCODE_SLE:
3510 exec_vector_binary(mach, inst, micro_sle, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3511 break;
3512
3513 case TGSI_OPCODE_SNE:
3514 exec_vector_binary(mach, inst, micro_sne, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3515 break;
3516
3517 case TGSI_OPCODE_STR:
3518 exec_vector(mach, inst, micro_str, TGSI_EXEC_DATA_FLOAT);
3519 break;
3520
3521 case TGSI_OPCODE_TEX:
3522 /* simple texture lookup */
3523 /* src[0] = texcoord */
3524 /* src[1] = sampler unit */
3525 exec_tex(mach, inst, TEX_MODIFIER_NONE);
3526 break;
3527
3528 case TGSI_OPCODE_TXB:
3529 /* Texture lookup with lod bias */
3530 /* src[0] = texcoord (src[0].w = LOD bias) */
3531 /* src[1] = sampler unit */
3532 exec_tex(mach, inst, TEX_MODIFIER_LOD_BIAS);
3533 break;
3534
3535 case TGSI_OPCODE_TXD:
3536 /* Texture lookup with explict partial derivatives */
3537 /* src[0] = texcoord */
3538 /* src[1] = d[strq]/dx */
3539 /* src[2] = d[strq]/dy */
3540 /* src[3] = sampler unit */
3541 exec_txd(mach, inst);
3542 break;
3543
3544 case TGSI_OPCODE_TXL:
3545 /* Texture lookup with explit LOD */
3546 /* src[0] = texcoord (src[0].w = LOD) */
3547 /* src[1] = sampler unit */
3548 exec_tex(mach, inst, TEX_MODIFIER_EXPLICIT_LOD);
3549 break;
3550
3551 case TGSI_OPCODE_TXP:
3552 /* Texture lookup with projection */
3553 /* src[0] = texcoord (src[0].w = projection) */
3554 /* src[1] = sampler unit */
3555 exec_tex(mach, inst, TEX_MODIFIER_PROJECTED);
3556 break;
3557
3558 case TGSI_OPCODE_UP2H:
3559 assert (0);
3560 break;
3561
3562 case TGSI_OPCODE_UP2US:
3563 assert (0);
3564 break;
3565
3566 case TGSI_OPCODE_UP4B:
3567 assert (0);
3568 break;
3569
3570 case TGSI_OPCODE_UP4UB:
3571 assert (0);
3572 break;
3573
3574 case TGSI_OPCODE_X2D:
3575 exec_x2d(mach, inst);
3576 break;
3577
3578 case TGSI_OPCODE_ARA:
3579 assert (0);
3580 break;
3581
3582 case TGSI_OPCODE_ARR:
3583 exec_vector_unary(mach, inst, micro_arr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT);
3584 break;
3585
3586 case TGSI_OPCODE_BRA:
3587 assert (0);
3588 break;
3589
3590 case TGSI_OPCODE_CAL:
3591 /* skip the call if no execution channels are enabled */
3592 if (mach->ExecMask) {
3593 /* do the call */
3594
3595 /* First, record the depths of the execution stacks.
3596 * This is important for deeply nested/looped return statements.
3597 * We have to unwind the stacks by the correct amount. For a
3598 * real code generator, we could determine the number of entries
3599 * to pop off each stack with simple static analysis and avoid
3600 * implementing this data structure at run time.
3601 */
3602 mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop;
3603 mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop;
3604 mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop;
3605 mach->CallStack[mach->CallStackTop].SwitchStackTop = mach->SwitchStackTop;
3606 mach->CallStack[mach->CallStackTop].BreakStackTop = mach->BreakStackTop;
3607 /* note that PC was already incremented above */
3608 mach->CallStack[mach->CallStackTop].ReturnAddr = *pc;
3609
3610 mach->CallStackTop++;
3611
3612 /* Second, push the Cond, Loop, Cont, Func stacks */
3613 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
3614 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3615 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3616 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING);
3617 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
3618 assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
3619
3620 mach->CondStack[mach->CondStackTop++] = mach->CondMask;
3621 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
3622 mach->ContStack[mach->ContStackTop++] = mach->ContMask;
3623 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch;
3624 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
3625 mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
3626
3627 /* Finally, jump to the subroutine */
3628 *pc = inst->Label.Label;
3629 }
3630 break;
3631
3632 case TGSI_OPCODE_RET:
3633 mach->FuncMask &= ~mach->ExecMask;
3634 UPDATE_EXEC_MASK(mach);
3635
3636 if (mach->FuncMask == 0x0) {
3637 /* really return now (otherwise, keep executing */
3638
3639 if (mach->CallStackTop == 0) {
3640 /* returning from main() */
3641 mach->CondStackTop = 0;
3642 mach->LoopStackTop = 0;
3643 *pc = -1;
3644 return;
3645 }
3646
3647 assert(mach->CallStackTop > 0);
3648 mach->CallStackTop--;
3649
3650 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
3651 mach->CondMask = mach->CondStack[mach->CondStackTop];
3652
3653 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
3654 mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
3655
3656 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
3657 mach->ContMask = mach->ContStack[mach->ContStackTop];
3658
3659 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop;
3660 mach->Switch = mach->SwitchStack[mach->SwitchStackTop];
3661
3662 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop;
3663 mach->BreakType = mach->BreakStack[mach->BreakStackTop];
3664
3665 assert(mach->FuncStackTop > 0);
3666 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
3667
3668 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
3669
3670 UPDATE_EXEC_MASK(mach);
3671 }
3672 break;
3673
3674 case TGSI_OPCODE_SSG:
3675 exec_vector_unary(mach, inst, micro_sgn, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3676 break;
3677
3678 case TGSI_OPCODE_CMP:
3679 exec_vector_trinary(mach, inst, micro_cmp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3680 break;
3681
3682 case TGSI_OPCODE_SCS:
3683 exec_scs(mach, inst);
3684 break;
3685
3686 case TGSI_OPCODE_NRM:
3687 exec_nrm3(mach, inst);
3688 break;
3689
3690 case TGSI_OPCODE_NRM4:
3691 exec_nrm4(mach, inst);
3692 break;
3693
3694 case TGSI_OPCODE_DIV:
3695 exec_vector_binary(mach, inst, micro_div, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3696 break;
3697
3698 case TGSI_OPCODE_DP2:
3699 exec_dp2(mach, inst);
3700 break;
3701
3702 case TGSI_OPCODE_IF:
3703 /* push CondMask */
3704 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
3705 mach->CondStack[mach->CondStackTop++] = mach->CondMask;
3706 FETCH( &r[0], 0, CHAN_X );
3707 /* update CondMask */
3708 if( ! r[0].u[0] ) {
3709 mach->CondMask &= ~0x1;
3710 }
3711 if( ! r[0].u[1] ) {
3712 mach->CondMask &= ~0x2;
3713 }
3714 if( ! r[0].u[2] ) {
3715 mach->CondMask &= ~0x4;
3716 }
3717 if( ! r[0].u[3] ) {
3718 mach->CondMask &= ~0x8;
3719 }
3720 UPDATE_EXEC_MASK(mach);
3721 /* Todo: If CondMask==0, jump to ELSE */
3722 break;
3723
3724 case TGSI_OPCODE_ELSE:
3725 /* invert CondMask wrt previous mask */
3726 {
3727 uint prevMask;
3728 assert(mach->CondStackTop > 0);
3729 prevMask = mach->CondStack[mach->CondStackTop - 1];
3730 mach->CondMask = ~mach->CondMask & prevMask;
3731 UPDATE_EXEC_MASK(mach);
3732 /* Todo: If CondMask==0, jump to ENDIF */
3733 }
3734 break;
3735
3736 case TGSI_OPCODE_ENDIF:
3737 /* pop CondMask */
3738 assert(mach->CondStackTop > 0);
3739 mach->CondMask = mach->CondStack[--mach->CondStackTop];
3740 UPDATE_EXEC_MASK(mach);
3741 break;
3742
3743 case TGSI_OPCODE_END:
3744 /* make sure we end primitives which haven't
3745 * been explicitly emitted */
3746 conditional_emit_primitive(mach);
3747 /* halt execution */
3748 *pc = -1;
3749 break;
3750
3751 case TGSI_OPCODE_PUSHA:
3752 assert (0);
3753 break;
3754
3755 case TGSI_OPCODE_POPA:
3756 assert (0);
3757 break;
3758
3759 case TGSI_OPCODE_CEIL:
3760 exec_vector_unary(mach, inst, micro_ceil, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3761 break;
3762
3763 case TGSI_OPCODE_I2F:
3764 exec_vector_unary(mach, inst, micro_i2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_INT);
3765 break;
3766
3767 case TGSI_OPCODE_NOT:
3768 exec_vector_unary(mach, inst, micro_not, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3769 break;
3770
3771 case TGSI_OPCODE_TRUNC:
3772 exec_vector_unary(mach, inst, micro_trunc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT);
3773 break;
3774
3775 case TGSI_OPCODE_SHL:
3776 exec_vector_binary(mach, inst, micro_shl, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3777 break;
3778
3779 case TGSI_OPCODE_AND:
3780 exec_vector_binary(mach, inst, micro_and, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3781 break;
3782
3783 case TGSI_OPCODE_OR:
3784 exec_vector_binary(mach, inst, micro_or, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3785 break;
3786
3787 case TGSI_OPCODE_MOD:
3788 exec_vector_binary(mach, inst, micro_mod, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3789 break;
3790
3791 case TGSI_OPCODE_XOR:
3792 exec_vector_binary(mach, inst, micro_xor, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3793 break;
3794
3795 case TGSI_OPCODE_SAD:
3796 assert (0);
3797 break;
3798
3799 case TGSI_OPCODE_TXF:
3800 exec_txf(mach, inst);
3801 break;
3802
3803 case TGSI_OPCODE_TXQ:
3804 exec_txq(mach, inst);
3805 break;
3806
3807 case TGSI_OPCODE_EMIT:
3808 emit_vertex(mach);
3809 break;
3810
3811 case TGSI_OPCODE_ENDPRIM:
3812 emit_primitive(mach);
3813 break;
3814
3815 case TGSI_OPCODE_BGNLOOP:
3816 /* push LoopMask and ContMasks */
3817 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3818 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3819 assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3820 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
3821
3822 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
3823 mach->ContStack[mach->ContStackTop++] = mach->ContMask;
3824 mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1;
3825 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
3826 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP;
3827 break;
3828
3829 case TGSI_OPCODE_ENDLOOP:
3830 /* Restore ContMask, but don't pop */
3831 assert(mach->ContStackTop > 0);
3832 mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
3833 UPDATE_EXEC_MASK(mach);
3834 if (mach->ExecMask) {
3835 /* repeat loop: jump to instruction just past BGNLOOP */
3836 assert(mach->LoopLabelStackTop > 0);
3837 *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1;
3838 }
3839 else {
3840 /* exit loop: pop LoopMask */
3841 assert(mach->LoopStackTop > 0);
3842 mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
3843 /* pop ContMask */
3844 assert(mach->ContStackTop > 0);
3845 mach->ContMask = mach->ContStack[--mach->ContStackTop];
3846 assert(mach->LoopLabelStackTop > 0);
3847 --mach->LoopLabelStackTop;
3848
3849 mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
3850 }
3851 UPDATE_EXEC_MASK(mach);
3852 break;
3853
3854 case TGSI_OPCODE_BRK:
3855 exec_break(mach);
3856 break;
3857
3858 case TGSI_OPCODE_CONT:
3859 /* turn off cont channels for each enabled exec channel */
3860 mach->ContMask &= ~mach->ExecMask;
3861 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3862 UPDATE_EXEC_MASK(mach);
3863 break;
3864
3865 case TGSI_OPCODE_BGNSUB:
3866 /* no-op */
3867 break;
3868
3869 case TGSI_OPCODE_ENDSUB:
3870 /*
3871 * XXX: This really should be a no-op. We should never reach this opcode.
3872 */
3873
3874 assert(mach->CallStackTop > 0);
3875 mach->CallStackTop--;
3876
3877 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
3878 mach->CondMask = mach->CondStack[mach->CondStackTop];
3879
3880 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
3881 mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
3882
3883 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
3884 mach->ContMask = mach->ContStack[mach->ContStackTop];
3885
3886 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop;
3887 mach->Switch = mach->SwitchStack[mach->SwitchStackTop];
3888
3889 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop;
3890 mach->BreakType = mach->BreakStack[mach->BreakStackTop];
3891
3892 assert(mach->FuncStackTop > 0);
3893 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
3894
3895 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
3896
3897 UPDATE_EXEC_MASK(mach);
3898 break;
3899
3900 case TGSI_OPCODE_NOP:
3901 break;
3902
3903 case TGSI_OPCODE_BREAKC:
3904 FETCH(&r[0], 0, CHAN_X);
3905 /* update CondMask */
3906 if (r[0].u[0] && (mach->ExecMask & 0x1)) {
3907 mach->LoopMask &= ~0x1;
3908 }
3909 if (r[0].u[1] && (mach->ExecMask & 0x2)) {
3910 mach->LoopMask &= ~0x2;
3911 }
3912 if (r[0].u[2] && (mach->ExecMask & 0x4)) {
3913 mach->LoopMask &= ~0x4;
3914 }
3915 if (r[0].u[3] && (mach->ExecMask & 0x8)) {
3916 mach->LoopMask &= ~0x8;
3917 }
3918 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3919 UPDATE_EXEC_MASK(mach);
3920 break;
3921
3922 case TGSI_OPCODE_F2I:
3923 exec_vector_unary(mach, inst, micro_f2i, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT);
3924 break;
3925
3926 case TGSI_OPCODE_IDIV:
3927 exec_vector_binary(mach, inst, micro_idiv, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3928 break;
3929
3930 case TGSI_OPCODE_IMAX:
3931 exec_vector_binary(mach, inst, micro_imax, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3932 break;
3933
3934 case TGSI_OPCODE_IMIN:
3935 exec_vector_binary(mach, inst, micro_imin, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3936 break;
3937
3938 case TGSI_OPCODE_INEG:
3939 exec_vector_unary(mach, inst, micro_ineg, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3940 break;
3941
3942 case TGSI_OPCODE_ISGE:
3943 exec_vector_binary(mach, inst, micro_isge, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3944 break;
3945
3946 case TGSI_OPCODE_ISHR:
3947 exec_vector_binary(mach, inst, micro_ishr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3948 break;
3949
3950 case TGSI_OPCODE_ISLT:
3951 exec_vector_binary(mach, inst, micro_islt, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT);
3952 break;
3953
3954 case TGSI_OPCODE_F2U:
3955 exec_vector_unary(mach, inst, micro_f2u, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT);
3956 break;
3957
3958 case TGSI_OPCODE_U2F:
3959 exec_vector_unary(mach, inst, micro_u2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_UINT);
3960 break;
3961
3962 case TGSI_OPCODE_UADD:
3963 exec_vector_binary(mach, inst, micro_uadd, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3964 break;
3965
3966 case TGSI_OPCODE_UDIV:
3967 exec_vector_binary(mach, inst, micro_udiv, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3968 break;
3969
3970 case TGSI_OPCODE_UMAD:
3971 exec_vector_trinary(mach, inst, micro_umad, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3972 break;
3973
3974 case TGSI_OPCODE_UMAX:
3975 exec_vector_binary(mach, inst, micro_umax, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3976 break;
3977
3978 case TGSI_OPCODE_UMIN:
3979 exec_vector_binary(mach, inst, micro_umin, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3980 break;
3981
3982 case TGSI_OPCODE_UMOD:
3983 exec_vector_binary(mach, inst, micro_umod, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3984 break;
3985
3986 case TGSI_OPCODE_UMUL:
3987 exec_vector_binary(mach, inst, micro_umul, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3988 break;
3989
3990 case TGSI_OPCODE_USEQ:
3991 exec_vector_binary(mach, inst, micro_useq, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3992 break;
3993
3994 case TGSI_OPCODE_USGE:
3995 exec_vector_binary(mach, inst, micro_usge, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
3996 break;
3997
3998 case TGSI_OPCODE_USHR:
3999 exec_vector_binary(mach, inst, micro_ushr, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
4000 break;
4001
4002 case TGSI_OPCODE_USLT:
4003 exec_vector_binary(mach, inst, micro_uslt, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
4004 break;
4005
4006 case TGSI_OPCODE_USNE:
4007 exec_vector_binary(mach, inst, micro_usne, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT);
4008 break;
4009
4010 case TGSI_OPCODE_SWITCH:
4011 exec_switch(mach, inst);
4012 break;
4013
4014 case TGSI_OPCODE_CASE:
4015 exec_case(mach, inst);
4016 break;
4017
4018 case TGSI_OPCODE_DEFAULT:
4019 exec_default(mach);
4020 break;
4021
4022 case TGSI_OPCODE_ENDSWITCH:
4023 exec_endswitch(mach);
4024 break;
4025
4026 case TGSI_OPCODE_LOAD:
4027 assert(0);
4028 break;
4029
4030 case TGSI_OPCODE_LOAD_MS:
4031 assert(0);
4032 break;
4033
4034 case TGSI_OPCODE_SAMPLE:
4035 exec_sample(mach, inst, TEX_MODIFIER_NONE);
4036 break;
4037
4038 case TGSI_OPCODE_SAMPLE_B:
4039 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS);
4040 break;
4041
4042 case TGSI_OPCODE_SAMPLE_C:
4043 exec_sample(mach, inst, TEX_MODIFIER_NONE);
4044 break;
4045
4046 case TGSI_OPCODE_SAMPLE_C_LZ:
4047 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS);
4048 break;
4049
4050 case TGSI_OPCODE_SAMPLE_D:
4051 exec_sample_d(mach, inst);
4052 break;
4053
4054 case TGSI_OPCODE_SAMPLE_L:
4055 exec_sample(mach, inst, TEX_MODIFIER_EXPLICIT_LOD);
4056 break;
4057
4058 case TGSI_OPCODE_GATHER4:
4059 assert(0);
4060 break;
4061
4062 case TGSI_OPCODE_RESINFO:
4063 assert(0);
4064 break;
4065
4066 case TGSI_OPCODE_SAMPLE_POS:
4067 assert(0);
4068 break;
4069
4070 case TGSI_OPCODE_SAMPLE_INFO:
4071 assert(0);
4072 break;
4073
4074 default:
4075 assert( 0 );
4076 }
4077 }
4078
4079
4080 #define DEBUG_EXECUTION 0
4081
4082
4083 /**
4084 * Run TGSI interpreter.
4085 * \return bitmask of "alive" quad components
4086 */
4087 uint
4088 tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
4089 {
4090 uint i;
4091 int pc = 0;
4092
4093 mach->CondMask = 0xf;
4094 mach->LoopMask = 0xf;
4095 mach->ContMask = 0xf;
4096 mach->FuncMask = 0xf;
4097 mach->ExecMask = 0xf;
4098
4099 mach->Switch.mask = 0xf;
4100
4101 assert(mach->CondStackTop == 0);
4102 assert(mach->LoopStackTop == 0);
4103 assert(mach->ContStackTop == 0);
4104 assert(mach->SwitchStackTop == 0);
4105 assert(mach->BreakStackTop == 0);
4106 assert(mach->CallStackTop == 0);
4107
4108 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0;
4109 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0;
4110
4111 if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) {
4112 mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0;
4113 mach->Primitives[0] = 0;
4114 }
4115
4116 /* execute declarations (interpolants) */
4117 for (i = 0; i < mach->NumDeclarations; i++) {
4118 exec_declaration( mach, mach->Declarations+i );
4119 }
4120
4121 {
4122 #if DEBUG_EXECUTION
4123 struct tgsi_exec_vector temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS];
4124 struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS];
4125 uint inst = 1;
4126
4127 memcpy(temps, mach->Temps, sizeof(temps));
4128 memcpy(outputs, mach->Outputs, sizeof(outputs));
4129 #endif
4130
4131 /* execute instructions, until pc is set to -1 */
4132 while (pc != -1) {
4133
4134 #if DEBUG_EXECUTION
4135 uint i;
4136
4137 tgsi_dump_instruction(&mach->Instructions[pc], inst++);
4138 #endif
4139
4140 assert(pc < (int) mach->NumInstructions);
4141 exec_instruction(mach, mach->Instructions + pc, &pc);
4142
4143 #if DEBUG_EXECUTION
4144 for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) {
4145 if (memcmp(&temps[i], &mach->Temps[i], sizeof(temps[i]))) {
4146 uint j;
4147
4148 memcpy(&temps[i], &mach->Temps[i], sizeof(temps[i]));
4149 debug_printf("TEMP[%2u] = ", i);
4150 for (j = 0; j < 4; j++) {
4151 if (j > 0) {
4152 debug_printf(" ");
4153 }
4154 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
4155 temps[i].xyzw[0].f[j], temps[i].xyzw[0].u[j],
4156 temps[i].xyzw[1].f[j], temps[i].xyzw[1].u[j],
4157 temps[i].xyzw[2].f[j], temps[i].xyzw[2].u[j],
4158 temps[i].xyzw[3].f[j], temps[i].xyzw[3].u[j]);
4159 }
4160 }
4161 }
4162 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {
4163 if (memcmp(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]))) {
4164 uint j;
4165
4166 memcpy(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]));
4167 debug_printf("OUT[%2u] = ", i);
4168 for (j = 0; j < 4; j++) {
4169 if (j > 0) {
4170 debug_printf(" ");
4171 }
4172 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
4173 outputs[i].xyzw[0].f[j], outputs[i].xyzw[0].u[j],
4174 outputs[i].xyzw[1].f[j], outputs[i].xyzw[1].u[j],
4175 outputs[i].xyzw[2].f[j], outputs[i].xyzw[2].u[j],
4176 outputs[i].xyzw[3].f[j], outputs[i].xyzw[3].u[j]);
4177 }
4178 }
4179 }
4180 #endif
4181 }
4182 }
4183
4184 #if 0
4185 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
4186 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
4187 /*
4188 * Scale back depth component.
4189 */
4190 for (i = 0; i < 4; i++)
4191 mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF;
4192 }
4193 #endif
4194
4195 /* Strictly speaking, these assertions aren't really needed but they
4196 * can potentially catch some bugs in the control flow code.
4197 */
4198 assert(mach->CondStackTop == 0);
4199 assert(mach->LoopStackTop == 0);
4200 assert(mach->ContStackTop == 0);
4201 assert(mach->SwitchStackTop == 0);
4202 assert(mach->BreakStackTop == 0);
4203 assert(mach->CallStackTop == 0);
4204
4205 return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
4206 }