1 /**************************************************************************
3 * Copyright 2007-2008 VMware, Inc.
5 * Copyright 2009-2010 VMware, Inc. All rights Reserved.
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:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
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 VMWARE 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.
27 **************************************************************************/
30 * TGSI interpreter/executor.
32 * Flow control information:
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.
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.
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).
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_half.h"
62 #include "util/u_memory.h"
63 #include "util/u_math.h"
64 #include "util/rounding.h"
67 #define DEBUG_EXECUTION 0
72 #define TILE_TOP_LEFT 0
73 #define TILE_TOP_RIGHT 1
74 #define TILE_BOTTOM_LEFT 2
75 #define TILE_BOTTOM_RIGHT 3
77 union tgsi_double_channel
{
78 double d
[TGSI_QUAD_SIZE
];
79 unsigned u
[TGSI_QUAD_SIZE
][2];
80 uint64_t u64
[TGSI_QUAD_SIZE
];
81 int64_t i64
[TGSI_QUAD_SIZE
];
84 struct tgsi_double_vector
{
85 union tgsi_double_channel xy
;
86 union tgsi_double_channel zw
;
90 micro_abs(union tgsi_exec_channel
*dst
,
91 const union tgsi_exec_channel
*src
)
93 dst
->f
[0] = fabsf(src
->f
[0]);
94 dst
->f
[1] = fabsf(src
->f
[1]);
95 dst
->f
[2] = fabsf(src
->f
[2]);
96 dst
->f
[3] = fabsf(src
->f
[3]);
100 micro_arl(union tgsi_exec_channel
*dst
,
101 const union tgsi_exec_channel
*src
)
103 dst
->i
[0] = (int)floorf(src
->f
[0]);
104 dst
->i
[1] = (int)floorf(src
->f
[1]);
105 dst
->i
[2] = (int)floorf(src
->f
[2]);
106 dst
->i
[3] = (int)floorf(src
->f
[3]);
110 micro_arr(union tgsi_exec_channel
*dst
,
111 const union tgsi_exec_channel
*src
)
113 dst
->i
[0] = (int)floorf(src
->f
[0] + 0.5f
);
114 dst
->i
[1] = (int)floorf(src
->f
[1] + 0.5f
);
115 dst
->i
[2] = (int)floorf(src
->f
[2] + 0.5f
);
116 dst
->i
[3] = (int)floorf(src
->f
[3] + 0.5f
);
120 micro_ceil(union tgsi_exec_channel
*dst
,
121 const union tgsi_exec_channel
*src
)
123 dst
->f
[0] = ceilf(src
->f
[0]);
124 dst
->f
[1] = ceilf(src
->f
[1]);
125 dst
->f
[2] = ceilf(src
->f
[2]);
126 dst
->f
[3] = ceilf(src
->f
[3]);
130 micro_cmp(union tgsi_exec_channel
*dst
,
131 const union tgsi_exec_channel
*src0
,
132 const union tgsi_exec_channel
*src1
,
133 const union tgsi_exec_channel
*src2
)
135 dst
->f
[0] = src0
->f
[0] < 0.0f
? src1
->f
[0] : src2
->f
[0];
136 dst
->f
[1] = src0
->f
[1] < 0.0f
? src1
->f
[1] : src2
->f
[1];
137 dst
->f
[2] = src0
->f
[2] < 0.0f
? src1
->f
[2] : src2
->f
[2];
138 dst
->f
[3] = src0
->f
[3] < 0.0f
? src1
->f
[3] : src2
->f
[3];
142 micro_cos(union tgsi_exec_channel
*dst
,
143 const union tgsi_exec_channel
*src
)
145 dst
->f
[0] = cosf(src
->f
[0]);
146 dst
->f
[1] = cosf(src
->f
[1]);
147 dst
->f
[2] = cosf(src
->f
[2]);
148 dst
->f
[3] = cosf(src
->f
[3]);
152 micro_d2f(union tgsi_exec_channel
*dst
,
153 const union tgsi_double_channel
*src
)
155 dst
->f
[0] = (float)src
->d
[0];
156 dst
->f
[1] = (float)src
->d
[1];
157 dst
->f
[2] = (float)src
->d
[2];
158 dst
->f
[3] = (float)src
->d
[3];
162 micro_d2i(union tgsi_exec_channel
*dst
,
163 const union tgsi_double_channel
*src
)
165 dst
->i
[0] = (int)src
->d
[0];
166 dst
->i
[1] = (int)src
->d
[1];
167 dst
->i
[2] = (int)src
->d
[2];
168 dst
->i
[3] = (int)src
->d
[3];
172 micro_d2u(union tgsi_exec_channel
*dst
,
173 const union tgsi_double_channel
*src
)
175 dst
->u
[0] = (unsigned)src
->d
[0];
176 dst
->u
[1] = (unsigned)src
->d
[1];
177 dst
->u
[2] = (unsigned)src
->d
[2];
178 dst
->u
[3] = (unsigned)src
->d
[3];
181 micro_dabs(union tgsi_double_channel
*dst
,
182 const union tgsi_double_channel
*src
)
184 dst
->d
[0] = src
->d
[0] >= 0.0 ? src
->d
[0] : -src
->d
[0];
185 dst
->d
[1] = src
->d
[1] >= 0.0 ? src
->d
[1] : -src
->d
[1];
186 dst
->d
[2] = src
->d
[2] >= 0.0 ? src
->d
[2] : -src
->d
[2];
187 dst
->d
[3] = src
->d
[3] >= 0.0 ? src
->d
[3] : -src
->d
[3];
191 micro_dadd(union tgsi_double_channel
*dst
,
192 const union tgsi_double_channel
*src
)
194 dst
->d
[0] = src
[0].d
[0] + src
[1].d
[0];
195 dst
->d
[1] = src
[0].d
[1] + src
[1].d
[1];
196 dst
->d
[2] = src
[0].d
[2] + src
[1].d
[2];
197 dst
->d
[3] = src
[0].d
[3] + src
[1].d
[3];
201 micro_ddiv(union tgsi_double_channel
*dst
,
202 const union tgsi_double_channel
*src
)
204 dst
->d
[0] = src
[0].d
[0] / src
[1].d
[0];
205 dst
->d
[1] = src
[0].d
[1] / src
[1].d
[1];
206 dst
->d
[2] = src
[0].d
[2] / src
[1].d
[2];
207 dst
->d
[3] = src
[0].d
[3] / src
[1].d
[3];
211 micro_ddx(union tgsi_exec_channel
*dst
,
212 const union tgsi_exec_channel
*src
)
217 dst
->f
[3] = src
->f
[TILE_BOTTOM_RIGHT
] - src
->f
[TILE_BOTTOM_LEFT
];
221 micro_ddy(union tgsi_exec_channel
*dst
,
222 const union tgsi_exec_channel
*src
)
227 dst
->f
[3] = src
->f
[TILE_BOTTOM_LEFT
] - src
->f
[TILE_TOP_LEFT
];
231 micro_dmul(union tgsi_double_channel
*dst
,
232 const union tgsi_double_channel
*src
)
234 dst
->d
[0] = src
[0].d
[0] * src
[1].d
[0];
235 dst
->d
[1] = src
[0].d
[1] * src
[1].d
[1];
236 dst
->d
[2] = src
[0].d
[2] * src
[1].d
[2];
237 dst
->d
[3] = src
[0].d
[3] * src
[1].d
[3];
241 micro_dmax(union tgsi_double_channel
*dst
,
242 const union tgsi_double_channel
*src
)
244 dst
->d
[0] = src
[0].d
[0] > src
[1].d
[0] ? src
[0].d
[0] : src
[1].d
[0];
245 dst
->d
[1] = src
[0].d
[1] > src
[1].d
[1] ? src
[0].d
[1] : src
[1].d
[1];
246 dst
->d
[2] = src
[0].d
[2] > src
[1].d
[2] ? src
[0].d
[2] : src
[1].d
[2];
247 dst
->d
[3] = src
[0].d
[3] > src
[1].d
[3] ? src
[0].d
[3] : src
[1].d
[3];
251 micro_dmin(union tgsi_double_channel
*dst
,
252 const union tgsi_double_channel
*src
)
254 dst
->d
[0] = src
[0].d
[0] < src
[1].d
[0] ? src
[0].d
[0] : src
[1].d
[0];
255 dst
->d
[1] = src
[0].d
[1] < src
[1].d
[1] ? src
[0].d
[1] : src
[1].d
[1];
256 dst
->d
[2] = src
[0].d
[2] < src
[1].d
[2] ? src
[0].d
[2] : src
[1].d
[2];
257 dst
->d
[3] = src
[0].d
[3] < src
[1].d
[3] ? src
[0].d
[3] : src
[1].d
[3];
261 micro_dneg(union tgsi_double_channel
*dst
,
262 const union tgsi_double_channel
*src
)
264 dst
->d
[0] = -src
->d
[0];
265 dst
->d
[1] = -src
->d
[1];
266 dst
->d
[2] = -src
->d
[2];
267 dst
->d
[3] = -src
->d
[3];
271 micro_dslt(union tgsi_double_channel
*dst
,
272 const union tgsi_double_channel
*src
)
274 dst
->u
[0][0] = src
[0].d
[0] < src
[1].d
[0] ? ~0U : 0U;
275 dst
->u
[1][0] = src
[0].d
[1] < src
[1].d
[1] ? ~0U : 0U;
276 dst
->u
[2][0] = src
[0].d
[2] < src
[1].d
[2] ? ~0U : 0U;
277 dst
->u
[3][0] = src
[0].d
[3] < src
[1].d
[3] ? ~0U : 0U;
281 micro_dsne(union tgsi_double_channel
*dst
,
282 const union tgsi_double_channel
*src
)
284 dst
->u
[0][0] = src
[0].d
[0] != src
[1].d
[0] ? ~0U : 0U;
285 dst
->u
[1][0] = src
[0].d
[1] != src
[1].d
[1] ? ~0U : 0U;
286 dst
->u
[2][0] = src
[0].d
[2] != src
[1].d
[2] ? ~0U : 0U;
287 dst
->u
[3][0] = src
[0].d
[3] != src
[1].d
[3] ? ~0U : 0U;
291 micro_dsge(union tgsi_double_channel
*dst
,
292 const union tgsi_double_channel
*src
)
294 dst
->u
[0][0] = src
[0].d
[0] >= src
[1].d
[0] ? ~0U : 0U;
295 dst
->u
[1][0] = src
[0].d
[1] >= src
[1].d
[1] ? ~0U : 0U;
296 dst
->u
[2][0] = src
[0].d
[2] >= src
[1].d
[2] ? ~0U : 0U;
297 dst
->u
[3][0] = src
[0].d
[3] >= src
[1].d
[3] ? ~0U : 0U;
301 micro_dseq(union tgsi_double_channel
*dst
,
302 const union tgsi_double_channel
*src
)
304 dst
->u
[0][0] = src
[0].d
[0] == src
[1].d
[0] ? ~0U : 0U;
305 dst
->u
[1][0] = src
[0].d
[1] == src
[1].d
[1] ? ~0U : 0U;
306 dst
->u
[2][0] = src
[0].d
[2] == src
[1].d
[2] ? ~0U : 0U;
307 dst
->u
[3][0] = src
[0].d
[3] == src
[1].d
[3] ? ~0U : 0U;
311 micro_drcp(union tgsi_double_channel
*dst
,
312 const union tgsi_double_channel
*src
)
314 dst
->d
[0] = 1.0 / src
->d
[0];
315 dst
->d
[1] = 1.0 / src
->d
[1];
316 dst
->d
[2] = 1.0 / src
->d
[2];
317 dst
->d
[3] = 1.0 / src
->d
[3];
321 micro_dsqrt(union tgsi_double_channel
*dst
,
322 const union tgsi_double_channel
*src
)
324 dst
->d
[0] = sqrt(src
->d
[0]);
325 dst
->d
[1] = sqrt(src
->d
[1]);
326 dst
->d
[2] = sqrt(src
->d
[2]);
327 dst
->d
[3] = sqrt(src
->d
[3]);
331 micro_drsq(union tgsi_double_channel
*dst
,
332 const union tgsi_double_channel
*src
)
334 dst
->d
[0] = 1.0 / sqrt(src
->d
[0]);
335 dst
->d
[1] = 1.0 / sqrt(src
->d
[1]);
336 dst
->d
[2] = 1.0 / sqrt(src
->d
[2]);
337 dst
->d
[3] = 1.0 / sqrt(src
->d
[3]);
341 micro_dmad(union tgsi_double_channel
*dst
,
342 const union tgsi_double_channel
*src
)
344 dst
->d
[0] = src
[0].d
[0] * src
[1].d
[0] + src
[2].d
[0];
345 dst
->d
[1] = src
[0].d
[1] * src
[1].d
[1] + src
[2].d
[1];
346 dst
->d
[2] = src
[0].d
[2] * src
[1].d
[2] + src
[2].d
[2];
347 dst
->d
[3] = src
[0].d
[3] * src
[1].d
[3] + src
[2].d
[3];
351 micro_dfrac(union tgsi_double_channel
*dst
,
352 const union tgsi_double_channel
*src
)
354 dst
->d
[0] = src
->d
[0] - floor(src
->d
[0]);
355 dst
->d
[1] = src
->d
[1] - floor(src
->d
[1]);
356 dst
->d
[2] = src
->d
[2] - floor(src
->d
[2]);
357 dst
->d
[3] = src
->d
[3] - floor(src
->d
[3]);
361 micro_dldexp(union tgsi_double_channel
*dst
,
362 const union tgsi_double_channel
*src0
,
363 union tgsi_exec_channel
*src1
)
365 dst
->d
[0] = ldexp(src0
->d
[0], src1
->i
[0]);
366 dst
->d
[1] = ldexp(src0
->d
[1], src1
->i
[1]);
367 dst
->d
[2] = ldexp(src0
->d
[2], src1
->i
[2]);
368 dst
->d
[3] = ldexp(src0
->d
[3], src1
->i
[3]);
372 micro_dfracexp(union tgsi_double_channel
*dst
,
373 union tgsi_exec_channel
*dst_exp
,
374 const union tgsi_double_channel
*src
)
376 dst
->d
[0] = frexp(src
->d
[0], &dst_exp
->i
[0]);
377 dst
->d
[1] = frexp(src
->d
[1], &dst_exp
->i
[1]);
378 dst
->d
[2] = frexp(src
->d
[2], &dst_exp
->i
[2]);
379 dst
->d
[3] = frexp(src
->d
[3], &dst_exp
->i
[3]);
383 micro_exp2(union tgsi_exec_channel
*dst
,
384 const union tgsi_exec_channel
*src
)
387 dst
->f
[0] = util_fast_exp2(src
->f
[0]);
388 dst
->f
[1] = util_fast_exp2(src
->f
[1]);
389 dst
->f
[2] = util_fast_exp2(src
->f
[2]);
390 dst
->f
[3] = util_fast_exp2(src
->f
[3]);
393 /* Inf is okay for this instruction, so clamp it to silence assertions. */
395 union tgsi_exec_channel clamped
;
397 for (i
= 0; i
< 4; i
++) {
398 if (src
->f
[i
] > 127.99999f
) {
399 clamped
.f
[i
] = 127.99999f
;
400 } else if (src
->f
[i
] < -126.99999f
) {
401 clamped
.f
[i
] = -126.99999f
;
403 clamped
.f
[i
] = src
->f
[i
];
409 dst
->f
[0] = powf(2.0f
, src
->f
[0]);
410 dst
->f
[1] = powf(2.0f
, src
->f
[1]);
411 dst
->f
[2] = powf(2.0f
, src
->f
[2]);
412 dst
->f
[3] = powf(2.0f
, src
->f
[3]);
413 #endif /* FAST_MATH */
417 micro_f2d(union tgsi_double_channel
*dst
,
418 const union tgsi_exec_channel
*src
)
420 dst
->d
[0] = (double)src
->f
[0];
421 dst
->d
[1] = (double)src
->f
[1];
422 dst
->d
[2] = (double)src
->f
[2];
423 dst
->d
[3] = (double)src
->f
[3];
427 micro_flr(union tgsi_exec_channel
*dst
,
428 const union tgsi_exec_channel
*src
)
430 dst
->f
[0] = floorf(src
->f
[0]);
431 dst
->f
[1] = floorf(src
->f
[1]);
432 dst
->f
[2] = floorf(src
->f
[2]);
433 dst
->f
[3] = floorf(src
->f
[3]);
437 micro_frc(union tgsi_exec_channel
*dst
,
438 const union tgsi_exec_channel
*src
)
440 dst
->f
[0] = src
->f
[0] - floorf(src
->f
[0]);
441 dst
->f
[1] = src
->f
[1] - floorf(src
->f
[1]);
442 dst
->f
[2] = src
->f
[2] - floorf(src
->f
[2]);
443 dst
->f
[3] = src
->f
[3] - floorf(src
->f
[3]);
447 micro_i2d(union tgsi_double_channel
*dst
,
448 const union tgsi_exec_channel
*src
)
450 dst
->d
[0] = (double)src
->i
[0];
451 dst
->d
[1] = (double)src
->i
[1];
452 dst
->d
[2] = (double)src
->i
[2];
453 dst
->d
[3] = (double)src
->i
[3];
457 micro_iabs(union tgsi_exec_channel
*dst
,
458 const union tgsi_exec_channel
*src
)
460 dst
->i
[0] = src
->i
[0] >= 0 ? src
->i
[0] : -src
->i
[0];
461 dst
->i
[1] = src
->i
[1] >= 0 ? src
->i
[1] : -src
->i
[1];
462 dst
->i
[2] = src
->i
[2] >= 0 ? src
->i
[2] : -src
->i
[2];
463 dst
->i
[3] = src
->i
[3] >= 0 ? src
->i
[3] : -src
->i
[3];
467 micro_ineg(union tgsi_exec_channel
*dst
,
468 const union tgsi_exec_channel
*src
)
470 dst
->i
[0] = -src
->i
[0];
471 dst
->i
[1] = -src
->i
[1];
472 dst
->i
[2] = -src
->i
[2];
473 dst
->i
[3] = -src
->i
[3];
477 micro_lg2(union tgsi_exec_channel
*dst
,
478 const union tgsi_exec_channel
*src
)
481 dst
->f
[0] = util_fast_log2(src
->f
[0]);
482 dst
->f
[1] = util_fast_log2(src
->f
[1]);
483 dst
->f
[2] = util_fast_log2(src
->f
[2]);
484 dst
->f
[3] = util_fast_log2(src
->f
[3]);
486 dst
->f
[0] = logf(src
->f
[0]) * 1.442695f
;
487 dst
->f
[1] = logf(src
->f
[1]) * 1.442695f
;
488 dst
->f
[2] = logf(src
->f
[2]) * 1.442695f
;
489 dst
->f
[3] = logf(src
->f
[3]) * 1.442695f
;
494 micro_lrp(union tgsi_exec_channel
*dst
,
495 const union tgsi_exec_channel
*src0
,
496 const union tgsi_exec_channel
*src1
,
497 const union tgsi_exec_channel
*src2
)
499 dst
->f
[0] = src0
->f
[0] * (src1
->f
[0] - src2
->f
[0]) + src2
->f
[0];
500 dst
->f
[1] = src0
->f
[1] * (src1
->f
[1] - src2
->f
[1]) + src2
->f
[1];
501 dst
->f
[2] = src0
->f
[2] * (src1
->f
[2] - src2
->f
[2]) + src2
->f
[2];
502 dst
->f
[3] = src0
->f
[3] * (src1
->f
[3] - src2
->f
[3]) + src2
->f
[3];
506 micro_mad(union tgsi_exec_channel
*dst
,
507 const union tgsi_exec_channel
*src0
,
508 const union tgsi_exec_channel
*src1
,
509 const union tgsi_exec_channel
*src2
)
511 dst
->f
[0] = src0
->f
[0] * src1
->f
[0] + src2
->f
[0];
512 dst
->f
[1] = src0
->f
[1] * src1
->f
[1] + src2
->f
[1];
513 dst
->f
[2] = src0
->f
[2] * src1
->f
[2] + src2
->f
[2];
514 dst
->f
[3] = src0
->f
[3] * src1
->f
[3] + src2
->f
[3];
518 micro_mov(union tgsi_exec_channel
*dst
,
519 const union tgsi_exec_channel
*src
)
521 dst
->u
[0] = src
->u
[0];
522 dst
->u
[1] = src
->u
[1];
523 dst
->u
[2] = src
->u
[2];
524 dst
->u
[3] = src
->u
[3];
528 micro_rcp(union tgsi_exec_channel
*dst
,
529 const union tgsi_exec_channel
*src
)
531 #if 0 /* for debugging */
532 assert(src
->f
[0] != 0.0f
);
533 assert(src
->f
[1] != 0.0f
);
534 assert(src
->f
[2] != 0.0f
);
535 assert(src
->f
[3] != 0.0f
);
537 dst
->f
[0] = 1.0f
/ src
->f
[0];
538 dst
->f
[1] = 1.0f
/ src
->f
[1];
539 dst
->f
[2] = 1.0f
/ src
->f
[2];
540 dst
->f
[3] = 1.0f
/ src
->f
[3];
544 micro_rnd(union tgsi_exec_channel
*dst
,
545 const union tgsi_exec_channel
*src
)
547 dst
->f
[0] = _mesa_roundevenf(src
->f
[0]);
548 dst
->f
[1] = _mesa_roundevenf(src
->f
[1]);
549 dst
->f
[2] = _mesa_roundevenf(src
->f
[2]);
550 dst
->f
[3] = _mesa_roundevenf(src
->f
[3]);
554 micro_rsq(union tgsi_exec_channel
*dst
,
555 const union tgsi_exec_channel
*src
)
557 #if 0 /* for debugging */
558 assert(src
->f
[0] != 0.0f
);
559 assert(src
->f
[1] != 0.0f
);
560 assert(src
->f
[2] != 0.0f
);
561 assert(src
->f
[3] != 0.0f
);
563 dst
->f
[0] = 1.0f
/ sqrtf(src
->f
[0]);
564 dst
->f
[1] = 1.0f
/ sqrtf(src
->f
[1]);
565 dst
->f
[2] = 1.0f
/ sqrtf(src
->f
[2]);
566 dst
->f
[3] = 1.0f
/ sqrtf(src
->f
[3]);
570 micro_sqrt(union tgsi_exec_channel
*dst
,
571 const union tgsi_exec_channel
*src
)
573 dst
->f
[0] = sqrtf(src
->f
[0]);
574 dst
->f
[1] = sqrtf(src
->f
[1]);
575 dst
->f
[2] = sqrtf(src
->f
[2]);
576 dst
->f
[3] = sqrtf(src
->f
[3]);
580 micro_seq(union tgsi_exec_channel
*dst
,
581 const union tgsi_exec_channel
*src0
,
582 const union tgsi_exec_channel
*src1
)
584 dst
->f
[0] = src0
->f
[0] == src1
->f
[0] ? 1.0f
: 0.0f
;
585 dst
->f
[1] = src0
->f
[1] == src1
->f
[1] ? 1.0f
: 0.0f
;
586 dst
->f
[2] = src0
->f
[2] == src1
->f
[2] ? 1.0f
: 0.0f
;
587 dst
->f
[3] = src0
->f
[3] == src1
->f
[3] ? 1.0f
: 0.0f
;
591 micro_sge(union tgsi_exec_channel
*dst
,
592 const union tgsi_exec_channel
*src0
,
593 const union tgsi_exec_channel
*src1
)
595 dst
->f
[0] = src0
->f
[0] >= src1
->f
[0] ? 1.0f
: 0.0f
;
596 dst
->f
[1] = src0
->f
[1] >= src1
->f
[1] ? 1.0f
: 0.0f
;
597 dst
->f
[2] = src0
->f
[2] >= src1
->f
[2] ? 1.0f
: 0.0f
;
598 dst
->f
[3] = src0
->f
[3] >= src1
->f
[3] ? 1.0f
: 0.0f
;
602 micro_sgn(union tgsi_exec_channel
*dst
,
603 const union tgsi_exec_channel
*src
)
605 dst
->f
[0] = src
->f
[0] < 0.0f
? -1.0f
: src
->f
[0] > 0.0f
? 1.0f
: 0.0f
;
606 dst
->f
[1] = src
->f
[1] < 0.0f
? -1.0f
: src
->f
[1] > 0.0f
? 1.0f
: 0.0f
;
607 dst
->f
[2] = src
->f
[2] < 0.0f
? -1.0f
: src
->f
[2] > 0.0f
? 1.0f
: 0.0f
;
608 dst
->f
[3] = src
->f
[3] < 0.0f
? -1.0f
: src
->f
[3] > 0.0f
? 1.0f
: 0.0f
;
612 micro_isgn(union tgsi_exec_channel
*dst
,
613 const union tgsi_exec_channel
*src
)
615 dst
->i
[0] = src
->i
[0] < 0 ? -1 : src
->i
[0] > 0 ? 1 : 0;
616 dst
->i
[1] = src
->i
[1] < 0 ? -1 : src
->i
[1] > 0 ? 1 : 0;
617 dst
->i
[2] = src
->i
[2] < 0 ? -1 : src
->i
[2] > 0 ? 1 : 0;
618 dst
->i
[3] = src
->i
[3] < 0 ? -1 : src
->i
[3] > 0 ? 1 : 0;
622 micro_sgt(union tgsi_exec_channel
*dst
,
623 const union tgsi_exec_channel
*src0
,
624 const union tgsi_exec_channel
*src1
)
626 dst
->f
[0] = src0
->f
[0] > src1
->f
[0] ? 1.0f
: 0.0f
;
627 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? 1.0f
: 0.0f
;
628 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? 1.0f
: 0.0f
;
629 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? 1.0f
: 0.0f
;
633 micro_sin(union tgsi_exec_channel
*dst
,
634 const union tgsi_exec_channel
*src
)
636 dst
->f
[0] = sinf(src
->f
[0]);
637 dst
->f
[1] = sinf(src
->f
[1]);
638 dst
->f
[2] = sinf(src
->f
[2]);
639 dst
->f
[3] = sinf(src
->f
[3]);
643 micro_sle(union tgsi_exec_channel
*dst
,
644 const union tgsi_exec_channel
*src0
,
645 const union tgsi_exec_channel
*src1
)
647 dst
->f
[0] = src0
->f
[0] <= src1
->f
[0] ? 1.0f
: 0.0f
;
648 dst
->f
[1] = src0
->f
[1] <= src1
->f
[1] ? 1.0f
: 0.0f
;
649 dst
->f
[2] = src0
->f
[2] <= src1
->f
[2] ? 1.0f
: 0.0f
;
650 dst
->f
[3] = src0
->f
[3] <= src1
->f
[3] ? 1.0f
: 0.0f
;
654 micro_slt(union tgsi_exec_channel
*dst
,
655 const union tgsi_exec_channel
*src0
,
656 const union tgsi_exec_channel
*src1
)
658 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? 1.0f
: 0.0f
;
659 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? 1.0f
: 0.0f
;
660 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? 1.0f
: 0.0f
;
661 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? 1.0f
: 0.0f
;
665 micro_sne(union tgsi_exec_channel
*dst
,
666 const union tgsi_exec_channel
*src0
,
667 const union tgsi_exec_channel
*src1
)
669 dst
->f
[0] = src0
->f
[0] != src1
->f
[0] ? 1.0f
: 0.0f
;
670 dst
->f
[1] = src0
->f
[1] != src1
->f
[1] ? 1.0f
: 0.0f
;
671 dst
->f
[2] = src0
->f
[2] != src1
->f
[2] ? 1.0f
: 0.0f
;
672 dst
->f
[3] = src0
->f
[3] != src1
->f
[3] ? 1.0f
: 0.0f
;
676 micro_trunc(union tgsi_exec_channel
*dst
,
677 const union tgsi_exec_channel
*src
)
679 dst
->f
[0] = truncf(src
->f
[0]);
680 dst
->f
[1] = truncf(src
->f
[1]);
681 dst
->f
[2] = truncf(src
->f
[2]);
682 dst
->f
[3] = truncf(src
->f
[3]);
686 micro_u2d(union tgsi_double_channel
*dst
,
687 const union tgsi_exec_channel
*src
)
689 dst
->d
[0] = (double)src
->u
[0];
690 dst
->d
[1] = (double)src
->u
[1];
691 dst
->d
[2] = (double)src
->u
[2];
692 dst
->d
[3] = (double)src
->u
[3];
696 micro_i64abs(union tgsi_double_channel
*dst
,
697 const union tgsi_double_channel
*src
)
699 dst
->i64
[0] = src
->i64
[0] >= 0.0 ? src
->i64
[0] : -src
->i64
[0];
700 dst
->i64
[1] = src
->i64
[1] >= 0.0 ? src
->i64
[1] : -src
->i64
[1];
701 dst
->i64
[2] = src
->i64
[2] >= 0.0 ? src
->i64
[2] : -src
->i64
[2];
702 dst
->i64
[3] = src
->i64
[3] >= 0.0 ? src
->i64
[3] : -src
->i64
[3];
706 micro_i64sgn(union tgsi_double_channel
*dst
,
707 const union tgsi_double_channel
*src
)
709 dst
->i64
[0] = src
->i64
[0] < 0 ? -1 : src
->i64
[0] > 0 ? 1 : 0;
710 dst
->i64
[1] = src
->i64
[1] < 0 ? -1 : src
->i64
[1] > 0 ? 1 : 0;
711 dst
->i64
[2] = src
->i64
[2] < 0 ? -1 : src
->i64
[2] > 0 ? 1 : 0;
712 dst
->i64
[3] = src
->i64
[3] < 0 ? -1 : src
->i64
[3] > 0 ? 1 : 0;
716 micro_i64neg(union tgsi_double_channel
*dst
,
717 const union tgsi_double_channel
*src
)
719 dst
->i64
[0] = -src
->i64
[0];
720 dst
->i64
[1] = -src
->i64
[1];
721 dst
->i64
[2] = -src
->i64
[2];
722 dst
->i64
[3] = -src
->i64
[3];
726 micro_u64seq(union tgsi_double_channel
*dst
,
727 const union tgsi_double_channel
*src
)
729 dst
->u
[0][0] = src
[0].u64
[0] == src
[1].u64
[0] ? ~0U : 0U;
730 dst
->u
[1][0] = src
[0].u64
[1] == src
[1].u64
[1] ? ~0U : 0U;
731 dst
->u
[2][0] = src
[0].u64
[2] == src
[1].u64
[2] ? ~0U : 0U;
732 dst
->u
[3][0] = src
[0].u64
[3] == src
[1].u64
[3] ? ~0U : 0U;
736 micro_u64sne(union tgsi_double_channel
*dst
,
737 const union tgsi_double_channel
*src
)
739 dst
->u
[0][0] = src
[0].u64
[0] != src
[1].u64
[0] ? ~0U : 0U;
740 dst
->u
[1][0] = src
[0].u64
[1] != src
[1].u64
[1] ? ~0U : 0U;
741 dst
->u
[2][0] = src
[0].u64
[2] != src
[1].u64
[2] ? ~0U : 0U;
742 dst
->u
[3][0] = src
[0].u64
[3] != src
[1].u64
[3] ? ~0U : 0U;
746 micro_i64slt(union tgsi_double_channel
*dst
,
747 const union tgsi_double_channel
*src
)
749 dst
->u
[0][0] = src
[0].i64
[0] < src
[1].i64
[0] ? ~0U : 0U;
750 dst
->u
[1][0] = src
[0].i64
[1] < src
[1].i64
[1] ? ~0U : 0U;
751 dst
->u
[2][0] = src
[0].i64
[2] < src
[1].i64
[2] ? ~0U : 0U;
752 dst
->u
[3][0] = src
[0].i64
[3] < src
[1].i64
[3] ? ~0U : 0U;
756 micro_u64slt(union tgsi_double_channel
*dst
,
757 const union tgsi_double_channel
*src
)
759 dst
->u
[0][0] = src
[0].u64
[0] < src
[1].u64
[0] ? ~0U : 0U;
760 dst
->u
[1][0] = src
[0].u64
[1] < src
[1].u64
[1] ? ~0U : 0U;
761 dst
->u
[2][0] = src
[0].u64
[2] < src
[1].u64
[2] ? ~0U : 0U;
762 dst
->u
[3][0] = src
[0].u64
[3] < src
[1].u64
[3] ? ~0U : 0U;
766 micro_i64sge(union tgsi_double_channel
*dst
,
767 const union tgsi_double_channel
*src
)
769 dst
->u
[0][0] = src
[0].i64
[0] >= src
[1].i64
[0] ? ~0U : 0U;
770 dst
->u
[1][0] = src
[0].i64
[1] >= src
[1].i64
[1] ? ~0U : 0U;
771 dst
->u
[2][0] = src
[0].i64
[2] >= src
[1].i64
[2] ? ~0U : 0U;
772 dst
->u
[3][0] = src
[0].i64
[3] >= src
[1].i64
[3] ? ~0U : 0U;
776 micro_u64sge(union tgsi_double_channel
*dst
,
777 const union tgsi_double_channel
*src
)
779 dst
->u
[0][0] = src
[0].u64
[0] >= src
[1].u64
[0] ? ~0U : 0U;
780 dst
->u
[1][0] = src
[0].u64
[1] >= src
[1].u64
[1] ? ~0U : 0U;
781 dst
->u
[2][0] = src
[0].u64
[2] >= src
[1].u64
[2] ? ~0U : 0U;
782 dst
->u
[3][0] = src
[0].u64
[3] >= src
[1].u64
[3] ? ~0U : 0U;
786 micro_u64max(union tgsi_double_channel
*dst
,
787 const union tgsi_double_channel
*src
)
789 dst
->u64
[0] = src
[0].u64
[0] > src
[1].u64
[0] ? src
[0].u64
[0] : src
[1].u64
[0];
790 dst
->u64
[1] = src
[0].u64
[1] > src
[1].u64
[1] ? src
[0].u64
[1] : src
[1].u64
[1];
791 dst
->u64
[2] = src
[0].u64
[2] > src
[1].u64
[2] ? src
[0].u64
[2] : src
[1].u64
[2];
792 dst
->u64
[3] = src
[0].u64
[3] > src
[1].u64
[3] ? src
[0].u64
[3] : src
[1].u64
[3];
796 micro_i64max(union tgsi_double_channel
*dst
,
797 const union tgsi_double_channel
*src
)
799 dst
->i64
[0] = src
[0].i64
[0] > src
[1].i64
[0] ? src
[0].i64
[0] : src
[1].i64
[0];
800 dst
->i64
[1] = src
[0].i64
[1] > src
[1].i64
[1] ? src
[0].i64
[1] : src
[1].i64
[1];
801 dst
->i64
[2] = src
[0].i64
[2] > src
[1].i64
[2] ? src
[0].i64
[2] : src
[1].i64
[2];
802 dst
->i64
[3] = src
[0].i64
[3] > src
[1].i64
[3] ? src
[0].i64
[3] : src
[1].i64
[3];
806 micro_u64min(union tgsi_double_channel
*dst
,
807 const union tgsi_double_channel
*src
)
809 dst
->u64
[0] = src
[0].u64
[0] < src
[1].u64
[0] ? src
[0].u64
[0] : src
[1].u64
[0];
810 dst
->u64
[1] = src
[0].u64
[1] < src
[1].u64
[1] ? src
[0].u64
[1] : src
[1].u64
[1];
811 dst
->u64
[2] = src
[0].u64
[2] < src
[1].u64
[2] ? src
[0].u64
[2] : src
[1].u64
[2];
812 dst
->u64
[3] = src
[0].u64
[3] < src
[1].u64
[3] ? src
[0].u64
[3] : src
[1].u64
[3];
816 micro_i64min(union tgsi_double_channel
*dst
,
817 const union tgsi_double_channel
*src
)
819 dst
->i64
[0] = src
[0].i64
[0] < src
[1].i64
[0] ? src
[0].i64
[0] : src
[1].i64
[0];
820 dst
->i64
[1] = src
[0].i64
[1] < src
[1].i64
[1] ? src
[0].i64
[1] : src
[1].i64
[1];
821 dst
->i64
[2] = src
[0].i64
[2] < src
[1].i64
[2] ? src
[0].i64
[2] : src
[1].i64
[2];
822 dst
->i64
[3] = src
[0].i64
[3] < src
[1].i64
[3] ? src
[0].i64
[3] : src
[1].i64
[3];
826 micro_u64add(union tgsi_double_channel
*dst
,
827 const union tgsi_double_channel
*src
)
829 dst
->u64
[0] = src
[0].u64
[0] + src
[1].u64
[0];
830 dst
->u64
[1] = src
[0].u64
[1] + src
[1].u64
[1];
831 dst
->u64
[2] = src
[0].u64
[2] + src
[1].u64
[2];
832 dst
->u64
[3] = src
[0].u64
[3] + src
[1].u64
[3];
836 micro_u64mul(union tgsi_double_channel
*dst
,
837 const union tgsi_double_channel
*src
)
839 dst
->u64
[0] = src
[0].u64
[0] * src
[1].u64
[0];
840 dst
->u64
[1] = src
[0].u64
[1] * src
[1].u64
[1];
841 dst
->u64
[2] = src
[0].u64
[2] * src
[1].u64
[2];
842 dst
->u64
[3] = src
[0].u64
[3] * src
[1].u64
[3];
846 micro_u64div(union tgsi_double_channel
*dst
,
847 const union tgsi_double_channel
*src
)
849 dst
->u64
[0] = src
[1].u64
[0] ? src
[0].u64
[0] / src
[1].u64
[0] : ~0ull;
850 dst
->u64
[1] = src
[1].u64
[1] ? src
[0].u64
[1] / src
[1].u64
[1] : ~0ull;
851 dst
->u64
[2] = src
[1].u64
[2] ? src
[0].u64
[2] / src
[1].u64
[2] : ~0ull;
852 dst
->u64
[3] = src
[1].u64
[3] ? src
[0].u64
[3] / src
[1].u64
[3] : ~0ull;
856 micro_i64div(union tgsi_double_channel
*dst
,
857 const union tgsi_double_channel
*src
)
859 dst
->i64
[0] = src
[1].i64
[0] ? src
[0].i64
[0] / src
[1].i64
[0] : 0;
860 dst
->i64
[1] = src
[1].i64
[1] ? src
[0].i64
[1] / src
[1].i64
[1] : 0;
861 dst
->i64
[2] = src
[1].i64
[2] ? src
[0].i64
[2] / src
[1].i64
[2] : 0;
862 dst
->i64
[3] = src
[1].i64
[3] ? src
[0].i64
[3] / src
[1].i64
[3] : 0;
866 micro_u64mod(union tgsi_double_channel
*dst
,
867 const union tgsi_double_channel
*src
)
869 dst
->u64
[0] = src
[1].u64
[0] ? src
[0].u64
[0] % src
[1].u64
[0] : ~0ull;
870 dst
->u64
[1] = src
[1].u64
[1] ? src
[0].u64
[1] % src
[1].u64
[1] : ~0ull;
871 dst
->u64
[2] = src
[1].u64
[2] ? src
[0].u64
[2] % src
[1].u64
[2] : ~0ull;
872 dst
->u64
[3] = src
[1].u64
[3] ? src
[0].u64
[3] % src
[1].u64
[3] : ~0ull;
876 micro_i64mod(union tgsi_double_channel
*dst
,
877 const union tgsi_double_channel
*src
)
879 dst
->i64
[0] = src
[1].i64
[0] ? src
[0].i64
[0] % src
[1].i64
[0] : ~0ll;
880 dst
->i64
[1] = src
[1].i64
[1] ? src
[0].i64
[1] % src
[1].i64
[1] : ~0ll;
881 dst
->i64
[2] = src
[1].i64
[2] ? src
[0].i64
[2] % src
[1].i64
[2] : ~0ll;
882 dst
->i64
[3] = src
[1].i64
[3] ? src
[0].i64
[3] % src
[1].i64
[3] : ~0ll;
886 micro_u64shl(union tgsi_double_channel
*dst
,
887 const union tgsi_double_channel
*src0
,
888 union tgsi_exec_channel
*src1
)
890 unsigned masked_count
;
891 masked_count
= src1
->u
[0] & 0x3f;
892 dst
->u64
[0] = src0
->u64
[0] << masked_count
;
893 masked_count
= src1
->u
[1] & 0x3f;
894 dst
->u64
[1] = src0
->u64
[1] << masked_count
;
895 masked_count
= src1
->u
[2] & 0x3f;
896 dst
->u64
[2] = src0
->u64
[2] << masked_count
;
897 masked_count
= src1
->u
[3] & 0x3f;
898 dst
->u64
[3] = src0
->u64
[3] << masked_count
;
902 micro_i64shr(union tgsi_double_channel
*dst
,
903 const union tgsi_double_channel
*src0
,
904 union tgsi_exec_channel
*src1
)
906 unsigned masked_count
;
907 masked_count
= src1
->u
[0] & 0x3f;
908 dst
->i64
[0] = src0
->i64
[0] >> masked_count
;
909 masked_count
= src1
->u
[1] & 0x3f;
910 dst
->i64
[1] = src0
->i64
[1] >> masked_count
;
911 masked_count
= src1
->u
[2] & 0x3f;
912 dst
->i64
[2] = src0
->i64
[2] >> masked_count
;
913 masked_count
= src1
->u
[3] & 0x3f;
914 dst
->i64
[3] = src0
->i64
[3] >> masked_count
;
918 micro_u64shr(union tgsi_double_channel
*dst
,
919 const union tgsi_double_channel
*src0
,
920 union tgsi_exec_channel
*src1
)
922 unsigned masked_count
;
923 masked_count
= src1
->u
[0] & 0x3f;
924 dst
->u64
[0] = src0
->u64
[0] >> masked_count
;
925 masked_count
= src1
->u
[1] & 0x3f;
926 dst
->u64
[1] = src0
->u64
[1] >> masked_count
;
927 masked_count
= src1
->u
[2] & 0x3f;
928 dst
->u64
[2] = src0
->u64
[2] >> masked_count
;
929 masked_count
= src1
->u
[3] & 0x3f;
930 dst
->u64
[3] = src0
->u64
[3] >> masked_count
;
933 enum tgsi_exec_datatype
{
934 TGSI_EXEC_DATA_FLOAT
,
937 TGSI_EXEC_DATA_DOUBLE
,
938 TGSI_EXEC_DATA_INT64
,
939 TGSI_EXEC_DATA_UINT64
,
943 * Shorthand locations of various utility registers (_I = Index, _C = Channel)
945 #define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I
946 #define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C
947 #define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I
948 #define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C
949 #define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I
950 #define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C
953 /** The execution mask depends on the conditional mask and the loop mask */
954 #define UPDATE_EXEC_MASK(MACH) \
955 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask
958 static const union tgsi_exec_channel ZeroVec
=
959 { { 0.0, 0.0, 0.0, 0.0 } };
961 static const union tgsi_exec_channel OneVec
= {
962 {1.0f
, 1.0f
, 1.0f
, 1.0f
}
965 static const union tgsi_exec_channel P128Vec
= {
966 {128.0f
, 128.0f
, 128.0f
, 128.0f
}
969 static const union tgsi_exec_channel M128Vec
= {
970 {-128.0f
, -128.0f
, -128.0f
, -128.0f
}
975 * Assert that none of the float values in 'chan' are infinite or NaN.
976 * NaN and Inf may occur normally during program execution and should
977 * not lead to crashes, etc. But when debugging, it's helpful to catch
981 check_inf_or_nan(const union tgsi_exec_channel
*chan
)
983 assert(!util_is_inf_or_nan((chan
)->f
[0]));
984 assert(!util_is_inf_or_nan((chan
)->f
[1]));
985 assert(!util_is_inf_or_nan((chan
)->f
[2]));
986 assert(!util_is_inf_or_nan((chan
)->f
[3]));
992 print_chan(const char *msg
, const union tgsi_exec_channel
*chan
)
994 debug_printf("%s = {%f, %f, %f, %f}\n",
995 msg
, chan
->f
[0], chan
->f
[1], chan
->f
[2], chan
->f
[3]);
1002 print_temp(const struct tgsi_exec_machine
*mach
, uint index
)
1004 const struct tgsi_exec_vector
*tmp
= &mach
->Temps
[index
];
1006 debug_printf("Temp[%u] =\n", index
);
1007 for (i
= 0; i
< 4; i
++) {
1008 debug_printf(" %c: { %f, %f, %f, %f }\n",
1020 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine
*mach
,
1023 const unsigned *buf_sizes
)
1027 for (i
= 0; i
< num_bufs
; i
++) {
1028 mach
->Consts
[i
] = bufs
[i
];
1029 mach
->ConstsSize
[i
] = buf_sizes
[i
];
1035 * Check if there's a potential src/dst register data dependency when
1036 * using SOA execution.
1039 * This would expand into:
1044 * The second instruction will have the wrong value for t0 if executed as-is.
1047 tgsi_check_soa_dependencies(const struct tgsi_full_instruction
*inst
)
1051 uint writemask
= inst
->Dst
[0].Register
.WriteMask
;
1052 if (writemask
== TGSI_WRITEMASK_X
||
1053 writemask
== TGSI_WRITEMASK_Y
||
1054 writemask
== TGSI_WRITEMASK_Z
||
1055 writemask
== TGSI_WRITEMASK_W
||
1056 writemask
== TGSI_WRITEMASK_NONE
) {
1057 /* no chance of data dependency */
1061 /* loop over src regs */
1062 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
1063 if ((inst
->Src
[i
].Register
.File
==
1064 inst
->Dst
[0].Register
.File
) &&
1065 ((inst
->Src
[i
].Register
.Index
==
1066 inst
->Dst
[0].Register
.Index
) ||
1067 inst
->Src
[i
].Register
.Indirect
||
1068 inst
->Dst
[0].Register
.Indirect
)) {
1069 /* loop over dest channels */
1070 uint channelsWritten
= 0x0;
1071 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
1072 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1073 /* check if we're reading a channel that's been written */
1074 uint swizzle
= tgsi_util_get_full_src_register_swizzle(&inst
->Src
[i
], chan
);
1075 if (channelsWritten
& (1 << swizzle
)) {
1079 channelsWritten
|= (1 << chan
);
1089 * Initialize machine state by expanding tokens to full instructions,
1090 * allocating temporary storage, setting up constants, etc.
1091 * After this, we can call tgsi_exec_machine_run() many times.
1094 tgsi_exec_machine_bind_shader(
1095 struct tgsi_exec_machine
*mach
,
1096 const struct tgsi_token
*tokens
,
1097 struct tgsi_sampler
*sampler
,
1098 struct tgsi_image
*image
,
1099 struct tgsi_buffer
*buffer
)
1102 struct tgsi_parse_context parse
;
1103 struct tgsi_full_instruction
*instructions
;
1104 struct tgsi_full_declaration
*declarations
;
1105 uint maxInstructions
= 10, numInstructions
= 0;
1106 uint maxDeclarations
= 10, numDeclarations
= 0;
1109 tgsi_dump(tokens
, 0);
1115 mach
->Tokens
= tokens
;
1116 mach
->Sampler
= sampler
;
1117 mach
->Image
= image
;
1118 mach
->Buffer
= buffer
;
1121 /* unbind and free all */
1122 FREE(mach
->Declarations
);
1123 mach
->Declarations
= NULL
;
1124 mach
->NumDeclarations
= 0;
1126 FREE(mach
->Instructions
);
1127 mach
->Instructions
= NULL
;
1128 mach
->NumInstructions
= 0;
1133 k
= tgsi_parse_init (&parse
, mach
->Tokens
);
1134 if (k
!= TGSI_PARSE_OK
) {
1135 debug_printf( "Problem parsing!\n" );
1140 mach
->NumOutputs
= 0;
1142 for (k
= 0; k
< TGSI_SEMANTIC_COUNT
; k
++)
1143 mach
->SysSemanticToIndex
[k
] = -1;
1145 if (mach
->ShaderType
== PIPE_SHADER_GEOMETRY
&&
1146 !mach
->UsedGeometryShader
) {
1147 struct tgsi_exec_vector
*inputs
;
1148 struct tgsi_exec_vector
*outputs
;
1150 inputs
= align_malloc(sizeof(struct tgsi_exec_vector
) *
1151 TGSI_MAX_PRIM_VERTICES
* PIPE_MAX_SHADER_INPUTS
,
1157 outputs
= align_malloc(sizeof(struct tgsi_exec_vector
) *
1158 TGSI_MAX_TOTAL_VERTICES
, 16);
1165 align_free(mach
->Inputs
);
1166 align_free(mach
->Outputs
);
1168 mach
->Inputs
= inputs
;
1169 mach
->Outputs
= outputs
;
1170 mach
->UsedGeometryShader
= TRUE
;
1173 declarations
= (struct tgsi_full_declaration
*)
1174 MALLOC( maxDeclarations
* sizeof(struct tgsi_full_declaration
) );
1176 if (!declarations
) {
1180 instructions
= (struct tgsi_full_instruction
*)
1181 MALLOC( maxInstructions
* sizeof(struct tgsi_full_instruction
) );
1183 if (!instructions
) {
1184 FREE( declarations
);
1188 while( !tgsi_parse_end_of_tokens( &parse
) ) {
1191 tgsi_parse_token( &parse
);
1192 switch( parse
.FullToken
.Token
.Type
) {
1193 case TGSI_TOKEN_TYPE_DECLARATION
:
1194 /* save expanded declaration */
1195 if (numDeclarations
== maxDeclarations
) {
1196 declarations
= REALLOC(declarations
,
1198 * sizeof(struct tgsi_full_declaration
),
1199 (maxDeclarations
+ 10)
1200 * sizeof(struct tgsi_full_declaration
));
1201 maxDeclarations
+= 10;
1203 if (parse
.FullToken
.FullDeclaration
.Declaration
.File
== TGSI_FILE_OUTPUT
) {
1205 for (reg
= parse
.FullToken
.FullDeclaration
.Range
.First
;
1206 reg
<= parse
.FullToken
.FullDeclaration
.Range
.Last
;
1211 else if (parse
.FullToken
.FullDeclaration
.Declaration
.File
== TGSI_FILE_SYSTEM_VALUE
) {
1212 const struct tgsi_full_declaration
*decl
= &parse
.FullToken
.FullDeclaration
;
1213 mach
->SysSemanticToIndex
[decl
->Semantic
.Name
] = decl
->Range
.First
;
1216 memcpy(declarations
+ numDeclarations
,
1217 &parse
.FullToken
.FullDeclaration
,
1218 sizeof(declarations
[0]));
1222 case TGSI_TOKEN_TYPE_IMMEDIATE
:
1224 uint size
= parse
.FullToken
.FullImmediate
.Immediate
.NrTokens
- 1;
1225 assert( size
<= 4 );
1226 assert( mach
->ImmLimit
+ 1 <= TGSI_EXEC_NUM_IMMEDIATES
);
1228 for( i
= 0; i
< size
; i
++ ) {
1229 mach
->Imms
[mach
->ImmLimit
][i
] =
1230 parse
.FullToken
.FullImmediate
.u
[i
].Float
;
1232 mach
->ImmLimit
+= 1;
1236 case TGSI_TOKEN_TYPE_INSTRUCTION
:
1238 /* save expanded instruction */
1239 if (numInstructions
== maxInstructions
) {
1240 instructions
= REALLOC(instructions
,
1242 * sizeof(struct tgsi_full_instruction
),
1243 (maxInstructions
+ 10)
1244 * sizeof(struct tgsi_full_instruction
));
1245 maxInstructions
+= 10;
1248 memcpy(instructions
+ numInstructions
,
1249 &parse
.FullToken
.FullInstruction
,
1250 sizeof(instructions
[0]));
1255 case TGSI_TOKEN_TYPE_PROPERTY
:
1256 if (mach
->ShaderType
== PIPE_SHADER_GEOMETRY
) {
1257 if (parse
.FullToken
.FullProperty
.Property
.PropertyName
== TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
) {
1258 mach
->MaxOutputVertices
= parse
.FullToken
.FullProperty
.u
[0].Data
;
1267 tgsi_parse_free (&parse
);
1269 FREE(mach
->Declarations
);
1270 mach
->Declarations
= declarations
;
1271 mach
->NumDeclarations
= numDeclarations
;
1273 FREE(mach
->Instructions
);
1274 mach
->Instructions
= instructions
;
1275 mach
->NumInstructions
= numInstructions
;
1279 struct tgsi_exec_machine
*
1280 tgsi_exec_machine_create(enum pipe_shader_type shader_type
)
1282 struct tgsi_exec_machine
*mach
;
1285 mach
= align_malloc( sizeof *mach
, 16 );
1289 memset(mach
, 0, sizeof(*mach
));
1291 mach
->ShaderType
= shader_type
;
1292 mach
->Addrs
= &mach
->Temps
[TGSI_EXEC_TEMP_ADDR
];
1293 mach
->MaxGeometryShaderOutputs
= TGSI_MAX_TOTAL_VERTICES
;
1295 if (shader_type
!= PIPE_SHADER_COMPUTE
) {
1296 mach
->Inputs
= align_malloc(sizeof(struct tgsi_exec_vector
) * PIPE_MAX_SHADER_INPUTS
, 16);
1297 mach
->Outputs
= align_malloc(sizeof(struct tgsi_exec_vector
) * PIPE_MAX_SHADER_OUTPUTS
, 16);
1298 if (!mach
->Inputs
|| !mach
->Outputs
)
1302 /* Setup constants needed by the SSE2 executor. */
1303 for( i
= 0; i
< 4; i
++ ) {
1304 mach
->Temps
[TGSI_EXEC_TEMP_00000000_I
].xyzw
[TGSI_EXEC_TEMP_00000000_C
].u
[i
] = 0x00000000;
1305 mach
->Temps
[TGSI_EXEC_TEMP_7FFFFFFF_I
].xyzw
[TGSI_EXEC_TEMP_7FFFFFFF_C
].u
[i
] = 0x7FFFFFFF;
1306 mach
->Temps
[TGSI_EXEC_TEMP_80000000_I
].xyzw
[TGSI_EXEC_TEMP_80000000_C
].u
[i
] = 0x80000000;
1307 mach
->Temps
[TGSI_EXEC_TEMP_FFFFFFFF_I
].xyzw
[TGSI_EXEC_TEMP_FFFFFFFF_C
].u
[i
] = 0xFFFFFFFF; /* not used */
1308 mach
->Temps
[TGSI_EXEC_TEMP_ONE_I
].xyzw
[TGSI_EXEC_TEMP_ONE_C
].f
[i
] = 1.0f
;
1309 mach
->Temps
[TGSI_EXEC_TEMP_TWO_I
].xyzw
[TGSI_EXEC_TEMP_TWO_C
].f
[i
] = 2.0f
; /* not used */
1310 mach
->Temps
[TGSI_EXEC_TEMP_128_I
].xyzw
[TGSI_EXEC_TEMP_128_C
].f
[i
] = 128.0f
;
1311 mach
->Temps
[TGSI_EXEC_TEMP_MINUS_128_I
].xyzw
[TGSI_EXEC_TEMP_MINUS_128_C
].f
[i
] = -128.0f
;
1312 mach
->Temps
[TGSI_EXEC_TEMP_THREE_I
].xyzw
[TGSI_EXEC_TEMP_THREE_C
].f
[i
] = 3.0f
;
1313 mach
->Temps
[TGSI_EXEC_TEMP_HALF_I
].xyzw
[TGSI_EXEC_TEMP_HALF_C
].f
[i
] = 0.5f
;
1317 /* silence warnings */
1326 align_free(mach
->Inputs
);
1327 align_free(mach
->Outputs
);
1335 tgsi_exec_machine_destroy(struct tgsi_exec_machine
*mach
)
1338 FREE(mach
->Instructions
);
1339 FREE(mach
->Declarations
);
1341 align_free(mach
->Inputs
);
1342 align_free(mach
->Outputs
);
1349 micro_add(union tgsi_exec_channel
*dst
,
1350 const union tgsi_exec_channel
*src0
,
1351 const union tgsi_exec_channel
*src1
)
1353 dst
->f
[0] = src0
->f
[0] + src1
->f
[0];
1354 dst
->f
[1] = src0
->f
[1] + src1
->f
[1];
1355 dst
->f
[2] = src0
->f
[2] + src1
->f
[2];
1356 dst
->f
[3] = src0
->f
[3] + src1
->f
[3];
1361 union tgsi_exec_channel
*dst
,
1362 const union tgsi_exec_channel
*src0
,
1363 const union tgsi_exec_channel
*src1
)
1365 if (src1
->f
[0] != 0) {
1366 dst
->f
[0] = src0
->f
[0] / src1
->f
[0];
1368 if (src1
->f
[1] != 0) {
1369 dst
->f
[1] = src0
->f
[1] / src1
->f
[1];
1371 if (src1
->f
[2] != 0) {
1372 dst
->f
[2] = src0
->f
[2] / src1
->f
[2];
1374 if (src1
->f
[3] != 0) {
1375 dst
->f
[3] = src0
->f
[3] / src1
->f
[3];
1381 union tgsi_exec_channel
*dst
,
1382 const union tgsi_exec_channel
*src0
,
1383 const union tgsi_exec_channel
*src1
,
1384 const union tgsi_exec_channel
*src2
,
1385 const union tgsi_exec_channel
*src3
)
1387 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
1388 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
1389 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
1390 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
1394 micro_max(union tgsi_exec_channel
*dst
,
1395 const union tgsi_exec_channel
*src0
,
1396 const union tgsi_exec_channel
*src1
)
1398 dst
->f
[0] = src0
->f
[0] > src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
1399 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
1400 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
1401 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
1405 micro_min(union tgsi_exec_channel
*dst
,
1406 const union tgsi_exec_channel
*src0
,
1407 const union tgsi_exec_channel
*src1
)
1409 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
1410 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
1411 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
1412 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
1416 micro_mul(union tgsi_exec_channel
*dst
,
1417 const union tgsi_exec_channel
*src0
,
1418 const union tgsi_exec_channel
*src1
)
1420 dst
->f
[0] = src0
->f
[0] * src1
->f
[0];
1421 dst
->f
[1] = src0
->f
[1] * src1
->f
[1];
1422 dst
->f
[2] = src0
->f
[2] * src1
->f
[2];
1423 dst
->f
[3] = src0
->f
[3] * src1
->f
[3];
1428 union tgsi_exec_channel
*dst
,
1429 const union tgsi_exec_channel
*src
)
1431 dst
->f
[0] = -src
->f
[0];
1432 dst
->f
[1] = -src
->f
[1];
1433 dst
->f
[2] = -src
->f
[2];
1434 dst
->f
[3] = -src
->f
[3];
1439 union tgsi_exec_channel
*dst
,
1440 const union tgsi_exec_channel
*src0
,
1441 const union tgsi_exec_channel
*src1
)
1444 dst
->f
[0] = util_fast_pow( src0
->f
[0], src1
->f
[0] );
1445 dst
->f
[1] = util_fast_pow( src0
->f
[1], src1
->f
[1] );
1446 dst
->f
[2] = util_fast_pow( src0
->f
[2], src1
->f
[2] );
1447 dst
->f
[3] = util_fast_pow( src0
->f
[3], src1
->f
[3] );
1449 dst
->f
[0] = powf( src0
->f
[0], src1
->f
[0] );
1450 dst
->f
[1] = powf( src0
->f
[1], src1
->f
[1] );
1451 dst
->f
[2] = powf( src0
->f
[2], src1
->f
[2] );
1452 dst
->f
[3] = powf( src0
->f
[3], src1
->f
[3] );
1457 micro_sub(union tgsi_exec_channel
*dst
,
1458 const union tgsi_exec_channel
*src0
,
1459 const union tgsi_exec_channel
*src1
)
1461 dst
->f
[0] = src0
->f
[0] - src1
->f
[0];
1462 dst
->f
[1] = src0
->f
[1] - src1
->f
[1];
1463 dst
->f
[2] = src0
->f
[2] - src1
->f
[2];
1464 dst
->f
[3] = src0
->f
[3] - src1
->f
[3];
1468 fetch_src_file_channel(const struct tgsi_exec_machine
*mach
,
1469 const uint chan_index
,
1472 const union tgsi_exec_channel
*index
,
1473 const union tgsi_exec_channel
*index2D
,
1474 union tgsi_exec_channel
*chan
)
1478 assert(swizzle
< 4);
1481 case TGSI_FILE_CONSTANT
:
1482 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1483 assert(index2D
->i
[i
] >= 0 && index2D
->i
[i
] < PIPE_MAX_CONSTANT_BUFFERS
);
1484 assert(mach
->Consts
[index2D
->i
[i
]]);
1486 if (index
->i
[i
] < 0) {
1489 /* NOTE: copying the const value as a uint instead of float */
1490 const uint constbuf
= index2D
->i
[i
];
1491 const uint
*buf
= (const uint
*)mach
->Consts
[constbuf
];
1492 const int pos
= index
->i
[i
] * 4 + swizzle
;
1493 /* const buffer bounds check */
1494 if (pos
< 0 || pos
>= (int) mach
->ConstsSize
[constbuf
]) {
1496 /* Debug: print warning */
1497 static int count
= 0;
1499 debug_printf("TGSI Exec: const buffer index %d"
1500 " out of bounds\n", pos
);
1505 chan
->u
[i
] = buf
[pos
];
1510 case TGSI_FILE_INPUT
:
1511 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1513 if (PIPE_SHADER_GEOMETRY == mach->ShaderType) {
1514 debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n",
1515 index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i],
1516 index2D->i[i], index->i[i]);
1518 int pos
= index2D
->i
[i
] * TGSI_EXEC_MAX_INPUT_ATTRIBS
+ index
->i
[i
];
1520 assert(pos
< TGSI_MAX_PRIM_VERTICES
* PIPE_MAX_ATTRIBS
);
1521 chan
->u
[i
] = mach
->Inputs
[pos
].xyzw
[swizzle
].u
[i
];
1525 case TGSI_FILE_SYSTEM_VALUE
:
1526 /* XXX no swizzling at this point. Will be needed if we put
1527 * gl_FragCoord, for example, in a sys value register.
1529 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1530 chan
->u
[i
] = mach
->SystemValue
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1534 case TGSI_FILE_TEMPORARY
:
1535 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1536 assert(index
->i
[i
] < TGSI_EXEC_NUM_TEMPS
);
1537 assert(index2D
->i
[i
] == 0);
1539 chan
->u
[i
] = mach
->Temps
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1543 case TGSI_FILE_IMMEDIATE
:
1544 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1545 assert(index
->i
[i
] >= 0 && index
->i
[i
] < (int)mach
->ImmLimit
);
1546 assert(index2D
->i
[i
] == 0);
1548 chan
->f
[i
] = mach
->Imms
[index
->i
[i
]][swizzle
];
1552 case TGSI_FILE_ADDRESS
:
1553 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1554 assert(index
->i
[i
] >= 0);
1555 assert(index2D
->i
[i
] == 0);
1557 chan
->u
[i
] = mach
->Addrs
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1561 case TGSI_FILE_OUTPUT
:
1562 /* vertex/fragment output vars can be read too */
1563 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1564 assert(index
->i
[i
] >= 0);
1565 assert(index2D
->i
[i
] == 0);
1567 chan
->u
[i
] = mach
->Outputs
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1573 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1580 fetch_source_d(const struct tgsi_exec_machine
*mach
,
1581 union tgsi_exec_channel
*chan
,
1582 const struct tgsi_full_src_register
*reg
,
1583 const uint chan_index
,
1584 enum tgsi_exec_datatype src_datatype
)
1586 union tgsi_exec_channel index
;
1587 union tgsi_exec_channel index2D
;
1590 /* We start with a direct index into a register file.
1594 * file = Register.File
1595 * [1] = Register.Index
1600 index
.i
[3] = reg
->Register
.Index
;
1602 /* There is an extra source register that indirectly subscripts
1603 * a register file. The direct index now becomes an offset
1604 * that is being added to the indirect register.
1608 * ind = Indirect.File
1609 * [2] = Indirect.Index
1610 * .x = Indirect.SwizzleX
1612 if (reg
->Register
.Indirect
) {
1613 union tgsi_exec_channel index2
;
1614 union tgsi_exec_channel indir_index
;
1615 const uint execmask
= mach
->ExecMask
;
1618 /* which address register (always zero now) */
1622 index2
.i
[3] = reg
->Indirect
.Index
;
1623 /* get current value of address register[swizzle] */
1624 swizzle
= reg
->Indirect
.Swizzle
;
1625 fetch_src_file_channel(mach
,
1633 /* add value of address register to the offset */
1634 index
.i
[0] += indir_index
.i
[0];
1635 index
.i
[1] += indir_index
.i
[1];
1636 index
.i
[2] += indir_index
.i
[2];
1637 index
.i
[3] += indir_index
.i
[3];
1639 /* for disabled execution channels, zero-out the index to
1640 * avoid using a potential garbage value.
1642 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1643 if ((execmask
& (1 << i
)) == 0)
1648 /* There is an extra source register that is a second
1649 * subscript to a register file. Effectively it means that
1650 * the register file is actually a 2D array of registers.
1654 * [3] = Dimension.Index
1656 if (reg
->Register
.Dimension
) {
1660 index2D
.i
[3] = reg
->Dimension
.Index
;
1662 /* Again, the second subscript index can be addressed indirectly
1663 * identically to the first one.
1664 * Nothing stops us from indirectly addressing the indirect register,
1665 * but there is no need for that, so we won't exercise it.
1667 * file[ind[4].y+3][1],
1669 * ind = DimIndirect.File
1670 * [4] = DimIndirect.Index
1671 * .y = DimIndirect.SwizzleX
1673 if (reg
->Dimension
.Indirect
) {
1674 union tgsi_exec_channel index2
;
1675 union tgsi_exec_channel indir_index
;
1676 const uint execmask
= mach
->ExecMask
;
1682 index2
.i
[3] = reg
->DimIndirect
.Index
;
1684 swizzle
= reg
->DimIndirect
.Swizzle
;
1685 fetch_src_file_channel(mach
,
1687 reg
->DimIndirect
.File
,
1693 index2D
.i
[0] += indir_index
.i
[0];
1694 index2D
.i
[1] += indir_index
.i
[1];
1695 index2D
.i
[2] += indir_index
.i
[2];
1696 index2D
.i
[3] += indir_index
.i
[3];
1698 /* for disabled execution channels, zero-out the index to
1699 * avoid using a potential garbage value.
1701 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1702 if ((execmask
& (1 << i
)) == 0) {
1708 /* If by any chance there was a need for a 3D array of register
1709 * files, we would have to check whether Dimension is followed
1710 * by a dimension register and continue the saga.
1719 swizzle
= tgsi_util_get_full_src_register_swizzle( reg
, chan_index
);
1720 fetch_src_file_channel(mach
,
1730 fetch_source(const struct tgsi_exec_machine
*mach
,
1731 union tgsi_exec_channel
*chan
,
1732 const struct tgsi_full_src_register
*reg
,
1733 const uint chan_index
,
1734 enum tgsi_exec_datatype src_datatype
)
1736 fetch_source_d(mach
, chan
, reg
, chan_index
, src_datatype
);
1738 if (reg
->Register
.Absolute
) {
1739 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1740 micro_abs(chan
, chan
);
1742 micro_iabs(chan
, chan
);
1746 if (reg
->Register
.Negate
) {
1747 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1748 micro_neg(chan
, chan
);
1750 micro_ineg(chan
, chan
);
1755 static union tgsi_exec_channel
*
1756 store_dest_dstret(struct tgsi_exec_machine
*mach
,
1757 const union tgsi_exec_channel
*chan
,
1758 const struct tgsi_full_dst_register
*reg
,
1759 const struct tgsi_full_instruction
*inst
,
1761 enum tgsi_exec_datatype dst_datatype
)
1763 static union tgsi_exec_channel null
;
1764 union tgsi_exec_channel
*dst
;
1765 union tgsi_exec_channel index2D
;
1766 int offset
= 0; /* indirection offset */
1770 if (0 && dst_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1771 check_inf_or_nan(chan
);
1774 /* There is an extra source register that indirectly subscripts
1775 * a register file. The direct index now becomes an offset
1776 * that is being added to the indirect register.
1780 * ind = Indirect.File
1781 * [2] = Indirect.Index
1782 * .x = Indirect.SwizzleX
1784 if (reg
->Register
.Indirect
) {
1785 union tgsi_exec_channel index
;
1786 union tgsi_exec_channel indir_index
;
1789 /* which address register (always zero for now) */
1793 index
.i
[3] = reg
->Indirect
.Index
;
1795 /* get current value of address register[swizzle] */
1796 swizzle
= reg
->Indirect
.Swizzle
;
1798 /* fetch values from the address/indirection register */
1799 fetch_src_file_channel(mach
,
1807 /* save indirection offset */
1808 offset
= indir_index
.i
[0];
1811 /* There is an extra source register that is a second
1812 * subscript to a register file. Effectively it means that
1813 * the register file is actually a 2D array of registers.
1817 * [3] = Dimension.Index
1819 if (reg
->Register
.Dimension
) {
1823 index2D
.i
[3] = reg
->Dimension
.Index
;
1825 /* Again, the second subscript index can be addressed indirectly
1826 * identically to the first one.
1827 * Nothing stops us from indirectly addressing the indirect register,
1828 * but there is no need for that, so we won't exercise it.
1830 * file[ind[4].y+3][1],
1832 * ind = DimIndirect.File
1833 * [4] = DimIndirect.Index
1834 * .y = DimIndirect.SwizzleX
1836 if (reg
->Dimension
.Indirect
) {
1837 union tgsi_exec_channel index2
;
1838 union tgsi_exec_channel indir_index
;
1839 const uint execmask
= mach
->ExecMask
;
1846 index2
.i
[3] = reg
->DimIndirect
.Index
;
1848 swizzle
= reg
->DimIndirect
.Swizzle
;
1849 fetch_src_file_channel(mach
,
1851 reg
->DimIndirect
.File
,
1857 index2D
.i
[0] += indir_index
.i
[0];
1858 index2D
.i
[1] += indir_index
.i
[1];
1859 index2D
.i
[2] += indir_index
.i
[2];
1860 index2D
.i
[3] += indir_index
.i
[3];
1862 /* for disabled execution channels, zero-out the index to
1863 * avoid using a potential garbage value.
1865 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1866 if ((execmask
& (1 << i
)) == 0) {
1872 /* If by any chance there was a need for a 3D array of register
1873 * files, we would have to check whether Dimension is followed
1874 * by a dimension register and continue the saga.
1883 switch (reg
->Register
.File
) {
1884 case TGSI_FILE_NULL
:
1888 case TGSI_FILE_OUTPUT
:
1889 index
= mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0]
1890 + reg
->Register
.Index
;
1891 dst
= &mach
->Outputs
[offset
+ index
].xyzw
[chan_index
];
1893 debug_printf("NumOutputs = %d, TEMP_O_C/I = %d, redindex = %d\n",
1894 mach
->NumOutputs
, mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0],
1895 reg
->Register
.Index
);
1896 if (PIPE_SHADER_GEOMETRY
== mach
->ShaderType
) {
1897 debug_printf("STORING OUT[%d] mask(%d), = (", offset
+ index
, execmask
);
1898 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1899 if (execmask
& (1 << i
))
1900 debug_printf("%f, ", chan
->f
[i
]);
1901 debug_printf(")\n");
1906 case TGSI_FILE_TEMPORARY
:
1907 index
= reg
->Register
.Index
;
1908 assert( index
< TGSI_EXEC_NUM_TEMPS
);
1909 dst
= &mach
->Temps
[offset
+ index
].xyzw
[chan_index
];
1912 case TGSI_FILE_ADDRESS
:
1913 index
= reg
->Register
.Index
;
1914 dst
= &mach
->Addrs
[index
].xyzw
[chan_index
];
1926 store_dest_double(struct tgsi_exec_machine
*mach
,
1927 const union tgsi_exec_channel
*chan
,
1928 const struct tgsi_full_dst_register
*reg
,
1929 const struct tgsi_full_instruction
*inst
,
1931 enum tgsi_exec_datatype dst_datatype
)
1933 union tgsi_exec_channel
*dst
;
1934 const uint execmask
= mach
->ExecMask
;
1937 dst
= store_dest_dstret(mach
, chan
, reg
, inst
, chan_index
,
1943 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1944 if (execmask
& (1 << i
))
1945 dst
->i
[i
] = chan
->i
[i
];
1949 store_dest(struct tgsi_exec_machine
*mach
,
1950 const union tgsi_exec_channel
*chan
,
1951 const struct tgsi_full_dst_register
*reg
,
1952 const struct tgsi_full_instruction
*inst
,
1954 enum tgsi_exec_datatype dst_datatype
)
1956 union tgsi_exec_channel
*dst
;
1957 const uint execmask
= mach
->ExecMask
;
1960 dst
= store_dest_dstret(mach
, chan
, reg
, inst
, chan_index
,
1965 if (!inst
->Instruction
.Saturate
) {
1966 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1967 if (execmask
& (1 << i
))
1968 dst
->i
[i
] = chan
->i
[i
];
1971 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1972 if (execmask
& (1 << i
)) {
1973 if (chan
->f
[i
] < 0.0f
)
1975 else if (chan
->f
[i
] > 1.0f
)
1978 dst
->i
[i
] = chan
->i
[i
];
1983 #define FETCH(VAL,INDEX,CHAN)\
1984 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT)
1986 #define IFETCH(VAL,INDEX,CHAN)\
1987 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT)
1991 * Execute ARB-style KIL which is predicated by a src register.
1992 * Kill fragment if any of the four values is less than zero.
1995 exec_kill_if(struct tgsi_exec_machine
*mach
,
1996 const struct tgsi_full_instruction
*inst
)
2000 uint kilmask
= 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
2001 union tgsi_exec_channel r
[1];
2003 /* This mask stores component bits that were already tested. */
2006 for (chan_index
= 0; chan_index
< 4; chan_index
++)
2011 /* unswizzle channel */
2012 swizzle
= tgsi_util_get_full_src_register_swizzle (
2016 /* check if the component has not been already tested */
2017 if (uniquemask
& (1 << swizzle
))
2019 uniquemask
|= 1 << swizzle
;
2021 FETCH(&r
[0], 0, chan_index
);
2022 for (i
= 0; i
< 4; i
++)
2023 if (r
[0].f
[i
] < 0.0f
)
2027 /* restrict to fragments currently executing */
2028 kilmask
&= mach
->ExecMask
;
2030 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
2034 * Unconditional fragment kill/discard.
2037 exec_kill(struct tgsi_exec_machine
*mach
,
2038 const struct tgsi_full_instruction
*inst
)
2040 uint kilmask
; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
2042 /* kill fragment for all fragments currently executing */
2043 kilmask
= mach
->ExecMask
;
2044 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
2048 emit_vertex(struct tgsi_exec_machine
*mach
)
2050 /* FIXME: check for exec mask correctly
2052 for (i = 0; i < TGSI_QUAD_SIZE; ++i) {
2053 if ((mach->ExecMask & (1 << i)))
2055 if (mach
->ExecMask
) {
2056 if (mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]] >= mach
->MaxOutputVertices
)
2059 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] += mach
->NumOutputs
;
2060 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]]++;
2065 emit_primitive(struct tgsi_exec_machine
*mach
)
2067 unsigned *prim_count
= &mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0];
2068 /* FIXME: check for exec mask correctly
2070 for (i = 0; i < TGSI_QUAD_SIZE; ++i) {
2071 if ((mach->ExecMask & (1 << i)))
2073 if (mach
->ExecMask
) {
2075 debug_assert((*prim_count
* mach
->NumOutputs
) < mach
->MaxGeometryShaderOutputs
);
2076 mach
->Primitives
[*prim_count
] = 0;
2081 conditional_emit_primitive(struct tgsi_exec_machine
*mach
)
2083 if (PIPE_SHADER_GEOMETRY
== mach
->ShaderType
) {
2085 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]];
2086 if (emitted_verts
) {
2087 emit_primitive(mach
);
2094 * Fetch four texture samples using STR texture coordinates.
2097 fetch_texel( struct tgsi_sampler
*sampler
,
2098 const unsigned sview_idx
,
2099 const unsigned sampler_idx
,
2100 const union tgsi_exec_channel
*s
,
2101 const union tgsi_exec_channel
*t
,
2102 const union tgsi_exec_channel
*p
,
2103 const union tgsi_exec_channel
*c0
,
2104 const union tgsi_exec_channel
*c1
,
2105 float derivs
[3][2][TGSI_QUAD_SIZE
],
2106 const int8_t offset
[3],
2107 enum tgsi_sampler_control control
,
2108 union tgsi_exec_channel
*r
,
2109 union tgsi_exec_channel
*g
,
2110 union tgsi_exec_channel
*b
,
2111 union tgsi_exec_channel
*a
)
2114 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2116 /* FIXME: handle explicit derivs, offsets */
2117 sampler
->get_samples(sampler
, sview_idx
, sampler_idx
,
2118 s
->f
, t
->f
, p
->f
, c0
->f
, c1
->f
, derivs
, offset
, control
, rgba
);
2120 for (j
= 0; j
< 4; j
++) {
2121 r
->f
[j
] = rgba
[0][j
];
2122 g
->f
[j
] = rgba
[1][j
];
2123 b
->f
[j
] = rgba
[2][j
];
2124 a
->f
[j
] = rgba
[3][j
];
2129 #define TEX_MODIFIER_NONE 0
2130 #define TEX_MODIFIER_PROJECTED 1
2131 #define TEX_MODIFIER_LOD_BIAS 2
2132 #define TEX_MODIFIER_EXPLICIT_LOD 3
2133 #define TEX_MODIFIER_LEVEL_ZERO 4
2134 #define TEX_MODIFIER_GATHER 5
2137 * Fetch all 3 (for s,t,r coords) texel offsets, put them into int array.
2140 fetch_texel_offsets(struct tgsi_exec_machine
*mach
,
2141 const struct tgsi_full_instruction
*inst
,
2144 if (inst
->Texture
.NumOffsets
== 1) {
2145 union tgsi_exec_channel index
;
2146 union tgsi_exec_channel offset
[3];
2147 index
.i
[0] = index
.i
[1] = index
.i
[2] = index
.i
[3] = inst
->TexOffsets
[0].Index
;
2148 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
2149 inst
->TexOffsets
[0].SwizzleX
, &index
, &ZeroVec
, &offset
[0]);
2150 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
2151 inst
->TexOffsets
[0].SwizzleY
, &index
, &ZeroVec
, &offset
[1]);
2152 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
2153 inst
->TexOffsets
[0].SwizzleZ
, &index
, &ZeroVec
, &offset
[2]);
2154 offsets
[0] = offset
[0].i
[0];
2155 offsets
[1] = offset
[1].i
[0];
2156 offsets
[2] = offset
[2].i
[0];
2158 assert(inst
->Texture
.NumOffsets
== 0);
2159 offsets
[0] = offsets
[1] = offsets
[2] = 0;
2165 * Fetch dx and dy values for one channel (s, t or r).
2166 * Put dx values into one float array, dy values into another.
2169 fetch_assign_deriv_channel(struct tgsi_exec_machine
*mach
,
2170 const struct tgsi_full_instruction
*inst
,
2173 float derivs
[2][TGSI_QUAD_SIZE
])
2175 union tgsi_exec_channel d
;
2176 FETCH(&d
, regdsrcx
, chan
);
2177 derivs
[0][0] = d
.f
[0];
2178 derivs
[0][1] = d
.f
[1];
2179 derivs
[0][2] = d
.f
[2];
2180 derivs
[0][3] = d
.f
[3];
2181 FETCH(&d
, regdsrcx
+ 1, chan
);
2182 derivs
[1][0] = d
.f
[0];
2183 derivs
[1][1] = d
.f
[1];
2184 derivs
[1][2] = d
.f
[2];
2185 derivs
[1][3] = d
.f
[3];
2189 fetch_sampler_unit(struct tgsi_exec_machine
*mach
,
2190 const struct tgsi_full_instruction
*inst
,
2195 if (inst
->Src
[sampler
].Register
.Indirect
) {
2196 const struct tgsi_full_src_register
*reg
= &inst
->Src
[sampler
];
2197 union tgsi_exec_channel indir_index
, index2
;
2198 const uint execmask
= mach
->ExecMask
;
2202 index2
.i
[3] = reg
->Indirect
.Index
;
2204 fetch_src_file_channel(mach
,
2207 reg
->Indirect
.Swizzle
,
2211 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2212 if (execmask
& (1 << i
)) {
2213 unit
= inst
->Src
[sampler
].Register
.Index
+ indir_index
.i
[i
];
2219 unit
= inst
->Src
[sampler
].Register
.Index
;
2225 * execute a texture instruction.
2227 * modifier is used to control the channel routing for the
2228 * instruction variants like proj, lod, and texture with lod bias.
2229 * sampler indicates which src register the sampler is contained in.
2232 exec_tex(struct tgsi_exec_machine
*mach
,
2233 const struct tgsi_full_instruction
*inst
,
2234 uint modifier
, uint sampler
)
2236 const union tgsi_exec_channel
*args
[5], *proj
= NULL
;
2237 union tgsi_exec_channel r
[5];
2238 enum tgsi_sampler_control control
= TGSI_SAMPLER_LOD_NONE
;
2242 int dim
, shadow_ref
, i
;
2244 unit
= fetch_sampler_unit(mach
, inst
, sampler
);
2245 /* always fetch all 3 offsets, overkill but keeps code simple */
2246 fetch_texel_offsets(mach
, inst
, offsets
);
2248 assert(modifier
!= TEX_MODIFIER_LEVEL_ZERO
);
2249 assert(inst
->Texture
.Texture
!= TGSI_TEXTURE_BUFFER
);
2251 dim
= tgsi_util_get_texture_coord_dim(inst
->Texture
.Texture
);
2252 shadow_ref
= tgsi_util_get_shadow_ref_src_index(inst
->Texture
.Texture
);
2255 if (shadow_ref
>= 0)
2256 assert(shadow_ref
>= dim
&& shadow_ref
< ARRAY_SIZE(args
));
2258 /* fetch modifier to the last argument */
2259 if (modifier
!= TEX_MODIFIER_NONE
) {
2260 const int last
= ARRAY_SIZE(args
) - 1;
2262 /* fetch modifier from src0.w or src1.x */
2264 assert(dim
<= TGSI_CHAN_W
&& shadow_ref
!= TGSI_CHAN_W
);
2265 FETCH(&r
[last
], 0, TGSI_CHAN_W
);
2268 assert(shadow_ref
!= 4);
2269 FETCH(&r
[last
], 1, TGSI_CHAN_X
);
2272 if (modifier
!= TEX_MODIFIER_PROJECTED
) {
2273 args
[last
] = &r
[last
];
2277 args
[last
] = &ZeroVec
;
2280 /* point unused arguments to zero vector */
2281 for (i
= dim
; i
< last
; i
++)
2284 if (modifier
== TEX_MODIFIER_EXPLICIT_LOD
)
2285 control
= TGSI_SAMPLER_LOD_EXPLICIT
;
2286 else if (modifier
== TEX_MODIFIER_LOD_BIAS
)
2287 control
= TGSI_SAMPLER_LOD_BIAS
;
2288 else if (modifier
== TEX_MODIFIER_GATHER
)
2289 control
= TGSI_SAMPLER_GATHER
;
2292 for (i
= dim
; i
< ARRAY_SIZE(args
); i
++)
2296 /* fetch coordinates */
2297 for (i
= 0; i
< dim
; i
++) {
2298 FETCH(&r
[i
], 0, TGSI_CHAN_X
+ i
);
2301 micro_div(&r
[i
], &r
[i
], proj
);
2306 /* fetch reference value */
2307 if (shadow_ref
>= 0) {
2308 FETCH(&r
[shadow_ref
], shadow_ref
/ 4, TGSI_CHAN_X
+ (shadow_ref
% 4));
2311 micro_div(&r
[shadow_ref
], &r
[shadow_ref
], proj
);
2313 args
[shadow_ref
] = &r
[shadow_ref
];
2316 fetch_texel(mach
->Sampler
, unit
, unit
,
2317 args
[0], args
[1], args
[2], args
[3], args
[4],
2318 NULL
, offsets
, control
,
2319 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2322 debug_printf("fetch r: %g %g %g %g\n",
2323 r
[0].f
[0], r
[0].f
[1], r
[0].f
[2], r
[0].f
[3]);
2324 debug_printf("fetch g: %g %g %g %g\n",
2325 r
[1].f
[0], r
[1].f
[1], r
[1].f
[2], r
[1].f
[3]);
2326 debug_printf("fetch b: %g %g %g %g\n",
2327 r
[2].f
[0], r
[2].f
[1], r
[2].f
[2], r
[2].f
[3]);
2328 debug_printf("fetch a: %g %g %g %g\n",
2329 r
[3].f
[0], r
[3].f
[1], r
[3].f
[2], r
[3].f
[3]);
2332 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2333 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2334 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2340 exec_lodq(struct tgsi_exec_machine
*mach
,
2341 const struct tgsi_full_instruction
*inst
)
2346 union tgsi_exec_channel coords
[4];
2347 const union tgsi_exec_channel
*args
[ARRAY_SIZE(coords
)];
2348 union tgsi_exec_channel r
[2];
2350 unit
= fetch_sampler_unit(mach
, inst
, 1);
2351 dim
= tgsi_util_get_texture_coord_dim(inst
->Texture
.Texture
);
2352 assert(dim
<= ARRAY_SIZE(coords
));
2353 /* fetch coordinates */
2354 for (i
= 0; i
< dim
; i
++) {
2355 FETCH(&coords
[i
], 0, TGSI_CHAN_X
+ i
);
2356 args
[i
] = &coords
[i
];
2358 for (i
= dim
; i
< ARRAY_SIZE(coords
); i
++) {
2361 mach
->Sampler
->query_lod(mach
->Sampler
, unit
, unit
,
2366 TGSI_SAMPLER_LOD_NONE
,
2370 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2371 store_dest(mach
, &r
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_X
,
2372 TGSI_EXEC_DATA_FLOAT
);
2374 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2375 store_dest(mach
, &r
[1], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
,
2376 TGSI_EXEC_DATA_FLOAT
);
2381 exec_txd(struct tgsi_exec_machine
*mach
,
2382 const struct tgsi_full_instruction
*inst
)
2384 union tgsi_exec_channel r
[4];
2385 float derivs
[3][2][TGSI_QUAD_SIZE
];
2390 unit
= fetch_sampler_unit(mach
, inst
, 3);
2391 /* always fetch all 3 offsets, overkill but keeps code simple */
2392 fetch_texel_offsets(mach
, inst
, offsets
);
2394 switch (inst
->Texture
.Texture
) {
2395 case TGSI_TEXTURE_1D
:
2396 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2398 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
2400 fetch_texel(mach
->Sampler
, unit
, unit
,
2401 &r
[0], &ZeroVec
, &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
2402 derivs
, offsets
, TGSI_SAMPLER_DERIVS_EXPLICIT
,
2403 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2406 case TGSI_TEXTURE_SHADOW1D
:
2407 case TGSI_TEXTURE_1D_ARRAY
:
2408 case TGSI_TEXTURE_SHADOW1D_ARRAY
:
2409 /* SHADOW1D/1D_ARRAY would not need Y/Z respectively, but don't bother */
2410 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2411 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2412 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2414 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
2416 fetch_texel(mach
->Sampler
, unit
, unit
,
2417 &r
[0], &r
[1], &r
[2], &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
2418 derivs
, offsets
, TGSI_SAMPLER_DERIVS_EXPLICIT
,
2419 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2422 case TGSI_TEXTURE_2D
:
2423 case TGSI_TEXTURE_RECT
:
2424 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2425 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2427 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
2428 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
2430 fetch_texel(mach
->Sampler
, unit
, unit
,
2431 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
2432 derivs
, offsets
, TGSI_SAMPLER_DERIVS_EXPLICIT
,
2433 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2437 case TGSI_TEXTURE_SHADOW2D
:
2438 case TGSI_TEXTURE_SHADOWRECT
:
2439 case TGSI_TEXTURE_2D_ARRAY
:
2440 case TGSI_TEXTURE_SHADOW2D_ARRAY
:
2441 /* only SHADOW2D_ARRAY actually needs W */
2442 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2443 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2444 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2445 FETCH(&r
[3], 0, TGSI_CHAN_W
);
2447 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
2448 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
2450 fetch_texel(mach
->Sampler
, unit
, unit
,
2451 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
, /* inputs */
2452 derivs
, offsets
, TGSI_SAMPLER_DERIVS_EXPLICIT
,
2453 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2456 case TGSI_TEXTURE_3D
:
2457 case TGSI_TEXTURE_CUBE
:
2458 case TGSI_TEXTURE_CUBE_ARRAY
:
2459 case TGSI_TEXTURE_SHADOWCUBE
:
2460 /* only TEXTURE_CUBE_ARRAY and TEXTURE_SHADOWCUBE actually need W */
2461 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2462 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2463 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2464 FETCH(&r
[3], 0, TGSI_CHAN_W
);
2466 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
2467 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
2468 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Z
, derivs
[2]);
2470 fetch_texel(mach
->Sampler
, unit
, unit
,
2471 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
, /* inputs */
2472 derivs
, offsets
, TGSI_SAMPLER_DERIVS_EXPLICIT
,
2473 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2480 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2481 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2482 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2489 exec_txf(struct tgsi_exec_machine
*mach
,
2490 const struct tgsi_full_instruction
*inst
)
2492 union tgsi_exec_channel r
[4];
2495 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2500 unit
= fetch_sampler_unit(mach
, inst
, 1);
2501 /* always fetch all 3 offsets, overkill but keeps code simple */
2502 fetch_texel_offsets(mach
, inst
, offsets
);
2504 IFETCH(&r
[3], 0, TGSI_CHAN_W
);
2506 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_SAMPLE_I
||
2507 inst
->Instruction
.Opcode
== TGSI_OPCODE_SAMPLE_I_MS
) {
2508 target
= mach
->SamplerViews
[unit
].Resource
;
2511 target
= inst
->Texture
.Texture
;
2514 case TGSI_TEXTURE_3D
:
2515 case TGSI_TEXTURE_2D_ARRAY
:
2516 case TGSI_TEXTURE_SHADOW2D_ARRAY
:
2517 case TGSI_TEXTURE_2D_ARRAY_MSAA
:
2518 IFETCH(&r
[2], 0, TGSI_CHAN_Z
);
2520 case TGSI_TEXTURE_2D
:
2521 case TGSI_TEXTURE_RECT
:
2522 case TGSI_TEXTURE_SHADOW1D_ARRAY
:
2523 case TGSI_TEXTURE_SHADOW2D
:
2524 case TGSI_TEXTURE_SHADOWRECT
:
2525 case TGSI_TEXTURE_1D_ARRAY
:
2526 case TGSI_TEXTURE_2D_MSAA
:
2527 IFETCH(&r
[1], 0, TGSI_CHAN_Y
);
2529 case TGSI_TEXTURE_BUFFER
:
2530 case TGSI_TEXTURE_1D
:
2531 case TGSI_TEXTURE_SHADOW1D
:
2532 IFETCH(&r
[0], 0, TGSI_CHAN_X
);
2539 mach
->Sampler
->get_texel(mach
->Sampler
, unit
, r
[0].i
, r
[1].i
, r
[2].i
, r
[3].i
,
2542 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2543 r
[0].f
[j
] = rgba
[0][j
];
2544 r
[1].f
[j
] = rgba
[1][j
];
2545 r
[2].f
[j
] = rgba
[2][j
];
2546 r
[3].f
[j
] = rgba
[3][j
];
2549 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_SAMPLE_I
||
2550 inst
->Instruction
.Opcode
== TGSI_OPCODE_SAMPLE_I_MS
) {
2551 unsigned char swizzles
[4];
2552 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2553 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2554 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2555 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2557 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2558 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2559 store_dest(mach
, &r
[swizzles
[chan
]],
2560 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2565 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2566 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2567 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2574 exec_txq(struct tgsi_exec_machine
*mach
,
2575 const struct tgsi_full_instruction
*inst
)
2578 union tgsi_exec_channel r
[4], src
;
2583 unit
= fetch_sampler_unit(mach
, inst
, 1);
2585 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_INT
);
2587 /* XXX: This interface can't return per-pixel values */
2588 mach
->Sampler
->get_dims(mach
->Sampler
, unit
, src
.i
[0], result
);
2590 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2591 for (j
= 0; j
< 4; j
++) {
2592 r
[j
].i
[i
] = result
[j
];
2596 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2597 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2598 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
,
2599 TGSI_EXEC_DATA_INT
);
2605 exec_sample(struct tgsi_exec_machine
*mach
,
2606 const struct tgsi_full_instruction
*inst
,
2607 uint modifier
, boolean compare
)
2609 const uint resource_unit
= inst
->Src
[1].Register
.Index
;
2610 const uint sampler_unit
= inst
->Src
[2].Register
.Index
;
2611 union tgsi_exec_channel r
[5], c1
;
2612 const union tgsi_exec_channel
*lod
= &ZeroVec
;
2613 enum tgsi_sampler_control control
= TGSI_SAMPLER_LOD_NONE
;
2615 unsigned char swizzles
[4];
2618 /* always fetch all 3 offsets, overkill but keeps code simple */
2619 fetch_texel_offsets(mach
, inst
, offsets
);
2621 assert(modifier
!= TEX_MODIFIER_PROJECTED
);
2623 if (modifier
!= TEX_MODIFIER_NONE
) {
2624 if (modifier
== TEX_MODIFIER_LOD_BIAS
) {
2625 FETCH(&c1
, 3, TGSI_CHAN_X
);
2627 control
= TGSI_SAMPLER_LOD_BIAS
;
2629 else if (modifier
== TEX_MODIFIER_EXPLICIT_LOD
) {
2630 FETCH(&c1
, 3, TGSI_CHAN_X
);
2632 control
= TGSI_SAMPLER_LOD_EXPLICIT
;
2634 else if (modifier
== TEX_MODIFIER_GATHER
) {
2635 control
= TGSI_SAMPLER_GATHER
;
2638 assert(modifier
== TEX_MODIFIER_LEVEL_ZERO
);
2639 control
= TGSI_SAMPLER_LOD_ZERO
;
2643 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2645 switch (mach
->SamplerViews
[resource_unit
].Resource
) {
2646 case TGSI_TEXTURE_1D
:
2648 FETCH(&r
[2], 3, TGSI_CHAN_X
);
2649 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2650 &r
[0], &ZeroVec
, &r
[2], &ZeroVec
, lod
, /* S, T, P, C, LOD */
2651 NULL
, offsets
, control
,
2652 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2655 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2656 &r
[0], &ZeroVec
, &ZeroVec
, &ZeroVec
, lod
, /* S, T, P, C, LOD */
2657 NULL
, offsets
, control
,
2658 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2662 case TGSI_TEXTURE_1D_ARRAY
:
2663 case TGSI_TEXTURE_2D
:
2664 case TGSI_TEXTURE_RECT
:
2665 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2667 FETCH(&r
[2], 3, TGSI_CHAN_X
);
2668 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2669 &r
[0], &r
[1], &r
[2], &ZeroVec
, lod
, /* S, T, P, C, LOD */
2670 NULL
, offsets
, control
,
2671 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2674 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2675 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, lod
, /* S, T, P, C, LOD */
2676 NULL
, offsets
, control
,
2677 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2681 case TGSI_TEXTURE_2D_ARRAY
:
2682 case TGSI_TEXTURE_3D
:
2683 case TGSI_TEXTURE_CUBE
:
2684 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2685 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2687 FETCH(&r
[3], 3, TGSI_CHAN_X
);
2688 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2689 &r
[0], &r
[1], &r
[2], &r
[3], lod
,
2690 NULL
, offsets
, control
,
2691 &r
[0], &r
[1], &r
[2], &r
[3]);
2694 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2695 &r
[0], &r
[1], &r
[2], &ZeroVec
, lod
,
2696 NULL
, offsets
, control
,
2697 &r
[0], &r
[1], &r
[2], &r
[3]);
2701 case TGSI_TEXTURE_CUBE_ARRAY
:
2702 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2703 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2704 FETCH(&r
[3], 0, TGSI_CHAN_W
);
2706 FETCH(&r
[4], 3, TGSI_CHAN_X
);
2707 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2708 &r
[0], &r
[1], &r
[2], &r
[3], &r
[4],
2709 NULL
, offsets
, control
,
2710 &r
[0], &r
[1], &r
[2], &r
[3]);
2713 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2714 &r
[0], &r
[1], &r
[2], &r
[3], lod
,
2715 NULL
, offsets
, control
,
2716 &r
[0], &r
[1], &r
[2], &r
[3]);
2725 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2726 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2727 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2728 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2730 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2731 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2732 store_dest(mach
, &r
[swizzles
[chan
]],
2733 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2739 exec_sample_d(struct tgsi_exec_machine
*mach
,
2740 const struct tgsi_full_instruction
*inst
)
2742 const uint resource_unit
= inst
->Src
[1].Register
.Index
;
2743 const uint sampler_unit
= inst
->Src
[2].Register
.Index
;
2744 union tgsi_exec_channel r
[4];
2745 float derivs
[3][2][TGSI_QUAD_SIZE
];
2747 unsigned char swizzles
[4];
2750 /* always fetch all 3 offsets, overkill but keeps code simple */
2751 fetch_texel_offsets(mach
, inst
, offsets
);
2753 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2755 switch (mach
->SamplerViews
[resource_unit
].Resource
) {
2756 case TGSI_TEXTURE_1D
:
2757 case TGSI_TEXTURE_1D_ARRAY
:
2758 /* only 1D array actually needs Y */
2759 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2761 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2763 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2764 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
2765 derivs
, offsets
, TGSI_SAMPLER_DERIVS_EXPLICIT
,
2766 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2769 case TGSI_TEXTURE_2D
:
2770 case TGSI_TEXTURE_RECT
:
2771 case TGSI_TEXTURE_2D_ARRAY
:
2772 /* only 2D array actually needs Z */
2773 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2774 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2776 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2777 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Y
, derivs
[1]);
2779 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2780 &r
[0], &r
[1], &r
[2], &ZeroVec
, &ZeroVec
, /* inputs */
2781 derivs
, offsets
, TGSI_SAMPLER_DERIVS_EXPLICIT
,
2782 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2785 case TGSI_TEXTURE_3D
:
2786 case TGSI_TEXTURE_CUBE
:
2787 case TGSI_TEXTURE_CUBE_ARRAY
:
2788 /* only cube array actually needs W */
2789 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2790 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2791 FETCH(&r
[3], 0, TGSI_CHAN_W
);
2793 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2794 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Y
, derivs
[1]);
2795 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Z
, derivs
[2]);
2797 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2798 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
,
2799 derivs
, offsets
, TGSI_SAMPLER_DERIVS_EXPLICIT
,
2800 &r
[0], &r
[1], &r
[2], &r
[3]);
2807 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2808 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2809 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2810 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2812 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2813 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2814 store_dest(mach
, &r
[swizzles
[chan
]],
2815 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2822 * Evaluate a constant-valued coefficient at the position of the
2827 struct tgsi_exec_machine
*mach
,
2833 for( i
= 0; i
< TGSI_QUAD_SIZE
; i
++ ) {
2834 mach
->Inputs
[attrib
].xyzw
[chan
].f
[i
] = mach
->InterpCoefs
[attrib
].a0
[chan
];
2839 * Evaluate a linear-valued coefficient at the position of the
2844 struct tgsi_exec_machine
*mach
,
2848 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
2849 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
2850 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
2851 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
2852 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
2853 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
;
2854 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = a0
+ dadx
;
2855 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = a0
+ dady
;
2856 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = a0
+ dadx
+ dady
;
2860 * Evaluate a perspective-valued coefficient at the position of the
2864 eval_perspective_coef(
2865 struct tgsi_exec_machine
*mach
,
2869 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
2870 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
2871 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
2872 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
2873 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
2874 const float *w
= mach
->QuadPos
.xyzw
[3].f
;
2875 /* divide by W here */
2876 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
/ w
[0];
2877 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = (a0
+ dadx
) / w
[1];
2878 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = (a0
+ dady
) / w
[2];
2879 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = (a0
+ dadx
+ dady
) / w
[3];
2883 typedef void (* eval_coef_func
)(
2884 struct tgsi_exec_machine
*mach
,
2889 exec_declaration(struct tgsi_exec_machine
*mach
,
2890 const struct tgsi_full_declaration
*decl
)
2892 if (decl
->Declaration
.File
== TGSI_FILE_SAMPLER_VIEW
) {
2893 mach
->SamplerViews
[decl
->Range
.First
] = decl
->SamplerView
;
2897 if (mach
->ShaderType
== PIPE_SHADER_FRAGMENT
) {
2898 if (decl
->Declaration
.File
== TGSI_FILE_INPUT
) {
2899 uint first
, last
, mask
;
2901 first
= decl
->Range
.First
;
2902 last
= decl
->Range
.Last
;
2903 mask
= decl
->Declaration
.UsageMask
;
2905 /* XXX we could remove this special-case code since
2906 * mach->InterpCoefs[first].a0 should already have the
2907 * front/back-face value. But we should first update the
2908 * ureg code to emit the right UsageMask value (WRITEMASK_X).
2909 * Then, we could remove the tgsi_exec_machine::Face field.
2911 /* XXX make FACE a system value */
2912 if (decl
->Semantic
.Name
== TGSI_SEMANTIC_FACE
) {
2915 assert(decl
->Semantic
.Index
== 0);
2916 assert(first
== last
);
2918 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2919 mach
->Inputs
[first
].xyzw
[0].f
[i
] = mach
->Face
;
2922 eval_coef_func eval
;
2925 switch (decl
->Interp
.Interpolate
) {
2926 case TGSI_INTERPOLATE_CONSTANT
:
2927 eval
= eval_constant_coef
;
2930 case TGSI_INTERPOLATE_LINEAR
:
2931 eval
= eval_linear_coef
;
2934 case TGSI_INTERPOLATE_PERSPECTIVE
:
2935 eval
= eval_perspective_coef
;
2938 case TGSI_INTERPOLATE_COLOR
:
2939 eval
= mach
->flatshade_color
? eval_constant_coef
: eval_perspective_coef
;
2947 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
2948 if (mask
& (1 << j
)) {
2949 for (i
= first
; i
<= last
; i
++) {
2956 if (DEBUG_EXECUTION
) {
2958 for (i
= first
; i
<= last
; ++i
) {
2959 debug_printf("IN[%2u] = ", i
);
2960 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
2964 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
2965 mach
->Inputs
[i
].xyzw
[0].f
[j
], mach
->Inputs
[i
].xyzw
[0].u
[j
],
2966 mach
->Inputs
[i
].xyzw
[1].f
[j
], mach
->Inputs
[i
].xyzw
[1].u
[j
],
2967 mach
->Inputs
[i
].xyzw
[2].f
[j
], mach
->Inputs
[i
].xyzw
[2].u
[j
],
2968 mach
->Inputs
[i
].xyzw
[3].f
[j
], mach
->Inputs
[i
].xyzw
[3].u
[j
]);
2977 typedef void (* micro_unary_op
)(union tgsi_exec_channel
*dst
,
2978 const union tgsi_exec_channel
*src
);
2981 exec_scalar_unary(struct tgsi_exec_machine
*mach
,
2982 const struct tgsi_full_instruction
*inst
,
2984 enum tgsi_exec_datatype dst_datatype
,
2985 enum tgsi_exec_datatype src_datatype
)
2988 union tgsi_exec_channel src
;
2989 union tgsi_exec_channel dst
;
2991 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, src_datatype
);
2993 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2994 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2995 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
3001 exec_vector_unary(struct tgsi_exec_machine
*mach
,
3002 const struct tgsi_full_instruction
*inst
,
3004 enum tgsi_exec_datatype dst_datatype
,
3005 enum tgsi_exec_datatype src_datatype
)
3008 struct tgsi_exec_vector dst
;
3010 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3011 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3012 union tgsi_exec_channel src
;
3014 fetch_source(mach
, &src
, &inst
->Src
[0], chan
, src_datatype
);
3015 op(&dst
.xyzw
[chan
], &src
);
3018 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3019 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3020 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
3025 typedef void (* micro_binary_op
)(union tgsi_exec_channel
*dst
,
3026 const union tgsi_exec_channel
*src0
,
3027 const union tgsi_exec_channel
*src1
);
3030 exec_scalar_binary(struct tgsi_exec_machine
*mach
,
3031 const struct tgsi_full_instruction
*inst
,
3033 enum tgsi_exec_datatype dst_datatype
,
3034 enum tgsi_exec_datatype src_datatype
)
3037 union tgsi_exec_channel src
[2];
3038 union tgsi_exec_channel dst
;
3040 fetch_source(mach
, &src
[0], &inst
->Src
[0], TGSI_CHAN_X
, src_datatype
);
3041 fetch_source(mach
, &src
[1], &inst
->Src
[1], TGSI_CHAN_X
, src_datatype
);
3042 op(&dst
, &src
[0], &src
[1]);
3043 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3044 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3045 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
3051 exec_vector_binary(struct tgsi_exec_machine
*mach
,
3052 const struct tgsi_full_instruction
*inst
,
3054 enum tgsi_exec_datatype dst_datatype
,
3055 enum tgsi_exec_datatype src_datatype
)
3058 struct tgsi_exec_vector dst
;
3060 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3061 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3062 union tgsi_exec_channel src
[2];
3064 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
3065 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
3066 op(&dst
.xyzw
[chan
], &src
[0], &src
[1]);
3069 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3070 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3071 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
3076 typedef void (* micro_trinary_op
)(union tgsi_exec_channel
*dst
,
3077 const union tgsi_exec_channel
*src0
,
3078 const union tgsi_exec_channel
*src1
,
3079 const union tgsi_exec_channel
*src2
);
3082 exec_vector_trinary(struct tgsi_exec_machine
*mach
,
3083 const struct tgsi_full_instruction
*inst
,
3084 micro_trinary_op op
,
3085 enum tgsi_exec_datatype dst_datatype
,
3086 enum tgsi_exec_datatype src_datatype
)
3089 struct tgsi_exec_vector dst
;
3091 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3092 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3093 union tgsi_exec_channel src
[3];
3095 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
3096 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
3097 fetch_source(mach
, &src
[2], &inst
->Src
[2], chan
, src_datatype
);
3098 op(&dst
.xyzw
[chan
], &src
[0], &src
[1], &src
[2]);
3101 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3102 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3103 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
3108 typedef void (* micro_quaternary_op
)(union tgsi_exec_channel
*dst
,
3109 const union tgsi_exec_channel
*src0
,
3110 const union tgsi_exec_channel
*src1
,
3111 const union tgsi_exec_channel
*src2
,
3112 const union tgsi_exec_channel
*src3
);
3115 exec_vector_quaternary(struct tgsi_exec_machine
*mach
,
3116 const struct tgsi_full_instruction
*inst
,
3117 micro_quaternary_op op
,
3118 enum tgsi_exec_datatype dst_datatype
,
3119 enum tgsi_exec_datatype src_datatype
)
3122 struct tgsi_exec_vector dst
;
3124 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3125 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3126 union tgsi_exec_channel src
[4];
3128 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
3129 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
3130 fetch_source(mach
, &src
[2], &inst
->Src
[2], chan
, src_datatype
);
3131 fetch_source(mach
, &src
[3], &inst
->Src
[3], chan
, src_datatype
);
3132 op(&dst
.xyzw
[chan
], &src
[0], &src
[1], &src
[2], &src
[3]);
3135 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3136 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3137 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
3143 exec_dp3(struct tgsi_exec_machine
*mach
,
3144 const struct tgsi_full_instruction
*inst
)
3147 union tgsi_exec_channel arg
[3];
3149 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3150 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3151 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
3153 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_Z
; chan
++) {
3154 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
3155 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
3156 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
3159 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3160 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3161 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
3167 exec_dp4(struct tgsi_exec_machine
*mach
,
3168 const struct tgsi_full_instruction
*inst
)
3171 union tgsi_exec_channel arg
[3];
3173 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3174 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3175 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
3177 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_W
; chan
++) {
3178 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
3179 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
3180 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
3183 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3184 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3185 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
3191 exec_dp2(struct tgsi_exec_machine
*mach
,
3192 const struct tgsi_full_instruction
*inst
)
3195 union tgsi_exec_channel arg
[3];
3197 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3198 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3199 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
3201 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3202 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3203 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
3205 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3206 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3207 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
3213 exec_pk2h(struct tgsi_exec_machine
*mach
,
3214 const struct tgsi_full_instruction
*inst
)
3217 union tgsi_exec_channel arg
[2], dst
;
3219 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3220 fetch_source(mach
, &arg
[1], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3221 for (chan
= 0; chan
< TGSI_QUAD_SIZE
; chan
++) {
3222 dst
.u
[chan
] = util_float_to_half(arg
[0].f
[chan
]) |
3223 (util_float_to_half(arg
[1].f
[chan
]) << 16);
3225 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3226 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3227 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_UINT
);
3233 exec_up2h(struct tgsi_exec_machine
*mach
,
3234 const struct tgsi_full_instruction
*inst
)
3237 union tgsi_exec_channel arg
, dst
[2];
3239 fetch_source(mach
, &arg
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_UINT
);
3240 for (chan
= 0; chan
< TGSI_QUAD_SIZE
; chan
++) {
3241 dst
[0].f
[chan
] = util_half_to_float(arg
.u
[chan
] & 0xffff);
3242 dst
[1].f
[chan
] = util_half_to_float(arg
.u
[chan
] >> 16);
3244 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3245 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3246 store_dest(mach
, &dst
[chan
& 1], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
3252 micro_ucmp(union tgsi_exec_channel
*dst
,
3253 const union tgsi_exec_channel
*src0
,
3254 const union tgsi_exec_channel
*src1
,
3255 const union tgsi_exec_channel
*src2
)
3257 dst
->f
[0] = src0
->u
[0] ? src1
->f
[0] : src2
->f
[0];
3258 dst
->f
[1] = src0
->u
[1] ? src1
->f
[1] : src2
->f
[1];
3259 dst
->f
[2] = src0
->u
[2] ? src1
->f
[2] : src2
->f
[2];
3260 dst
->f
[3] = src0
->u
[3] ? src1
->f
[3] : src2
->f
[3];
3264 exec_ucmp(struct tgsi_exec_machine
*mach
,
3265 const struct tgsi_full_instruction
*inst
)
3268 struct tgsi_exec_vector dst
;
3270 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3271 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3272 union tgsi_exec_channel src
[3];
3274 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
,
3275 TGSI_EXEC_DATA_UINT
);
3276 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
,
3277 TGSI_EXEC_DATA_FLOAT
);
3278 fetch_source(mach
, &src
[2], &inst
->Src
[2], chan
,
3279 TGSI_EXEC_DATA_FLOAT
);
3280 micro_ucmp(&dst
.xyzw
[chan
], &src
[0], &src
[1], &src
[2]);
3283 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3284 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3285 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
,
3286 TGSI_EXEC_DATA_FLOAT
);
3292 exec_dst(struct tgsi_exec_machine
*mach
,
3293 const struct tgsi_full_instruction
*inst
)
3295 union tgsi_exec_channel r
[2];
3296 union tgsi_exec_channel d
[4];
3298 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3299 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3300 fetch_source(mach
, &r
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3301 micro_mul(&d
[TGSI_CHAN_Y
], &r
[0], &r
[1]);
3303 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3304 fetch_source(mach
, &d
[TGSI_CHAN_Z
], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3306 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3307 fetch_source(mach
, &d
[TGSI_CHAN_W
], &inst
->Src
[1], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3310 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3311 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3313 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3314 store_dest(mach
, &d
[TGSI_CHAN_Y
], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3316 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3317 store_dest(mach
, &d
[TGSI_CHAN_Z
], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3319 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3320 store_dest(mach
, &d
[TGSI_CHAN_W
], &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3325 exec_log(struct tgsi_exec_machine
*mach
,
3326 const struct tgsi_full_instruction
*inst
)
3328 union tgsi_exec_channel r
[3];
3330 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3331 micro_abs(&r
[2], &r
[0]); /* r2 = abs(r0) */
3332 micro_lg2(&r
[1], &r
[2]); /* r1 = lg2(r2) */
3333 micro_flr(&r
[0], &r
[1]); /* r0 = floor(r1) */
3334 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3335 store_dest(mach
, &r
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3337 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3338 micro_exp2(&r
[0], &r
[0]); /* r0 = 2 ^ r0 */
3339 micro_div(&r
[0], &r
[2], &r
[0]); /* r0 = r2 / r0 */
3340 store_dest(mach
, &r
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3342 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3343 store_dest(mach
, &r
[1], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3345 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3346 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3351 exec_exp(struct tgsi_exec_machine
*mach
,
3352 const struct tgsi_full_instruction
*inst
)
3354 union tgsi_exec_channel r
[3];
3356 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3357 micro_flr(&r
[1], &r
[0]); /* r1 = floor(r0) */
3358 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3359 micro_exp2(&r
[2], &r
[1]); /* r2 = 2 ^ r1 */
3360 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3362 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3363 micro_sub(&r
[2], &r
[0], &r
[1]); /* r2 = r0 - r1 */
3364 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3366 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3367 micro_exp2(&r
[2], &r
[0]); /* r2 = 2 ^ r0 */
3368 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3370 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3371 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3376 exec_lit(struct tgsi_exec_machine
*mach
,
3377 const struct tgsi_full_instruction
*inst
)
3379 union tgsi_exec_channel r
[3];
3380 union tgsi_exec_channel d
[3];
3382 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_YZ
) {
3383 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3384 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3385 fetch_source(mach
, &r
[1], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3386 micro_max(&r
[1], &r
[1], &ZeroVec
);
3388 fetch_source(mach
, &r
[2], &inst
->Src
[0], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3389 micro_min(&r
[2], &r
[2], &P128Vec
);
3390 micro_max(&r
[2], &r
[2], &M128Vec
);
3391 micro_pow(&r
[1], &r
[1], &r
[2]);
3392 micro_lt(&d
[TGSI_CHAN_Z
], &ZeroVec
, &r
[0], &r
[1], &ZeroVec
);
3393 store_dest(mach
, &d
[TGSI_CHAN_Z
], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3395 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3396 micro_max(&d
[TGSI_CHAN_Y
], &r
[0], &ZeroVec
);
3397 store_dest(mach
, &d
[TGSI_CHAN_Y
], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3400 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3401 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3404 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3405 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3410 exec_break(struct tgsi_exec_machine
*mach
)
3412 if (mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_LOOP
) {
3413 /* turn off loop channels for each enabled exec channel */
3414 mach
->LoopMask
&= ~mach
->ExecMask
;
3415 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3416 UPDATE_EXEC_MASK(mach
);
3418 assert(mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_SWITCH
);
3420 mach
->Switch
.mask
= 0x0;
3422 UPDATE_EXEC_MASK(mach
);
3427 exec_switch(struct tgsi_exec_machine
*mach
,
3428 const struct tgsi_full_instruction
*inst
)
3430 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
3431 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
3433 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
3434 fetch_source(mach
, &mach
->Switch
.selector
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_UINT
);
3435 mach
->Switch
.mask
= 0x0;
3436 mach
->Switch
.defaultMask
= 0x0;
3438 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
3439 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_SWITCH
;
3441 UPDATE_EXEC_MASK(mach
);
3445 exec_case(struct tgsi_exec_machine
*mach
,
3446 const struct tgsi_full_instruction
*inst
)
3448 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
3449 union tgsi_exec_channel src
;
3452 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_UINT
);
3454 if (mach
->Switch
.selector
.u
[0] == src
.u
[0]) {
3457 if (mach
->Switch
.selector
.u
[1] == src
.u
[1]) {
3460 if (mach
->Switch
.selector
.u
[2] == src
.u
[2]) {
3463 if (mach
->Switch
.selector
.u
[3] == src
.u
[3]) {
3467 mach
->Switch
.defaultMask
|= mask
;
3469 mach
->Switch
.mask
|= mask
& prevMask
;
3471 UPDATE_EXEC_MASK(mach
);
3474 /* FIXME: this will only work if default is last */
3476 exec_default(struct tgsi_exec_machine
*mach
)
3478 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
3480 mach
->Switch
.mask
|= ~mach
->Switch
.defaultMask
& prevMask
;
3482 UPDATE_EXEC_MASK(mach
);
3486 exec_endswitch(struct tgsi_exec_machine
*mach
)
3488 mach
->Switch
= mach
->SwitchStack
[--mach
->SwitchStackTop
];
3489 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
3491 UPDATE_EXEC_MASK(mach
);
3494 typedef void (* micro_dop
)(union tgsi_double_channel
*dst
,
3495 const union tgsi_double_channel
*src
);
3497 typedef void (* micro_dop_sop
)(union tgsi_double_channel
*dst
,
3498 const union tgsi_double_channel
*src0
,
3499 union tgsi_exec_channel
*src1
);
3501 typedef void (* micro_dop_s
)(union tgsi_double_channel
*dst
,
3502 const union tgsi_exec_channel
*src
);
3504 typedef void (* micro_sop_d
)(union tgsi_exec_channel
*dst
,
3505 const union tgsi_double_channel
*src
);
3508 fetch_double_channel(struct tgsi_exec_machine
*mach
,
3509 union tgsi_double_channel
*chan
,
3510 const struct tgsi_full_src_register
*reg
,
3514 union tgsi_exec_channel src
[2];
3517 fetch_source_d(mach
, &src
[0], reg
, chan_0
, TGSI_EXEC_DATA_UINT
);
3518 fetch_source_d(mach
, &src
[1], reg
, chan_1
, TGSI_EXEC_DATA_UINT
);
3520 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3521 chan
->u
[i
][0] = src
[0].u
[i
];
3522 chan
->u
[i
][1] = src
[1].u
[i
];
3524 if (reg
->Register
.Absolute
) {
3525 micro_dabs(chan
, chan
);
3527 if (reg
->Register
.Negate
) {
3528 micro_dneg(chan
, chan
);
3533 store_double_channel(struct tgsi_exec_machine
*mach
,
3534 const union tgsi_double_channel
*chan
,
3535 const struct tgsi_full_dst_register
*reg
,
3536 const struct tgsi_full_instruction
*inst
,
3540 union tgsi_exec_channel dst
[2];
3542 union tgsi_double_channel temp
;
3543 const uint execmask
= mach
->ExecMask
;
3545 if (!inst
->Instruction
.Saturate
) {
3546 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
3547 if (execmask
& (1 << i
)) {
3548 dst
[0].u
[i
] = chan
->u
[i
][0];
3549 dst
[1].u
[i
] = chan
->u
[i
][1];
3553 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
3554 if (execmask
& (1 << i
)) {
3555 if (chan
->d
[i
] < 0.0)
3557 else if (chan
->d
[i
] > 1.0)
3560 temp
.d
[i
] = chan
->d
[i
];
3562 dst
[0].u
[i
] = temp
.u
[i
][0];
3563 dst
[1].u
[i
] = temp
.u
[i
][1];
3567 store_dest_double(mach
, &dst
[0], reg
, inst
, chan_0
, TGSI_EXEC_DATA_UINT
);
3569 store_dest_double(mach
, &dst
[1], reg
, inst
, chan_1
, TGSI_EXEC_DATA_UINT
);
3573 exec_double_unary(struct tgsi_exec_machine
*mach
,
3574 const struct tgsi_full_instruction
*inst
,
3577 union tgsi_double_channel src
;
3578 union tgsi_double_channel dst
;
3580 if ((inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XY
) == TGSI_WRITEMASK_XY
) {
3581 fetch_double_channel(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3583 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_CHAN_Y
);
3585 if ((inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_ZW
) == TGSI_WRITEMASK_ZW
) {
3586 fetch_double_channel(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_CHAN_W
);
3588 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_CHAN_W
);
3593 exec_double_binary(struct tgsi_exec_machine
*mach
,
3594 const struct tgsi_full_instruction
*inst
,
3596 enum tgsi_exec_datatype dst_datatype
)
3598 union tgsi_double_channel src
[2];
3599 union tgsi_double_channel dst
;
3600 int first_dest_chan
, second_dest_chan
;
3603 wmask
= inst
->Dst
[0].Register
.WriteMask
;
3604 /* these are & because of the way DSLT etc store their destinations */
3605 if (wmask
& TGSI_WRITEMASK_XY
) {
3606 first_dest_chan
= TGSI_CHAN_X
;
3607 second_dest_chan
= TGSI_CHAN_Y
;
3608 if (dst_datatype
== TGSI_EXEC_DATA_UINT
) {
3609 first_dest_chan
= (wmask
& TGSI_WRITEMASK_X
) ? TGSI_CHAN_X
: TGSI_CHAN_Y
;
3610 second_dest_chan
= -1;
3613 fetch_double_channel(mach
, &src
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3614 fetch_double_channel(mach
, &src
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3616 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, first_dest_chan
, second_dest_chan
);
3619 if (wmask
& TGSI_WRITEMASK_ZW
) {
3620 first_dest_chan
= TGSI_CHAN_Z
;
3621 second_dest_chan
= TGSI_CHAN_W
;
3622 if (dst_datatype
== TGSI_EXEC_DATA_UINT
) {
3623 first_dest_chan
= (wmask
& TGSI_WRITEMASK_Z
) ? TGSI_CHAN_Z
: TGSI_CHAN_W
;
3624 second_dest_chan
= -1;
3627 fetch_double_channel(mach
, &src
[0], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_CHAN_W
);
3628 fetch_double_channel(mach
, &src
[1], &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_CHAN_W
);
3630 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, first_dest_chan
, second_dest_chan
);
3635 exec_double_trinary(struct tgsi_exec_machine
*mach
,
3636 const struct tgsi_full_instruction
*inst
,
3639 union tgsi_double_channel src
[3];
3640 union tgsi_double_channel dst
;
3642 if ((inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XY
) == TGSI_WRITEMASK_XY
) {
3643 fetch_double_channel(mach
, &src
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3644 fetch_double_channel(mach
, &src
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3645 fetch_double_channel(mach
, &src
[2], &inst
->Src
[2], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3647 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_CHAN_Y
);
3649 if ((inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_ZW
) == TGSI_WRITEMASK_ZW
) {
3650 fetch_double_channel(mach
, &src
[0], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_CHAN_W
);
3651 fetch_double_channel(mach
, &src
[1], &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_CHAN_W
);
3652 fetch_double_channel(mach
, &src
[2], &inst
->Src
[2], TGSI_CHAN_Z
, TGSI_CHAN_W
);
3654 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_CHAN_W
);
3659 exec_dldexp(struct tgsi_exec_machine
*mach
,
3660 const struct tgsi_full_instruction
*inst
)
3662 union tgsi_double_channel src0
;
3663 union tgsi_exec_channel src1
;
3664 union tgsi_double_channel dst
;
3667 wmask
= inst
->Dst
[0].Register
.WriteMask
;
3668 if (wmask
& TGSI_WRITEMASK_XY
) {
3669 fetch_double_channel(mach
, &src0
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3670 fetch_source(mach
, &src1
, &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_INT
);
3671 micro_dldexp(&dst
, &src0
, &src1
);
3672 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_CHAN_Y
);
3675 if (wmask
& TGSI_WRITEMASK_ZW
) {
3676 fetch_double_channel(mach
, &src0
, &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_CHAN_W
);
3677 fetch_source(mach
, &src1
, &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_EXEC_DATA_INT
);
3678 micro_dldexp(&dst
, &src0
, &src1
);
3679 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_CHAN_W
);
3684 exec_dfracexp(struct tgsi_exec_machine
*mach
,
3685 const struct tgsi_full_instruction
*inst
)
3687 union tgsi_double_channel src
;
3688 union tgsi_double_channel dst
;
3689 union tgsi_exec_channel dst_exp
;
3691 fetch_double_channel(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3692 micro_dfracexp(&dst
, &dst_exp
, &src
);
3693 if ((inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XY
) == TGSI_WRITEMASK_XY
)
3694 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_CHAN_Y
);
3695 if ((inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_ZW
) == TGSI_WRITEMASK_ZW
)
3696 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_CHAN_W
);
3697 for (unsigned chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3698 if (inst
->Dst
[1].Register
.WriteMask
& (1 << chan
))
3699 store_dest(mach
, &dst_exp
, &inst
->Dst
[1], inst
, chan
, TGSI_EXEC_DATA_INT
);
3704 exec_arg0_64_arg1_32(struct tgsi_exec_machine
*mach
,
3705 const struct tgsi_full_instruction
*inst
,
3708 union tgsi_double_channel src0
;
3709 union tgsi_exec_channel src1
;
3710 union tgsi_double_channel dst
;
3713 wmask
= inst
->Dst
[0].Register
.WriteMask
;
3714 if (wmask
& TGSI_WRITEMASK_XY
) {
3715 fetch_double_channel(mach
, &src0
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_CHAN_Y
);
3716 fetch_source(mach
, &src1
, &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_INT
);
3717 op(&dst
, &src0
, &src1
);
3718 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_CHAN_Y
);
3721 if (wmask
& TGSI_WRITEMASK_ZW
) {
3722 fetch_double_channel(mach
, &src0
, &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_CHAN_W
);
3723 fetch_source(mach
, &src1
, &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_EXEC_DATA_INT
);
3724 op(&dst
, &src0
, &src1
);
3725 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_CHAN_W
);
3730 get_image_coord_dim(unsigned tgsi_tex
)
3734 case TGSI_TEXTURE_BUFFER
:
3735 case TGSI_TEXTURE_1D
:
3738 case TGSI_TEXTURE_2D
:
3739 case TGSI_TEXTURE_RECT
:
3740 case TGSI_TEXTURE_1D_ARRAY
:
3741 case TGSI_TEXTURE_2D_MSAA
:
3744 case TGSI_TEXTURE_3D
:
3745 case TGSI_TEXTURE_CUBE
:
3746 case TGSI_TEXTURE_2D_ARRAY
:
3747 case TGSI_TEXTURE_2D_ARRAY_MSAA
:
3748 case TGSI_TEXTURE_CUBE_ARRAY
:
3752 assert(!"unknown texture target");
3761 get_image_coord_sample(unsigned tgsi_tex
)
3765 case TGSI_TEXTURE_2D_MSAA
:
3768 case TGSI_TEXTURE_2D_ARRAY_MSAA
:
3778 exec_load_img(struct tgsi_exec_machine
*mach
,
3779 const struct tgsi_full_instruction
*inst
)
3781 union tgsi_exec_channel r
[4], sample_r
;
3787 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3788 struct tgsi_image_params params
;
3789 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
3791 unit
= fetch_sampler_unit(mach
, inst
, 0);
3792 dim
= get_image_coord_dim(inst
->Memory
.Texture
);
3793 sample
= get_image_coord_sample(inst
->Memory
.Texture
);
3796 params
.execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
3798 params
.tgsi_tex_instr
= inst
->Memory
.Texture
;
3799 params
.format
= inst
->Memory
.Format
;
3801 for (i
= 0; i
< dim
; i
++) {
3802 IFETCH(&r
[i
], 1, TGSI_CHAN_X
+ i
);
3806 IFETCH(&sample_r
, 1, TGSI_CHAN_X
+ sample
);
3808 mach
->Image
->load(mach
->Image
, ¶ms
,
3809 r
[0].i
, r
[1].i
, r
[2].i
, sample_r
.i
,
3811 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3812 r
[0].f
[j
] = rgba
[0][j
];
3813 r
[1].f
[j
] = rgba
[1][j
];
3814 r
[2].f
[j
] = rgba
[2][j
];
3815 r
[3].f
[j
] = rgba
[3][j
];
3817 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3818 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3819 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
3825 exec_load_buf(struct tgsi_exec_machine
*mach
,
3826 const struct tgsi_full_instruction
*inst
)
3828 union tgsi_exec_channel r
[4];
3832 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3833 struct tgsi_buffer_params params
;
3834 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
3836 unit
= fetch_sampler_unit(mach
, inst
, 0);
3838 params
.execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
3840 IFETCH(&r
[0], 1, TGSI_CHAN_X
);
3842 mach
->Buffer
->load(mach
->Buffer
, ¶ms
,
3844 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3845 r
[0].f
[j
] = rgba
[0][j
];
3846 r
[1].f
[j
] = rgba
[1][j
];
3847 r
[2].f
[j
] = rgba
[2][j
];
3848 r
[3].f
[j
] = rgba
[3][j
];
3850 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3851 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3852 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
3858 exec_load_mem(struct tgsi_exec_machine
*mach
,
3859 const struct tgsi_full_instruction
*inst
)
3861 union tgsi_exec_channel r
[4];
3863 char *ptr
= mach
->LocalMem
;
3867 IFETCH(&r
[0], 1, TGSI_CHAN_X
);
3868 if (r
[0].u
[0] >= mach
->LocalMemSize
)
3874 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3875 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3876 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3877 memcpy(&r
[chan
].u
[j
], ptr
+ (4 * chan
), 4);
3882 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
3883 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
3884 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
3890 exec_load(struct tgsi_exec_machine
*mach
,
3891 const struct tgsi_full_instruction
*inst
)
3893 if (inst
->Src
[0].Register
.File
== TGSI_FILE_IMAGE
)
3894 exec_load_img(mach
, inst
);
3895 else if (inst
->Src
[0].Register
.File
== TGSI_FILE_BUFFER
)
3896 exec_load_buf(mach
, inst
);
3897 else if (inst
->Src
[0].Register
.File
== TGSI_FILE_MEMORY
)
3898 exec_load_mem(mach
, inst
);
3902 exec_store_img(struct tgsi_exec_machine
*mach
,
3903 const struct tgsi_full_instruction
*inst
)
3905 union tgsi_exec_channel r
[3], sample_r
;
3906 union tgsi_exec_channel value
[4];
3907 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3908 struct tgsi_image_params params
;
3913 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
3914 unit
= inst
->Dst
[0].Register
.Index
;
3915 dim
= get_image_coord_dim(inst
->Memory
.Texture
);
3916 sample
= get_image_coord_sample(inst
->Memory
.Texture
);
3919 params
.execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
3921 params
.tgsi_tex_instr
= inst
->Memory
.Texture
;
3922 params
.format
= inst
->Memory
.Format
;
3924 for (i
= 0; i
< dim
; i
++) {
3925 IFETCH(&r
[i
], 0, TGSI_CHAN_X
+ i
);
3928 for (i
= 0; i
< 4; i
++) {
3929 FETCH(&value
[i
], 1, TGSI_CHAN_X
+ i
);
3932 IFETCH(&sample_r
, 0, TGSI_CHAN_X
+ sample
);
3934 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3935 rgba
[0][j
] = value
[0].f
[j
];
3936 rgba
[1][j
] = value
[1].f
[j
];
3937 rgba
[2][j
] = value
[2].f
[j
];
3938 rgba
[3][j
] = value
[3].f
[j
];
3941 mach
->Image
->store(mach
->Image
, ¶ms
,
3942 r
[0].i
, r
[1].i
, r
[2].i
, sample_r
.i
,
3947 exec_store_buf(struct tgsi_exec_machine
*mach
,
3948 const struct tgsi_full_instruction
*inst
)
3950 union tgsi_exec_channel r
[3];
3951 union tgsi_exec_channel value
[4];
3952 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3953 struct tgsi_buffer_params params
;
3956 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
3958 unit
= inst
->Dst
[0].Register
.Index
;
3960 params
.execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
3962 params
.writemask
= inst
->Dst
[0].Register
.WriteMask
;
3964 IFETCH(&r
[0], 0, TGSI_CHAN_X
);
3965 for (i
= 0; i
< 4; i
++) {
3966 FETCH(&value
[i
], 1, TGSI_CHAN_X
+ i
);
3969 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3970 rgba
[0][j
] = value
[0].f
[j
];
3971 rgba
[1][j
] = value
[1].f
[j
];
3972 rgba
[2][j
] = value
[2].f
[j
];
3973 rgba
[3][j
] = value
[3].f
[j
];
3976 mach
->Buffer
->store(mach
->Buffer
, ¶ms
,
3982 exec_store_mem(struct tgsi_exec_machine
*mach
,
3983 const struct tgsi_full_instruction
*inst
)
3985 union tgsi_exec_channel r
[3];
3986 union tgsi_exec_channel value
[4];
3988 char *ptr
= mach
->LocalMem
;
3989 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
3990 int execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
3992 IFETCH(&r
[0], 0, TGSI_CHAN_X
);
3994 for (i
= 0; i
< 4; i
++) {
3995 FETCH(&value
[i
], 1, TGSI_CHAN_X
+ i
);
3998 if (r
[0].u
[0] >= mach
->LocalMemSize
)
4002 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
4003 if (execmask
& (1 << i
)) {
4004 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
4005 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
4006 memcpy(ptr
+ (chan
* 4), &value
[chan
].u
[0], 4);
4014 exec_store(struct tgsi_exec_machine
*mach
,
4015 const struct tgsi_full_instruction
*inst
)
4017 if (inst
->Dst
[0].Register
.File
== TGSI_FILE_IMAGE
)
4018 exec_store_img(mach
, inst
);
4019 else if (inst
->Dst
[0].Register
.File
== TGSI_FILE_BUFFER
)
4020 exec_store_buf(mach
, inst
);
4021 else if (inst
->Dst
[0].Register
.File
== TGSI_FILE_MEMORY
)
4022 exec_store_mem(mach
, inst
);
4026 exec_atomop_img(struct tgsi_exec_machine
*mach
,
4027 const struct tgsi_full_instruction
*inst
)
4029 union tgsi_exec_channel r
[4], sample_r
;
4030 union tgsi_exec_channel value
[4], value2
[4];
4031 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
4032 float rgba2
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
4033 struct tgsi_image_params params
;
4038 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
4039 unit
= fetch_sampler_unit(mach
, inst
, 0);
4040 dim
= get_image_coord_dim(inst
->Memory
.Texture
);
4041 sample
= get_image_coord_sample(inst
->Memory
.Texture
);
4044 params
.execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
4046 params
.tgsi_tex_instr
= inst
->Memory
.Texture
;
4047 params
.format
= inst
->Memory
.Format
;
4049 for (i
= 0; i
< dim
; i
++) {
4050 IFETCH(&r
[i
], 1, TGSI_CHAN_X
+ i
);
4053 for (i
= 0; i
< 4; i
++) {
4054 FETCH(&value
[i
], 2, TGSI_CHAN_X
+ i
);
4055 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_ATOMCAS
)
4056 FETCH(&value2
[i
], 3, TGSI_CHAN_X
+ i
);
4059 IFETCH(&sample_r
, 1, TGSI_CHAN_X
+ sample
);
4061 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
4062 rgba
[0][j
] = value
[0].f
[j
];
4063 rgba
[1][j
] = value
[1].f
[j
];
4064 rgba
[2][j
] = value
[2].f
[j
];
4065 rgba
[3][j
] = value
[3].f
[j
];
4067 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_ATOMCAS
) {
4068 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
4069 rgba2
[0][j
] = value2
[0].f
[j
];
4070 rgba2
[1][j
] = value2
[1].f
[j
];
4071 rgba2
[2][j
] = value2
[2].f
[j
];
4072 rgba2
[3][j
] = value2
[3].f
[j
];
4076 mach
->Image
->op(mach
->Image
, ¶ms
, inst
->Instruction
.Opcode
,
4077 r
[0].i
, r
[1].i
, r
[2].i
, sample_r
.i
,
4080 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
4081 r
[0].f
[j
] = rgba
[0][j
];
4082 r
[1].f
[j
] = rgba
[1][j
];
4083 r
[2].f
[j
] = rgba
[2][j
];
4084 r
[3].f
[j
] = rgba
[3][j
];
4086 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
4087 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
4088 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
4094 exec_atomop_buf(struct tgsi_exec_machine
*mach
,
4095 const struct tgsi_full_instruction
*inst
)
4097 union tgsi_exec_channel r
[4];
4098 union tgsi_exec_channel value
[4], value2
[4];
4099 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
4100 float rgba2
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
4101 struct tgsi_buffer_params params
;
4104 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
4106 unit
= fetch_sampler_unit(mach
, inst
, 0);
4108 params
.execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
4110 params
.writemask
= inst
->Dst
[0].Register
.WriteMask
;
4112 IFETCH(&r
[0], 1, TGSI_CHAN_X
);
4114 for (i
= 0; i
< 4; i
++) {
4115 FETCH(&value
[i
], 2, TGSI_CHAN_X
+ i
);
4116 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_ATOMCAS
)
4117 FETCH(&value2
[i
], 3, TGSI_CHAN_X
+ i
);
4120 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
4121 rgba
[0][j
] = value
[0].f
[j
];
4122 rgba
[1][j
] = value
[1].f
[j
];
4123 rgba
[2][j
] = value
[2].f
[j
];
4124 rgba
[3][j
] = value
[3].f
[j
];
4126 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_ATOMCAS
) {
4127 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
4128 rgba2
[0][j
] = value2
[0].f
[j
];
4129 rgba2
[1][j
] = value2
[1].f
[j
];
4130 rgba2
[2][j
] = value2
[2].f
[j
];
4131 rgba2
[3][j
] = value2
[3].f
[j
];
4135 mach
->Buffer
->op(mach
->Buffer
, ¶ms
, inst
->Instruction
.Opcode
,
4139 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
4140 r
[0].f
[j
] = rgba
[0][j
];
4141 r
[1].f
[j
] = rgba
[1][j
];
4142 r
[2].f
[j
] = rgba
[2][j
];
4143 r
[3].f
[j
] = rgba
[3][j
];
4145 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
4146 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
4147 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
4153 exec_atomop_mem(struct tgsi_exec_machine
*mach
,
4154 const struct tgsi_full_instruction
*inst
)
4156 union tgsi_exec_channel r
[4];
4157 union tgsi_exec_channel value
[4], value2
[4];
4158 char *ptr
= mach
->LocalMem
;
4162 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
4163 int execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
4164 IFETCH(&r
[0], 1, TGSI_CHAN_X
);
4166 if (r
[0].u
[0] >= mach
->LocalMemSize
)
4171 for (i
= 0; i
< 4; i
++) {
4172 FETCH(&value
[i
], 2, TGSI_CHAN_X
+ i
);
4173 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_ATOMCAS
)
4174 FETCH(&value2
[i
], 3, TGSI_CHAN_X
+ i
);
4177 memcpy(&r
[0].u
[0], ptr
, 4);
4179 switch (inst
->Instruction
.Opcode
) {
4180 case TGSI_OPCODE_ATOMUADD
:
4181 val
+= value
[0].u
[0];
4183 case TGSI_OPCODE_ATOMXOR
:
4184 val
^= value
[0].u
[0];
4186 case TGSI_OPCODE_ATOMOR
:
4187 val
|= value
[0].u
[0];
4189 case TGSI_OPCODE_ATOMAND
:
4190 val
&= value
[0].u
[0];
4192 case TGSI_OPCODE_ATOMUMIN
:
4193 val
= MIN2(val
, value
[0].u
[0]);
4195 case TGSI_OPCODE_ATOMUMAX
:
4196 val
= MAX2(val
, value
[0].u
[0]);
4198 case TGSI_OPCODE_ATOMIMIN
:
4199 val
= MIN2(r
[0].i
[0], value
[0].i
[0]);
4201 case TGSI_OPCODE_ATOMIMAX
:
4202 val
= MAX2(r
[0].i
[0], value
[0].i
[0]);
4204 case TGSI_OPCODE_ATOMXCHG
:
4205 val
= value
[0].i
[0];
4207 case TGSI_OPCODE_ATOMCAS
:
4208 if (val
== value
[0].u
[0])
4209 val
= value2
[0].u
[0];
4214 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
4215 if (execmask
& (1 << i
))
4216 memcpy(ptr
, &val
, 4);
4218 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
4219 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
4220 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
4226 exec_atomop(struct tgsi_exec_machine
*mach
,
4227 const struct tgsi_full_instruction
*inst
)
4229 if (inst
->Src
[0].Register
.File
== TGSI_FILE_IMAGE
)
4230 exec_atomop_img(mach
, inst
);
4231 else if (inst
->Src
[0].Register
.File
== TGSI_FILE_BUFFER
)
4232 exec_atomop_buf(mach
, inst
);
4233 else if (inst
->Src
[0].Register
.File
== TGSI_FILE_MEMORY
)
4234 exec_atomop_mem(mach
, inst
);
4238 exec_resq_img(struct tgsi_exec_machine
*mach
,
4239 const struct tgsi_full_instruction
*inst
)
4242 union tgsi_exec_channel r
[4];
4245 struct tgsi_image_params params
;
4246 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
4248 unit
= fetch_sampler_unit(mach
, inst
, 0);
4250 params
.execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
4252 params
.tgsi_tex_instr
= inst
->Memory
.Texture
;
4253 params
.format
= inst
->Memory
.Format
;
4255 mach
->Image
->get_dims(mach
->Image
, ¶ms
, result
);
4257 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
4258 for (j
= 0; j
< 4; j
++) {
4259 r
[j
].i
[i
] = result
[j
];
4263 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
4264 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
4265 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
,
4266 TGSI_EXEC_DATA_INT
);
4272 exec_resq_buf(struct tgsi_exec_machine
*mach
,
4273 const struct tgsi_full_instruction
*inst
)
4276 union tgsi_exec_channel r
[4];
4279 struct tgsi_buffer_params params
;
4280 int kilmask
= mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];
4282 unit
= fetch_sampler_unit(mach
, inst
, 0);
4284 params
.execmask
= mach
->ExecMask
& mach
->NonHelperMask
& ~kilmask
;
4287 mach
->Buffer
->get_dims(mach
->Buffer
, ¶ms
, &result
);
4289 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
4293 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
4294 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
4295 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
,
4296 TGSI_EXEC_DATA_INT
);
4302 exec_resq(struct tgsi_exec_machine
*mach
,
4303 const struct tgsi_full_instruction
*inst
)
4305 if (inst
->Src
[0].Register
.File
== TGSI_FILE_IMAGE
)
4306 exec_resq_img(mach
, inst
);
4308 exec_resq_buf(mach
, inst
);
4312 micro_f2u64(union tgsi_double_channel
*dst
,
4313 const union tgsi_exec_channel
*src
)
4315 dst
->u64
[0] = (uint64_t)src
->f
[0];
4316 dst
->u64
[1] = (uint64_t)src
->f
[1];
4317 dst
->u64
[2] = (uint64_t)src
->f
[2];
4318 dst
->u64
[3] = (uint64_t)src
->f
[3];
4322 micro_f2i64(union tgsi_double_channel
*dst
,
4323 const union tgsi_exec_channel
*src
)
4325 dst
->i64
[0] = (int64_t)src
->f
[0];
4326 dst
->i64
[1] = (int64_t)src
->f
[1];
4327 dst
->i64
[2] = (int64_t)src
->f
[2];
4328 dst
->i64
[3] = (int64_t)src
->f
[3];
4332 micro_u2i64(union tgsi_double_channel
*dst
,
4333 const union tgsi_exec_channel
*src
)
4335 dst
->u64
[0] = (uint64_t)src
->u
[0];
4336 dst
->u64
[1] = (uint64_t)src
->u
[1];
4337 dst
->u64
[2] = (uint64_t)src
->u
[2];
4338 dst
->u64
[3] = (uint64_t)src
->u
[3];
4342 micro_i2i64(union tgsi_double_channel
*dst
,
4343 const union tgsi_exec_channel
*src
)
4345 dst
->i64
[0] = (int64_t)src
->i
[0];
4346 dst
->i64
[1] = (int64_t)src
->i
[1];
4347 dst
->i64
[2] = (int64_t)src
->i
[2];
4348 dst
->i64
[3] = (int64_t)src
->i
[3];
4352 micro_d2u64(union tgsi_double_channel
*dst
,
4353 const union tgsi_double_channel
*src
)
4355 dst
->u64
[0] = (uint64_t)src
->d
[0];
4356 dst
->u64
[1] = (uint64_t)src
->d
[1];
4357 dst
->u64
[2] = (uint64_t)src
->d
[2];
4358 dst
->u64
[3] = (uint64_t)src
->d
[3];
4362 micro_d2i64(union tgsi_double_channel
*dst
,
4363 const union tgsi_double_channel
*src
)
4365 dst
->i64
[0] = (int64_t)src
->d
[0];
4366 dst
->i64
[1] = (int64_t)src
->d
[1];
4367 dst
->i64
[2] = (int64_t)src
->d
[2];
4368 dst
->i64
[3] = (int64_t)src
->d
[3];
4372 micro_u642d(union tgsi_double_channel
*dst
,
4373 const union tgsi_double_channel
*src
)
4375 dst
->d
[0] = (double)src
->u64
[0];
4376 dst
->d
[1] = (double)src
->u64
[1];
4377 dst
->d
[2] = (double)src
->u64
[2];
4378 dst
->d
[3] = (double)src
->u64
[3];
4382 micro_i642d(union tgsi_double_channel
*dst
,
4383 const union tgsi_double_channel
*src
)
4385 dst
->d
[0] = (double)src
->i64
[0];
4386 dst
->d
[1] = (double)src
->i64
[1];
4387 dst
->d
[2] = (double)src
->i64
[2];
4388 dst
->d
[3] = (double)src
->i64
[3];
4392 micro_u642f(union tgsi_exec_channel
*dst
,
4393 const union tgsi_double_channel
*src
)
4395 dst
->f
[0] = (float)src
->u64
[0];
4396 dst
->f
[1] = (float)src
->u64
[1];
4397 dst
->f
[2] = (float)src
->u64
[2];
4398 dst
->f
[3] = (float)src
->u64
[3];
4402 micro_i642f(union tgsi_exec_channel
*dst
,
4403 const union tgsi_double_channel
*src
)
4405 dst
->f
[0] = (float)src
->i64
[0];
4406 dst
->f
[1] = (float)src
->i64
[1];
4407 dst
->f
[2] = (float)src
->i64
[2];
4408 dst
->f
[3] = (float)src
->i64
[3];
4412 exec_t_2_64(struct tgsi_exec_machine
*mach
,
4413 const struct tgsi_full_instruction
*inst
,
4415 enum tgsi_exec_datatype src_datatype
)
4417 union tgsi_exec_channel src
;
4418 union tgsi_double_channel dst
;
4420 if ((inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XY
) == TGSI_WRITEMASK_XY
) {
4421 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, src_datatype
);
4423 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_CHAN_Y
);
4425 if ((inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_ZW
) == TGSI_WRITEMASK_ZW
) {
4426 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_Y
, src_datatype
);
4428 store_double_channel(mach
, &dst
, &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_CHAN_W
);
4433 exec_64_2_t(struct tgsi_exec_machine
*mach
,
4434 const struct tgsi_full_instruction
*inst
,
4436 enum tgsi_exec_datatype dst_datatype
)
4438 union tgsi_double_channel src
;
4439 union tgsi_exec_channel dst
;
4440 int wm
= inst
->Dst
[0].Register
.WriteMask
;
4443 for (i
= 0; i
< 2; i
++) {
4446 wm
&= ~(1 << (bit
- 1));
4448 fetch_double_channel(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_CHAN_Y
);
4450 fetch_double_channel(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_CHAN_W
);
4452 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, bit
- 1, dst_datatype
);
4458 micro_i2f(union tgsi_exec_channel
*dst
,
4459 const union tgsi_exec_channel
*src
)
4461 dst
->f
[0] = (float)src
->i
[0];
4462 dst
->f
[1] = (float)src
->i
[1];
4463 dst
->f
[2] = (float)src
->i
[2];
4464 dst
->f
[3] = (float)src
->i
[3];
4468 micro_not(union tgsi_exec_channel
*dst
,
4469 const union tgsi_exec_channel
*src
)
4471 dst
->u
[0] = ~src
->u
[0];
4472 dst
->u
[1] = ~src
->u
[1];
4473 dst
->u
[2] = ~src
->u
[2];
4474 dst
->u
[3] = ~src
->u
[3];
4478 micro_shl(union tgsi_exec_channel
*dst
,
4479 const union tgsi_exec_channel
*src0
,
4480 const union tgsi_exec_channel
*src1
)
4482 unsigned masked_count
;
4483 masked_count
= src1
->u
[0] & 0x1f;
4484 dst
->u
[0] = src0
->u
[0] << masked_count
;
4485 masked_count
= src1
->u
[1] & 0x1f;
4486 dst
->u
[1] = src0
->u
[1] << masked_count
;
4487 masked_count
= src1
->u
[2] & 0x1f;
4488 dst
->u
[2] = src0
->u
[2] << masked_count
;
4489 masked_count
= src1
->u
[3] & 0x1f;
4490 dst
->u
[3] = src0
->u
[3] << masked_count
;
4494 micro_and(union tgsi_exec_channel
*dst
,
4495 const union tgsi_exec_channel
*src0
,
4496 const union tgsi_exec_channel
*src1
)
4498 dst
->u
[0] = src0
->u
[0] & src1
->u
[0];
4499 dst
->u
[1] = src0
->u
[1] & src1
->u
[1];
4500 dst
->u
[2] = src0
->u
[2] & src1
->u
[2];
4501 dst
->u
[3] = src0
->u
[3] & src1
->u
[3];
4505 micro_or(union tgsi_exec_channel
*dst
,
4506 const union tgsi_exec_channel
*src0
,
4507 const union tgsi_exec_channel
*src1
)
4509 dst
->u
[0] = src0
->u
[0] | src1
->u
[0];
4510 dst
->u
[1] = src0
->u
[1] | src1
->u
[1];
4511 dst
->u
[2] = src0
->u
[2] | src1
->u
[2];
4512 dst
->u
[3] = src0
->u
[3] | src1
->u
[3];
4516 micro_xor(union tgsi_exec_channel
*dst
,
4517 const union tgsi_exec_channel
*src0
,
4518 const union tgsi_exec_channel
*src1
)
4520 dst
->u
[0] = src0
->u
[0] ^ src1
->u
[0];
4521 dst
->u
[1] = src0
->u
[1] ^ src1
->u
[1];
4522 dst
->u
[2] = src0
->u
[2] ^ src1
->u
[2];
4523 dst
->u
[3] = src0
->u
[3] ^ src1
->u
[3];
4527 micro_mod(union tgsi_exec_channel
*dst
,
4528 const union tgsi_exec_channel
*src0
,
4529 const union tgsi_exec_channel
*src1
)
4531 dst
->i
[0] = src1
->i
[0] ? src0
->i
[0] % src1
->i
[0] : ~0;
4532 dst
->i
[1] = src1
->i
[1] ? src0
->i
[1] % src1
->i
[1] : ~0;
4533 dst
->i
[2] = src1
->i
[2] ? src0
->i
[2] % src1
->i
[2] : ~0;
4534 dst
->i
[3] = src1
->i
[3] ? src0
->i
[3] % src1
->i
[3] : ~0;
4538 micro_f2i(union tgsi_exec_channel
*dst
,
4539 const union tgsi_exec_channel
*src
)
4541 dst
->i
[0] = (int)src
->f
[0];
4542 dst
->i
[1] = (int)src
->f
[1];
4543 dst
->i
[2] = (int)src
->f
[2];
4544 dst
->i
[3] = (int)src
->f
[3];
4548 micro_fseq(union tgsi_exec_channel
*dst
,
4549 const union tgsi_exec_channel
*src0
,
4550 const union tgsi_exec_channel
*src1
)
4552 dst
->u
[0] = src0
->f
[0] == src1
->f
[0] ? ~0 : 0;
4553 dst
->u
[1] = src0
->f
[1] == src1
->f
[1] ? ~0 : 0;
4554 dst
->u
[2] = src0
->f
[2] == src1
->f
[2] ? ~0 : 0;
4555 dst
->u
[3] = src0
->f
[3] == src1
->f
[3] ? ~0 : 0;
4559 micro_fsge(union tgsi_exec_channel
*dst
,
4560 const union tgsi_exec_channel
*src0
,
4561 const union tgsi_exec_channel
*src1
)
4563 dst
->u
[0] = src0
->f
[0] >= src1
->f
[0] ? ~0 : 0;
4564 dst
->u
[1] = src0
->f
[1] >= src1
->f
[1] ? ~0 : 0;
4565 dst
->u
[2] = src0
->f
[2] >= src1
->f
[2] ? ~0 : 0;
4566 dst
->u
[3] = src0
->f
[3] >= src1
->f
[3] ? ~0 : 0;
4570 micro_fslt(union tgsi_exec_channel
*dst
,
4571 const union tgsi_exec_channel
*src0
,
4572 const union tgsi_exec_channel
*src1
)
4574 dst
->u
[0] = src0
->f
[0] < src1
->f
[0] ? ~0 : 0;
4575 dst
->u
[1] = src0
->f
[1] < src1
->f
[1] ? ~0 : 0;
4576 dst
->u
[2] = src0
->f
[2] < src1
->f
[2] ? ~0 : 0;
4577 dst
->u
[3] = src0
->f
[3] < src1
->f
[3] ? ~0 : 0;
4581 micro_fsne(union tgsi_exec_channel
*dst
,
4582 const union tgsi_exec_channel
*src0
,
4583 const union tgsi_exec_channel
*src1
)
4585 dst
->u
[0] = src0
->f
[0] != src1
->f
[0] ? ~0 : 0;
4586 dst
->u
[1] = src0
->f
[1] != src1
->f
[1] ? ~0 : 0;
4587 dst
->u
[2] = src0
->f
[2] != src1
->f
[2] ? ~0 : 0;
4588 dst
->u
[3] = src0
->f
[3] != src1
->f
[3] ? ~0 : 0;
4592 micro_idiv(union tgsi_exec_channel
*dst
,
4593 const union tgsi_exec_channel
*src0
,
4594 const union tgsi_exec_channel
*src1
)
4596 dst
->i
[0] = src1
->i
[0] ? src0
->i
[0] / src1
->i
[0] : 0;
4597 dst
->i
[1] = src1
->i
[1] ? src0
->i
[1] / src1
->i
[1] : 0;
4598 dst
->i
[2] = src1
->i
[2] ? src0
->i
[2] / src1
->i
[2] : 0;
4599 dst
->i
[3] = src1
->i
[3] ? src0
->i
[3] / src1
->i
[3] : 0;
4603 micro_imax(union tgsi_exec_channel
*dst
,
4604 const union tgsi_exec_channel
*src0
,
4605 const union tgsi_exec_channel
*src1
)
4607 dst
->i
[0] = src0
->i
[0] > src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
4608 dst
->i
[1] = src0
->i
[1] > src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
4609 dst
->i
[2] = src0
->i
[2] > src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
4610 dst
->i
[3] = src0
->i
[3] > src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
4614 micro_imin(union tgsi_exec_channel
*dst
,
4615 const union tgsi_exec_channel
*src0
,
4616 const union tgsi_exec_channel
*src1
)
4618 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
4619 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
4620 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
4621 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
4625 micro_isge(union tgsi_exec_channel
*dst
,
4626 const union tgsi_exec_channel
*src0
,
4627 const union tgsi_exec_channel
*src1
)
4629 dst
->i
[0] = src0
->i
[0] >= src1
->i
[0] ? -1 : 0;
4630 dst
->i
[1] = src0
->i
[1] >= src1
->i
[1] ? -1 : 0;
4631 dst
->i
[2] = src0
->i
[2] >= src1
->i
[2] ? -1 : 0;
4632 dst
->i
[3] = src0
->i
[3] >= src1
->i
[3] ? -1 : 0;
4636 micro_ishr(union tgsi_exec_channel
*dst
,
4637 const union tgsi_exec_channel
*src0
,
4638 const union tgsi_exec_channel
*src1
)
4640 unsigned masked_count
;
4641 masked_count
= src1
->i
[0] & 0x1f;
4642 dst
->i
[0] = src0
->i
[0] >> masked_count
;
4643 masked_count
= src1
->i
[1] & 0x1f;
4644 dst
->i
[1] = src0
->i
[1] >> masked_count
;
4645 masked_count
= src1
->i
[2] & 0x1f;
4646 dst
->i
[2] = src0
->i
[2] >> masked_count
;
4647 masked_count
= src1
->i
[3] & 0x1f;
4648 dst
->i
[3] = src0
->i
[3] >> masked_count
;
4652 micro_islt(union tgsi_exec_channel
*dst
,
4653 const union tgsi_exec_channel
*src0
,
4654 const union tgsi_exec_channel
*src1
)
4656 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? -1 : 0;
4657 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? -1 : 0;
4658 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? -1 : 0;
4659 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? -1 : 0;
4663 micro_f2u(union tgsi_exec_channel
*dst
,
4664 const union tgsi_exec_channel
*src
)
4666 dst
->u
[0] = (uint
)src
->f
[0];
4667 dst
->u
[1] = (uint
)src
->f
[1];
4668 dst
->u
[2] = (uint
)src
->f
[2];
4669 dst
->u
[3] = (uint
)src
->f
[3];
4673 micro_u2f(union tgsi_exec_channel
*dst
,
4674 const union tgsi_exec_channel
*src
)
4676 dst
->f
[0] = (float)src
->u
[0];
4677 dst
->f
[1] = (float)src
->u
[1];
4678 dst
->f
[2] = (float)src
->u
[2];
4679 dst
->f
[3] = (float)src
->u
[3];
4683 micro_uadd(union tgsi_exec_channel
*dst
,
4684 const union tgsi_exec_channel
*src0
,
4685 const union tgsi_exec_channel
*src1
)
4687 dst
->u
[0] = src0
->u
[0] + src1
->u
[0];
4688 dst
->u
[1] = src0
->u
[1] + src1
->u
[1];
4689 dst
->u
[2] = src0
->u
[2] + src1
->u
[2];
4690 dst
->u
[3] = src0
->u
[3] + src1
->u
[3];
4694 micro_udiv(union tgsi_exec_channel
*dst
,
4695 const union tgsi_exec_channel
*src0
,
4696 const union tgsi_exec_channel
*src1
)
4698 dst
->u
[0] = src1
->u
[0] ? src0
->u
[0] / src1
->u
[0] : ~0u;
4699 dst
->u
[1] = src1
->u
[1] ? src0
->u
[1] / src1
->u
[1] : ~0u;
4700 dst
->u
[2] = src1
->u
[2] ? src0
->u
[2] / src1
->u
[2] : ~0u;
4701 dst
->u
[3] = src1
->u
[3] ? src0
->u
[3] / src1
->u
[3] : ~0u;
4705 micro_umad(union tgsi_exec_channel
*dst
,
4706 const union tgsi_exec_channel
*src0
,
4707 const union tgsi_exec_channel
*src1
,
4708 const union tgsi_exec_channel
*src2
)
4710 dst
->u
[0] = src0
->u
[0] * src1
->u
[0] + src2
->u
[0];
4711 dst
->u
[1] = src0
->u
[1] * src1
->u
[1] + src2
->u
[1];
4712 dst
->u
[2] = src0
->u
[2] * src1
->u
[2] + src2
->u
[2];
4713 dst
->u
[3] = src0
->u
[3] * src1
->u
[3] + src2
->u
[3];
4717 micro_umax(union tgsi_exec_channel
*dst
,
4718 const union tgsi_exec_channel
*src0
,
4719 const union tgsi_exec_channel
*src1
)
4721 dst
->u
[0] = src0
->u
[0] > src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
4722 dst
->u
[1] = src0
->u
[1] > src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
4723 dst
->u
[2] = src0
->u
[2] > src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
4724 dst
->u
[3] = src0
->u
[3] > src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
4728 micro_umin(union tgsi_exec_channel
*dst
,
4729 const union tgsi_exec_channel
*src0
,
4730 const union tgsi_exec_channel
*src1
)
4732 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
4733 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
4734 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
4735 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
4739 micro_umod(union tgsi_exec_channel
*dst
,
4740 const union tgsi_exec_channel
*src0
,
4741 const union tgsi_exec_channel
*src1
)
4743 dst
->u
[0] = src1
->u
[0] ? src0
->u
[0] % src1
->u
[0] : ~0u;
4744 dst
->u
[1] = src1
->u
[1] ? src0
->u
[1] % src1
->u
[1] : ~0u;
4745 dst
->u
[2] = src1
->u
[2] ? src0
->u
[2] % src1
->u
[2] : ~0u;
4746 dst
->u
[3] = src1
->u
[3] ? src0
->u
[3] % src1
->u
[3] : ~0u;
4750 micro_umul(union tgsi_exec_channel
*dst
,
4751 const union tgsi_exec_channel
*src0
,
4752 const union tgsi_exec_channel
*src1
)
4754 dst
->u
[0] = src0
->u
[0] * src1
->u
[0];
4755 dst
->u
[1] = src0
->u
[1] * src1
->u
[1];
4756 dst
->u
[2] = src0
->u
[2] * src1
->u
[2];
4757 dst
->u
[3] = src0
->u
[3] * src1
->u
[3];
4761 micro_imul_hi(union tgsi_exec_channel
*dst
,
4762 const union tgsi_exec_channel
*src0
,
4763 const union tgsi_exec_channel
*src1
)
4765 #define I64M(x, y) ((((int64_t)x) * ((int64_t)y)) >> 32)
4766 dst
->i
[0] = I64M(src0
->i
[0], src1
->i
[0]);
4767 dst
->i
[1] = I64M(src0
->i
[1], src1
->i
[1]);
4768 dst
->i
[2] = I64M(src0
->i
[2], src1
->i
[2]);
4769 dst
->i
[3] = I64M(src0
->i
[3], src1
->i
[3]);
4774 micro_umul_hi(union tgsi_exec_channel
*dst
,
4775 const union tgsi_exec_channel
*src0
,
4776 const union tgsi_exec_channel
*src1
)
4778 #define U64M(x, y) ((((uint64_t)x) * ((uint64_t)y)) >> 32)
4779 dst
->u
[0] = U64M(src0
->u
[0], src1
->u
[0]);
4780 dst
->u
[1] = U64M(src0
->u
[1], src1
->u
[1]);
4781 dst
->u
[2] = U64M(src0
->u
[2], src1
->u
[2]);
4782 dst
->u
[3] = U64M(src0
->u
[3], src1
->u
[3]);
4787 micro_useq(union tgsi_exec_channel
*dst
,
4788 const union tgsi_exec_channel
*src0
,
4789 const union tgsi_exec_channel
*src1
)
4791 dst
->u
[0] = src0
->u
[0] == src1
->u
[0] ? ~0 : 0;
4792 dst
->u
[1] = src0
->u
[1] == src1
->u
[1] ? ~0 : 0;
4793 dst
->u
[2] = src0
->u
[2] == src1
->u
[2] ? ~0 : 0;
4794 dst
->u
[3] = src0
->u
[3] == src1
->u
[3] ? ~0 : 0;
4798 micro_usge(union tgsi_exec_channel
*dst
,
4799 const union tgsi_exec_channel
*src0
,
4800 const union tgsi_exec_channel
*src1
)
4802 dst
->u
[0] = src0
->u
[0] >= src1
->u
[0] ? ~0 : 0;
4803 dst
->u
[1] = src0
->u
[1] >= src1
->u
[1] ? ~0 : 0;
4804 dst
->u
[2] = src0
->u
[2] >= src1
->u
[2] ? ~0 : 0;
4805 dst
->u
[3] = src0
->u
[3] >= src1
->u
[3] ? ~0 : 0;
4809 micro_ushr(union tgsi_exec_channel
*dst
,
4810 const union tgsi_exec_channel
*src0
,
4811 const union tgsi_exec_channel
*src1
)
4813 unsigned masked_count
;
4814 masked_count
= src1
->u
[0] & 0x1f;
4815 dst
->u
[0] = src0
->u
[0] >> masked_count
;
4816 masked_count
= src1
->u
[1] & 0x1f;
4817 dst
->u
[1] = src0
->u
[1] >> masked_count
;
4818 masked_count
= src1
->u
[2] & 0x1f;
4819 dst
->u
[2] = src0
->u
[2] >> masked_count
;
4820 masked_count
= src1
->u
[3] & 0x1f;
4821 dst
->u
[3] = src0
->u
[3] >> masked_count
;
4825 micro_uslt(union tgsi_exec_channel
*dst
,
4826 const union tgsi_exec_channel
*src0
,
4827 const union tgsi_exec_channel
*src1
)
4829 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? ~0 : 0;
4830 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? ~0 : 0;
4831 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? ~0 : 0;
4832 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? ~0 : 0;
4836 micro_usne(union tgsi_exec_channel
*dst
,
4837 const union tgsi_exec_channel
*src0
,
4838 const union tgsi_exec_channel
*src1
)
4840 dst
->u
[0] = src0
->u
[0] != src1
->u
[0] ? ~0 : 0;
4841 dst
->u
[1] = src0
->u
[1] != src1
->u
[1] ? ~0 : 0;
4842 dst
->u
[2] = src0
->u
[2] != src1
->u
[2] ? ~0 : 0;
4843 dst
->u
[3] = src0
->u
[3] != src1
->u
[3] ? ~0 : 0;
4847 micro_uarl(union tgsi_exec_channel
*dst
,
4848 const union tgsi_exec_channel
*src
)
4850 dst
->i
[0] = src
->u
[0];
4851 dst
->i
[1] = src
->u
[1];
4852 dst
->i
[2] = src
->u
[2];
4853 dst
->i
[3] = src
->u
[3];
4857 * Signed bitfield extract (i.e. sign-extend the extracted bits)
4860 micro_ibfe(union tgsi_exec_channel
*dst
,
4861 const union tgsi_exec_channel
*src0
,
4862 const union tgsi_exec_channel
*src1
,
4863 const union tgsi_exec_channel
*src2
)
4866 for (i
= 0; i
< 4; i
++) {
4867 int width
= src2
->i
[i
] & 0x1f;
4868 int offset
= src1
->i
[i
] & 0x1f;
4871 else if (width
+ offset
< 32)
4872 dst
->i
[i
] = (src0
->i
[i
] << (32 - width
- offset
)) >> (32 - width
);
4874 dst
->i
[i
] = src0
->i
[i
] >> offset
;
4879 * Unsigned bitfield extract
4882 micro_ubfe(union tgsi_exec_channel
*dst
,
4883 const union tgsi_exec_channel
*src0
,
4884 const union tgsi_exec_channel
*src1
,
4885 const union tgsi_exec_channel
*src2
)
4888 for (i
= 0; i
< 4; i
++) {
4889 int width
= src2
->u
[i
] & 0x1f;
4890 int offset
= src1
->u
[i
] & 0x1f;
4893 else if (width
+ offset
< 32)
4894 dst
->u
[i
] = (src0
->u
[i
] << (32 - width
- offset
)) >> (32 - width
);
4896 dst
->u
[i
] = src0
->u
[i
] >> offset
;
4901 * Bitfield insert: copy low bits from src1 into a region of src0.
4904 micro_bfi(union tgsi_exec_channel
*dst
,
4905 const union tgsi_exec_channel
*src0
,
4906 const union tgsi_exec_channel
*src1
,
4907 const union tgsi_exec_channel
*src2
,
4908 const union tgsi_exec_channel
*src3
)
4911 for (i
= 0; i
< 4; i
++) {
4912 int width
= src3
->u
[i
] & 0x1f;
4913 int offset
= src2
->u
[i
] & 0x1f;
4914 int bitmask
= ((1 << width
) - 1) << offset
;
4915 dst
->u
[i
] = ((src1
->u
[i
] << offset
) & bitmask
) | (src0
->u
[i
] & ~bitmask
);
4920 micro_brev(union tgsi_exec_channel
*dst
,
4921 const union tgsi_exec_channel
*src
)
4923 dst
->u
[0] = util_bitreverse(src
->u
[0]);
4924 dst
->u
[1] = util_bitreverse(src
->u
[1]);
4925 dst
->u
[2] = util_bitreverse(src
->u
[2]);
4926 dst
->u
[3] = util_bitreverse(src
->u
[3]);
4930 micro_popc(union tgsi_exec_channel
*dst
,
4931 const union tgsi_exec_channel
*src
)
4933 dst
->u
[0] = util_bitcount(src
->u
[0]);
4934 dst
->u
[1] = util_bitcount(src
->u
[1]);
4935 dst
->u
[2] = util_bitcount(src
->u
[2]);
4936 dst
->u
[3] = util_bitcount(src
->u
[3]);
4940 micro_lsb(union tgsi_exec_channel
*dst
,
4941 const union tgsi_exec_channel
*src
)
4943 dst
->i
[0] = ffs(src
->u
[0]) - 1;
4944 dst
->i
[1] = ffs(src
->u
[1]) - 1;
4945 dst
->i
[2] = ffs(src
->u
[2]) - 1;
4946 dst
->i
[3] = ffs(src
->u
[3]) - 1;
4950 micro_imsb(union tgsi_exec_channel
*dst
,
4951 const union tgsi_exec_channel
*src
)
4953 dst
->i
[0] = util_last_bit_signed(src
->i
[0]) - 1;
4954 dst
->i
[1] = util_last_bit_signed(src
->i
[1]) - 1;
4955 dst
->i
[2] = util_last_bit_signed(src
->i
[2]) - 1;
4956 dst
->i
[3] = util_last_bit_signed(src
->i
[3]) - 1;
4960 micro_umsb(union tgsi_exec_channel
*dst
,
4961 const union tgsi_exec_channel
*src
)
4963 dst
->i
[0] = util_last_bit(src
->u
[0]) - 1;
4964 dst
->i
[1] = util_last_bit(src
->u
[1]) - 1;
4965 dst
->i
[2] = util_last_bit(src
->u
[2]) - 1;
4966 dst
->i
[3] = util_last_bit(src
->u
[3]) - 1;
4970 * Execute a TGSI instruction.
4971 * Returns TRUE if a barrier instruction is hit,
4976 struct tgsi_exec_machine
*mach
,
4977 const struct tgsi_full_instruction
*inst
,
4980 union tgsi_exec_channel r
[10];
4984 switch (inst
->Instruction
.Opcode
) {
4985 case TGSI_OPCODE_ARL
:
4986 exec_vector_unary(mach
, inst
, micro_arl
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
4989 case TGSI_OPCODE_MOV
:
4990 exec_vector_unary(mach
, inst
, micro_mov
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4993 case TGSI_OPCODE_LIT
:
4994 exec_lit(mach
, inst
);
4997 case TGSI_OPCODE_RCP
:
4998 exec_scalar_unary(mach
, inst
, micro_rcp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5001 case TGSI_OPCODE_RSQ
:
5002 exec_scalar_unary(mach
, inst
, micro_rsq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5005 case TGSI_OPCODE_EXP
:
5006 exec_exp(mach
, inst
);
5009 case TGSI_OPCODE_LOG
:
5010 exec_log(mach
, inst
);
5013 case TGSI_OPCODE_MUL
:
5014 exec_vector_binary(mach
, inst
, micro_mul
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5017 case TGSI_OPCODE_ADD
:
5018 exec_vector_binary(mach
, inst
, micro_add
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5021 case TGSI_OPCODE_DP3
:
5022 exec_dp3(mach
, inst
);
5025 case TGSI_OPCODE_DP4
:
5026 exec_dp4(mach
, inst
);
5029 case TGSI_OPCODE_DST
:
5030 exec_dst(mach
, inst
);
5033 case TGSI_OPCODE_MIN
:
5034 exec_vector_binary(mach
, inst
, micro_min
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5037 case TGSI_OPCODE_MAX
:
5038 exec_vector_binary(mach
, inst
, micro_max
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5041 case TGSI_OPCODE_SLT
:
5042 exec_vector_binary(mach
, inst
, micro_slt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5045 case TGSI_OPCODE_SGE
:
5046 exec_vector_binary(mach
, inst
, micro_sge
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5049 case TGSI_OPCODE_MAD
:
5050 exec_vector_trinary(mach
, inst
, micro_mad
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5053 case TGSI_OPCODE_LRP
:
5054 exec_vector_trinary(mach
, inst
, micro_lrp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5057 case TGSI_OPCODE_SQRT
:
5058 exec_scalar_unary(mach
, inst
, micro_sqrt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5061 case TGSI_OPCODE_FRC
:
5062 exec_vector_unary(mach
, inst
, micro_frc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5065 case TGSI_OPCODE_FLR
:
5066 exec_vector_unary(mach
, inst
, micro_flr
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5069 case TGSI_OPCODE_ROUND
:
5070 exec_vector_unary(mach
, inst
, micro_rnd
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5073 case TGSI_OPCODE_EX2
:
5074 exec_scalar_unary(mach
, inst
, micro_exp2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5077 case TGSI_OPCODE_LG2
:
5078 exec_scalar_unary(mach
, inst
, micro_lg2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5081 case TGSI_OPCODE_POW
:
5082 exec_scalar_binary(mach
, inst
, micro_pow
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5085 case TGSI_OPCODE_COS
:
5086 exec_scalar_unary(mach
, inst
, micro_cos
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5089 case TGSI_OPCODE_DDX
:
5090 exec_vector_unary(mach
, inst
, micro_ddx
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5093 case TGSI_OPCODE_DDY
:
5094 exec_vector_unary(mach
, inst
, micro_ddy
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5097 case TGSI_OPCODE_KILL
:
5098 exec_kill (mach
, inst
);
5101 case TGSI_OPCODE_KILL_IF
:
5102 exec_kill_if (mach
, inst
);
5105 case TGSI_OPCODE_PK2H
:
5106 exec_pk2h(mach
, inst
);
5109 case TGSI_OPCODE_PK2US
:
5113 case TGSI_OPCODE_PK4B
:
5117 case TGSI_OPCODE_PK4UB
:
5121 case TGSI_OPCODE_SEQ
:
5122 exec_vector_binary(mach
, inst
, micro_seq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5125 case TGSI_OPCODE_SGT
:
5126 exec_vector_binary(mach
, inst
, micro_sgt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5129 case TGSI_OPCODE_SIN
:
5130 exec_scalar_unary(mach
, inst
, micro_sin
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5133 case TGSI_OPCODE_SLE
:
5134 exec_vector_binary(mach
, inst
, micro_sle
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5137 case TGSI_OPCODE_SNE
:
5138 exec_vector_binary(mach
, inst
, micro_sne
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5141 case TGSI_OPCODE_TEX
:
5142 /* simple texture lookup */
5143 /* src[0] = texcoord */
5144 /* src[1] = sampler unit */
5145 exec_tex(mach
, inst
, TEX_MODIFIER_NONE
, 1);
5148 case TGSI_OPCODE_TXB
:
5149 /* Texture lookup with lod bias */
5150 /* src[0] = texcoord (src[0].w = LOD bias) */
5151 /* src[1] = sampler unit */
5152 exec_tex(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, 1);
5155 case TGSI_OPCODE_TXD
:
5156 /* Texture lookup with explict partial derivatives */
5157 /* src[0] = texcoord */
5158 /* src[1] = d[strq]/dx */
5159 /* src[2] = d[strq]/dy */
5160 /* src[3] = sampler unit */
5161 exec_txd(mach
, inst
);
5164 case TGSI_OPCODE_TXL
:
5165 /* Texture lookup with explit LOD */
5166 /* src[0] = texcoord (src[0].w = LOD) */
5167 /* src[1] = sampler unit */
5168 exec_tex(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, 1);
5171 case TGSI_OPCODE_TXP
:
5172 /* Texture lookup with projection */
5173 /* src[0] = texcoord (src[0].w = projection) */
5174 /* src[1] = sampler unit */
5175 exec_tex(mach
, inst
, TEX_MODIFIER_PROJECTED
, 1);
5178 case TGSI_OPCODE_TG4
:
5179 /* src[0] = texcoord */
5180 /* src[1] = component */
5181 /* src[2] = sampler unit */
5182 exec_tex(mach
, inst
, TEX_MODIFIER_GATHER
, 2);
5185 case TGSI_OPCODE_LODQ
:
5186 /* src[0] = texcoord */
5187 /* src[1] = sampler unit */
5188 exec_lodq(mach
, inst
);
5191 case TGSI_OPCODE_UP2H
:
5192 exec_up2h(mach
, inst
);
5195 case TGSI_OPCODE_UP2US
:
5199 case TGSI_OPCODE_UP4B
:
5203 case TGSI_OPCODE_UP4UB
:
5207 case TGSI_OPCODE_ARR
:
5208 exec_vector_unary(mach
, inst
, micro_arr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
5211 case TGSI_OPCODE_CAL
:
5212 /* skip the call if no execution channels are enabled */
5213 if (mach
->ExecMask
) {
5216 /* First, record the depths of the execution stacks.
5217 * This is important for deeply nested/looped return statements.
5218 * We have to unwind the stacks by the correct amount. For a
5219 * real code generator, we could determine the number of entries
5220 * to pop off each stack with simple static analysis and avoid
5221 * implementing this data structure at run time.
5223 mach
->CallStack
[mach
->CallStackTop
].CondStackTop
= mach
->CondStackTop
;
5224 mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
= mach
->LoopStackTop
;
5225 mach
->CallStack
[mach
->CallStackTop
].ContStackTop
= mach
->ContStackTop
;
5226 mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
= mach
->SwitchStackTop
;
5227 mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
= mach
->BreakStackTop
;
5228 /* note that PC was already incremented above */
5229 mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
= *pc
;
5231 mach
->CallStackTop
++;
5233 /* Second, push the Cond, Loop, Cont, Func stacks */
5234 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
5235 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
5236 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
5237 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
5238 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
5239 assert(mach
->FuncStackTop
< TGSI_EXEC_MAX_CALL_NESTING
);
5241 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
5242 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
5243 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
5244 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
5245 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
5246 mach
->FuncStack
[mach
->FuncStackTop
++] = mach
->FuncMask
;
5248 /* Finally, jump to the subroutine. The label is a pointer
5249 * (an instruction number) to the BGNSUB instruction.
5251 *pc
= inst
->Label
.Label
;
5252 assert(mach
->Instructions
[*pc
].Instruction
.Opcode
5253 == TGSI_OPCODE_BGNSUB
);
5257 case TGSI_OPCODE_RET
:
5258 mach
->FuncMask
&= ~mach
->ExecMask
;
5259 UPDATE_EXEC_MASK(mach
);
5261 if (mach
->FuncMask
== 0x0) {
5262 /* really return now (otherwise, keep executing */
5264 if (mach
->CallStackTop
== 0) {
5265 /* returning from main() */
5266 mach
->CondStackTop
= 0;
5267 mach
->LoopStackTop
= 0;
5268 mach
->ContStackTop
= 0;
5269 mach
->LoopLabelStackTop
= 0;
5270 mach
->SwitchStackTop
= 0;
5271 mach
->BreakStackTop
= 0;
5276 assert(mach
->CallStackTop
> 0);
5277 mach
->CallStackTop
--;
5279 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
5280 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
5282 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
5283 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
5285 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
5286 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
5288 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
5289 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
5291 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
5292 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
5294 assert(mach
->FuncStackTop
> 0);
5295 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
5297 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
5299 UPDATE_EXEC_MASK(mach
);
5303 case TGSI_OPCODE_SSG
:
5304 exec_vector_unary(mach
, inst
, micro_sgn
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5307 case TGSI_OPCODE_CMP
:
5308 exec_vector_trinary(mach
, inst
, micro_cmp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5311 case TGSI_OPCODE_DIV
:
5312 exec_vector_binary(mach
, inst
, micro_div
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5315 case TGSI_OPCODE_DP2
:
5316 exec_dp2(mach
, inst
);
5319 case TGSI_OPCODE_IF
:
5321 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
5322 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
5323 FETCH( &r
[0], 0, TGSI_CHAN_X
);
5324 /* update CondMask */
5326 mach
->CondMask
&= ~0x1;
5329 mach
->CondMask
&= ~0x2;
5332 mach
->CondMask
&= ~0x4;
5335 mach
->CondMask
&= ~0x8;
5337 UPDATE_EXEC_MASK(mach
);
5338 /* Todo: If CondMask==0, jump to ELSE */
5341 case TGSI_OPCODE_UIF
:
5343 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
5344 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
5345 IFETCH( &r
[0], 0, TGSI_CHAN_X
);
5346 /* update CondMask */
5348 mach
->CondMask
&= ~0x1;
5351 mach
->CondMask
&= ~0x2;
5354 mach
->CondMask
&= ~0x4;
5357 mach
->CondMask
&= ~0x8;
5359 UPDATE_EXEC_MASK(mach
);
5360 /* Todo: If CondMask==0, jump to ELSE */
5363 case TGSI_OPCODE_ELSE
:
5364 /* invert CondMask wrt previous mask */
5367 assert(mach
->CondStackTop
> 0);
5368 prevMask
= mach
->CondStack
[mach
->CondStackTop
- 1];
5369 mach
->CondMask
= ~mach
->CondMask
& prevMask
;
5370 UPDATE_EXEC_MASK(mach
);
5371 /* Todo: If CondMask==0, jump to ENDIF */
5375 case TGSI_OPCODE_ENDIF
:
5377 assert(mach
->CondStackTop
> 0);
5378 mach
->CondMask
= mach
->CondStack
[--mach
->CondStackTop
];
5379 UPDATE_EXEC_MASK(mach
);
5382 case TGSI_OPCODE_END
:
5383 /* make sure we end primitives which haven't
5384 * been explicitly emitted */
5385 conditional_emit_primitive(mach
);
5386 /* halt execution */
5390 case TGSI_OPCODE_CEIL
:
5391 exec_vector_unary(mach
, inst
, micro_ceil
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5394 case TGSI_OPCODE_I2F
:
5395 exec_vector_unary(mach
, inst
, micro_i2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_INT
);
5398 case TGSI_OPCODE_NOT
:
5399 exec_vector_unary(mach
, inst
, micro_not
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5402 case TGSI_OPCODE_TRUNC
:
5403 exec_vector_unary(mach
, inst
, micro_trunc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
5406 case TGSI_OPCODE_SHL
:
5407 exec_vector_binary(mach
, inst
, micro_shl
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5410 case TGSI_OPCODE_AND
:
5411 exec_vector_binary(mach
, inst
, micro_and
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5414 case TGSI_OPCODE_OR
:
5415 exec_vector_binary(mach
, inst
, micro_or
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5418 case TGSI_OPCODE_MOD
:
5419 exec_vector_binary(mach
, inst
, micro_mod
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5422 case TGSI_OPCODE_XOR
:
5423 exec_vector_binary(mach
, inst
, micro_xor
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5426 case TGSI_OPCODE_TXF
:
5427 exec_txf(mach
, inst
);
5430 case TGSI_OPCODE_TXQ
:
5431 exec_txq(mach
, inst
);
5434 case TGSI_OPCODE_EMIT
:
5438 case TGSI_OPCODE_ENDPRIM
:
5439 emit_primitive(mach
);
5442 case TGSI_OPCODE_BGNLOOP
:
5443 /* push LoopMask and ContMasks */
5444 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
5445 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
5446 assert(mach
->LoopLabelStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
5447 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
5449 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
5450 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
5451 mach
->LoopLabelStack
[mach
->LoopLabelStackTop
++] = *pc
- 1;
5452 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
5453 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_LOOP
;
5456 case TGSI_OPCODE_ENDLOOP
:
5457 /* Restore ContMask, but don't pop */
5458 assert(mach
->ContStackTop
> 0);
5459 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
- 1];
5460 UPDATE_EXEC_MASK(mach
);
5461 if (mach
->ExecMask
) {
5462 /* repeat loop: jump to instruction just past BGNLOOP */
5463 assert(mach
->LoopLabelStackTop
> 0);
5464 *pc
= mach
->LoopLabelStack
[mach
->LoopLabelStackTop
- 1] + 1;
5467 /* exit loop: pop LoopMask */
5468 assert(mach
->LoopStackTop
> 0);
5469 mach
->LoopMask
= mach
->LoopStack
[--mach
->LoopStackTop
];
5471 assert(mach
->ContStackTop
> 0);
5472 mach
->ContMask
= mach
->ContStack
[--mach
->ContStackTop
];
5473 assert(mach
->LoopLabelStackTop
> 0);
5474 --mach
->LoopLabelStackTop
;
5476 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
5478 UPDATE_EXEC_MASK(mach
);
5481 case TGSI_OPCODE_BRK
:
5485 case TGSI_OPCODE_CONT
:
5486 /* turn off cont channels for each enabled exec channel */
5487 mach
->ContMask
&= ~mach
->ExecMask
;
5488 /* Todo: if mach->LoopMask == 0, jump to end of loop */
5489 UPDATE_EXEC_MASK(mach
);
5492 case TGSI_OPCODE_BGNSUB
:
5496 case TGSI_OPCODE_ENDSUB
:
5498 * XXX: This really should be a no-op. We should never reach this opcode.
5501 assert(mach
->CallStackTop
> 0);
5502 mach
->CallStackTop
--;
5504 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
5505 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
5507 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
5508 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
5510 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
5511 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
5513 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
5514 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
5516 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
5517 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
5519 assert(mach
->FuncStackTop
> 0);
5520 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
5522 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
5524 UPDATE_EXEC_MASK(mach
);
5527 case TGSI_OPCODE_NOP
:
5530 case TGSI_OPCODE_F2I
:
5531 exec_vector_unary(mach
, inst
, micro_f2i
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
5534 case TGSI_OPCODE_FSEQ
:
5535 exec_vector_binary(mach
, inst
, micro_fseq
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
5538 case TGSI_OPCODE_FSGE
:
5539 exec_vector_binary(mach
, inst
, micro_fsge
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
5542 case TGSI_OPCODE_FSLT
:
5543 exec_vector_binary(mach
, inst
, micro_fslt
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
5546 case TGSI_OPCODE_FSNE
:
5547 exec_vector_binary(mach
, inst
, micro_fsne
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
5550 case TGSI_OPCODE_IDIV
:
5551 exec_vector_binary(mach
, inst
, micro_idiv
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5554 case TGSI_OPCODE_IMAX
:
5555 exec_vector_binary(mach
, inst
, micro_imax
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5558 case TGSI_OPCODE_IMIN
:
5559 exec_vector_binary(mach
, inst
, micro_imin
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5562 case TGSI_OPCODE_INEG
:
5563 exec_vector_unary(mach
, inst
, micro_ineg
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5566 case TGSI_OPCODE_ISGE
:
5567 exec_vector_binary(mach
, inst
, micro_isge
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5570 case TGSI_OPCODE_ISHR
:
5571 exec_vector_binary(mach
, inst
, micro_ishr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5574 case TGSI_OPCODE_ISLT
:
5575 exec_vector_binary(mach
, inst
, micro_islt
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5578 case TGSI_OPCODE_F2U
:
5579 exec_vector_unary(mach
, inst
, micro_f2u
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
5582 case TGSI_OPCODE_U2F
:
5583 exec_vector_unary(mach
, inst
, micro_u2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_UINT
);
5586 case TGSI_OPCODE_UADD
:
5587 exec_vector_binary(mach
, inst
, micro_uadd
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5590 case TGSI_OPCODE_UDIV
:
5591 exec_vector_binary(mach
, inst
, micro_udiv
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5594 case TGSI_OPCODE_UMAD
:
5595 exec_vector_trinary(mach
, inst
, micro_umad
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5598 case TGSI_OPCODE_UMAX
:
5599 exec_vector_binary(mach
, inst
, micro_umax
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5602 case TGSI_OPCODE_UMIN
:
5603 exec_vector_binary(mach
, inst
, micro_umin
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5606 case TGSI_OPCODE_UMOD
:
5607 exec_vector_binary(mach
, inst
, micro_umod
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5610 case TGSI_OPCODE_UMUL
:
5611 exec_vector_binary(mach
, inst
, micro_umul
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5614 case TGSI_OPCODE_IMUL_HI
:
5615 exec_vector_binary(mach
, inst
, micro_imul_hi
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5618 case TGSI_OPCODE_UMUL_HI
:
5619 exec_vector_binary(mach
, inst
, micro_umul_hi
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5622 case TGSI_OPCODE_USEQ
:
5623 exec_vector_binary(mach
, inst
, micro_useq
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5626 case TGSI_OPCODE_USGE
:
5627 exec_vector_binary(mach
, inst
, micro_usge
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5630 case TGSI_OPCODE_USHR
:
5631 exec_vector_binary(mach
, inst
, micro_ushr
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5634 case TGSI_OPCODE_USLT
:
5635 exec_vector_binary(mach
, inst
, micro_uslt
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5638 case TGSI_OPCODE_USNE
:
5639 exec_vector_binary(mach
, inst
, micro_usne
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5642 case TGSI_OPCODE_SWITCH
:
5643 exec_switch(mach
, inst
);
5646 case TGSI_OPCODE_CASE
:
5647 exec_case(mach
, inst
);
5650 case TGSI_OPCODE_DEFAULT
:
5654 case TGSI_OPCODE_ENDSWITCH
:
5655 exec_endswitch(mach
);
5658 case TGSI_OPCODE_SAMPLE_I
:
5659 exec_txf(mach
, inst
);
5662 case TGSI_OPCODE_SAMPLE_I_MS
:
5663 exec_txf(mach
, inst
);
5666 case TGSI_OPCODE_SAMPLE
:
5667 exec_sample(mach
, inst
, TEX_MODIFIER_NONE
, FALSE
);
5670 case TGSI_OPCODE_SAMPLE_B
:
5671 exec_sample(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, FALSE
);
5674 case TGSI_OPCODE_SAMPLE_C
:
5675 exec_sample(mach
, inst
, TEX_MODIFIER_NONE
, TRUE
);
5678 case TGSI_OPCODE_SAMPLE_C_LZ
:
5679 exec_sample(mach
, inst
, TEX_MODIFIER_LEVEL_ZERO
, TRUE
);
5682 case TGSI_OPCODE_SAMPLE_D
:
5683 exec_sample_d(mach
, inst
);
5686 case TGSI_OPCODE_SAMPLE_L
:
5687 exec_sample(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, FALSE
);
5690 case TGSI_OPCODE_GATHER4
:
5691 exec_sample(mach
, inst
, TEX_MODIFIER_GATHER
, FALSE
);
5694 case TGSI_OPCODE_SVIEWINFO
:
5695 exec_txq(mach
, inst
);
5698 case TGSI_OPCODE_SAMPLE_POS
:
5702 case TGSI_OPCODE_SAMPLE_INFO
:
5706 case TGSI_OPCODE_UARL
:
5707 exec_vector_unary(mach
, inst
, micro_uarl
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_UINT
);
5710 case TGSI_OPCODE_UCMP
:
5711 exec_ucmp(mach
, inst
);
5714 case TGSI_OPCODE_IABS
:
5715 exec_vector_unary(mach
, inst
, micro_iabs
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5718 case TGSI_OPCODE_ISSG
:
5719 exec_vector_unary(mach
, inst
, micro_isgn
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5722 case TGSI_OPCODE_TEX2
:
5723 /* simple texture lookup */
5724 /* src[0] = texcoord */
5725 /* src[1] = compare */
5726 /* src[2] = sampler unit */
5727 exec_tex(mach
, inst
, TEX_MODIFIER_NONE
, 2);
5729 case TGSI_OPCODE_TXB2
:
5730 /* simple texture lookup */
5731 /* src[0] = texcoord */
5733 /* src[2] = sampler unit */
5734 exec_tex(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, 2);
5736 case TGSI_OPCODE_TXL2
:
5737 /* simple texture lookup */
5738 /* src[0] = texcoord */
5740 /* src[2] = sampler unit */
5741 exec_tex(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, 2);
5744 case TGSI_OPCODE_IBFE
:
5745 exec_vector_trinary(mach
, inst
, micro_ibfe
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5747 case TGSI_OPCODE_UBFE
:
5748 exec_vector_trinary(mach
, inst
, micro_ubfe
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5750 case TGSI_OPCODE_BFI
:
5751 exec_vector_quaternary(mach
, inst
, micro_bfi
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5753 case TGSI_OPCODE_BREV
:
5754 exec_vector_unary(mach
, inst
, micro_brev
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5756 case TGSI_OPCODE_POPC
:
5757 exec_vector_unary(mach
, inst
, micro_popc
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
5759 case TGSI_OPCODE_LSB
:
5760 exec_vector_unary(mach
, inst
, micro_lsb
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_UINT
);
5762 case TGSI_OPCODE_IMSB
:
5763 exec_vector_unary(mach
, inst
, micro_imsb
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
5765 case TGSI_OPCODE_UMSB
:
5766 exec_vector_unary(mach
, inst
, micro_umsb
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_UINT
);
5769 case TGSI_OPCODE_F2D
:
5770 exec_t_2_64(mach
, inst
, micro_f2d
, TGSI_EXEC_DATA_FLOAT
);
5773 case TGSI_OPCODE_D2F
:
5774 exec_64_2_t(mach
, inst
, micro_d2f
, TGSI_EXEC_DATA_FLOAT
);
5777 case TGSI_OPCODE_DABS
:
5778 exec_double_unary(mach
, inst
, micro_dabs
);
5781 case TGSI_OPCODE_DNEG
:
5782 exec_double_unary(mach
, inst
, micro_dneg
);
5785 case TGSI_OPCODE_DADD
:
5786 exec_double_binary(mach
, inst
, micro_dadd
, TGSI_EXEC_DATA_DOUBLE
);
5789 case TGSI_OPCODE_DDIV
:
5790 exec_double_binary(mach
, inst
, micro_ddiv
, TGSI_EXEC_DATA_DOUBLE
);
5793 case TGSI_OPCODE_DMUL
:
5794 exec_double_binary(mach
, inst
, micro_dmul
, TGSI_EXEC_DATA_DOUBLE
);
5797 case TGSI_OPCODE_DMAX
:
5798 exec_double_binary(mach
, inst
, micro_dmax
, TGSI_EXEC_DATA_DOUBLE
);
5801 case TGSI_OPCODE_DMIN
:
5802 exec_double_binary(mach
, inst
, micro_dmin
, TGSI_EXEC_DATA_DOUBLE
);
5805 case TGSI_OPCODE_DSLT
:
5806 exec_double_binary(mach
, inst
, micro_dslt
, TGSI_EXEC_DATA_UINT
);
5809 case TGSI_OPCODE_DSGE
:
5810 exec_double_binary(mach
, inst
, micro_dsge
, TGSI_EXEC_DATA_UINT
);
5813 case TGSI_OPCODE_DSEQ
:
5814 exec_double_binary(mach
, inst
, micro_dseq
, TGSI_EXEC_DATA_UINT
);
5817 case TGSI_OPCODE_DSNE
:
5818 exec_double_binary(mach
, inst
, micro_dsne
, TGSI_EXEC_DATA_UINT
);
5821 case TGSI_OPCODE_DRCP
:
5822 exec_double_unary(mach
, inst
, micro_drcp
);
5825 case TGSI_OPCODE_DSQRT
:
5826 exec_double_unary(mach
, inst
, micro_dsqrt
);
5829 case TGSI_OPCODE_DRSQ
:
5830 exec_double_unary(mach
, inst
, micro_drsq
);
5833 case TGSI_OPCODE_DMAD
:
5834 exec_double_trinary(mach
, inst
, micro_dmad
);
5837 case TGSI_OPCODE_DFRAC
:
5838 exec_double_unary(mach
, inst
, micro_dfrac
);
5841 case TGSI_OPCODE_DLDEXP
:
5842 exec_dldexp(mach
, inst
);
5845 case TGSI_OPCODE_DFRACEXP
:
5846 exec_dfracexp(mach
, inst
);
5849 case TGSI_OPCODE_I2D
:
5850 exec_t_2_64(mach
, inst
, micro_i2d
, TGSI_EXEC_DATA_INT
);
5853 case TGSI_OPCODE_D2I
:
5854 exec_64_2_t(mach
, inst
, micro_d2i
, TGSI_EXEC_DATA_INT
);
5857 case TGSI_OPCODE_U2D
:
5858 exec_t_2_64(mach
, inst
, micro_u2d
, TGSI_EXEC_DATA_UINT
);
5861 case TGSI_OPCODE_D2U
:
5862 exec_64_2_t(mach
, inst
, micro_d2u
, TGSI_EXEC_DATA_INT
);
5865 case TGSI_OPCODE_LOAD
:
5866 exec_load(mach
, inst
);
5869 case TGSI_OPCODE_STORE
:
5870 exec_store(mach
, inst
);
5873 case TGSI_OPCODE_ATOMUADD
:
5874 case TGSI_OPCODE_ATOMXCHG
:
5875 case TGSI_OPCODE_ATOMCAS
:
5876 case TGSI_OPCODE_ATOMAND
:
5877 case TGSI_OPCODE_ATOMOR
:
5878 case TGSI_OPCODE_ATOMXOR
:
5879 case TGSI_OPCODE_ATOMUMIN
:
5880 case TGSI_OPCODE_ATOMUMAX
:
5881 case TGSI_OPCODE_ATOMIMIN
:
5882 case TGSI_OPCODE_ATOMIMAX
:
5883 exec_atomop(mach
, inst
);
5886 case TGSI_OPCODE_RESQ
:
5887 exec_resq(mach
, inst
);
5889 case TGSI_OPCODE_BARRIER
:
5890 case TGSI_OPCODE_MEMBAR
:
5894 case TGSI_OPCODE_I64ABS
:
5895 exec_double_unary(mach
, inst
, micro_i64abs
);
5898 case TGSI_OPCODE_I64SSG
:
5899 exec_double_unary(mach
, inst
, micro_i64sgn
);
5902 case TGSI_OPCODE_I64NEG
:
5903 exec_double_unary(mach
, inst
, micro_i64neg
);
5906 case TGSI_OPCODE_U64SEQ
:
5907 exec_double_binary(mach
, inst
, micro_u64seq
, TGSI_EXEC_DATA_UINT
);
5910 case TGSI_OPCODE_U64SNE
:
5911 exec_double_binary(mach
, inst
, micro_u64sne
, TGSI_EXEC_DATA_UINT
);
5914 case TGSI_OPCODE_I64SLT
:
5915 exec_double_binary(mach
, inst
, micro_i64slt
, TGSI_EXEC_DATA_UINT
);
5917 case TGSI_OPCODE_U64SLT
:
5918 exec_double_binary(mach
, inst
, micro_u64slt
, TGSI_EXEC_DATA_UINT
);
5921 case TGSI_OPCODE_I64SGE
:
5922 exec_double_binary(mach
, inst
, micro_i64sge
, TGSI_EXEC_DATA_UINT
);
5924 case TGSI_OPCODE_U64SGE
:
5925 exec_double_binary(mach
, inst
, micro_u64sge
, TGSI_EXEC_DATA_UINT
);
5928 case TGSI_OPCODE_I64MIN
:
5929 exec_double_binary(mach
, inst
, micro_i64min
, TGSI_EXEC_DATA_INT64
);
5931 case TGSI_OPCODE_U64MIN
:
5932 exec_double_binary(mach
, inst
, micro_u64min
, TGSI_EXEC_DATA_UINT64
);
5934 case TGSI_OPCODE_I64MAX
:
5935 exec_double_binary(mach
, inst
, micro_i64max
, TGSI_EXEC_DATA_INT64
);
5937 case TGSI_OPCODE_U64MAX
:
5938 exec_double_binary(mach
, inst
, micro_u64max
, TGSI_EXEC_DATA_UINT64
);
5940 case TGSI_OPCODE_U64ADD
:
5941 exec_double_binary(mach
, inst
, micro_u64add
, TGSI_EXEC_DATA_UINT64
);
5943 case TGSI_OPCODE_U64MUL
:
5944 exec_double_binary(mach
, inst
, micro_u64mul
, TGSI_EXEC_DATA_UINT64
);
5946 case TGSI_OPCODE_U64SHL
:
5947 exec_arg0_64_arg1_32(mach
, inst
, micro_u64shl
);
5949 case TGSI_OPCODE_I64SHR
:
5950 exec_arg0_64_arg1_32(mach
, inst
, micro_i64shr
);
5952 case TGSI_OPCODE_U64SHR
:
5953 exec_arg0_64_arg1_32(mach
, inst
, micro_u64shr
);
5955 case TGSI_OPCODE_U64DIV
:
5956 exec_double_binary(mach
, inst
, micro_u64div
, TGSI_EXEC_DATA_UINT64
);
5958 case TGSI_OPCODE_I64DIV
:
5959 exec_double_binary(mach
, inst
, micro_i64div
, TGSI_EXEC_DATA_INT64
);
5961 case TGSI_OPCODE_U64MOD
:
5962 exec_double_binary(mach
, inst
, micro_u64mod
, TGSI_EXEC_DATA_UINT64
);
5964 case TGSI_OPCODE_I64MOD
:
5965 exec_double_binary(mach
, inst
, micro_i64mod
, TGSI_EXEC_DATA_INT64
);
5968 case TGSI_OPCODE_F2U64
:
5969 exec_t_2_64(mach
, inst
, micro_f2u64
, TGSI_EXEC_DATA_FLOAT
);
5972 case TGSI_OPCODE_F2I64
:
5973 exec_t_2_64(mach
, inst
, micro_f2i64
, TGSI_EXEC_DATA_FLOAT
);
5976 case TGSI_OPCODE_U2I64
:
5977 exec_t_2_64(mach
, inst
, micro_u2i64
, TGSI_EXEC_DATA_INT
);
5979 case TGSI_OPCODE_I2I64
:
5980 exec_t_2_64(mach
, inst
, micro_i2i64
, TGSI_EXEC_DATA_INT
);
5983 case TGSI_OPCODE_D2U64
:
5984 exec_double_unary(mach
, inst
, micro_d2u64
);
5987 case TGSI_OPCODE_D2I64
:
5988 exec_double_unary(mach
, inst
, micro_d2i64
);
5991 case TGSI_OPCODE_U642F
:
5992 exec_64_2_t(mach
, inst
, micro_u642f
, TGSI_EXEC_DATA_FLOAT
);
5994 case TGSI_OPCODE_I642F
:
5995 exec_64_2_t(mach
, inst
, micro_i642f
, TGSI_EXEC_DATA_FLOAT
);
5998 case TGSI_OPCODE_U642D
:
5999 exec_double_unary(mach
, inst
, micro_u642d
);
6001 case TGSI_OPCODE_I642D
:
6002 exec_double_unary(mach
, inst
, micro_i642d
);
6012 tgsi_exec_machine_setup_masks(struct tgsi_exec_machine
*mach
)
6014 uint default_mask
= 0xf;
6016 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] = 0;
6017 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] = 0;
6019 if (mach
->ShaderType
== PIPE_SHADER_GEOMETRY
) {
6020 mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0] = 0;
6021 mach
->Primitives
[0] = 0;
6022 /* GS runs on a single primitive for now */
6026 if (mach
->NonHelperMask
== 0)
6027 mach
->NonHelperMask
= default_mask
;
6028 mach
->CondMask
= default_mask
;
6029 mach
->LoopMask
= default_mask
;
6030 mach
->ContMask
= default_mask
;
6031 mach
->FuncMask
= default_mask
;
6032 mach
->ExecMask
= default_mask
;
6034 mach
->Switch
.mask
= default_mask
;
6036 assert(mach
->CondStackTop
== 0);
6037 assert(mach
->LoopStackTop
== 0);
6038 assert(mach
->ContStackTop
== 0);
6039 assert(mach
->SwitchStackTop
== 0);
6040 assert(mach
->BreakStackTop
== 0);
6041 assert(mach
->CallStackTop
== 0);
6045 * Run TGSI interpreter.
6046 * \return bitmask of "alive" quad components
6049 tgsi_exec_machine_run( struct tgsi_exec_machine
*mach
, int start_pc
)
6053 mach
->pc
= start_pc
;
6056 tgsi_exec_machine_setup_masks(mach
);
6058 /* execute declarations (interpolants) */
6059 for (i
= 0; i
< mach
->NumDeclarations
; i
++) {
6060 exec_declaration( mach
, mach
->Declarations
+i
);
6066 struct tgsi_exec_vector temps
[TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
];
6067 struct tgsi_exec_vector outputs
[PIPE_MAX_ATTRIBS
];
6071 memset(mach
->Temps
, 0, sizeof(temps
));
6073 memset(mach
->Outputs
, 0, sizeof(outputs
));
6074 memset(temps
, 0, sizeof(temps
));
6075 memset(outputs
, 0, sizeof(outputs
));
6079 /* execute instructions, until pc is set to -1 */
6080 while (mach
->pc
!= -1) {
6081 boolean barrier_hit
;
6085 tgsi_dump_instruction(&mach
->Instructions
[mach
->pc
], inst
++);
6088 assert(mach
->pc
< (int) mach
->NumInstructions
);
6089 barrier_hit
= exec_instruction(mach
, mach
->Instructions
+ mach
->pc
, &mach
->pc
);
6091 /* for compute shaders if we hit a barrier return now for later rescheduling */
6092 if (barrier_hit
&& mach
->ShaderType
== PIPE_SHADER_COMPUTE
)
6096 for (i
= 0; i
< TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
; i
++) {
6097 if (memcmp(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]))) {
6100 memcpy(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]));
6101 debug_printf("TEMP[%2u] = ", i
);
6102 for (j
= 0; j
< 4; j
++) {
6106 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
6107 temps
[i
].xyzw
[0].f
[j
], temps
[i
].xyzw
[0].u
[j
],
6108 temps
[i
].xyzw
[1].f
[j
], temps
[i
].xyzw
[1].u
[j
],
6109 temps
[i
].xyzw
[2].f
[j
], temps
[i
].xyzw
[2].u
[j
],
6110 temps
[i
].xyzw
[3].f
[j
], temps
[i
].xyzw
[3].u
[j
]);
6114 if (mach
->Outputs
) {
6115 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
6116 if (memcmp(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]))) {
6119 memcpy(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]));
6120 debug_printf("OUT[%2u] = ", i
);
6121 for (j
= 0; j
< 4; j
++) {
6125 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
6126 outputs
[i
].xyzw
[0].f
[j
], outputs
[i
].xyzw
[0].u
[j
],
6127 outputs
[i
].xyzw
[1].f
[j
], outputs
[i
].xyzw
[1].u
[j
],
6128 outputs
[i
].xyzw
[2].f
[j
], outputs
[i
].xyzw
[2].u
[j
],
6129 outputs
[i
].xyzw
[3].f
[j
], outputs
[i
].xyzw
[3].u
[j
]);
6139 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
6140 if (mach
->ShaderType
== PIPE_SHADER_FRAGMENT
) {
6142 * Scale back depth component.
6144 for (i
= 0; i
< 4; i
++)
6145 mach
->Outputs
[0].xyzw
[2].f
[i
] *= ctx
->DrawBuffer
->_DepthMaxF
;
6149 /* Strictly speaking, these assertions aren't really needed but they
6150 * can potentially catch some bugs in the control flow code.
6152 assert(mach
->CondStackTop
== 0);
6153 assert(mach
->LoopStackTop
== 0);
6154 assert(mach
->ContStackTop
== 0);
6155 assert(mach
->SwitchStackTop
== 0);
6156 assert(mach
->BreakStackTop
== 0);
6157 assert(mach
->CallStackTop
== 0);
6159 return ~mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];