1 /**************************************************************************
3 * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
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 TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
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_memory.h"
62 #include "util/u_math.h"
65 #define DEBUG_EXECUTION 0
70 #define TILE_TOP_LEFT 0
71 #define TILE_TOP_RIGHT 1
72 #define TILE_BOTTOM_LEFT 2
73 #define TILE_BOTTOM_RIGHT 3
76 micro_abs(union tgsi_exec_channel
*dst
,
77 const union tgsi_exec_channel
*src
)
79 dst
->f
[0] = fabsf(src
->f
[0]);
80 dst
->f
[1] = fabsf(src
->f
[1]);
81 dst
->f
[2] = fabsf(src
->f
[2]);
82 dst
->f
[3] = fabsf(src
->f
[3]);
86 micro_arl(union tgsi_exec_channel
*dst
,
87 const union tgsi_exec_channel
*src
)
89 dst
->i
[0] = (int)floorf(src
->f
[0]);
90 dst
->i
[1] = (int)floorf(src
->f
[1]);
91 dst
->i
[2] = (int)floorf(src
->f
[2]);
92 dst
->i
[3] = (int)floorf(src
->f
[3]);
96 micro_arr(union tgsi_exec_channel
*dst
,
97 const union tgsi_exec_channel
*src
)
99 dst
->i
[0] = (int)floorf(src
->f
[0] + 0.5f
);
100 dst
->i
[1] = (int)floorf(src
->f
[1] + 0.5f
);
101 dst
->i
[2] = (int)floorf(src
->f
[2] + 0.5f
);
102 dst
->i
[3] = (int)floorf(src
->f
[3] + 0.5f
);
106 micro_ceil(union tgsi_exec_channel
*dst
,
107 const union tgsi_exec_channel
*src
)
109 dst
->f
[0] = ceilf(src
->f
[0]);
110 dst
->f
[1] = ceilf(src
->f
[1]);
111 dst
->f
[2] = ceilf(src
->f
[2]);
112 dst
->f
[3] = ceilf(src
->f
[3]);
116 micro_clamp(union tgsi_exec_channel
*dst
,
117 const union tgsi_exec_channel
*src0
,
118 const union tgsi_exec_channel
*src1
,
119 const union tgsi_exec_channel
*src2
)
121 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src1
->f
[0] : src0
->f
[0] > src2
->f
[0] ? src2
->f
[0] : src0
->f
[0];
122 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src1
->f
[1] : src0
->f
[1] > src2
->f
[1] ? src2
->f
[1] : src0
->f
[1];
123 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src1
->f
[2] : src0
->f
[2] > src2
->f
[2] ? src2
->f
[2] : src0
->f
[2];
124 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src1
->f
[3] : src0
->f
[3] > src2
->f
[3] ? src2
->f
[3] : src0
->f
[3];
128 micro_cmp(union tgsi_exec_channel
*dst
,
129 const union tgsi_exec_channel
*src0
,
130 const union tgsi_exec_channel
*src1
,
131 const union tgsi_exec_channel
*src2
)
133 dst
->f
[0] = src0
->f
[0] < 0.0f
? src1
->f
[0] : src2
->f
[0];
134 dst
->f
[1] = src0
->f
[1] < 0.0f
? src1
->f
[1] : src2
->f
[1];
135 dst
->f
[2] = src0
->f
[2] < 0.0f
? src1
->f
[2] : src2
->f
[2];
136 dst
->f
[3] = src0
->f
[3] < 0.0f
? src1
->f
[3] : src2
->f
[3];
140 micro_cnd(union tgsi_exec_channel
*dst
,
141 const union tgsi_exec_channel
*src0
,
142 const union tgsi_exec_channel
*src1
,
143 const union tgsi_exec_channel
*src2
)
145 dst
->f
[0] = src2
->f
[0] > 0.5f
? src0
->f
[0] : src1
->f
[0];
146 dst
->f
[1] = src2
->f
[1] > 0.5f
? src0
->f
[1] : src1
->f
[1];
147 dst
->f
[2] = src2
->f
[2] > 0.5f
? src0
->f
[2] : src1
->f
[2];
148 dst
->f
[3] = src2
->f
[3] > 0.5f
? src0
->f
[3] : src1
->f
[3];
152 micro_cos(union tgsi_exec_channel
*dst
,
153 const union tgsi_exec_channel
*src
)
155 dst
->f
[0] = cosf(src
->f
[0]);
156 dst
->f
[1] = cosf(src
->f
[1]);
157 dst
->f
[2] = cosf(src
->f
[2]);
158 dst
->f
[3] = cosf(src
->f
[3]);
162 micro_ddx(union tgsi_exec_channel
*dst
,
163 const union tgsi_exec_channel
*src
)
168 dst
->f
[3] = src
->f
[TILE_BOTTOM_RIGHT
] - src
->f
[TILE_BOTTOM_LEFT
];
172 micro_ddy(union tgsi_exec_channel
*dst
,
173 const union tgsi_exec_channel
*src
)
178 dst
->f
[3] = src
->f
[TILE_BOTTOM_LEFT
] - src
->f
[TILE_TOP_LEFT
];
182 micro_exp2(union tgsi_exec_channel
*dst
,
183 const union tgsi_exec_channel
*src
)
186 dst
->f
[0] = util_fast_exp2(src
->f
[0]);
187 dst
->f
[1] = util_fast_exp2(src
->f
[1]);
188 dst
->f
[2] = util_fast_exp2(src
->f
[2]);
189 dst
->f
[3] = util_fast_exp2(src
->f
[3]);
192 /* Inf is okay for this instruction, so clamp it to silence assertions. */
194 union tgsi_exec_channel clamped
;
196 for (i
= 0; i
< 4; i
++) {
197 if (src
->f
[i
] > 127.99999f
) {
198 clamped
.f
[i
] = 127.99999f
;
199 } else if (src
->f
[i
] < -126.99999f
) {
200 clamped
.f
[i
] = -126.99999f
;
202 clamped
.f
[i
] = src
->f
[i
];
208 dst
->f
[0] = powf(2.0f
, src
->f
[0]);
209 dst
->f
[1] = powf(2.0f
, src
->f
[1]);
210 dst
->f
[2] = powf(2.0f
, src
->f
[2]);
211 dst
->f
[3] = powf(2.0f
, src
->f
[3]);
212 #endif /* FAST_MATH */
216 micro_flr(union tgsi_exec_channel
*dst
,
217 const union tgsi_exec_channel
*src
)
219 dst
->f
[0] = floorf(src
->f
[0]);
220 dst
->f
[1] = floorf(src
->f
[1]);
221 dst
->f
[2] = floorf(src
->f
[2]);
222 dst
->f
[3] = floorf(src
->f
[3]);
226 micro_frc(union tgsi_exec_channel
*dst
,
227 const union tgsi_exec_channel
*src
)
229 dst
->f
[0] = src
->f
[0] - floorf(src
->f
[0]);
230 dst
->f
[1] = src
->f
[1] - floorf(src
->f
[1]);
231 dst
->f
[2] = src
->f
[2] - floorf(src
->f
[2]);
232 dst
->f
[3] = src
->f
[3] - floorf(src
->f
[3]);
236 micro_iabs(union tgsi_exec_channel
*dst
,
237 const union tgsi_exec_channel
*src
)
239 dst
->i
[0] = src
->i
[0] >= 0 ? src
->i
[0] : -src
->i
[0];
240 dst
->i
[1] = src
->i
[1] >= 0 ? src
->i
[1] : -src
->i
[1];
241 dst
->i
[2] = src
->i
[2] >= 0 ? src
->i
[2] : -src
->i
[2];
242 dst
->i
[3] = src
->i
[3] >= 0 ? src
->i
[3] : -src
->i
[3];
246 micro_ineg(union tgsi_exec_channel
*dst
,
247 const union tgsi_exec_channel
*src
)
249 dst
->i
[0] = -src
->i
[0];
250 dst
->i
[1] = -src
->i
[1];
251 dst
->i
[2] = -src
->i
[2];
252 dst
->i
[3] = -src
->i
[3];
256 micro_lg2(union tgsi_exec_channel
*dst
,
257 const union tgsi_exec_channel
*src
)
260 dst
->f
[0] = util_fast_log2(src
->f
[0]);
261 dst
->f
[1] = util_fast_log2(src
->f
[1]);
262 dst
->f
[2] = util_fast_log2(src
->f
[2]);
263 dst
->f
[3] = util_fast_log2(src
->f
[3]);
265 dst
->f
[0] = logf(src
->f
[0]) * 1.442695f
;
266 dst
->f
[1] = logf(src
->f
[1]) * 1.442695f
;
267 dst
->f
[2] = logf(src
->f
[2]) * 1.442695f
;
268 dst
->f
[3] = logf(src
->f
[3]) * 1.442695f
;
273 micro_lrp(union tgsi_exec_channel
*dst
,
274 const union tgsi_exec_channel
*src0
,
275 const union tgsi_exec_channel
*src1
,
276 const union tgsi_exec_channel
*src2
)
278 dst
->f
[0] = src0
->f
[0] * (src1
->f
[0] - src2
->f
[0]) + src2
->f
[0];
279 dst
->f
[1] = src0
->f
[1] * (src1
->f
[1] - src2
->f
[1]) + src2
->f
[1];
280 dst
->f
[2] = src0
->f
[2] * (src1
->f
[2] - src2
->f
[2]) + src2
->f
[2];
281 dst
->f
[3] = src0
->f
[3] * (src1
->f
[3] - src2
->f
[3]) + src2
->f
[3];
285 micro_mad(union tgsi_exec_channel
*dst
,
286 const union tgsi_exec_channel
*src0
,
287 const union tgsi_exec_channel
*src1
,
288 const union tgsi_exec_channel
*src2
)
290 dst
->f
[0] = src0
->f
[0] * src1
->f
[0] + src2
->f
[0];
291 dst
->f
[1] = src0
->f
[1] * src1
->f
[1] + src2
->f
[1];
292 dst
->f
[2] = src0
->f
[2] * src1
->f
[2] + src2
->f
[2];
293 dst
->f
[3] = src0
->f
[3] * src1
->f
[3] + src2
->f
[3];
297 micro_mov(union tgsi_exec_channel
*dst
,
298 const union tgsi_exec_channel
*src
)
300 dst
->u
[0] = src
->u
[0];
301 dst
->u
[1] = src
->u
[1];
302 dst
->u
[2] = src
->u
[2];
303 dst
->u
[3] = src
->u
[3];
307 micro_rcp(union tgsi_exec_channel
*dst
,
308 const union tgsi_exec_channel
*src
)
310 #if 0 /* for debugging */
311 assert(src
->f
[0] != 0.0f
);
312 assert(src
->f
[1] != 0.0f
);
313 assert(src
->f
[2] != 0.0f
);
314 assert(src
->f
[3] != 0.0f
);
316 dst
->f
[0] = 1.0f
/ src
->f
[0];
317 dst
->f
[1] = 1.0f
/ src
->f
[1];
318 dst
->f
[2] = 1.0f
/ src
->f
[2];
319 dst
->f
[3] = 1.0f
/ src
->f
[3];
323 micro_rnd(union tgsi_exec_channel
*dst
,
324 const union tgsi_exec_channel
*src
)
326 dst
->f
[0] = floorf(src
->f
[0] + 0.5f
);
327 dst
->f
[1] = floorf(src
->f
[1] + 0.5f
);
328 dst
->f
[2] = floorf(src
->f
[2] + 0.5f
);
329 dst
->f
[3] = floorf(src
->f
[3] + 0.5f
);
333 micro_rsq(union tgsi_exec_channel
*dst
,
334 const union tgsi_exec_channel
*src
)
336 #if 0 /* for debugging */
337 assert(src
->f
[0] != 0.0f
);
338 assert(src
->f
[1] != 0.0f
);
339 assert(src
->f
[2] != 0.0f
);
340 assert(src
->f
[3] != 0.0f
);
342 dst
->f
[0] = 1.0f
/ sqrtf(src
->f
[0]);
343 dst
->f
[1] = 1.0f
/ sqrtf(src
->f
[1]);
344 dst
->f
[2] = 1.0f
/ sqrtf(src
->f
[2]);
345 dst
->f
[3] = 1.0f
/ sqrtf(src
->f
[3]);
349 micro_sqrt(union tgsi_exec_channel
*dst
,
350 const union tgsi_exec_channel
*src
)
352 dst
->f
[0] = sqrtf(src
->f
[0]);
353 dst
->f
[1] = sqrtf(src
->f
[1]);
354 dst
->f
[2] = sqrtf(src
->f
[2]);
355 dst
->f
[3] = sqrtf(src
->f
[3]);
359 micro_seq(union tgsi_exec_channel
*dst
,
360 const union tgsi_exec_channel
*src0
,
361 const union tgsi_exec_channel
*src1
)
363 dst
->f
[0] = src0
->f
[0] == src1
->f
[0] ? 1.0f
: 0.0f
;
364 dst
->f
[1] = src0
->f
[1] == src1
->f
[1] ? 1.0f
: 0.0f
;
365 dst
->f
[2] = src0
->f
[2] == src1
->f
[2] ? 1.0f
: 0.0f
;
366 dst
->f
[3] = src0
->f
[3] == src1
->f
[3] ? 1.0f
: 0.0f
;
370 micro_sge(union tgsi_exec_channel
*dst
,
371 const union tgsi_exec_channel
*src0
,
372 const union tgsi_exec_channel
*src1
)
374 dst
->f
[0] = src0
->f
[0] >= src1
->f
[0] ? 1.0f
: 0.0f
;
375 dst
->f
[1] = src0
->f
[1] >= src1
->f
[1] ? 1.0f
: 0.0f
;
376 dst
->f
[2] = src0
->f
[2] >= src1
->f
[2] ? 1.0f
: 0.0f
;
377 dst
->f
[3] = src0
->f
[3] >= src1
->f
[3] ? 1.0f
: 0.0f
;
381 micro_sgn(union tgsi_exec_channel
*dst
,
382 const union tgsi_exec_channel
*src
)
384 dst
->f
[0] = src
->f
[0] < 0.0f
? -1.0f
: src
->f
[0] > 0.0f
? 1.0f
: 0.0f
;
385 dst
->f
[1] = src
->f
[1] < 0.0f
? -1.0f
: src
->f
[1] > 0.0f
? 1.0f
: 0.0f
;
386 dst
->f
[2] = src
->f
[2] < 0.0f
? -1.0f
: src
->f
[2] > 0.0f
? 1.0f
: 0.0f
;
387 dst
->f
[3] = src
->f
[3] < 0.0f
? -1.0f
: src
->f
[3] > 0.0f
? 1.0f
: 0.0f
;
391 micro_isgn(union tgsi_exec_channel
*dst
,
392 const union tgsi_exec_channel
*src
)
394 dst
->i
[0] = src
->i
[0] < 0 ? -1 : src
->i
[0] > 0 ? 1 : 0;
395 dst
->i
[1] = src
->i
[1] < 0 ? -1 : src
->i
[1] > 0 ? 1 : 0;
396 dst
->i
[2] = src
->i
[2] < 0 ? -1 : src
->i
[2] > 0 ? 1 : 0;
397 dst
->i
[3] = src
->i
[3] < 0 ? -1 : src
->i
[3] > 0 ? 1 : 0;
401 micro_sgt(union tgsi_exec_channel
*dst
,
402 const union tgsi_exec_channel
*src0
,
403 const union tgsi_exec_channel
*src1
)
405 dst
->f
[0] = src0
->f
[0] > src1
->f
[0] ? 1.0f
: 0.0f
;
406 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? 1.0f
: 0.0f
;
407 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? 1.0f
: 0.0f
;
408 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? 1.0f
: 0.0f
;
412 micro_sin(union tgsi_exec_channel
*dst
,
413 const union tgsi_exec_channel
*src
)
415 dst
->f
[0] = sinf(src
->f
[0]);
416 dst
->f
[1] = sinf(src
->f
[1]);
417 dst
->f
[2] = sinf(src
->f
[2]);
418 dst
->f
[3] = sinf(src
->f
[3]);
422 micro_sle(union tgsi_exec_channel
*dst
,
423 const union tgsi_exec_channel
*src0
,
424 const union tgsi_exec_channel
*src1
)
426 dst
->f
[0] = src0
->f
[0] <= src1
->f
[0] ? 1.0f
: 0.0f
;
427 dst
->f
[1] = src0
->f
[1] <= src1
->f
[1] ? 1.0f
: 0.0f
;
428 dst
->f
[2] = src0
->f
[2] <= src1
->f
[2] ? 1.0f
: 0.0f
;
429 dst
->f
[3] = src0
->f
[3] <= src1
->f
[3] ? 1.0f
: 0.0f
;
433 micro_slt(union tgsi_exec_channel
*dst
,
434 const union tgsi_exec_channel
*src0
,
435 const union tgsi_exec_channel
*src1
)
437 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? 1.0f
: 0.0f
;
438 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? 1.0f
: 0.0f
;
439 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? 1.0f
: 0.0f
;
440 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? 1.0f
: 0.0f
;
444 micro_sne(union tgsi_exec_channel
*dst
,
445 const union tgsi_exec_channel
*src0
,
446 const union tgsi_exec_channel
*src1
)
448 dst
->f
[0] = src0
->f
[0] != src1
->f
[0] ? 1.0f
: 0.0f
;
449 dst
->f
[1] = src0
->f
[1] != src1
->f
[1] ? 1.0f
: 0.0f
;
450 dst
->f
[2] = src0
->f
[2] != src1
->f
[2] ? 1.0f
: 0.0f
;
451 dst
->f
[3] = src0
->f
[3] != src1
->f
[3] ? 1.0f
: 0.0f
;
455 micro_sfl(union tgsi_exec_channel
*dst
)
464 micro_str(union tgsi_exec_channel
*dst
)
473 micro_trunc(union tgsi_exec_channel
*dst
,
474 const union tgsi_exec_channel
*src
)
476 dst
->f
[0] = (float)(int)src
->f
[0];
477 dst
->f
[1] = (float)(int)src
->f
[1];
478 dst
->f
[2] = (float)(int)src
->f
[2];
479 dst
->f
[3] = (float)(int)src
->f
[3];
483 enum tgsi_exec_datatype
{
484 TGSI_EXEC_DATA_FLOAT
,
490 * Shorthand locations of various utility registers (_I = Index, _C = Channel)
492 #define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I
493 #define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C
494 #define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I
495 #define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C
496 #define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I
497 #define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C
500 /** The execution mask depends on the conditional mask and the loop mask */
501 #define UPDATE_EXEC_MASK(MACH) \
502 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask
505 static const union tgsi_exec_channel ZeroVec
=
506 { { 0.0, 0.0, 0.0, 0.0 } };
508 static const union tgsi_exec_channel OneVec
= {
509 {1.0f
, 1.0f
, 1.0f
, 1.0f
}
512 static const union tgsi_exec_channel P128Vec
= {
513 {128.0f
, 128.0f
, 128.0f
, 128.0f
}
516 static const union tgsi_exec_channel M128Vec
= {
517 {-128.0f
, -128.0f
, -128.0f
, -128.0f
}
522 * Assert that none of the float values in 'chan' are infinite or NaN.
523 * NaN and Inf may occur normally during program execution and should
524 * not lead to crashes, etc. But when debugging, it's helpful to catch
528 check_inf_or_nan(const union tgsi_exec_channel
*chan
)
530 assert(!util_is_inf_or_nan((chan
)->f
[0]));
531 assert(!util_is_inf_or_nan((chan
)->f
[1]));
532 assert(!util_is_inf_or_nan((chan
)->f
[2]));
533 assert(!util_is_inf_or_nan((chan
)->f
[3]));
539 print_chan(const char *msg
, const union tgsi_exec_channel
*chan
)
541 debug_printf("%s = {%f, %f, %f, %f}\n",
542 msg
, chan
->f
[0], chan
->f
[1], chan
->f
[2], chan
->f
[3]);
549 print_temp(const struct tgsi_exec_machine
*mach
, uint index
)
551 const struct tgsi_exec_vector
*tmp
= &mach
->Temps
[index
];
553 debug_printf("Temp[%u] =\n", index
);
554 for (i
= 0; i
< 4; i
++) {
555 debug_printf(" %c: { %f, %f, %f, %f }\n",
567 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine
*mach
,
570 const unsigned *buf_sizes
)
574 for (i
= 0; i
< num_bufs
; i
++) {
575 mach
->Consts
[i
] = bufs
[i
];
576 mach
->ConstsSize
[i
] = buf_sizes
[i
];
582 * Check if there's a potential src/dst register data dependency when
583 * using SOA execution.
586 * This would expand into:
591 * The second instruction will have the wrong value for t0 if executed as-is.
594 tgsi_check_soa_dependencies(const struct tgsi_full_instruction
*inst
)
598 uint writemask
= inst
->Dst
[0].Register
.WriteMask
;
599 if (writemask
== TGSI_WRITEMASK_X
||
600 writemask
== TGSI_WRITEMASK_Y
||
601 writemask
== TGSI_WRITEMASK_Z
||
602 writemask
== TGSI_WRITEMASK_W
||
603 writemask
== TGSI_WRITEMASK_NONE
) {
604 /* no chance of data dependency */
608 /* loop over src regs */
609 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
610 if ((inst
->Src
[i
].Register
.File
==
611 inst
->Dst
[0].Register
.File
) &&
612 ((inst
->Src
[i
].Register
.Index
==
613 inst
->Dst
[0].Register
.Index
) ||
614 inst
->Src
[i
].Register
.Indirect
||
615 inst
->Dst
[0].Register
.Indirect
)) {
616 /* loop over dest channels */
617 uint channelsWritten
= 0x0;
618 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
619 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
620 /* check if we're reading a channel that's been written */
621 uint swizzle
= tgsi_util_get_full_src_register_swizzle(&inst
->Src
[i
], chan
);
622 if (channelsWritten
& (1 << swizzle
)) {
626 channelsWritten
|= (1 << chan
);
636 * Initialize machine state by expanding tokens to full instructions,
637 * allocating temporary storage, setting up constants, etc.
638 * After this, we can call tgsi_exec_machine_run() many times.
641 tgsi_exec_machine_bind_shader(
642 struct tgsi_exec_machine
*mach
,
643 const struct tgsi_token
*tokens
,
644 struct tgsi_sampler
*sampler
)
647 struct tgsi_parse_context parse
;
648 struct tgsi_full_instruction
*instructions
;
649 struct tgsi_full_declaration
*declarations
;
650 uint maxInstructions
= 10, numInstructions
= 0;
651 uint maxDeclarations
= 10, numDeclarations
= 0;
654 tgsi_dump(tokens
, 0);
660 mach
->Tokens
= tokens
;
661 mach
->Sampler
= sampler
;
664 /* unbind and free all */
665 FREE(mach
->Declarations
);
666 mach
->Declarations
= NULL
;
667 mach
->NumDeclarations
= 0;
669 FREE(mach
->Instructions
);
670 mach
->Instructions
= NULL
;
671 mach
->NumInstructions
= 0;
676 k
= tgsi_parse_init (&parse
, mach
->Tokens
);
677 if (k
!= TGSI_PARSE_OK
) {
678 debug_printf( "Problem parsing!\n" );
682 mach
->Processor
= parse
.FullHeader
.Processor
.Processor
;
684 mach
->NumOutputs
= 0;
686 if (mach
->Processor
== TGSI_PROCESSOR_GEOMETRY
&&
687 !mach
->UsedGeometryShader
) {
688 struct tgsi_exec_vector
*inputs
;
689 struct tgsi_exec_vector
*outputs
;
691 inputs
= align_malloc(sizeof(struct tgsi_exec_vector
) *
692 TGSI_MAX_PRIM_VERTICES
* PIPE_MAX_ATTRIBS
,
698 outputs
= align_malloc(sizeof(struct tgsi_exec_vector
) *
699 TGSI_MAX_TOTAL_VERTICES
, 16);
706 align_free(mach
->Inputs
);
707 align_free(mach
->Outputs
);
709 mach
->Inputs
= inputs
;
710 mach
->Outputs
= outputs
;
711 mach
->UsedGeometryShader
= TRUE
;
714 declarations
= (struct tgsi_full_declaration
*)
715 MALLOC( maxDeclarations
* sizeof(struct tgsi_full_declaration
) );
721 instructions
= (struct tgsi_full_instruction
*)
722 MALLOC( maxInstructions
* sizeof(struct tgsi_full_instruction
) );
725 FREE( declarations
);
729 while( !tgsi_parse_end_of_tokens( &parse
) ) {
732 tgsi_parse_token( &parse
);
733 switch( parse
.FullToken
.Token
.Type
) {
734 case TGSI_TOKEN_TYPE_DECLARATION
:
735 /* save expanded declaration */
736 if (numDeclarations
== maxDeclarations
) {
737 declarations
= REALLOC(declarations
,
739 * sizeof(struct tgsi_full_declaration
),
740 (maxDeclarations
+ 10)
741 * sizeof(struct tgsi_full_declaration
));
742 maxDeclarations
+= 10;
744 if (parse
.FullToken
.FullDeclaration
.Declaration
.File
== TGSI_FILE_OUTPUT
) {
746 for (reg
= parse
.FullToken
.FullDeclaration
.Range
.First
;
747 reg
<= parse
.FullToken
.FullDeclaration
.Range
.Last
;
752 memcpy(declarations
+ numDeclarations
,
753 &parse
.FullToken
.FullDeclaration
,
754 sizeof(declarations
[0]));
758 case TGSI_TOKEN_TYPE_IMMEDIATE
:
760 uint size
= parse
.FullToken
.FullImmediate
.Immediate
.NrTokens
- 1;
762 assert( mach
->ImmLimit
+ 1 <= TGSI_EXEC_NUM_IMMEDIATES
);
764 for( i
= 0; i
< size
; i
++ ) {
765 mach
->Imms
[mach
->ImmLimit
][i
] =
766 parse
.FullToken
.FullImmediate
.u
[i
].Float
;
772 case TGSI_TOKEN_TYPE_INSTRUCTION
:
774 /* save expanded instruction */
775 if (numInstructions
== maxInstructions
) {
776 instructions
= REALLOC(instructions
,
778 * sizeof(struct tgsi_full_instruction
),
779 (maxInstructions
+ 10)
780 * sizeof(struct tgsi_full_instruction
));
781 maxInstructions
+= 10;
784 memcpy(instructions
+ numInstructions
,
785 &parse
.FullToken
.FullInstruction
,
786 sizeof(instructions
[0]));
791 case TGSI_TOKEN_TYPE_PROPERTY
:
798 tgsi_parse_free (&parse
);
800 FREE(mach
->Declarations
);
801 mach
->Declarations
= declarations
;
802 mach
->NumDeclarations
= numDeclarations
;
804 FREE(mach
->Instructions
);
805 mach
->Instructions
= instructions
;
806 mach
->NumInstructions
= numInstructions
;
810 struct tgsi_exec_machine
*
811 tgsi_exec_machine_create( void )
813 struct tgsi_exec_machine
*mach
;
816 mach
= align_malloc( sizeof *mach
, 16 );
820 memset(mach
, 0, sizeof(*mach
));
822 mach
->Addrs
= &mach
->Temps
[TGSI_EXEC_TEMP_ADDR
];
823 mach
->MaxGeometryShaderOutputs
= TGSI_MAX_TOTAL_VERTICES
;
824 mach
->Predicates
= &mach
->Temps
[TGSI_EXEC_TEMP_P0
];
826 mach
->Inputs
= align_malloc(sizeof(struct tgsi_exec_vector
) * PIPE_MAX_ATTRIBS
, 16);
827 mach
->Outputs
= align_malloc(sizeof(struct tgsi_exec_vector
) * PIPE_MAX_ATTRIBS
, 16);
828 if (!mach
->Inputs
|| !mach
->Outputs
)
831 /* Setup constants needed by the SSE2 executor. */
832 for( i
= 0; i
< 4; i
++ ) {
833 mach
->Temps
[TGSI_EXEC_TEMP_00000000_I
].xyzw
[TGSI_EXEC_TEMP_00000000_C
].u
[i
] = 0x00000000;
834 mach
->Temps
[TGSI_EXEC_TEMP_7FFFFFFF_I
].xyzw
[TGSI_EXEC_TEMP_7FFFFFFF_C
].u
[i
] = 0x7FFFFFFF;
835 mach
->Temps
[TGSI_EXEC_TEMP_80000000_I
].xyzw
[TGSI_EXEC_TEMP_80000000_C
].u
[i
] = 0x80000000;
836 mach
->Temps
[TGSI_EXEC_TEMP_FFFFFFFF_I
].xyzw
[TGSI_EXEC_TEMP_FFFFFFFF_C
].u
[i
] = 0xFFFFFFFF; /* not used */
837 mach
->Temps
[TGSI_EXEC_TEMP_ONE_I
].xyzw
[TGSI_EXEC_TEMP_ONE_C
].f
[i
] = 1.0f
;
838 mach
->Temps
[TGSI_EXEC_TEMP_TWO_I
].xyzw
[TGSI_EXEC_TEMP_TWO_C
].f
[i
] = 2.0f
; /* not used */
839 mach
->Temps
[TGSI_EXEC_TEMP_128_I
].xyzw
[TGSI_EXEC_TEMP_128_C
].f
[i
] = 128.0f
;
840 mach
->Temps
[TGSI_EXEC_TEMP_MINUS_128_I
].xyzw
[TGSI_EXEC_TEMP_MINUS_128_C
].f
[i
] = -128.0f
;
841 mach
->Temps
[TGSI_EXEC_TEMP_THREE_I
].xyzw
[TGSI_EXEC_TEMP_THREE_C
].f
[i
] = 3.0f
;
842 mach
->Temps
[TGSI_EXEC_TEMP_HALF_I
].xyzw
[TGSI_EXEC_TEMP_HALF_C
].f
[i
] = 0.5f
;
846 /* silence warnings */
855 align_free(mach
->Inputs
);
856 align_free(mach
->Outputs
);
864 tgsi_exec_machine_destroy(struct tgsi_exec_machine
*mach
)
867 FREE(mach
->Instructions
);
868 FREE(mach
->Declarations
);
870 align_free(mach
->Inputs
);
871 align_free(mach
->Outputs
);
878 micro_add(union tgsi_exec_channel
*dst
,
879 const union tgsi_exec_channel
*src0
,
880 const union tgsi_exec_channel
*src1
)
882 dst
->f
[0] = src0
->f
[0] + src1
->f
[0];
883 dst
->f
[1] = src0
->f
[1] + src1
->f
[1];
884 dst
->f
[2] = src0
->f
[2] + src1
->f
[2];
885 dst
->f
[3] = src0
->f
[3] + src1
->f
[3];
890 union tgsi_exec_channel
*dst
,
891 const union tgsi_exec_channel
*src0
,
892 const union tgsi_exec_channel
*src1
)
894 if (src1
->f
[0] != 0) {
895 dst
->f
[0] = src0
->f
[0] / src1
->f
[0];
897 if (src1
->f
[1] != 0) {
898 dst
->f
[1] = src0
->f
[1] / src1
->f
[1];
900 if (src1
->f
[2] != 0) {
901 dst
->f
[2] = src0
->f
[2] / src1
->f
[2];
903 if (src1
->f
[3] != 0) {
904 dst
->f
[3] = src0
->f
[3] / src1
->f
[3];
909 micro_rcc(union tgsi_exec_channel
*dst
,
910 const union tgsi_exec_channel
*src
)
914 for (i
= 0; i
< 4; i
++) {
915 float recip
= 1.0f
/ src
->f
[i
];
918 if (recip
> 1.884467e+019f
) {
919 dst
->f
[i
] = 1.884467e+019f
;
921 else if (recip
< 5.42101e-020f
) {
922 dst
->f
[i
] = 5.42101e-020f
;
929 if (recip
< -1.884467e+019f
) {
930 dst
->f
[i
] = -1.884467e+019f
;
932 else if (recip
> -5.42101e-020f
) {
933 dst
->f
[i
] = -5.42101e-020f
;
944 union tgsi_exec_channel
*dst
,
945 const union tgsi_exec_channel
*src0
,
946 const union tgsi_exec_channel
*src1
,
947 const union tgsi_exec_channel
*src2
,
948 const union tgsi_exec_channel
*src3
)
950 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
951 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
952 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
953 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
957 micro_max(union tgsi_exec_channel
*dst
,
958 const union tgsi_exec_channel
*src0
,
959 const union tgsi_exec_channel
*src1
)
961 dst
->f
[0] = src0
->f
[0] > src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
962 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
963 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
964 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
968 micro_min(union tgsi_exec_channel
*dst
,
969 const union tgsi_exec_channel
*src0
,
970 const union tgsi_exec_channel
*src1
)
972 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
973 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
974 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
975 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
979 micro_mul(union tgsi_exec_channel
*dst
,
980 const union tgsi_exec_channel
*src0
,
981 const union tgsi_exec_channel
*src1
)
983 dst
->f
[0] = src0
->f
[0] * src1
->f
[0];
984 dst
->f
[1] = src0
->f
[1] * src1
->f
[1];
985 dst
->f
[2] = src0
->f
[2] * src1
->f
[2];
986 dst
->f
[3] = src0
->f
[3] * src1
->f
[3];
991 union tgsi_exec_channel
*dst
,
992 const union tgsi_exec_channel
*src
)
994 dst
->f
[0] = -src
->f
[0];
995 dst
->f
[1] = -src
->f
[1];
996 dst
->f
[2] = -src
->f
[2];
997 dst
->f
[3] = -src
->f
[3];
1002 union tgsi_exec_channel
*dst
,
1003 const union tgsi_exec_channel
*src0
,
1004 const union tgsi_exec_channel
*src1
)
1007 dst
->f
[0] = util_fast_pow( src0
->f
[0], src1
->f
[0] );
1008 dst
->f
[1] = util_fast_pow( src0
->f
[1], src1
->f
[1] );
1009 dst
->f
[2] = util_fast_pow( src0
->f
[2], src1
->f
[2] );
1010 dst
->f
[3] = util_fast_pow( src0
->f
[3], src1
->f
[3] );
1012 dst
->f
[0] = powf( src0
->f
[0], src1
->f
[0] );
1013 dst
->f
[1] = powf( src0
->f
[1], src1
->f
[1] );
1014 dst
->f
[2] = powf( src0
->f
[2], src1
->f
[2] );
1015 dst
->f
[3] = powf( src0
->f
[3], src1
->f
[3] );
1020 micro_sub(union tgsi_exec_channel
*dst
,
1021 const union tgsi_exec_channel
*src0
,
1022 const union tgsi_exec_channel
*src1
)
1024 dst
->f
[0] = src0
->f
[0] - src1
->f
[0];
1025 dst
->f
[1] = src0
->f
[1] - src1
->f
[1];
1026 dst
->f
[2] = src0
->f
[2] - src1
->f
[2];
1027 dst
->f
[3] = src0
->f
[3] - src1
->f
[3];
1031 fetch_src_file_channel(const struct tgsi_exec_machine
*mach
,
1032 const uint chan_index
,
1035 const union tgsi_exec_channel
*index
,
1036 const union tgsi_exec_channel
*index2D
,
1037 union tgsi_exec_channel
*chan
)
1041 assert(swizzle
< 4);
1044 case TGSI_FILE_CONSTANT
:
1045 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1046 assert(index2D
->i
[i
] >= 0 && index2D
->i
[i
] < PIPE_MAX_CONSTANT_BUFFERS
);
1047 assert(mach
->Consts
[index2D
->i
[i
]]);
1049 if (index
->i
[i
] < 0) {
1052 /* NOTE: copying the const value as a uint instead of float */
1053 const uint constbuf
= index2D
->i
[i
];
1054 const uint
*buf
= (const uint
*)mach
->Consts
[constbuf
];
1055 const int pos
= index
->i
[i
] * 4 + swizzle
;
1056 /* const buffer bounds check */
1057 if (pos
< 0 || pos
>= (int) mach
->ConstsSize
[constbuf
]) {
1059 /* Debug: print warning */
1060 static int count
= 0;
1062 debug_printf("TGSI Exec: const buffer index %d"
1063 " out of bounds\n", pos
);
1068 chan
->u
[i
] = buf
[pos
];
1073 case TGSI_FILE_INPUT
:
1074 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1076 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
1077 debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n",
1078 index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i],
1079 index2D->i[i], index->i[i]);
1081 int pos
= index2D
->i
[i
] * TGSI_EXEC_MAX_INPUT_ATTRIBS
+ index
->i
[i
];
1083 assert(pos
< TGSI_MAX_PRIM_VERTICES
* PIPE_MAX_ATTRIBS
);
1084 chan
->u
[i
] = mach
->Inputs
[pos
].xyzw
[swizzle
].u
[i
];
1088 case TGSI_FILE_SYSTEM_VALUE
:
1089 /* XXX no swizzling at this point. Will be needed if we put
1090 * gl_FragCoord, for example, in a sys value register.
1092 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1093 chan
->u
[i
] = mach
->SystemValue
[index
->i
[i
]].u
[i
];
1097 case TGSI_FILE_TEMPORARY
:
1098 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1099 assert(index
->i
[i
] < TGSI_EXEC_NUM_TEMPS
);
1100 assert(index2D
->i
[i
] == 0);
1102 chan
->u
[i
] = mach
->Temps
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1106 case TGSI_FILE_IMMEDIATE
:
1107 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1108 assert(index
->i
[i
] >= 0 && index
->i
[i
] < (int)mach
->ImmLimit
);
1109 assert(index2D
->i
[i
] == 0);
1111 chan
->f
[i
] = mach
->Imms
[index
->i
[i
]][swizzle
];
1115 case TGSI_FILE_ADDRESS
:
1116 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1117 assert(index
->i
[i
] >= 0);
1118 assert(index2D
->i
[i
] == 0);
1120 chan
->u
[i
] = mach
->Addrs
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1124 case TGSI_FILE_PREDICATE
:
1125 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1126 assert(index
->i
[i
] >= 0 && index
->i
[i
] < TGSI_EXEC_NUM_PREDS
);
1127 assert(index2D
->i
[i
] == 0);
1129 chan
->u
[i
] = mach
->Predicates
[0].xyzw
[swizzle
].u
[i
];
1133 case TGSI_FILE_OUTPUT
:
1134 /* vertex/fragment output vars can be read too */
1135 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1136 assert(index
->i
[i
] >= 0);
1137 assert(index2D
->i
[i
] == 0);
1139 chan
->u
[i
] = mach
->Outputs
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1145 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1152 fetch_source(const struct tgsi_exec_machine
*mach
,
1153 union tgsi_exec_channel
*chan
,
1154 const struct tgsi_full_src_register
*reg
,
1155 const uint chan_index
,
1156 enum tgsi_exec_datatype src_datatype
)
1158 union tgsi_exec_channel index
;
1159 union tgsi_exec_channel index2D
;
1162 /* We start with a direct index into a register file.
1166 * file = Register.File
1167 * [1] = Register.Index
1172 index
.i
[3] = reg
->Register
.Index
;
1174 /* There is an extra source register that indirectly subscripts
1175 * a register file. The direct index now becomes an offset
1176 * that is being added to the indirect register.
1180 * ind = Indirect.File
1181 * [2] = Indirect.Index
1182 * .x = Indirect.SwizzleX
1184 if (reg
->Register
.Indirect
) {
1185 union tgsi_exec_channel index2
;
1186 union tgsi_exec_channel indir_index
;
1187 const uint execmask
= mach
->ExecMask
;
1190 /* which address register (always zero now) */
1194 index2
.i
[3] = reg
->Indirect
.Index
;
1195 /* get current value of address register[swizzle] */
1196 swizzle
= reg
->Indirect
.Swizzle
;
1197 fetch_src_file_channel(mach
,
1205 /* add value of address register to the offset */
1206 index
.i
[0] += indir_index
.i
[0];
1207 index
.i
[1] += indir_index
.i
[1];
1208 index
.i
[2] += indir_index
.i
[2];
1209 index
.i
[3] += indir_index
.i
[3];
1211 /* for disabled execution channels, zero-out the index to
1212 * avoid using a potential garbage value.
1214 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1215 if ((execmask
& (1 << i
)) == 0)
1220 /* There is an extra source register that is a second
1221 * subscript to a register file. Effectively it means that
1222 * the register file is actually a 2D array of registers.
1226 * [3] = Dimension.Index
1228 if (reg
->Register
.Dimension
) {
1232 index2D
.i
[3] = reg
->Dimension
.Index
;
1234 /* Again, the second subscript index can be addressed indirectly
1235 * identically to the first one.
1236 * Nothing stops us from indirectly addressing the indirect register,
1237 * but there is no need for that, so we won't exercise it.
1239 * file[ind[4].y+3][1],
1241 * ind = DimIndirect.File
1242 * [4] = DimIndirect.Index
1243 * .y = DimIndirect.SwizzleX
1245 if (reg
->Dimension
.Indirect
) {
1246 union tgsi_exec_channel index2
;
1247 union tgsi_exec_channel indir_index
;
1248 const uint execmask
= mach
->ExecMask
;
1254 index2
.i
[3] = reg
->DimIndirect
.Index
;
1256 swizzle
= reg
->DimIndirect
.Swizzle
;
1257 fetch_src_file_channel(mach
,
1259 reg
->DimIndirect
.File
,
1265 index2D
.i
[0] += indir_index
.i
[0];
1266 index2D
.i
[1] += indir_index
.i
[1];
1267 index2D
.i
[2] += indir_index
.i
[2];
1268 index2D
.i
[3] += indir_index
.i
[3];
1270 /* for disabled execution channels, zero-out the index to
1271 * avoid using a potential garbage value.
1273 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1274 if ((execmask
& (1 << i
)) == 0) {
1280 /* If by any chance there was a need for a 3D array of register
1281 * files, we would have to check whether Dimension is followed
1282 * by a dimension register and continue the saga.
1291 swizzle
= tgsi_util_get_full_src_register_swizzle( reg
, chan_index
);
1292 fetch_src_file_channel(mach
,
1300 if (reg
->Register
.Absolute
) {
1301 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1302 micro_abs(chan
, chan
);
1304 micro_iabs(chan
, chan
);
1308 if (reg
->Register
.Negate
) {
1309 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1310 micro_neg(chan
, chan
);
1312 micro_ineg(chan
, chan
);
1318 store_dest(struct tgsi_exec_machine
*mach
,
1319 const union tgsi_exec_channel
*chan
,
1320 const struct tgsi_full_dst_register
*reg
,
1321 const struct tgsi_full_instruction
*inst
,
1323 enum tgsi_exec_datatype dst_datatype
)
1326 union tgsi_exec_channel null
;
1327 union tgsi_exec_channel
*dst
;
1328 union tgsi_exec_channel index2D
;
1329 uint execmask
= mach
->ExecMask
;
1330 int offset
= 0; /* indirection offset */
1334 if (0 && dst_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1335 check_inf_or_nan(chan
);
1338 /* There is an extra source register that indirectly subscripts
1339 * a register file. The direct index now becomes an offset
1340 * that is being added to the indirect register.
1344 * ind = Indirect.File
1345 * [2] = Indirect.Index
1346 * .x = Indirect.SwizzleX
1348 if (reg
->Register
.Indirect
) {
1349 union tgsi_exec_channel index
;
1350 union tgsi_exec_channel indir_index
;
1353 /* which address register (always zero for now) */
1357 index
.i
[3] = reg
->Indirect
.Index
;
1359 /* get current value of address register[swizzle] */
1360 swizzle
= reg
->Indirect
.Swizzle
;
1362 /* fetch values from the address/indirection register */
1363 fetch_src_file_channel(mach
,
1371 /* save indirection offset */
1372 offset
= indir_index
.i
[0];
1375 /* There is an extra source register that is a second
1376 * subscript to a register file. Effectively it means that
1377 * the register file is actually a 2D array of registers.
1381 * [3] = Dimension.Index
1383 if (reg
->Register
.Dimension
) {
1387 index2D
.i
[3] = reg
->Dimension
.Index
;
1389 /* Again, the second subscript index can be addressed indirectly
1390 * identically to the first one.
1391 * Nothing stops us from indirectly addressing the indirect register,
1392 * but there is no need for that, so we won't exercise it.
1394 * file[ind[4].y+3][1],
1396 * ind = DimIndirect.File
1397 * [4] = DimIndirect.Index
1398 * .y = DimIndirect.SwizzleX
1400 if (reg
->Dimension
.Indirect
) {
1401 union tgsi_exec_channel index2
;
1402 union tgsi_exec_channel indir_index
;
1403 const uint execmask
= mach
->ExecMask
;
1410 index2
.i
[3] = reg
->DimIndirect
.Index
;
1412 swizzle
= reg
->DimIndirect
.Swizzle
;
1413 fetch_src_file_channel(mach
,
1415 reg
->DimIndirect
.File
,
1421 index2D
.i
[0] += indir_index
.i
[0];
1422 index2D
.i
[1] += indir_index
.i
[1];
1423 index2D
.i
[2] += indir_index
.i
[2];
1424 index2D
.i
[3] += indir_index
.i
[3];
1426 /* for disabled execution channels, zero-out the index to
1427 * avoid using a potential garbage value.
1429 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1430 if ((execmask
& (1 << i
)) == 0) {
1436 /* If by any chance there was a need for a 3D array of register
1437 * files, we would have to check whether Dimension is followed
1438 * by a dimension register and continue the saga.
1447 switch (reg
->Register
.File
) {
1448 case TGSI_FILE_NULL
:
1452 case TGSI_FILE_OUTPUT
:
1453 index
= mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0]
1454 + reg
->Register
.Index
;
1455 dst
= &mach
->Outputs
[offset
+ index
].xyzw
[chan_index
];
1457 debug_printf("NumOutputs = %d, TEMP_O_C/I = %d, redindex = %d\n",
1458 mach
->NumOutputs
, mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0],
1459 reg
->Register
.Index
);
1460 if (TGSI_PROCESSOR_GEOMETRY
== mach
->Processor
) {
1461 debug_printf("STORING OUT[%d] mask(%d), = (", offset
+ index
, execmask
);
1462 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1463 if (execmask
& (1 << i
))
1464 debug_printf("%f, ", chan
->f
[i
]);
1465 debug_printf(")\n");
1470 case TGSI_FILE_TEMPORARY
:
1471 index
= reg
->Register
.Index
;
1472 assert( index
< TGSI_EXEC_NUM_TEMPS
);
1473 dst
= &mach
->Temps
[offset
+ index
].xyzw
[chan_index
];
1476 case TGSI_FILE_ADDRESS
:
1477 index
= reg
->Register
.Index
;
1478 dst
= &mach
->Addrs
[index
].xyzw
[chan_index
];
1481 case TGSI_FILE_PREDICATE
:
1482 index
= reg
->Register
.Index
;
1483 assert(index
< TGSI_EXEC_NUM_PREDS
);
1484 dst
= &mach
->Predicates
[index
].xyzw
[chan_index
];
1492 if (inst
->Instruction
.Predicate
) {
1494 union tgsi_exec_channel
*pred
;
1496 switch (chan_index
) {
1498 swizzle
= inst
->Predicate
.SwizzleX
;
1501 swizzle
= inst
->Predicate
.SwizzleY
;
1504 swizzle
= inst
->Predicate
.SwizzleZ
;
1507 swizzle
= inst
->Predicate
.SwizzleW
;
1514 assert(inst
->Predicate
.Index
== 0);
1516 pred
= &mach
->Predicates
[inst
->Predicate
.Index
].xyzw
[swizzle
];
1518 if (inst
->Predicate
.Negate
) {
1519 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1521 execmask
&= ~(1 << i
);
1525 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1527 execmask
&= ~(1 << i
);
1533 switch (inst
->Instruction
.Saturate
) {
1535 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1536 if (execmask
& (1 << i
))
1537 dst
->i
[i
] = chan
->i
[i
];
1540 case TGSI_SAT_ZERO_ONE
:
1541 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1542 if (execmask
& (1 << i
)) {
1543 if (chan
->f
[i
] < 0.0f
)
1545 else if (chan
->f
[i
] > 1.0f
)
1548 dst
->i
[i
] = chan
->i
[i
];
1552 case TGSI_SAT_MINUS_PLUS_ONE
:
1553 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1554 if (execmask
& (1 << i
)) {
1555 if (chan
->f
[i
] < -1.0f
)
1557 else if (chan
->f
[i
] > 1.0f
)
1560 dst
->i
[i
] = chan
->i
[i
];
1569 #define FETCH(VAL,INDEX,CHAN)\
1570 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT)
1572 #define IFETCH(VAL,INDEX,CHAN)\
1573 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT)
1577 * Execute ARB-style KIL which is predicated by a src register.
1578 * Kill fragment if any of the four values is less than zero.
1581 exec_kill_if(struct tgsi_exec_machine
*mach
,
1582 const struct tgsi_full_instruction
*inst
)
1586 uint kilmask
= 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1587 union tgsi_exec_channel r
[1];
1589 /* This mask stores component bits that were already tested. */
1592 for (chan_index
= 0; chan_index
< 4; chan_index
++)
1597 /* unswizzle channel */
1598 swizzle
= tgsi_util_get_full_src_register_swizzle (
1602 /* check if the component has not been already tested */
1603 if (uniquemask
& (1 << swizzle
))
1605 uniquemask
|= 1 << swizzle
;
1607 FETCH(&r
[0], 0, chan_index
);
1608 for (i
= 0; i
< 4; i
++)
1609 if (r
[0].f
[i
] < 0.0f
)
1613 /* restrict to fragments currently executing */
1614 kilmask
&= mach
->ExecMask
;
1616 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1620 * Unconditional fragment kill/discard.
1623 exec_kill(struct tgsi_exec_machine
*mach
,
1624 const struct tgsi_full_instruction
*inst
)
1626 uint kilmask
; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1628 /* kill fragment for all fragments currently executing */
1629 kilmask
= mach
->ExecMask
;
1630 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1634 emit_vertex(struct tgsi_exec_machine
*mach
)
1636 /* FIXME: check for exec mask correctly
1638 for (i = 0; i < TGSI_QUAD_SIZE; ++i) {
1639 if ((mach->ExecMask & (1 << i)))
1641 if (mach
->ExecMask
) {
1642 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] += mach
->NumOutputs
;
1643 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]]++;
1648 emit_primitive(struct tgsi_exec_machine
*mach
)
1650 unsigned *prim_count
= &mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0];
1651 /* FIXME: check for exec mask correctly
1653 for (i = 0; i < TGSI_QUAD_SIZE; ++i) {
1654 if ((mach->ExecMask & (1 << i)))
1656 if (mach
->ExecMask
) {
1658 debug_assert((*prim_count
* mach
->NumOutputs
) < mach
->MaxGeometryShaderOutputs
);
1659 mach
->Primitives
[*prim_count
] = 0;
1664 conditional_emit_primitive(struct tgsi_exec_machine
*mach
)
1666 if (TGSI_PROCESSOR_GEOMETRY
== mach
->Processor
) {
1668 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]];
1669 if (emitted_verts
) {
1670 emit_primitive(mach
);
1677 * Fetch four texture samples using STR texture coordinates.
1680 fetch_texel( struct tgsi_sampler
*sampler
,
1681 const unsigned sview_idx
,
1682 const unsigned sampler_idx
,
1683 const union tgsi_exec_channel
*s
,
1684 const union tgsi_exec_channel
*t
,
1685 const union tgsi_exec_channel
*p
,
1686 const union tgsi_exec_channel
*c0
,
1687 const union tgsi_exec_channel
*c1
,
1688 float derivs
[3][2][TGSI_QUAD_SIZE
],
1689 const int8_t offset
[3],
1690 enum tgsi_sampler_control control
,
1691 union tgsi_exec_channel
*r
,
1692 union tgsi_exec_channel
*g
,
1693 union tgsi_exec_channel
*b
,
1694 union tgsi_exec_channel
*a
)
1697 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
1699 /* FIXME: handle explicit derivs, offsets */
1700 sampler
->get_samples(sampler
, sview_idx
, sampler_idx
,
1701 s
->f
, t
->f
, p
->f
, c0
->f
, c1
->f
, derivs
, offset
, control
, rgba
);
1703 for (j
= 0; j
< 4; j
++) {
1704 r
->f
[j
] = rgba
[0][j
];
1705 g
->f
[j
] = rgba
[1][j
];
1706 b
->f
[j
] = rgba
[2][j
];
1707 a
->f
[j
] = rgba
[3][j
];
1712 #define TEX_MODIFIER_NONE 0
1713 #define TEX_MODIFIER_PROJECTED 1
1714 #define TEX_MODIFIER_LOD_BIAS 2
1715 #define TEX_MODIFIER_EXPLICIT_LOD 3
1716 #define TEX_MODIFIER_LEVEL_ZERO 4
1720 * Fetch all 3 (for s,t,r coords) texel offsets, put them into int array.
1723 fetch_texel_offsets(struct tgsi_exec_machine
*mach
,
1724 const struct tgsi_full_instruction
*inst
,
1727 if (inst
->Texture
.NumOffsets
== 1) {
1728 union tgsi_exec_channel index
;
1729 union tgsi_exec_channel offset
[3];
1730 index
.i
[0] = index
.i
[1] = index
.i
[2] = index
.i
[3] = inst
->TexOffsets
[0].Index
;
1731 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
1732 inst
->TexOffsets
[0].SwizzleX
, &index
, &ZeroVec
, &offset
[0]);
1733 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
1734 inst
->TexOffsets
[0].SwizzleY
, &index
, &ZeroVec
, &offset
[1]);
1735 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
1736 inst
->TexOffsets
[0].SwizzleZ
, &index
, &ZeroVec
, &offset
[2]);
1737 offsets
[0] = offset
[0].i
[0];
1738 offsets
[1] = offset
[1].i
[0];
1739 offsets
[2] = offset
[2].i
[0];
1741 assert(inst
->Texture
.NumOffsets
== 0);
1742 offsets
[0] = offsets
[1] = offsets
[2] = 0;
1748 * Fetch dx and dy values for one channel (s, t or r).
1749 * Put dx values into one float array, dy values into another.
1752 fetch_assign_deriv_channel(struct tgsi_exec_machine
*mach
,
1753 const struct tgsi_full_instruction
*inst
,
1756 float derivs
[2][TGSI_QUAD_SIZE
])
1758 union tgsi_exec_channel d
;
1759 FETCH(&d
, regdsrcx
, chan
);
1760 derivs
[0][0] = d
.f
[0];
1761 derivs
[0][1] = d
.f
[1];
1762 derivs
[0][2] = d
.f
[2];
1763 derivs
[0][3] = d
.f
[3];
1764 FETCH(&d
, regdsrcx
+ 1, chan
);
1765 derivs
[1][0] = d
.f
[0];
1766 derivs
[1][1] = d
.f
[1];
1767 derivs
[1][2] = d
.f
[2];
1768 derivs
[1][3] = d
.f
[3];
1773 * execute a texture instruction.
1775 * modifier is used to control the channel routing for the\
1776 * instruction variants like proj, lod, and texture with lod bias.
1777 * sampler indicates which src register the sampler is contained in.
1780 exec_tex(struct tgsi_exec_machine
*mach
,
1781 const struct tgsi_full_instruction
*inst
,
1782 uint modifier
, uint sampler
)
1784 const uint unit
= inst
->Src
[sampler
].Register
.Index
;
1785 const union tgsi_exec_channel
*args
[5], *proj
= NULL
;
1786 union tgsi_exec_channel r
[5];
1787 enum tgsi_sampler_control control
= tgsi_sampler_lod_none
;
1790 int dim
, shadow_ref
, i
;
1792 /* always fetch all 3 offsets, overkill but keeps code simple */
1793 fetch_texel_offsets(mach
, inst
, offsets
);
1795 assert(modifier
!= TEX_MODIFIER_LEVEL_ZERO
);
1796 assert(inst
->Texture
.Texture
!= TGSI_TEXTURE_BUFFER
);
1798 dim
= tgsi_util_get_texture_coord_dim(inst
->Texture
.Texture
, &shadow_ref
);
1801 if (shadow_ref
>= 0)
1802 assert(shadow_ref
>= dim
&& shadow_ref
< Elements(args
));
1804 /* fetch modifier to the last argument */
1805 if (modifier
!= TEX_MODIFIER_NONE
) {
1806 const int last
= Elements(args
) - 1;
1808 /* fetch modifier from src0.w or src1.x */
1810 assert(dim
<= TGSI_CHAN_W
&& shadow_ref
!= TGSI_CHAN_W
);
1811 FETCH(&r
[last
], 0, TGSI_CHAN_W
);
1814 assert(shadow_ref
!= 4);
1815 FETCH(&r
[last
], 1, TGSI_CHAN_X
);
1818 if (modifier
!= TEX_MODIFIER_PROJECTED
) {
1819 args
[last
] = &r
[last
];
1823 args
[last
] = &ZeroVec
;
1826 /* point unused arguments to zero vector */
1827 for (i
= dim
; i
< last
; i
++)
1830 if (modifier
== TEX_MODIFIER_EXPLICIT_LOD
)
1831 control
= tgsi_sampler_lod_explicit
;
1832 else if (modifier
== TEX_MODIFIER_LOD_BIAS
)
1833 control
= tgsi_sampler_lod_bias
;
1836 for (i
= dim
; i
< Elements(args
); i
++)
1840 /* fetch coordinates */
1841 for (i
= 0; i
< dim
; i
++) {
1842 FETCH(&r
[i
], 0, TGSI_CHAN_X
+ i
);
1845 micro_div(&r
[i
], &r
[i
], proj
);
1850 /* fetch reference value */
1851 if (shadow_ref
>= 0) {
1852 FETCH(&r
[shadow_ref
], shadow_ref
/ 4, TGSI_CHAN_X
+ (shadow_ref
% 4));
1855 micro_div(&r
[shadow_ref
], &r
[shadow_ref
], proj
);
1857 args
[shadow_ref
] = &r
[shadow_ref
];
1860 fetch_texel(mach
->Sampler
, unit
, unit
,
1861 args
[0], args
[1], args
[2], args
[3], args
[4],
1862 NULL
, offsets
, control
,
1863 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1866 debug_printf("fetch r: %g %g %g %g\n",
1867 r
[0].f
[0], r
[0].f
[1], r
[0].f
[2], r
[0].f
[3]);
1868 debug_printf("fetch g: %g %g %g %g\n",
1869 r
[1].f
[0], r
[1].f
[1], r
[1].f
[2], r
[1].f
[3]);
1870 debug_printf("fetch b: %g %g %g %g\n",
1871 r
[2].f
[0], r
[2].f
[1], r
[2].f
[2], r
[2].f
[3]);
1872 debug_printf("fetch a: %g %g %g %g\n",
1873 r
[3].f
[0], r
[3].f
[1], r
[3].f
[2], r
[3].f
[3]);
1876 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
1877 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1878 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
1885 exec_txd(struct tgsi_exec_machine
*mach
,
1886 const struct tgsi_full_instruction
*inst
)
1888 const uint unit
= inst
->Src
[3].Register
.Index
;
1889 union tgsi_exec_channel r
[4];
1890 float derivs
[3][2][TGSI_QUAD_SIZE
];
1894 /* always fetch all 3 offsets, overkill but keeps code simple */
1895 fetch_texel_offsets(mach
, inst
, offsets
);
1897 switch (inst
->Texture
.Texture
) {
1898 case TGSI_TEXTURE_1D
:
1899 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1901 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1903 fetch_texel(mach
->Sampler
, unit
, unit
,
1904 &r
[0], &ZeroVec
, &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
1905 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1906 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1909 case TGSI_TEXTURE_SHADOW1D
:
1910 case TGSI_TEXTURE_1D_ARRAY
:
1911 case TGSI_TEXTURE_SHADOW1D_ARRAY
:
1912 /* SHADOW1D/1D_ARRAY would not need Y/Z respectively, but don't bother */
1913 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1914 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
1915 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
1917 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1919 fetch_texel(mach
->Sampler
, unit
, unit
,
1920 &r
[0], &r
[1], &r
[2], &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
1921 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1922 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1925 case TGSI_TEXTURE_2D
:
1926 case TGSI_TEXTURE_RECT
:
1927 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1928 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
1930 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1931 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
1933 fetch_texel(mach
->Sampler
, unit
, unit
,
1934 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
1935 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1936 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1940 case TGSI_TEXTURE_SHADOW2D
:
1941 case TGSI_TEXTURE_SHADOWRECT
:
1942 case TGSI_TEXTURE_2D_ARRAY
:
1943 case TGSI_TEXTURE_SHADOW2D_ARRAY
:
1944 /* only SHADOW2D_ARRAY actually needs W */
1945 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1946 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
1947 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
1948 FETCH(&r
[3], 0, TGSI_CHAN_W
);
1950 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1951 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
1953 fetch_texel(mach
->Sampler
, unit
, unit
,
1954 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
, /* inputs */
1955 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1956 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1959 case TGSI_TEXTURE_3D
:
1960 case TGSI_TEXTURE_CUBE
:
1961 case TGSI_TEXTURE_CUBE_ARRAY
:
1962 /* only TEXTURE_CUBE_ARRAY actually needs W */
1963 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1964 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
1965 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
1966 FETCH(&r
[3], 0, TGSI_CHAN_W
);
1968 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1969 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
1970 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Z
, derivs
[2]);
1972 fetch_texel(mach
->Sampler
, unit
, unit
,
1973 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
, /* inputs */
1974 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1975 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1982 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
1983 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1984 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
1991 exec_txf(struct tgsi_exec_machine
*mach
,
1992 const struct tgsi_full_instruction
*inst
)
1994 const uint unit
= inst
->Src
[1].Register
.Index
;
1995 union tgsi_exec_channel r
[4];
1997 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2002 /* always fetch all 3 offsets, overkill but keeps code simple */
2003 fetch_texel_offsets(mach
, inst
, offsets
);
2005 IFETCH(&r
[3], 0, TGSI_CHAN_W
);
2007 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_SAMPLE_I
) {
2008 target
= mach
->SamplerViews
[unit
].Resource
;
2011 target
= inst
->Texture
.Texture
;
2014 case TGSI_TEXTURE_3D
:
2015 case TGSI_TEXTURE_2D_ARRAY
:
2016 case TGSI_TEXTURE_SHADOW2D_ARRAY
:
2017 IFETCH(&r
[2], 0, TGSI_CHAN_Z
);
2019 case TGSI_TEXTURE_2D
:
2020 case TGSI_TEXTURE_RECT
:
2021 case TGSI_TEXTURE_SHADOW1D_ARRAY
:
2022 case TGSI_TEXTURE_SHADOW2D
:
2023 case TGSI_TEXTURE_SHADOWRECT
:
2024 case TGSI_TEXTURE_1D_ARRAY
:
2025 IFETCH(&r
[1], 0, TGSI_CHAN_Y
);
2027 case TGSI_TEXTURE_BUFFER
:
2028 case TGSI_TEXTURE_1D
:
2029 case TGSI_TEXTURE_SHADOW1D
:
2030 IFETCH(&r
[0], 0, TGSI_CHAN_X
);
2037 mach
->Sampler
->get_texel(mach
->Sampler
, unit
, r
[0].i
, r
[1].i
, r
[2].i
, r
[3].i
,
2040 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2041 r
[0].f
[j
] = rgba
[0][j
];
2042 r
[1].f
[j
] = rgba
[1][j
];
2043 r
[2].f
[j
] = rgba
[2][j
];
2044 r
[3].f
[j
] = rgba
[3][j
];
2047 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_SAMPLE_I
) {
2048 unsigned char swizzles
[4];
2049 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2050 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2051 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2052 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2054 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2055 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2056 store_dest(mach
, &r
[swizzles
[chan
]],
2057 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2062 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2063 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2064 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2071 exec_txq(struct tgsi_exec_machine
*mach
,
2072 const struct tgsi_full_instruction
*inst
)
2074 const uint unit
= inst
->Src
[1].Register
.Index
;
2076 union tgsi_exec_channel r
[4], src
;
2080 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_INT
);
2082 /* XXX: This interface can't return per-pixel values */
2083 mach
->Sampler
->get_dims(mach
->Sampler
, unit
, src
.i
[0], result
);
2085 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2086 for (j
= 0; j
< 4; j
++) {
2087 r
[j
].i
[i
] = result
[j
];
2091 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2092 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2093 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
,
2094 TGSI_EXEC_DATA_INT
);
2100 exec_sample(struct tgsi_exec_machine
*mach
,
2101 const struct tgsi_full_instruction
*inst
,
2102 uint modifier
, boolean compare
)
2104 const uint resource_unit
= inst
->Src
[1].Register
.Index
;
2105 const uint sampler_unit
= inst
->Src
[2].Register
.Index
;
2106 union tgsi_exec_channel r
[4], c1
;
2107 const union tgsi_exec_channel
*lod
= &ZeroVec
;
2108 enum tgsi_sampler_control control
= tgsi_sampler_lod_none
;
2110 unsigned char swizzles
[4];
2113 /* always fetch all 3 offsets, overkill but keeps code simple */
2114 fetch_texel_offsets(mach
, inst
, offsets
);
2116 assert(modifier
!= TEX_MODIFIER_PROJECTED
);
2118 if (modifier
!= TEX_MODIFIER_NONE
) {
2119 if (modifier
== TEX_MODIFIER_LOD_BIAS
) {
2120 FETCH(&c1
, 3, TGSI_CHAN_X
);
2122 control
= tgsi_sampler_lod_bias
;
2124 else if (modifier
== TEX_MODIFIER_EXPLICIT_LOD
) {
2125 FETCH(&c1
, 3, TGSI_CHAN_X
);
2127 control
= tgsi_sampler_lod_explicit
;
2130 assert(modifier
== TEX_MODIFIER_LEVEL_ZERO
);
2131 control
= tgsi_sampler_lod_zero
;
2135 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2137 switch (mach
->SamplerViews
[resource_unit
].Resource
) {
2138 case TGSI_TEXTURE_1D
:
2140 FETCH(&r
[2], 3, TGSI_CHAN_X
);
2141 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2142 &r
[0], &ZeroVec
, &r
[2], &ZeroVec
, lod
, /* S, T, P, C, LOD */
2143 NULL
, offsets
, control
,
2144 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2147 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2148 &r
[0], &ZeroVec
, &ZeroVec
, &ZeroVec
, lod
, /* S, T, P, C, LOD */
2149 NULL
, offsets
, control
,
2150 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2154 case TGSI_TEXTURE_1D_ARRAY
:
2155 case TGSI_TEXTURE_2D
:
2156 case TGSI_TEXTURE_RECT
:
2157 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2159 FETCH(&r
[2], 3, TGSI_CHAN_X
);
2160 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2161 &r
[0], &r
[1], &r
[2], &ZeroVec
, lod
, /* S, T, P, C, LOD */
2162 NULL
, offsets
, control
,
2163 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2166 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2167 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, lod
, /* S, T, P, C, LOD */
2168 NULL
, offsets
, control
,
2169 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2173 case TGSI_TEXTURE_2D_ARRAY
:
2174 case TGSI_TEXTURE_3D
:
2175 case TGSI_TEXTURE_CUBE
:
2176 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2177 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2179 FETCH(&r
[3], 3, TGSI_CHAN_X
);
2180 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2181 &r
[0], &r
[1], &r
[2], &r
[3], lod
,
2182 NULL
, offsets
, control
,
2183 &r
[0], &r
[1], &r
[2], &r
[3]);
2186 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2187 &r
[0], &r
[1], &r
[2], &ZeroVec
, lod
,
2188 NULL
, offsets
, control
,
2189 &r
[0], &r
[1], &r
[2], &r
[3]);
2193 case TGSI_TEXTURE_CUBE_ARRAY
:
2194 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2195 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2196 FETCH(&r
[3], 0, TGSI_CHAN_W
);
2198 FETCH(&r
[4], 3, TGSI_CHAN_X
);
2199 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2200 &r
[0], &r
[1], &r
[2], &r
[3], &r
[4],
2201 NULL
, offsets
, control
,
2202 &r
[0], &r
[1], &r
[2], &r
[3]);
2205 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2206 &r
[0], &r
[1], &r
[2], &r
[3], lod
,
2207 NULL
, offsets
, control
,
2208 &r
[0], &r
[1], &r
[2], &r
[3]);
2217 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2218 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2219 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2220 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2222 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2223 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2224 store_dest(mach
, &r
[swizzles
[chan
]],
2225 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2231 exec_sample_d(struct tgsi_exec_machine
*mach
,
2232 const struct tgsi_full_instruction
*inst
)
2234 const uint resource_unit
= inst
->Src
[1].Register
.Index
;
2235 const uint sampler_unit
= inst
->Src
[2].Register
.Index
;
2236 union tgsi_exec_channel r
[4];
2237 float derivs
[3][2][TGSI_QUAD_SIZE
];
2239 unsigned char swizzles
[4];
2242 /* always fetch all 3 offsets, overkill but keeps code simple */
2243 fetch_texel_offsets(mach
, inst
, offsets
);
2245 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2247 switch (mach
->SamplerViews
[resource_unit
].Resource
) {
2248 case TGSI_TEXTURE_1D
:
2249 case TGSI_TEXTURE_1D_ARRAY
:
2250 /* only 1D array actually needs Y */
2251 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2253 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2255 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2256 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
2257 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
2258 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2261 case TGSI_TEXTURE_2D
:
2262 case TGSI_TEXTURE_RECT
:
2263 case TGSI_TEXTURE_2D_ARRAY
:
2264 /* only 2D array actually needs Z */
2265 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2266 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2268 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2269 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Y
, derivs
[1]);
2271 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2272 &r
[0], &r
[1], &r
[2], &ZeroVec
, &ZeroVec
, /* inputs */
2273 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
2274 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2277 case TGSI_TEXTURE_3D
:
2278 case TGSI_TEXTURE_CUBE
:
2279 case TGSI_TEXTURE_CUBE_ARRAY
:
2280 /* only cube array actually needs W */
2281 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2282 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2283 FETCH(&r
[3], 0, TGSI_CHAN_W
);
2285 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2286 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Y
, derivs
[1]);
2287 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Z
, derivs
[2]);
2289 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2290 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
,
2291 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
2292 &r
[0], &r
[1], &r
[2], &r
[3]);
2299 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2300 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2301 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2302 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2304 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2305 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2306 store_dest(mach
, &r
[swizzles
[chan
]],
2307 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2314 * Evaluate a constant-valued coefficient at the position of the
2319 struct tgsi_exec_machine
*mach
,
2325 for( i
= 0; i
< TGSI_QUAD_SIZE
; i
++ ) {
2326 mach
->Inputs
[attrib
].xyzw
[chan
].f
[i
] = mach
->InterpCoefs
[attrib
].a0
[chan
];
2331 * Evaluate a linear-valued coefficient at the position of the
2336 struct tgsi_exec_machine
*mach
,
2340 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
2341 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
2342 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
2343 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
2344 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
2345 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
;
2346 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = a0
+ dadx
;
2347 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = a0
+ dady
;
2348 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = a0
+ dadx
+ dady
;
2352 * Evaluate a perspective-valued coefficient at the position of the
2356 eval_perspective_coef(
2357 struct tgsi_exec_machine
*mach
,
2361 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
2362 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
2363 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
2364 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
2365 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
2366 const float *w
= mach
->QuadPos
.xyzw
[3].f
;
2367 /* divide by W here */
2368 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
/ w
[0];
2369 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = (a0
+ dadx
) / w
[1];
2370 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = (a0
+ dady
) / w
[2];
2371 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = (a0
+ dadx
+ dady
) / w
[3];
2375 typedef void (* eval_coef_func
)(
2376 struct tgsi_exec_machine
*mach
,
2381 exec_declaration(struct tgsi_exec_machine
*mach
,
2382 const struct tgsi_full_declaration
*decl
)
2384 if (decl
->Declaration
.File
== TGSI_FILE_SAMPLER_VIEW
) {
2385 mach
->SamplerViews
[decl
->Range
.First
] = decl
->SamplerView
;
2389 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
2390 if (decl
->Declaration
.File
== TGSI_FILE_INPUT
) {
2391 uint first
, last
, mask
;
2393 first
= decl
->Range
.First
;
2394 last
= decl
->Range
.Last
;
2395 mask
= decl
->Declaration
.UsageMask
;
2397 /* XXX we could remove this special-case code since
2398 * mach->InterpCoefs[first].a0 should already have the
2399 * front/back-face value. But we should first update the
2400 * ureg code to emit the right UsageMask value (WRITEMASK_X).
2401 * Then, we could remove the tgsi_exec_machine::Face field.
2403 /* XXX make FACE a system value */
2404 if (decl
->Semantic
.Name
== TGSI_SEMANTIC_FACE
) {
2407 assert(decl
->Semantic
.Index
== 0);
2408 assert(first
== last
);
2410 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2411 mach
->Inputs
[first
].xyzw
[0].f
[i
] = mach
->Face
;
2414 eval_coef_func eval
;
2417 switch (decl
->Interp
.Interpolate
) {
2418 case TGSI_INTERPOLATE_CONSTANT
:
2419 eval
= eval_constant_coef
;
2422 case TGSI_INTERPOLATE_LINEAR
:
2423 eval
= eval_linear_coef
;
2426 case TGSI_INTERPOLATE_PERSPECTIVE
:
2427 eval
= eval_perspective_coef
;
2430 case TGSI_INTERPOLATE_COLOR
:
2431 eval
= mach
->flatshade_color
? eval_constant_coef
: eval_perspective_coef
;
2439 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
2440 if (mask
& (1 << j
)) {
2441 for (i
= first
; i
<= last
; i
++) {
2448 if (DEBUG_EXECUTION
) {
2450 for (i
= first
; i
<= last
; ++i
) {
2451 debug_printf("IN[%2u] = ", i
);
2452 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
2456 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
2457 mach
->Inputs
[i
].xyzw
[0].f
[j
], mach
->Inputs
[i
].xyzw
[0].u
[j
],
2458 mach
->Inputs
[i
].xyzw
[1].f
[j
], mach
->Inputs
[i
].xyzw
[1].u
[j
],
2459 mach
->Inputs
[i
].xyzw
[2].f
[j
], mach
->Inputs
[i
].xyzw
[2].u
[j
],
2460 mach
->Inputs
[i
].xyzw
[3].f
[j
], mach
->Inputs
[i
].xyzw
[3].u
[j
]);
2467 if (decl
->Declaration
.File
== TGSI_FILE_SYSTEM_VALUE
) {
2468 mach
->SysSemanticToIndex
[decl
->Declaration
.Semantic
] = decl
->Range
.First
;
2473 typedef void (* micro_op
)(union tgsi_exec_channel
*dst
);
2476 exec_vector(struct tgsi_exec_machine
*mach
,
2477 const struct tgsi_full_instruction
*inst
,
2479 enum tgsi_exec_datatype dst_datatype
)
2483 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2484 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2485 union tgsi_exec_channel dst
;
2488 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2493 typedef void (* micro_unary_op
)(union tgsi_exec_channel
*dst
,
2494 const union tgsi_exec_channel
*src
);
2497 exec_scalar_unary(struct tgsi_exec_machine
*mach
,
2498 const struct tgsi_full_instruction
*inst
,
2500 enum tgsi_exec_datatype dst_datatype
,
2501 enum tgsi_exec_datatype src_datatype
)
2504 union tgsi_exec_channel src
;
2505 union tgsi_exec_channel dst
;
2507 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, src_datatype
);
2509 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2510 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2511 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2517 exec_vector_unary(struct tgsi_exec_machine
*mach
,
2518 const struct tgsi_full_instruction
*inst
,
2520 enum tgsi_exec_datatype dst_datatype
,
2521 enum tgsi_exec_datatype src_datatype
)
2524 struct tgsi_exec_vector dst
;
2526 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2527 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2528 union tgsi_exec_channel src
;
2530 fetch_source(mach
, &src
, &inst
->Src
[0], chan
, src_datatype
);
2531 op(&dst
.xyzw
[chan
], &src
);
2534 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2535 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2536 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2541 typedef void (* micro_binary_op
)(union tgsi_exec_channel
*dst
,
2542 const union tgsi_exec_channel
*src0
,
2543 const union tgsi_exec_channel
*src1
);
2546 exec_scalar_binary(struct tgsi_exec_machine
*mach
,
2547 const struct tgsi_full_instruction
*inst
,
2549 enum tgsi_exec_datatype dst_datatype
,
2550 enum tgsi_exec_datatype src_datatype
)
2553 union tgsi_exec_channel src
[2];
2554 union tgsi_exec_channel dst
;
2556 fetch_source(mach
, &src
[0], &inst
->Src
[0], TGSI_CHAN_X
, src_datatype
);
2557 fetch_source(mach
, &src
[1], &inst
->Src
[1], TGSI_CHAN_X
, src_datatype
);
2558 op(&dst
, &src
[0], &src
[1]);
2559 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2560 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2561 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2567 exec_vector_binary(struct tgsi_exec_machine
*mach
,
2568 const struct tgsi_full_instruction
*inst
,
2570 enum tgsi_exec_datatype dst_datatype
,
2571 enum tgsi_exec_datatype src_datatype
)
2574 struct tgsi_exec_vector dst
;
2576 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2577 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2578 union tgsi_exec_channel src
[2];
2580 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
2581 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
2582 op(&dst
.xyzw
[chan
], &src
[0], &src
[1]);
2585 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2586 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2587 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2592 typedef void (* micro_trinary_op
)(union tgsi_exec_channel
*dst
,
2593 const union tgsi_exec_channel
*src0
,
2594 const union tgsi_exec_channel
*src1
,
2595 const union tgsi_exec_channel
*src2
);
2598 exec_vector_trinary(struct tgsi_exec_machine
*mach
,
2599 const struct tgsi_full_instruction
*inst
,
2600 micro_trinary_op op
,
2601 enum tgsi_exec_datatype dst_datatype
,
2602 enum tgsi_exec_datatype src_datatype
)
2605 struct tgsi_exec_vector dst
;
2607 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2608 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2609 union tgsi_exec_channel src
[3];
2611 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
2612 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
2613 fetch_source(mach
, &src
[2], &inst
->Src
[2], chan
, src_datatype
);
2614 op(&dst
.xyzw
[chan
], &src
[0], &src
[1], &src
[2]);
2617 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2618 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2619 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2625 exec_dp3(struct tgsi_exec_machine
*mach
,
2626 const struct tgsi_full_instruction
*inst
)
2629 union tgsi_exec_channel arg
[3];
2631 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2632 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2633 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2635 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_Z
; chan
++) {
2636 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2637 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
2638 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2641 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2642 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2643 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2649 exec_dp4(struct tgsi_exec_machine
*mach
,
2650 const struct tgsi_full_instruction
*inst
)
2653 union tgsi_exec_channel arg
[3];
2655 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2656 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2657 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2659 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_W
; chan
++) {
2660 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2661 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
2662 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2665 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2666 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2667 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2673 exec_dp2a(struct tgsi_exec_machine
*mach
,
2674 const struct tgsi_full_instruction
*inst
)
2677 union tgsi_exec_channel arg
[3];
2679 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2680 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2681 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2683 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2684 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2685 micro_mad(&arg
[0], &arg
[0], &arg
[1], &arg
[2]);
2687 fetch_source(mach
, &arg
[1], &inst
->Src
[2], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2688 micro_add(&arg
[0], &arg
[0], &arg
[1]);
2690 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2691 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2692 store_dest(mach
, &arg
[0], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2698 exec_dph(struct tgsi_exec_machine
*mach
,
2699 const struct tgsi_full_instruction
*inst
)
2702 union tgsi_exec_channel arg
[3];
2704 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2705 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2706 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2708 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2709 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2710 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2712 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2713 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2714 micro_mad(&arg
[0], &arg
[0], &arg
[1], &arg
[2]);
2716 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2717 micro_add(&arg
[0], &arg
[0], &arg
[1]);
2719 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2720 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2721 store_dest(mach
, &arg
[0], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2727 exec_dp2(struct tgsi_exec_machine
*mach
,
2728 const struct tgsi_full_instruction
*inst
)
2731 union tgsi_exec_channel arg
[3];
2733 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2734 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2735 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2737 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2738 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2739 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2741 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2742 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2743 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2749 exec_nrm4(struct tgsi_exec_machine
*mach
,
2750 const struct tgsi_full_instruction
*inst
)
2753 union tgsi_exec_channel arg
[4];
2754 union tgsi_exec_channel scale
;
2756 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2757 micro_mul(&scale
, &arg
[0], &arg
[0]);
2759 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_W
; chan
++) {
2760 union tgsi_exec_channel product
;
2762 fetch_source(mach
, &arg
[chan
], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2763 micro_mul(&product
, &arg
[chan
], &arg
[chan
]);
2764 micro_add(&scale
, &scale
, &product
);
2767 micro_rsq(&scale
, &scale
);
2769 for (chan
= TGSI_CHAN_X
; chan
<= TGSI_CHAN_W
; chan
++) {
2770 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2771 micro_mul(&arg
[chan
], &arg
[chan
], &scale
);
2772 store_dest(mach
, &arg
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2778 exec_nrm3(struct tgsi_exec_machine
*mach
,
2779 const struct tgsi_full_instruction
*inst
)
2781 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XYZ
) {
2783 union tgsi_exec_channel arg
[3];
2784 union tgsi_exec_channel scale
;
2786 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2787 micro_mul(&scale
, &arg
[0], &arg
[0]);
2789 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_Z
; chan
++) {
2790 union tgsi_exec_channel product
;
2792 fetch_source(mach
, &arg
[chan
], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2793 micro_mul(&product
, &arg
[chan
], &arg
[chan
]);
2794 micro_add(&scale
, &scale
, &product
);
2797 micro_rsq(&scale
, &scale
);
2799 for (chan
= TGSI_CHAN_X
; chan
<= TGSI_CHAN_Z
; chan
++) {
2800 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2801 micro_mul(&arg
[chan
], &arg
[chan
], &scale
);
2802 store_dest(mach
, &arg
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2807 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2808 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2813 exec_scs(struct tgsi_exec_machine
*mach
,
2814 const struct tgsi_full_instruction
*inst
)
2816 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XY
) {
2817 union tgsi_exec_channel arg
;
2818 union tgsi_exec_channel result
;
2820 fetch_source(mach
, &arg
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2822 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2823 micro_cos(&result
, &arg
);
2824 store_dest(mach
, &result
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2826 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2827 micro_sin(&result
, &arg
);
2828 store_dest(mach
, &result
, &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2831 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2832 store_dest(mach
, &ZeroVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2834 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2835 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2840 exec_x2d(struct tgsi_exec_machine
*mach
,
2841 const struct tgsi_full_instruction
*inst
)
2843 union tgsi_exec_channel r
[4];
2844 union tgsi_exec_channel d
[2];
2846 fetch_source(mach
, &r
[0], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2847 fetch_source(mach
, &r
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2848 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XZ
) {
2849 fetch_source(mach
, &r
[2], &inst
->Src
[2], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2850 micro_mul(&r
[2], &r
[2], &r
[0]);
2851 fetch_source(mach
, &r
[3], &inst
->Src
[2], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2852 micro_mul(&r
[3], &r
[3], &r
[1]);
2853 micro_add(&r
[2], &r
[2], &r
[3]);
2854 fetch_source(mach
, &r
[3], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2855 micro_add(&d
[0], &r
[2], &r
[3]);
2857 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_YW
) {
2858 fetch_source(mach
, &r
[2], &inst
->Src
[2], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2859 micro_mul(&r
[2], &r
[2], &r
[0]);
2860 fetch_source(mach
, &r
[3], &inst
->Src
[2], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2861 micro_mul(&r
[3], &r
[3], &r
[1]);
2862 micro_add(&r
[2], &r
[2], &r
[3]);
2863 fetch_source(mach
, &r
[3], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2864 micro_add(&d
[1], &r
[2], &r
[3]);
2866 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2867 store_dest(mach
, &d
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2869 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2870 store_dest(mach
, &d
[1], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2872 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2873 store_dest(mach
, &d
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2875 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2876 store_dest(mach
, &d
[1], &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2881 exec_rfl(struct tgsi_exec_machine
*mach
,
2882 const struct tgsi_full_instruction
*inst
)
2884 union tgsi_exec_channel r
[9];
2886 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XYZ
) {
2887 /* r0 = dp3(src0, src0) */
2888 fetch_source(mach
, &r
[2], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2889 micro_mul(&r
[0], &r
[2], &r
[2]);
2890 fetch_source(mach
, &r
[4], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2891 micro_mul(&r
[8], &r
[4], &r
[4]);
2892 micro_add(&r
[0], &r
[0], &r
[8]);
2893 fetch_source(mach
, &r
[6], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2894 micro_mul(&r
[8], &r
[6], &r
[6]);
2895 micro_add(&r
[0], &r
[0], &r
[8]);
2897 /* r1 = dp3(src0, src1) */
2898 fetch_source(mach
, &r
[3], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2899 micro_mul(&r
[1], &r
[2], &r
[3]);
2900 fetch_source(mach
, &r
[5], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2901 micro_mul(&r
[8], &r
[4], &r
[5]);
2902 micro_add(&r
[1], &r
[1], &r
[8]);
2903 fetch_source(mach
, &r
[7], &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2904 micro_mul(&r
[8], &r
[6], &r
[7]);
2905 micro_add(&r
[1], &r
[1], &r
[8]);
2907 /* r1 = 2 * r1 / r0 */
2908 micro_add(&r
[1], &r
[1], &r
[1]);
2909 micro_div(&r
[1], &r
[1], &r
[0]);
2911 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2912 micro_mul(&r
[2], &r
[2], &r
[1]);
2913 micro_sub(&r
[2], &r
[2], &r
[3]);
2914 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2916 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2917 micro_mul(&r
[4], &r
[4], &r
[1]);
2918 micro_sub(&r
[4], &r
[4], &r
[5]);
2919 store_dest(mach
, &r
[4], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2921 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2922 micro_mul(&r
[6], &r
[6], &r
[1]);
2923 micro_sub(&r
[6], &r
[6], &r
[7]);
2924 store_dest(mach
, &r
[6], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2927 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2928 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2933 exec_xpd(struct tgsi_exec_machine
*mach
,
2934 const struct tgsi_full_instruction
*inst
)
2936 union tgsi_exec_channel r
[6];
2937 union tgsi_exec_channel d
[3];
2939 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2940 fetch_source(mach
, &r
[1], &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2942 micro_mul(&r
[2], &r
[0], &r
[1]);
2944 fetch_source(mach
, &r
[3], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2945 fetch_source(mach
, &r
[4], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2947 micro_mul(&r
[5], &r
[3], &r
[4] );
2948 micro_sub(&d
[TGSI_CHAN_X
], &r
[2], &r
[5]);
2950 fetch_source(mach
, &r
[2], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2952 micro_mul(&r
[3], &r
[3], &r
[2]);
2954 fetch_source(mach
, &r
[5], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2956 micro_mul(&r
[1], &r
[1], &r
[5]);
2957 micro_sub(&d
[TGSI_CHAN_Y
], &r
[3], &r
[1]);
2959 micro_mul(&r
[5], &r
[5], &r
[4]);
2960 micro_mul(&r
[0], &r
[0], &r
[2]);
2961 micro_sub(&d
[TGSI_CHAN_Z
], &r
[5], &r
[0]);
2963 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2964 store_dest(mach
, &d
[TGSI_CHAN_X
], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2966 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2967 store_dest(mach
, &d
[TGSI_CHAN_Y
], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2969 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2970 store_dest(mach
, &d
[TGSI_CHAN_Z
], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2972 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2973 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2978 exec_dst(struct tgsi_exec_machine
*mach
,
2979 const struct tgsi_full_instruction
*inst
)
2981 union tgsi_exec_channel r
[2];
2982 union tgsi_exec_channel d
[4];
2984 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2985 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2986 fetch_source(mach
, &r
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2987 micro_mul(&d
[TGSI_CHAN_Y
], &r
[0], &r
[1]);
2989 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2990 fetch_source(mach
, &d
[TGSI_CHAN_Z
], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2992 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2993 fetch_source(mach
, &d
[TGSI_CHAN_W
], &inst
->Src
[1], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2996 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2997 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2999 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3000 store_dest(mach
, &d
[TGSI_CHAN_Y
], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3002 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3003 store_dest(mach
, &d
[TGSI_CHAN_Z
], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3005 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3006 store_dest(mach
, &d
[TGSI_CHAN_W
], &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3011 exec_log(struct tgsi_exec_machine
*mach
,
3012 const struct tgsi_full_instruction
*inst
)
3014 union tgsi_exec_channel r
[3];
3016 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3017 micro_abs(&r
[2], &r
[0]); /* r2 = abs(r0) */
3018 micro_lg2(&r
[1], &r
[2]); /* r1 = lg2(r2) */
3019 micro_flr(&r
[0], &r
[1]); /* r0 = floor(r1) */
3020 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3021 store_dest(mach
, &r
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3023 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3024 micro_exp2(&r
[0], &r
[0]); /* r0 = 2 ^ r0 */
3025 micro_div(&r
[0], &r
[2], &r
[0]); /* r0 = r2 / r0 */
3026 store_dest(mach
, &r
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3028 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3029 store_dest(mach
, &r
[1], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3031 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3032 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3037 exec_exp(struct tgsi_exec_machine
*mach
,
3038 const struct tgsi_full_instruction
*inst
)
3040 union tgsi_exec_channel r
[3];
3042 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3043 micro_flr(&r
[1], &r
[0]); /* r1 = floor(r0) */
3044 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3045 micro_exp2(&r
[2], &r
[1]); /* r2 = 2 ^ r1 */
3046 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3048 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3049 micro_sub(&r
[2], &r
[0], &r
[1]); /* r2 = r0 - r1 */
3050 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3052 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3053 micro_exp2(&r
[2], &r
[0]); /* r2 = 2 ^ r0 */
3054 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3056 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3057 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3062 exec_lit(struct tgsi_exec_machine
*mach
,
3063 const struct tgsi_full_instruction
*inst
)
3065 union tgsi_exec_channel r
[3];
3066 union tgsi_exec_channel d
[3];
3068 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_YZ
) {
3069 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3070 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3071 fetch_source(mach
, &r
[1], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3072 micro_max(&r
[1], &r
[1], &ZeroVec
);
3074 fetch_source(mach
, &r
[2], &inst
->Src
[0], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3075 micro_min(&r
[2], &r
[2], &P128Vec
);
3076 micro_max(&r
[2], &r
[2], &M128Vec
);
3077 micro_pow(&r
[1], &r
[1], &r
[2]);
3078 micro_lt(&d
[TGSI_CHAN_Z
], &ZeroVec
, &r
[0], &r
[1], &ZeroVec
);
3079 store_dest(mach
, &d
[TGSI_CHAN_Z
], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3081 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3082 micro_max(&d
[TGSI_CHAN_Y
], &r
[0], &ZeroVec
);
3083 store_dest(mach
, &d
[TGSI_CHAN_Y
], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3086 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3087 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3090 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3091 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3096 exec_break(struct tgsi_exec_machine
*mach
)
3098 if (mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_LOOP
) {
3099 /* turn off loop channels for each enabled exec channel */
3100 mach
->LoopMask
&= ~mach
->ExecMask
;
3101 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3102 UPDATE_EXEC_MASK(mach
);
3104 assert(mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_SWITCH
);
3106 mach
->Switch
.mask
= 0x0;
3108 UPDATE_EXEC_MASK(mach
);
3113 exec_switch(struct tgsi_exec_machine
*mach
,
3114 const struct tgsi_full_instruction
*inst
)
3116 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
3117 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
3119 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
3120 fetch_source(mach
, &mach
->Switch
.selector
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_UINT
);
3121 mach
->Switch
.mask
= 0x0;
3122 mach
->Switch
.defaultMask
= 0x0;
3124 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
3125 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_SWITCH
;
3127 UPDATE_EXEC_MASK(mach
);
3131 exec_case(struct tgsi_exec_machine
*mach
,
3132 const struct tgsi_full_instruction
*inst
)
3134 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
3135 union tgsi_exec_channel src
;
3138 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_UINT
);
3140 if (mach
->Switch
.selector
.u
[0] == src
.u
[0]) {
3143 if (mach
->Switch
.selector
.u
[1] == src
.u
[1]) {
3146 if (mach
->Switch
.selector
.u
[2] == src
.u
[2]) {
3149 if (mach
->Switch
.selector
.u
[3] == src
.u
[3]) {
3153 mach
->Switch
.defaultMask
|= mask
;
3155 mach
->Switch
.mask
|= mask
& prevMask
;
3157 UPDATE_EXEC_MASK(mach
);
3160 /* FIXME: this will only work if default is last */
3162 exec_default(struct tgsi_exec_machine
*mach
)
3164 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
3166 mach
->Switch
.mask
|= ~mach
->Switch
.defaultMask
& prevMask
;
3168 UPDATE_EXEC_MASK(mach
);
3172 exec_endswitch(struct tgsi_exec_machine
*mach
)
3174 mach
->Switch
= mach
->SwitchStack
[--mach
->SwitchStackTop
];
3175 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
3177 UPDATE_EXEC_MASK(mach
);
3181 micro_i2f(union tgsi_exec_channel
*dst
,
3182 const union tgsi_exec_channel
*src
)
3184 dst
->f
[0] = (float)src
->i
[0];
3185 dst
->f
[1] = (float)src
->i
[1];
3186 dst
->f
[2] = (float)src
->i
[2];
3187 dst
->f
[3] = (float)src
->i
[3];
3191 micro_not(union tgsi_exec_channel
*dst
,
3192 const union tgsi_exec_channel
*src
)
3194 dst
->u
[0] = ~src
->u
[0];
3195 dst
->u
[1] = ~src
->u
[1];
3196 dst
->u
[2] = ~src
->u
[2];
3197 dst
->u
[3] = ~src
->u
[3];
3201 micro_shl(union tgsi_exec_channel
*dst
,
3202 const union tgsi_exec_channel
*src0
,
3203 const union tgsi_exec_channel
*src1
)
3205 unsigned masked_count
;
3206 masked_count
= src1
->u
[0] & 0x1f;
3207 dst
->u
[0] = src0
->u
[0] << masked_count
;
3208 masked_count
= src1
->u
[1] & 0x1f;
3209 dst
->u
[1] = src0
->u
[1] << masked_count
;
3210 masked_count
= src1
->u
[2] & 0x1f;
3211 dst
->u
[2] = src0
->u
[2] << masked_count
;
3212 masked_count
= src1
->u
[3] & 0x1f;
3213 dst
->u
[3] = src0
->u
[3] << masked_count
;
3217 micro_and(union tgsi_exec_channel
*dst
,
3218 const union tgsi_exec_channel
*src0
,
3219 const union tgsi_exec_channel
*src1
)
3221 dst
->u
[0] = src0
->u
[0] & src1
->u
[0];
3222 dst
->u
[1] = src0
->u
[1] & src1
->u
[1];
3223 dst
->u
[2] = src0
->u
[2] & src1
->u
[2];
3224 dst
->u
[3] = src0
->u
[3] & src1
->u
[3];
3228 micro_or(union tgsi_exec_channel
*dst
,
3229 const union tgsi_exec_channel
*src0
,
3230 const union tgsi_exec_channel
*src1
)
3232 dst
->u
[0] = src0
->u
[0] | src1
->u
[0];
3233 dst
->u
[1] = src0
->u
[1] | src1
->u
[1];
3234 dst
->u
[2] = src0
->u
[2] | src1
->u
[2];
3235 dst
->u
[3] = src0
->u
[3] | src1
->u
[3];
3239 micro_xor(union tgsi_exec_channel
*dst
,
3240 const union tgsi_exec_channel
*src0
,
3241 const union tgsi_exec_channel
*src1
)
3243 dst
->u
[0] = src0
->u
[0] ^ src1
->u
[0];
3244 dst
->u
[1] = src0
->u
[1] ^ src1
->u
[1];
3245 dst
->u
[2] = src0
->u
[2] ^ src1
->u
[2];
3246 dst
->u
[3] = src0
->u
[3] ^ src1
->u
[3];
3250 micro_mod(union tgsi_exec_channel
*dst
,
3251 const union tgsi_exec_channel
*src0
,
3252 const union tgsi_exec_channel
*src1
)
3254 dst
->i
[0] = src0
->i
[0] % src1
->i
[0];
3255 dst
->i
[1] = src0
->i
[1] % src1
->i
[1];
3256 dst
->i
[2] = src0
->i
[2] % src1
->i
[2];
3257 dst
->i
[3] = src0
->i
[3] % src1
->i
[3];
3261 micro_f2i(union tgsi_exec_channel
*dst
,
3262 const union tgsi_exec_channel
*src
)
3264 dst
->i
[0] = (int)src
->f
[0];
3265 dst
->i
[1] = (int)src
->f
[1];
3266 dst
->i
[2] = (int)src
->f
[2];
3267 dst
->i
[3] = (int)src
->f
[3];
3271 micro_fseq(union tgsi_exec_channel
*dst
,
3272 const union tgsi_exec_channel
*src0
,
3273 const union tgsi_exec_channel
*src1
)
3275 dst
->u
[0] = src0
->f
[0] == src1
->f
[0] ? ~0 : 0;
3276 dst
->u
[1] = src0
->f
[1] == src1
->f
[1] ? ~0 : 0;
3277 dst
->u
[2] = src0
->f
[2] == src1
->f
[2] ? ~0 : 0;
3278 dst
->u
[3] = src0
->f
[3] == src1
->f
[3] ? ~0 : 0;
3282 micro_fsge(union tgsi_exec_channel
*dst
,
3283 const union tgsi_exec_channel
*src0
,
3284 const union tgsi_exec_channel
*src1
)
3286 dst
->u
[0] = src0
->f
[0] >= src1
->f
[0] ? ~0 : 0;
3287 dst
->u
[1] = src0
->f
[1] >= src1
->f
[1] ? ~0 : 0;
3288 dst
->u
[2] = src0
->f
[2] >= src1
->f
[2] ? ~0 : 0;
3289 dst
->u
[3] = src0
->f
[3] >= src1
->f
[3] ? ~0 : 0;
3293 micro_fslt(union tgsi_exec_channel
*dst
,
3294 const union tgsi_exec_channel
*src0
,
3295 const union tgsi_exec_channel
*src1
)
3297 dst
->u
[0] = src0
->f
[0] < src1
->f
[0] ? ~0 : 0;
3298 dst
->u
[1] = src0
->f
[1] < src1
->f
[1] ? ~0 : 0;
3299 dst
->u
[2] = src0
->f
[2] < src1
->f
[2] ? ~0 : 0;
3300 dst
->u
[3] = src0
->f
[3] < src1
->f
[3] ? ~0 : 0;
3304 micro_fsne(union tgsi_exec_channel
*dst
,
3305 const union tgsi_exec_channel
*src0
,
3306 const union tgsi_exec_channel
*src1
)
3308 dst
->u
[0] = src0
->f
[0] != src1
->f
[0] ? ~0 : 0;
3309 dst
->u
[1] = src0
->f
[1] != src1
->f
[1] ? ~0 : 0;
3310 dst
->u
[2] = src0
->f
[2] != src1
->f
[2] ? ~0 : 0;
3311 dst
->u
[3] = src0
->f
[3] != src1
->f
[3] ? ~0 : 0;
3315 micro_idiv(union tgsi_exec_channel
*dst
,
3316 const union tgsi_exec_channel
*src0
,
3317 const union tgsi_exec_channel
*src1
)
3319 dst
->i
[0] = src0
->i
[0] / src1
->i
[0];
3320 dst
->i
[1] = src0
->i
[1] / src1
->i
[1];
3321 dst
->i
[2] = src0
->i
[2] / src1
->i
[2];
3322 dst
->i
[3] = src0
->i
[3] / src1
->i
[3];
3326 micro_imax(union tgsi_exec_channel
*dst
,
3327 const union tgsi_exec_channel
*src0
,
3328 const union tgsi_exec_channel
*src1
)
3330 dst
->i
[0] = src0
->i
[0] > src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
3331 dst
->i
[1] = src0
->i
[1] > src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
3332 dst
->i
[2] = src0
->i
[2] > src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
3333 dst
->i
[3] = src0
->i
[3] > src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
3337 micro_imin(union tgsi_exec_channel
*dst
,
3338 const union tgsi_exec_channel
*src0
,
3339 const union tgsi_exec_channel
*src1
)
3341 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
3342 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
3343 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
3344 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
3348 micro_isge(union tgsi_exec_channel
*dst
,
3349 const union tgsi_exec_channel
*src0
,
3350 const union tgsi_exec_channel
*src1
)
3352 dst
->i
[0] = src0
->i
[0] >= src1
->i
[0] ? -1 : 0;
3353 dst
->i
[1] = src0
->i
[1] >= src1
->i
[1] ? -1 : 0;
3354 dst
->i
[2] = src0
->i
[2] >= src1
->i
[2] ? -1 : 0;
3355 dst
->i
[3] = src0
->i
[3] >= src1
->i
[3] ? -1 : 0;
3359 micro_ishr(union tgsi_exec_channel
*dst
,
3360 const union tgsi_exec_channel
*src0
,
3361 const union tgsi_exec_channel
*src1
)
3363 unsigned masked_count
;
3364 masked_count
= src1
->i
[0] & 0x1f;
3365 dst
->i
[0] = src0
->i
[0] >> masked_count
;
3366 masked_count
= src1
->i
[1] & 0x1f;
3367 dst
->i
[1] = src0
->i
[1] >> masked_count
;
3368 masked_count
= src1
->i
[2] & 0x1f;
3369 dst
->i
[2] = src0
->i
[2] >> masked_count
;
3370 masked_count
= src1
->i
[3] & 0x1f;
3371 dst
->i
[3] = src0
->i
[3] >> masked_count
;
3375 micro_islt(union tgsi_exec_channel
*dst
,
3376 const union tgsi_exec_channel
*src0
,
3377 const union tgsi_exec_channel
*src1
)
3379 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? -1 : 0;
3380 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? -1 : 0;
3381 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? -1 : 0;
3382 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? -1 : 0;
3386 micro_f2u(union tgsi_exec_channel
*dst
,
3387 const union tgsi_exec_channel
*src
)
3389 dst
->u
[0] = (uint
)src
->f
[0];
3390 dst
->u
[1] = (uint
)src
->f
[1];
3391 dst
->u
[2] = (uint
)src
->f
[2];
3392 dst
->u
[3] = (uint
)src
->f
[3];
3396 micro_u2f(union tgsi_exec_channel
*dst
,
3397 const union tgsi_exec_channel
*src
)
3399 dst
->f
[0] = (float)src
->u
[0];
3400 dst
->f
[1] = (float)src
->u
[1];
3401 dst
->f
[2] = (float)src
->u
[2];
3402 dst
->f
[3] = (float)src
->u
[3];
3406 micro_uadd(union tgsi_exec_channel
*dst
,
3407 const union tgsi_exec_channel
*src0
,
3408 const union tgsi_exec_channel
*src1
)
3410 dst
->u
[0] = src0
->u
[0] + src1
->u
[0];
3411 dst
->u
[1] = src0
->u
[1] + src1
->u
[1];
3412 dst
->u
[2] = src0
->u
[2] + src1
->u
[2];
3413 dst
->u
[3] = src0
->u
[3] + src1
->u
[3];
3417 micro_udiv(union tgsi_exec_channel
*dst
,
3418 const union tgsi_exec_channel
*src0
,
3419 const union tgsi_exec_channel
*src1
)
3421 dst
->u
[0] = src1
->u
[0] ? src0
->u
[0] / src1
->u
[0] : ~0u;
3422 dst
->u
[1] = src1
->u
[1] ? src0
->u
[1] / src1
->u
[1] : ~0u;
3423 dst
->u
[2] = src1
->u
[2] ? src0
->u
[2] / src1
->u
[2] : ~0u;
3424 dst
->u
[3] = src1
->u
[3] ? src0
->u
[3] / src1
->u
[3] : ~0u;
3428 micro_umad(union tgsi_exec_channel
*dst
,
3429 const union tgsi_exec_channel
*src0
,
3430 const union tgsi_exec_channel
*src1
,
3431 const union tgsi_exec_channel
*src2
)
3433 dst
->u
[0] = src0
->u
[0] * src1
->u
[0] + src2
->u
[0];
3434 dst
->u
[1] = src0
->u
[1] * src1
->u
[1] + src2
->u
[1];
3435 dst
->u
[2] = src0
->u
[2] * src1
->u
[2] + src2
->u
[2];
3436 dst
->u
[3] = src0
->u
[3] * src1
->u
[3] + src2
->u
[3];
3440 micro_umax(union tgsi_exec_channel
*dst
,
3441 const union tgsi_exec_channel
*src0
,
3442 const union tgsi_exec_channel
*src1
)
3444 dst
->u
[0] = src0
->u
[0] > src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
3445 dst
->u
[1] = src0
->u
[1] > src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
3446 dst
->u
[2] = src0
->u
[2] > src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
3447 dst
->u
[3] = src0
->u
[3] > src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
3451 micro_umin(union tgsi_exec_channel
*dst
,
3452 const union tgsi_exec_channel
*src0
,
3453 const union tgsi_exec_channel
*src1
)
3455 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
3456 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
3457 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
3458 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
3462 micro_umod(union tgsi_exec_channel
*dst
,
3463 const union tgsi_exec_channel
*src0
,
3464 const union tgsi_exec_channel
*src1
)
3466 dst
->u
[0] = src1
->u
[0] ? src0
->u
[0] % src1
->u
[0] : ~0u;
3467 dst
->u
[1] = src1
->u
[1] ? src0
->u
[1] % src1
->u
[1] : ~0u;
3468 dst
->u
[2] = src1
->u
[2] ? src0
->u
[2] % src1
->u
[2] : ~0u;
3469 dst
->u
[3] = src1
->u
[3] ? src0
->u
[3] % src1
->u
[3] : ~0u;
3473 micro_umul(union tgsi_exec_channel
*dst
,
3474 const union tgsi_exec_channel
*src0
,
3475 const union tgsi_exec_channel
*src1
)
3477 dst
->u
[0] = src0
->u
[0] * src1
->u
[0];
3478 dst
->u
[1] = src0
->u
[1] * src1
->u
[1];
3479 dst
->u
[2] = src0
->u
[2] * src1
->u
[2];
3480 dst
->u
[3] = src0
->u
[3] * src1
->u
[3];
3484 micro_imul_hi(union tgsi_exec_channel
*dst
,
3485 const union tgsi_exec_channel
*src0
,
3486 const union tgsi_exec_channel
*src1
)
3488 #define I64M(x, y) ((((int64_t)x) * ((int64_t)y)) >> 32)
3489 dst
->i
[0] = I64M(src0
->i
[0], src1
->i
[0]);
3490 dst
->i
[1] = I64M(src0
->i
[1], src1
->i
[1]);
3491 dst
->i
[2] = I64M(src0
->i
[2], src1
->i
[2]);
3492 dst
->i
[3] = I64M(src0
->i
[3], src1
->i
[3]);
3497 micro_umul_hi(union tgsi_exec_channel
*dst
,
3498 const union tgsi_exec_channel
*src0
,
3499 const union tgsi_exec_channel
*src1
)
3501 #define U64M(x, y) ((((uint64_t)x) * ((uint64_t)y)) >> 32)
3502 dst
->u
[0] = U64M(src0
->u
[0], src1
->u
[0]);
3503 dst
->u
[1] = U64M(src0
->u
[1], src1
->u
[1]);
3504 dst
->u
[2] = U64M(src0
->u
[2], src1
->u
[2]);
3505 dst
->u
[3] = U64M(src0
->u
[3], src1
->u
[3]);
3510 micro_useq(union tgsi_exec_channel
*dst
,
3511 const union tgsi_exec_channel
*src0
,
3512 const union tgsi_exec_channel
*src1
)
3514 dst
->u
[0] = src0
->u
[0] == src1
->u
[0] ? ~0 : 0;
3515 dst
->u
[1] = src0
->u
[1] == src1
->u
[1] ? ~0 : 0;
3516 dst
->u
[2] = src0
->u
[2] == src1
->u
[2] ? ~0 : 0;
3517 dst
->u
[3] = src0
->u
[3] == src1
->u
[3] ? ~0 : 0;
3521 micro_usge(union tgsi_exec_channel
*dst
,
3522 const union tgsi_exec_channel
*src0
,
3523 const union tgsi_exec_channel
*src1
)
3525 dst
->u
[0] = src0
->u
[0] >= src1
->u
[0] ? ~0 : 0;
3526 dst
->u
[1] = src0
->u
[1] >= src1
->u
[1] ? ~0 : 0;
3527 dst
->u
[2] = src0
->u
[2] >= src1
->u
[2] ? ~0 : 0;
3528 dst
->u
[3] = src0
->u
[3] >= src1
->u
[3] ? ~0 : 0;
3532 micro_ushr(union tgsi_exec_channel
*dst
,
3533 const union tgsi_exec_channel
*src0
,
3534 const union tgsi_exec_channel
*src1
)
3536 unsigned masked_count
;
3537 masked_count
= src1
->u
[0] & 0x1f;
3538 dst
->u
[0] = src0
->u
[0] >> masked_count
;
3539 masked_count
= src1
->u
[1] & 0x1f;
3540 dst
->u
[1] = src0
->u
[1] >> masked_count
;
3541 masked_count
= src1
->u
[2] & 0x1f;
3542 dst
->u
[2] = src0
->u
[2] >> masked_count
;
3543 masked_count
= src1
->u
[3] & 0x1f;
3544 dst
->u
[3] = src0
->u
[3] >> masked_count
;
3548 micro_uslt(union tgsi_exec_channel
*dst
,
3549 const union tgsi_exec_channel
*src0
,
3550 const union tgsi_exec_channel
*src1
)
3552 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? ~0 : 0;
3553 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? ~0 : 0;
3554 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? ~0 : 0;
3555 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? ~0 : 0;
3559 micro_usne(union tgsi_exec_channel
*dst
,
3560 const union tgsi_exec_channel
*src0
,
3561 const union tgsi_exec_channel
*src1
)
3563 dst
->u
[0] = src0
->u
[0] != src1
->u
[0] ? ~0 : 0;
3564 dst
->u
[1] = src0
->u
[1] != src1
->u
[1] ? ~0 : 0;
3565 dst
->u
[2] = src0
->u
[2] != src1
->u
[2] ? ~0 : 0;
3566 dst
->u
[3] = src0
->u
[3] != src1
->u
[3] ? ~0 : 0;
3570 micro_uarl(union tgsi_exec_channel
*dst
,
3571 const union tgsi_exec_channel
*src
)
3573 dst
->i
[0] = src
->u
[0];
3574 dst
->i
[1] = src
->u
[1];
3575 dst
->i
[2] = src
->u
[2];
3576 dst
->i
[3] = src
->u
[3];
3580 micro_ucmp(union tgsi_exec_channel
*dst
,
3581 const union tgsi_exec_channel
*src0
,
3582 const union tgsi_exec_channel
*src1
,
3583 const union tgsi_exec_channel
*src2
)
3585 dst
->u
[0] = src0
->u
[0] ? src1
->u
[0] : src2
->u
[0];
3586 dst
->u
[1] = src0
->u
[1] ? src1
->u
[1] : src2
->u
[1];
3587 dst
->u
[2] = src0
->u
[2] ? src1
->u
[2] : src2
->u
[2];
3588 dst
->u
[3] = src0
->u
[3] ? src1
->u
[3] : src2
->u
[3];
3593 struct tgsi_exec_machine
*mach
,
3594 const struct tgsi_full_instruction
*inst
,
3597 union tgsi_exec_channel r
[10];
3601 switch (inst
->Instruction
.Opcode
) {
3602 case TGSI_OPCODE_ARL
:
3603 exec_vector_unary(mach
, inst
, micro_arl
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
3606 case TGSI_OPCODE_MOV
:
3607 exec_vector_unary(mach
, inst
, micro_mov
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
3610 case TGSI_OPCODE_LIT
:
3611 exec_lit(mach
, inst
);
3614 case TGSI_OPCODE_RCP
:
3615 exec_scalar_unary(mach
, inst
, micro_rcp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3618 case TGSI_OPCODE_RSQ
:
3619 exec_scalar_unary(mach
, inst
, micro_rsq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3622 case TGSI_OPCODE_EXP
:
3623 exec_exp(mach
, inst
);
3626 case TGSI_OPCODE_LOG
:
3627 exec_log(mach
, inst
);
3630 case TGSI_OPCODE_MUL
:
3631 exec_vector_binary(mach
, inst
, micro_mul
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3634 case TGSI_OPCODE_ADD
:
3635 exec_vector_binary(mach
, inst
, micro_add
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3638 case TGSI_OPCODE_DP3
:
3639 exec_dp3(mach
, inst
);
3642 case TGSI_OPCODE_DP4
:
3643 exec_dp4(mach
, inst
);
3646 case TGSI_OPCODE_DST
:
3647 exec_dst(mach
, inst
);
3650 case TGSI_OPCODE_MIN
:
3651 exec_vector_binary(mach
, inst
, micro_min
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3654 case TGSI_OPCODE_MAX
:
3655 exec_vector_binary(mach
, inst
, micro_max
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3658 case TGSI_OPCODE_SLT
:
3659 exec_vector_binary(mach
, inst
, micro_slt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3662 case TGSI_OPCODE_SGE
:
3663 exec_vector_binary(mach
, inst
, micro_sge
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3666 case TGSI_OPCODE_MAD
:
3667 exec_vector_trinary(mach
, inst
, micro_mad
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3670 case TGSI_OPCODE_SUB
:
3671 exec_vector_binary(mach
, inst
, micro_sub
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3674 case TGSI_OPCODE_LRP
:
3675 exec_vector_trinary(mach
, inst
, micro_lrp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3678 case TGSI_OPCODE_CND
:
3679 exec_vector_trinary(mach
, inst
, micro_cnd
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3682 case TGSI_OPCODE_SQRT
:
3683 exec_scalar_unary(mach
, inst
, micro_sqrt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3686 case TGSI_OPCODE_DP2A
:
3687 exec_dp2a(mach
, inst
);
3690 case TGSI_OPCODE_FRC
:
3691 exec_vector_unary(mach
, inst
, micro_frc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3694 case TGSI_OPCODE_CLAMP
:
3695 exec_vector_trinary(mach
, inst
, micro_clamp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3698 case TGSI_OPCODE_FLR
:
3699 exec_vector_unary(mach
, inst
, micro_flr
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3702 case TGSI_OPCODE_ROUND
:
3703 exec_vector_unary(mach
, inst
, micro_rnd
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3706 case TGSI_OPCODE_EX2
:
3707 exec_scalar_unary(mach
, inst
, micro_exp2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3710 case TGSI_OPCODE_LG2
:
3711 exec_scalar_unary(mach
, inst
, micro_lg2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3714 case TGSI_OPCODE_POW
:
3715 exec_scalar_binary(mach
, inst
, micro_pow
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3718 case TGSI_OPCODE_XPD
:
3719 exec_xpd(mach
, inst
);
3722 case TGSI_OPCODE_ABS
:
3723 exec_vector_unary(mach
, inst
, micro_abs
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3726 case TGSI_OPCODE_RCC
:
3727 exec_scalar_unary(mach
, inst
, micro_rcc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3730 case TGSI_OPCODE_DPH
:
3731 exec_dph(mach
, inst
);
3734 case TGSI_OPCODE_COS
:
3735 exec_scalar_unary(mach
, inst
, micro_cos
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3738 case TGSI_OPCODE_DDX
:
3739 exec_vector_unary(mach
, inst
, micro_ddx
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3742 case TGSI_OPCODE_DDY
:
3743 exec_vector_unary(mach
, inst
, micro_ddy
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3746 case TGSI_OPCODE_KILL
:
3747 exec_kill (mach
, inst
);
3750 case TGSI_OPCODE_KILL_IF
:
3751 exec_kill_if (mach
, inst
);
3754 case TGSI_OPCODE_PK2H
:
3758 case TGSI_OPCODE_PK2US
:
3762 case TGSI_OPCODE_PK4B
:
3766 case TGSI_OPCODE_PK4UB
:
3770 case TGSI_OPCODE_RFL
:
3771 exec_rfl(mach
, inst
);
3774 case TGSI_OPCODE_SEQ
:
3775 exec_vector_binary(mach
, inst
, micro_seq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3778 case TGSI_OPCODE_SFL
:
3779 exec_vector(mach
, inst
, micro_sfl
, TGSI_EXEC_DATA_FLOAT
);
3782 case TGSI_OPCODE_SGT
:
3783 exec_vector_binary(mach
, inst
, micro_sgt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3786 case TGSI_OPCODE_SIN
:
3787 exec_scalar_unary(mach
, inst
, micro_sin
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3790 case TGSI_OPCODE_SLE
:
3791 exec_vector_binary(mach
, inst
, micro_sle
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3794 case TGSI_OPCODE_SNE
:
3795 exec_vector_binary(mach
, inst
, micro_sne
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3798 case TGSI_OPCODE_STR
:
3799 exec_vector(mach
, inst
, micro_str
, TGSI_EXEC_DATA_FLOAT
);
3802 case TGSI_OPCODE_TEX
:
3803 /* simple texture lookup */
3804 /* src[0] = texcoord */
3805 /* src[1] = sampler unit */
3806 exec_tex(mach
, inst
, TEX_MODIFIER_NONE
, 1);
3809 case TGSI_OPCODE_TXB
:
3810 /* Texture lookup with lod bias */
3811 /* src[0] = texcoord (src[0].w = LOD bias) */
3812 /* src[1] = sampler unit */
3813 exec_tex(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, 1);
3816 case TGSI_OPCODE_TXD
:
3817 /* Texture lookup with explict partial derivatives */
3818 /* src[0] = texcoord */
3819 /* src[1] = d[strq]/dx */
3820 /* src[2] = d[strq]/dy */
3821 /* src[3] = sampler unit */
3822 exec_txd(mach
, inst
);
3825 case TGSI_OPCODE_TXL
:
3826 /* Texture lookup with explit LOD */
3827 /* src[0] = texcoord (src[0].w = LOD) */
3828 /* src[1] = sampler unit */
3829 exec_tex(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, 1);
3832 case TGSI_OPCODE_TXP
:
3833 /* Texture lookup with projection */
3834 /* src[0] = texcoord (src[0].w = projection) */
3835 /* src[1] = sampler unit */
3836 exec_tex(mach
, inst
, TEX_MODIFIER_PROJECTED
, 1);
3839 case TGSI_OPCODE_UP2H
:
3843 case TGSI_OPCODE_UP2US
:
3847 case TGSI_OPCODE_UP4B
:
3851 case TGSI_OPCODE_UP4UB
:
3855 case TGSI_OPCODE_X2D
:
3856 exec_x2d(mach
, inst
);
3859 case TGSI_OPCODE_ARA
:
3863 case TGSI_OPCODE_ARR
:
3864 exec_vector_unary(mach
, inst
, micro_arr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
3867 case TGSI_OPCODE_BRA
:
3871 case TGSI_OPCODE_CAL
:
3872 /* skip the call if no execution channels are enabled */
3873 if (mach
->ExecMask
) {
3876 /* First, record the depths of the execution stacks.
3877 * This is important for deeply nested/looped return statements.
3878 * We have to unwind the stacks by the correct amount. For a
3879 * real code generator, we could determine the number of entries
3880 * to pop off each stack with simple static analysis and avoid
3881 * implementing this data structure at run time.
3883 mach
->CallStack
[mach
->CallStackTop
].CondStackTop
= mach
->CondStackTop
;
3884 mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
= mach
->LoopStackTop
;
3885 mach
->CallStack
[mach
->CallStackTop
].ContStackTop
= mach
->ContStackTop
;
3886 mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
= mach
->SwitchStackTop
;
3887 mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
= mach
->BreakStackTop
;
3888 /* note that PC was already incremented above */
3889 mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
= *pc
;
3891 mach
->CallStackTop
++;
3893 /* Second, push the Cond, Loop, Cont, Func stacks */
3894 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
3895 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3896 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3897 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
3898 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
3899 assert(mach
->FuncStackTop
< TGSI_EXEC_MAX_CALL_NESTING
);
3901 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
3902 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
3903 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
3904 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
3905 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
3906 mach
->FuncStack
[mach
->FuncStackTop
++] = mach
->FuncMask
;
3908 /* Finally, jump to the subroutine */
3909 *pc
= inst
->Label
.Label
;
3913 case TGSI_OPCODE_RET
:
3914 mach
->FuncMask
&= ~mach
->ExecMask
;
3915 UPDATE_EXEC_MASK(mach
);
3917 if (mach
->FuncMask
== 0x0) {
3918 /* really return now (otherwise, keep executing */
3920 if (mach
->CallStackTop
== 0) {
3921 /* returning from main() */
3922 mach
->CondStackTop
= 0;
3923 mach
->LoopStackTop
= 0;
3928 assert(mach
->CallStackTop
> 0);
3929 mach
->CallStackTop
--;
3931 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
3932 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
3934 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
3935 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
3937 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
3938 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
3940 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
3941 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
3943 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
3944 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
3946 assert(mach
->FuncStackTop
> 0);
3947 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
3949 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
3951 UPDATE_EXEC_MASK(mach
);
3955 case TGSI_OPCODE_SSG
:
3956 exec_vector_unary(mach
, inst
, micro_sgn
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3959 case TGSI_OPCODE_CMP
:
3960 exec_vector_trinary(mach
, inst
, micro_cmp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3963 case TGSI_OPCODE_SCS
:
3964 exec_scs(mach
, inst
);
3967 case TGSI_OPCODE_NRM
:
3968 exec_nrm3(mach
, inst
);
3971 case TGSI_OPCODE_NRM4
:
3972 exec_nrm4(mach
, inst
);
3975 case TGSI_OPCODE_DIV
:
3976 exec_vector_binary(mach
, inst
, micro_div
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3979 case TGSI_OPCODE_DP2
:
3980 exec_dp2(mach
, inst
);
3983 case TGSI_OPCODE_IF
:
3985 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
3986 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
3987 FETCH( &r
[0], 0, TGSI_CHAN_X
);
3988 /* update CondMask */
3990 mach
->CondMask
&= ~0x1;
3993 mach
->CondMask
&= ~0x2;
3996 mach
->CondMask
&= ~0x4;
3999 mach
->CondMask
&= ~0x8;
4001 UPDATE_EXEC_MASK(mach
);
4002 /* Todo: If CondMask==0, jump to ELSE */
4005 case TGSI_OPCODE_UIF
:
4007 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
4008 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
4009 IFETCH( &r
[0], 0, TGSI_CHAN_X
);
4010 /* update CondMask */
4012 mach
->CondMask
&= ~0x1;
4015 mach
->CondMask
&= ~0x2;
4018 mach
->CondMask
&= ~0x4;
4021 mach
->CondMask
&= ~0x8;
4023 UPDATE_EXEC_MASK(mach
);
4024 /* Todo: If CondMask==0, jump to ELSE */
4027 case TGSI_OPCODE_ELSE
:
4028 /* invert CondMask wrt previous mask */
4031 assert(mach
->CondStackTop
> 0);
4032 prevMask
= mach
->CondStack
[mach
->CondStackTop
- 1];
4033 mach
->CondMask
= ~mach
->CondMask
& prevMask
;
4034 UPDATE_EXEC_MASK(mach
);
4035 /* Todo: If CondMask==0, jump to ENDIF */
4039 case TGSI_OPCODE_ENDIF
:
4041 assert(mach
->CondStackTop
> 0);
4042 mach
->CondMask
= mach
->CondStack
[--mach
->CondStackTop
];
4043 UPDATE_EXEC_MASK(mach
);
4046 case TGSI_OPCODE_END
:
4047 /* make sure we end primitives which haven't
4048 * been explicitly emitted */
4049 conditional_emit_primitive(mach
);
4050 /* halt execution */
4054 case TGSI_OPCODE_PUSHA
:
4058 case TGSI_OPCODE_POPA
:
4062 case TGSI_OPCODE_CEIL
:
4063 exec_vector_unary(mach
, inst
, micro_ceil
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
4066 case TGSI_OPCODE_I2F
:
4067 exec_vector_unary(mach
, inst
, micro_i2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_INT
);
4070 case TGSI_OPCODE_NOT
:
4071 exec_vector_unary(mach
, inst
, micro_not
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4074 case TGSI_OPCODE_TRUNC
:
4075 exec_vector_unary(mach
, inst
, micro_trunc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
4078 case TGSI_OPCODE_SHL
:
4079 exec_vector_binary(mach
, inst
, micro_shl
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4082 case TGSI_OPCODE_AND
:
4083 exec_vector_binary(mach
, inst
, micro_and
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4086 case TGSI_OPCODE_OR
:
4087 exec_vector_binary(mach
, inst
, micro_or
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4090 case TGSI_OPCODE_MOD
:
4091 exec_vector_binary(mach
, inst
, micro_mod
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4094 case TGSI_OPCODE_XOR
:
4095 exec_vector_binary(mach
, inst
, micro_xor
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4098 case TGSI_OPCODE_SAD
:
4102 case TGSI_OPCODE_TXF
:
4103 exec_txf(mach
, inst
);
4106 case TGSI_OPCODE_TXQ
:
4107 exec_txq(mach
, inst
);
4110 case TGSI_OPCODE_EMIT
:
4114 case TGSI_OPCODE_ENDPRIM
:
4115 emit_primitive(mach
);
4118 case TGSI_OPCODE_BGNLOOP
:
4119 /* push LoopMask and ContMasks */
4120 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
4121 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
4122 assert(mach
->LoopLabelStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
4123 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
4125 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
4126 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
4127 mach
->LoopLabelStack
[mach
->LoopLabelStackTop
++] = *pc
- 1;
4128 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
4129 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_LOOP
;
4132 case TGSI_OPCODE_ENDLOOP
:
4133 /* Restore ContMask, but don't pop */
4134 assert(mach
->ContStackTop
> 0);
4135 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
- 1];
4136 UPDATE_EXEC_MASK(mach
);
4137 if (mach
->ExecMask
) {
4138 /* repeat loop: jump to instruction just past BGNLOOP */
4139 assert(mach
->LoopLabelStackTop
> 0);
4140 *pc
= mach
->LoopLabelStack
[mach
->LoopLabelStackTop
- 1] + 1;
4143 /* exit loop: pop LoopMask */
4144 assert(mach
->LoopStackTop
> 0);
4145 mach
->LoopMask
= mach
->LoopStack
[--mach
->LoopStackTop
];
4147 assert(mach
->ContStackTop
> 0);
4148 mach
->ContMask
= mach
->ContStack
[--mach
->ContStackTop
];
4149 assert(mach
->LoopLabelStackTop
> 0);
4150 --mach
->LoopLabelStackTop
;
4152 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
4154 UPDATE_EXEC_MASK(mach
);
4157 case TGSI_OPCODE_BRK
:
4161 case TGSI_OPCODE_CONT
:
4162 /* turn off cont channels for each enabled exec channel */
4163 mach
->ContMask
&= ~mach
->ExecMask
;
4164 /* Todo: if mach->LoopMask == 0, jump to end of loop */
4165 UPDATE_EXEC_MASK(mach
);
4168 case TGSI_OPCODE_BGNSUB
:
4172 case TGSI_OPCODE_ENDSUB
:
4174 * XXX: This really should be a no-op. We should never reach this opcode.
4177 assert(mach
->CallStackTop
> 0);
4178 mach
->CallStackTop
--;
4180 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
4181 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
4183 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
4184 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
4186 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
4187 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
4189 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
4190 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
4192 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
4193 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
4195 assert(mach
->FuncStackTop
> 0);
4196 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
4198 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
4200 UPDATE_EXEC_MASK(mach
);
4203 case TGSI_OPCODE_NOP
:
4206 case TGSI_OPCODE_BREAKC
:
4207 IFETCH(&r
[0], 0, TGSI_CHAN_X
);
4208 /* update CondMask */
4209 if (r
[0].u
[0] && (mach
->ExecMask
& 0x1)) {
4210 mach
->LoopMask
&= ~0x1;
4212 if (r
[0].u
[1] && (mach
->ExecMask
& 0x2)) {
4213 mach
->LoopMask
&= ~0x2;
4215 if (r
[0].u
[2] && (mach
->ExecMask
& 0x4)) {
4216 mach
->LoopMask
&= ~0x4;
4218 if (r
[0].u
[3] && (mach
->ExecMask
& 0x8)) {
4219 mach
->LoopMask
&= ~0x8;
4221 /* Todo: if mach->LoopMask == 0, jump to end of loop */
4222 UPDATE_EXEC_MASK(mach
);
4225 case TGSI_OPCODE_F2I
:
4226 exec_vector_unary(mach
, inst
, micro_f2i
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
4229 case TGSI_OPCODE_FSEQ
:
4230 exec_vector_binary(mach
, inst
, micro_fseq
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4233 case TGSI_OPCODE_FSGE
:
4234 exec_vector_binary(mach
, inst
, micro_fsge
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4237 case TGSI_OPCODE_FSLT
:
4238 exec_vector_binary(mach
, inst
, micro_fslt
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4241 case TGSI_OPCODE_FSNE
:
4242 exec_vector_binary(mach
, inst
, micro_fsne
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4245 case TGSI_OPCODE_IDIV
:
4246 exec_vector_binary(mach
, inst
, micro_idiv
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4249 case TGSI_OPCODE_IMAX
:
4250 exec_vector_binary(mach
, inst
, micro_imax
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4253 case TGSI_OPCODE_IMIN
:
4254 exec_vector_binary(mach
, inst
, micro_imin
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4257 case TGSI_OPCODE_INEG
:
4258 exec_vector_unary(mach
, inst
, micro_ineg
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4261 case TGSI_OPCODE_ISGE
:
4262 exec_vector_binary(mach
, inst
, micro_isge
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4265 case TGSI_OPCODE_ISHR
:
4266 exec_vector_binary(mach
, inst
, micro_ishr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4269 case TGSI_OPCODE_ISLT
:
4270 exec_vector_binary(mach
, inst
, micro_islt
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4273 case TGSI_OPCODE_F2U
:
4274 exec_vector_unary(mach
, inst
, micro_f2u
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4277 case TGSI_OPCODE_U2F
:
4278 exec_vector_unary(mach
, inst
, micro_u2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_UINT
);
4281 case TGSI_OPCODE_UADD
:
4282 exec_vector_binary(mach
, inst
, micro_uadd
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4285 case TGSI_OPCODE_UDIV
:
4286 exec_vector_binary(mach
, inst
, micro_udiv
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4289 case TGSI_OPCODE_UMAD
:
4290 exec_vector_trinary(mach
, inst
, micro_umad
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4293 case TGSI_OPCODE_UMAX
:
4294 exec_vector_binary(mach
, inst
, micro_umax
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4297 case TGSI_OPCODE_UMIN
:
4298 exec_vector_binary(mach
, inst
, micro_umin
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4301 case TGSI_OPCODE_UMOD
:
4302 exec_vector_binary(mach
, inst
, micro_umod
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4305 case TGSI_OPCODE_UMUL
:
4306 exec_vector_binary(mach
, inst
, micro_umul
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4309 case TGSI_OPCODE_IMUL_HI
:
4310 exec_vector_binary(mach
, inst
, micro_imul_hi
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4313 case TGSI_OPCODE_UMUL_HI
:
4314 exec_vector_binary(mach
, inst
, micro_umul_hi
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4317 case TGSI_OPCODE_USEQ
:
4318 exec_vector_binary(mach
, inst
, micro_useq
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4321 case TGSI_OPCODE_USGE
:
4322 exec_vector_binary(mach
, inst
, micro_usge
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4325 case TGSI_OPCODE_USHR
:
4326 exec_vector_binary(mach
, inst
, micro_ushr
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4329 case TGSI_OPCODE_USLT
:
4330 exec_vector_binary(mach
, inst
, micro_uslt
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4333 case TGSI_OPCODE_USNE
:
4334 exec_vector_binary(mach
, inst
, micro_usne
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4337 case TGSI_OPCODE_SWITCH
:
4338 exec_switch(mach
, inst
);
4341 case TGSI_OPCODE_CASE
:
4342 exec_case(mach
, inst
);
4345 case TGSI_OPCODE_DEFAULT
:
4349 case TGSI_OPCODE_ENDSWITCH
:
4350 exec_endswitch(mach
);
4353 case TGSI_OPCODE_SAMPLE_I
:
4354 exec_txf(mach
, inst
);
4357 case TGSI_OPCODE_SAMPLE_I_MS
:
4361 case TGSI_OPCODE_SAMPLE
:
4362 exec_sample(mach
, inst
, TEX_MODIFIER_NONE
, FALSE
);
4365 case TGSI_OPCODE_SAMPLE_B
:
4366 exec_sample(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, FALSE
);
4369 case TGSI_OPCODE_SAMPLE_C
:
4370 exec_sample(mach
, inst
, TEX_MODIFIER_NONE
, TRUE
);
4373 case TGSI_OPCODE_SAMPLE_C_LZ
:
4374 exec_sample(mach
, inst
, TEX_MODIFIER_LEVEL_ZERO
, TRUE
);
4377 case TGSI_OPCODE_SAMPLE_D
:
4378 exec_sample_d(mach
, inst
);
4381 case TGSI_OPCODE_SAMPLE_L
:
4382 exec_sample(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, FALSE
);
4385 case TGSI_OPCODE_GATHER4
:
4389 case TGSI_OPCODE_SVIEWINFO
:
4390 exec_txq(mach
, inst
);
4393 case TGSI_OPCODE_SAMPLE_POS
:
4397 case TGSI_OPCODE_SAMPLE_INFO
:
4401 case TGSI_OPCODE_UARL
:
4402 exec_vector_unary(mach
, inst
, micro_uarl
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_UINT
);
4405 case TGSI_OPCODE_UCMP
:
4406 exec_vector_trinary(mach
, inst
, micro_ucmp
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4409 case TGSI_OPCODE_IABS
:
4410 exec_vector_unary(mach
, inst
, micro_iabs
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4413 case TGSI_OPCODE_ISSG
:
4414 exec_vector_unary(mach
, inst
, micro_isgn
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4417 case TGSI_OPCODE_TEX2
:
4418 /* simple texture lookup */
4419 /* src[0] = texcoord */
4420 /* src[1] = compare */
4421 /* src[2] = sampler unit */
4422 exec_tex(mach
, inst
, TEX_MODIFIER_NONE
, 2);
4424 case TGSI_OPCODE_TXB2
:
4425 /* simple texture lookup */
4426 /* src[0] = texcoord */
4428 /* src[2] = sampler unit */
4429 exec_tex(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, 2);
4431 case TGSI_OPCODE_TXL2
:
4432 /* simple texture lookup */
4433 /* src[0] = texcoord */
4435 /* src[2] = sampler unit */
4436 exec_tex(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, 2);
4445 * Run TGSI interpreter.
4446 * \return bitmask of "alive" quad components
4449 tgsi_exec_machine_run( struct tgsi_exec_machine
*mach
)
4453 uint default_mask
= 0xf;
4455 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] = 0;
4456 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] = 0;
4458 if( mach
->Processor
== TGSI_PROCESSOR_GEOMETRY
) {
4459 mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0] = 0;
4460 mach
->Primitives
[0] = 0;
4461 /* GS runs on a single primitive for now */
4465 mach
->CondMask
= default_mask
;
4466 mach
->LoopMask
= default_mask
;
4467 mach
->ContMask
= default_mask
;
4468 mach
->FuncMask
= default_mask
;
4469 mach
->ExecMask
= default_mask
;
4471 mach
->Switch
.mask
= default_mask
;
4473 assert(mach
->CondStackTop
== 0);
4474 assert(mach
->LoopStackTop
== 0);
4475 assert(mach
->ContStackTop
== 0);
4476 assert(mach
->SwitchStackTop
== 0);
4477 assert(mach
->BreakStackTop
== 0);
4478 assert(mach
->CallStackTop
== 0);
4481 /* execute declarations (interpolants) */
4482 for (i
= 0; i
< mach
->NumDeclarations
; i
++) {
4483 exec_declaration( mach
, mach
->Declarations
+i
);
4488 struct tgsi_exec_vector temps
[TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
];
4489 struct tgsi_exec_vector outputs
[PIPE_MAX_ATTRIBS
];
4492 memset(mach
->Temps
, 0, sizeof(temps
));
4493 memset(mach
->Outputs
, 0, sizeof(outputs
));
4494 memset(temps
, 0, sizeof(temps
));
4495 memset(outputs
, 0, sizeof(outputs
));
4498 /* execute instructions, until pc is set to -1 */
4504 tgsi_dump_instruction(&mach
->Instructions
[pc
], inst
++);
4507 assert(pc
< (int) mach
->NumInstructions
);
4508 exec_instruction(mach
, mach
->Instructions
+ pc
, &pc
);
4511 for (i
= 0; i
< TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
; i
++) {
4512 if (memcmp(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]))) {
4515 memcpy(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]));
4516 debug_printf("TEMP[%2u] = ", i
);
4517 for (j
= 0; j
< 4; j
++) {
4521 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
4522 temps
[i
].xyzw
[0].f
[j
], temps
[i
].xyzw
[0].u
[j
],
4523 temps
[i
].xyzw
[1].f
[j
], temps
[i
].xyzw
[1].u
[j
],
4524 temps
[i
].xyzw
[2].f
[j
], temps
[i
].xyzw
[2].u
[j
],
4525 temps
[i
].xyzw
[3].f
[j
], temps
[i
].xyzw
[3].u
[j
]);
4529 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
4530 if (memcmp(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]))) {
4533 memcpy(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]));
4534 debug_printf("OUT[%2u] = ", i
);
4535 for (j
= 0; j
< 4; j
++) {
4539 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
4540 outputs
[i
].xyzw
[0].f
[j
], outputs
[i
].xyzw
[0].u
[j
],
4541 outputs
[i
].xyzw
[1].f
[j
], outputs
[i
].xyzw
[1].u
[j
],
4542 outputs
[i
].xyzw
[2].f
[j
], outputs
[i
].xyzw
[2].u
[j
],
4543 outputs
[i
].xyzw
[3].f
[j
], outputs
[i
].xyzw
[3].u
[j
]);
4552 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
4553 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
4555 * Scale back depth component.
4557 for (i
= 0; i
< 4; i
++)
4558 mach
->Outputs
[0].xyzw
[2].f
[i
] *= ctx
->DrawBuffer
->_DepthMaxF
;
4562 /* Strictly speaking, these assertions aren't really needed but they
4563 * can potentially catch some bugs in the control flow code.
4565 assert(mach
->CondStackTop
== 0);
4566 assert(mach
->LoopStackTop
== 0);
4567 assert(mach
->ContStackTop
== 0);
4568 assert(mach
->SwitchStackTop
== 0);
4569 assert(mach
->BreakStackTop
== 0);
4570 assert(mach
->CallStackTop
== 0);
4572 return ~mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];