1 /**************************************************************************
3 * Copyright 2007-2008 VMware, Inc.
5 * Copyright 2009-2010 VMware, Inc. All rights Reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
30 * TGSI interpreter/executor.
32 * Flow control information:
34 * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
35 * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
36 * care since a condition may be true for some quad components but false
37 * for other components.
39 * We basically execute all statements (even if they're in the part of
40 * an IF/ELSE clause that's "not taken") and use a special mask to
41 * control writing to destination registers. This is the ExecMask.
44 * The ExecMask is computed from three other masks (CondMask, LoopMask and
45 * ContMask) which are controlled by the flow control instructions (namely:
46 * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
54 #include "pipe/p_compiler.h"
55 #include "pipe/p_state.h"
56 #include "pipe/p_shader_tokens.h"
57 #include "tgsi/tgsi_dump.h"
58 #include "tgsi/tgsi_parse.h"
59 #include "tgsi/tgsi_util.h"
60 #include "tgsi_exec.h"
61 #include "util/u_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_SHADER_INPUTS
,
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
:
792 if (mach
->Processor
== TGSI_PROCESSOR_GEOMETRY
) {
793 if (parse
.FullToken
.FullProperty
.Property
.PropertyName
== TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES
) {
794 mach
->MaxOutputVertices
= parse
.FullToken
.FullProperty
.u
[0].Data
;
803 tgsi_parse_free (&parse
);
805 FREE(mach
->Declarations
);
806 mach
->Declarations
= declarations
;
807 mach
->NumDeclarations
= numDeclarations
;
809 FREE(mach
->Instructions
);
810 mach
->Instructions
= instructions
;
811 mach
->NumInstructions
= numInstructions
;
815 struct tgsi_exec_machine
*
816 tgsi_exec_machine_create( void )
818 struct tgsi_exec_machine
*mach
;
821 mach
= align_malloc( sizeof *mach
, 16 );
825 memset(mach
, 0, sizeof(*mach
));
827 mach
->Addrs
= &mach
->Temps
[TGSI_EXEC_TEMP_ADDR
];
828 mach
->MaxGeometryShaderOutputs
= TGSI_MAX_TOTAL_VERTICES
;
829 mach
->Predicates
= &mach
->Temps
[TGSI_EXEC_TEMP_P0
];
831 mach
->Inputs
= align_malloc(sizeof(struct tgsi_exec_vector
) * PIPE_MAX_SHADER_INPUTS
, 16);
832 mach
->Outputs
= align_malloc(sizeof(struct tgsi_exec_vector
) * PIPE_MAX_SHADER_OUTPUTS
, 16);
833 if (!mach
->Inputs
|| !mach
->Outputs
)
836 /* Setup constants needed by the SSE2 executor. */
837 for( i
= 0; i
< 4; i
++ ) {
838 mach
->Temps
[TGSI_EXEC_TEMP_00000000_I
].xyzw
[TGSI_EXEC_TEMP_00000000_C
].u
[i
] = 0x00000000;
839 mach
->Temps
[TGSI_EXEC_TEMP_7FFFFFFF_I
].xyzw
[TGSI_EXEC_TEMP_7FFFFFFF_C
].u
[i
] = 0x7FFFFFFF;
840 mach
->Temps
[TGSI_EXEC_TEMP_80000000_I
].xyzw
[TGSI_EXEC_TEMP_80000000_C
].u
[i
] = 0x80000000;
841 mach
->Temps
[TGSI_EXEC_TEMP_FFFFFFFF_I
].xyzw
[TGSI_EXEC_TEMP_FFFFFFFF_C
].u
[i
] = 0xFFFFFFFF; /* not used */
842 mach
->Temps
[TGSI_EXEC_TEMP_ONE_I
].xyzw
[TGSI_EXEC_TEMP_ONE_C
].f
[i
] = 1.0f
;
843 mach
->Temps
[TGSI_EXEC_TEMP_TWO_I
].xyzw
[TGSI_EXEC_TEMP_TWO_C
].f
[i
] = 2.0f
; /* not used */
844 mach
->Temps
[TGSI_EXEC_TEMP_128_I
].xyzw
[TGSI_EXEC_TEMP_128_C
].f
[i
] = 128.0f
;
845 mach
->Temps
[TGSI_EXEC_TEMP_MINUS_128_I
].xyzw
[TGSI_EXEC_TEMP_MINUS_128_C
].f
[i
] = -128.0f
;
846 mach
->Temps
[TGSI_EXEC_TEMP_THREE_I
].xyzw
[TGSI_EXEC_TEMP_THREE_C
].f
[i
] = 3.0f
;
847 mach
->Temps
[TGSI_EXEC_TEMP_HALF_I
].xyzw
[TGSI_EXEC_TEMP_HALF_C
].f
[i
] = 0.5f
;
851 /* silence warnings */
860 align_free(mach
->Inputs
);
861 align_free(mach
->Outputs
);
869 tgsi_exec_machine_destroy(struct tgsi_exec_machine
*mach
)
872 FREE(mach
->Instructions
);
873 FREE(mach
->Declarations
);
875 align_free(mach
->Inputs
);
876 align_free(mach
->Outputs
);
883 micro_add(union tgsi_exec_channel
*dst
,
884 const union tgsi_exec_channel
*src0
,
885 const union tgsi_exec_channel
*src1
)
887 dst
->f
[0] = src0
->f
[0] + src1
->f
[0];
888 dst
->f
[1] = src0
->f
[1] + src1
->f
[1];
889 dst
->f
[2] = src0
->f
[2] + src1
->f
[2];
890 dst
->f
[3] = src0
->f
[3] + src1
->f
[3];
895 union tgsi_exec_channel
*dst
,
896 const union tgsi_exec_channel
*src0
,
897 const union tgsi_exec_channel
*src1
)
899 if (src1
->f
[0] != 0) {
900 dst
->f
[0] = src0
->f
[0] / src1
->f
[0];
902 if (src1
->f
[1] != 0) {
903 dst
->f
[1] = src0
->f
[1] / src1
->f
[1];
905 if (src1
->f
[2] != 0) {
906 dst
->f
[2] = src0
->f
[2] / src1
->f
[2];
908 if (src1
->f
[3] != 0) {
909 dst
->f
[3] = src0
->f
[3] / src1
->f
[3];
914 micro_rcc(union tgsi_exec_channel
*dst
,
915 const union tgsi_exec_channel
*src
)
919 for (i
= 0; i
< 4; i
++) {
920 float recip
= 1.0f
/ src
->f
[i
];
923 dst
->f
[i
] = CLAMP(recip
, 5.42101e-020f
, 1.84467e+019f
);
925 dst
->f
[i
] = CLAMP(recip
, -1.84467e+019f
, -5.42101e-020f
);
931 union tgsi_exec_channel
*dst
,
932 const union tgsi_exec_channel
*src0
,
933 const union tgsi_exec_channel
*src1
,
934 const union tgsi_exec_channel
*src2
,
935 const union tgsi_exec_channel
*src3
)
937 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
938 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
939 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
940 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
944 micro_max(union tgsi_exec_channel
*dst
,
945 const union tgsi_exec_channel
*src0
,
946 const union tgsi_exec_channel
*src1
)
948 dst
->f
[0] = src0
->f
[0] > src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
949 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
950 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
951 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
955 micro_min(union tgsi_exec_channel
*dst
,
956 const union tgsi_exec_channel
*src0
,
957 const union tgsi_exec_channel
*src1
)
959 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
960 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
961 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
962 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
966 micro_mul(union tgsi_exec_channel
*dst
,
967 const union tgsi_exec_channel
*src0
,
968 const union tgsi_exec_channel
*src1
)
970 dst
->f
[0] = src0
->f
[0] * src1
->f
[0];
971 dst
->f
[1] = src0
->f
[1] * src1
->f
[1];
972 dst
->f
[2] = src0
->f
[2] * src1
->f
[2];
973 dst
->f
[3] = src0
->f
[3] * src1
->f
[3];
978 union tgsi_exec_channel
*dst
,
979 const union tgsi_exec_channel
*src
)
981 dst
->f
[0] = -src
->f
[0];
982 dst
->f
[1] = -src
->f
[1];
983 dst
->f
[2] = -src
->f
[2];
984 dst
->f
[3] = -src
->f
[3];
989 union tgsi_exec_channel
*dst
,
990 const union tgsi_exec_channel
*src0
,
991 const union tgsi_exec_channel
*src1
)
994 dst
->f
[0] = util_fast_pow( src0
->f
[0], src1
->f
[0] );
995 dst
->f
[1] = util_fast_pow( src0
->f
[1], src1
->f
[1] );
996 dst
->f
[2] = util_fast_pow( src0
->f
[2], src1
->f
[2] );
997 dst
->f
[3] = util_fast_pow( src0
->f
[3], src1
->f
[3] );
999 dst
->f
[0] = powf( src0
->f
[0], src1
->f
[0] );
1000 dst
->f
[1] = powf( src0
->f
[1], src1
->f
[1] );
1001 dst
->f
[2] = powf( src0
->f
[2], src1
->f
[2] );
1002 dst
->f
[3] = powf( src0
->f
[3], src1
->f
[3] );
1007 micro_sub(union tgsi_exec_channel
*dst
,
1008 const union tgsi_exec_channel
*src0
,
1009 const union tgsi_exec_channel
*src1
)
1011 dst
->f
[0] = src0
->f
[0] - src1
->f
[0];
1012 dst
->f
[1] = src0
->f
[1] - src1
->f
[1];
1013 dst
->f
[2] = src0
->f
[2] - src1
->f
[2];
1014 dst
->f
[3] = src0
->f
[3] - src1
->f
[3];
1018 fetch_src_file_channel(const struct tgsi_exec_machine
*mach
,
1019 const uint chan_index
,
1022 const union tgsi_exec_channel
*index
,
1023 const union tgsi_exec_channel
*index2D
,
1024 union tgsi_exec_channel
*chan
)
1028 assert(swizzle
< 4);
1031 case TGSI_FILE_CONSTANT
:
1032 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1033 assert(index2D
->i
[i
] >= 0 && index2D
->i
[i
] < PIPE_MAX_CONSTANT_BUFFERS
);
1034 assert(mach
->Consts
[index2D
->i
[i
]]);
1036 if (index
->i
[i
] < 0) {
1039 /* NOTE: copying the const value as a uint instead of float */
1040 const uint constbuf
= index2D
->i
[i
];
1041 const uint
*buf
= (const uint
*)mach
->Consts
[constbuf
];
1042 const int pos
= index
->i
[i
] * 4 + swizzle
;
1043 /* const buffer bounds check */
1044 if (pos
< 0 || pos
>= (int) mach
->ConstsSize
[constbuf
]) {
1046 /* Debug: print warning */
1047 static int count
= 0;
1049 debug_printf("TGSI Exec: const buffer index %d"
1050 " out of bounds\n", pos
);
1055 chan
->u
[i
] = buf
[pos
];
1060 case TGSI_FILE_INPUT
:
1061 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1063 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
1064 debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n",
1065 index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i],
1066 index2D->i[i], index->i[i]);
1068 int pos
= index2D
->i
[i
] * TGSI_EXEC_MAX_INPUT_ATTRIBS
+ index
->i
[i
];
1070 assert(pos
< TGSI_MAX_PRIM_VERTICES
* PIPE_MAX_ATTRIBS
);
1071 chan
->u
[i
] = mach
->Inputs
[pos
].xyzw
[swizzle
].u
[i
];
1075 case TGSI_FILE_SYSTEM_VALUE
:
1076 /* XXX no swizzling at this point. Will be needed if we put
1077 * gl_FragCoord, for example, in a sys value register.
1079 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1080 chan
->u
[i
] = mach
->SystemValue
[index
->i
[i
]].u
[i
];
1084 case TGSI_FILE_TEMPORARY
:
1085 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1086 assert(index
->i
[i
] < TGSI_EXEC_NUM_TEMPS
);
1087 assert(index2D
->i
[i
] == 0);
1089 chan
->u
[i
] = mach
->Temps
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1093 case TGSI_FILE_IMMEDIATE
:
1094 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1095 assert(index
->i
[i
] >= 0 && index
->i
[i
] < (int)mach
->ImmLimit
);
1096 assert(index2D
->i
[i
] == 0);
1098 chan
->f
[i
] = mach
->Imms
[index
->i
[i
]][swizzle
];
1102 case TGSI_FILE_ADDRESS
:
1103 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1104 assert(index
->i
[i
] >= 0);
1105 assert(index2D
->i
[i
] == 0);
1107 chan
->u
[i
] = mach
->Addrs
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1111 case TGSI_FILE_PREDICATE
:
1112 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1113 assert(index
->i
[i
] >= 0 && index
->i
[i
] < TGSI_EXEC_NUM_PREDS
);
1114 assert(index2D
->i
[i
] == 0);
1116 chan
->u
[i
] = mach
->Predicates
[0].xyzw
[swizzle
].u
[i
];
1120 case TGSI_FILE_OUTPUT
:
1121 /* vertex/fragment output vars can be read too */
1122 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1123 assert(index
->i
[i
] >= 0);
1124 assert(index2D
->i
[i
] == 0);
1126 chan
->u
[i
] = mach
->Outputs
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1132 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1139 fetch_source(const struct tgsi_exec_machine
*mach
,
1140 union tgsi_exec_channel
*chan
,
1141 const struct tgsi_full_src_register
*reg
,
1142 const uint chan_index
,
1143 enum tgsi_exec_datatype src_datatype
)
1145 union tgsi_exec_channel index
;
1146 union tgsi_exec_channel index2D
;
1149 /* We start with a direct index into a register file.
1153 * file = Register.File
1154 * [1] = Register.Index
1159 index
.i
[3] = reg
->Register
.Index
;
1161 /* There is an extra source register that indirectly subscripts
1162 * a register file. The direct index now becomes an offset
1163 * that is being added to the indirect register.
1167 * ind = Indirect.File
1168 * [2] = Indirect.Index
1169 * .x = Indirect.SwizzleX
1171 if (reg
->Register
.Indirect
) {
1172 union tgsi_exec_channel index2
;
1173 union tgsi_exec_channel indir_index
;
1174 const uint execmask
= mach
->ExecMask
;
1177 /* which address register (always zero now) */
1181 index2
.i
[3] = reg
->Indirect
.Index
;
1182 /* get current value of address register[swizzle] */
1183 swizzle
= reg
->Indirect
.Swizzle
;
1184 fetch_src_file_channel(mach
,
1192 /* add value of address register to the offset */
1193 index
.i
[0] += indir_index
.i
[0];
1194 index
.i
[1] += indir_index
.i
[1];
1195 index
.i
[2] += indir_index
.i
[2];
1196 index
.i
[3] += indir_index
.i
[3];
1198 /* for disabled execution channels, zero-out the index to
1199 * avoid using a potential garbage value.
1201 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1202 if ((execmask
& (1 << i
)) == 0)
1207 /* There is an extra source register that is a second
1208 * subscript to a register file. Effectively it means that
1209 * the register file is actually a 2D array of registers.
1213 * [3] = Dimension.Index
1215 if (reg
->Register
.Dimension
) {
1219 index2D
.i
[3] = reg
->Dimension
.Index
;
1221 /* Again, the second subscript index can be addressed indirectly
1222 * identically to the first one.
1223 * Nothing stops us from indirectly addressing the indirect register,
1224 * but there is no need for that, so we won't exercise it.
1226 * file[ind[4].y+3][1],
1228 * ind = DimIndirect.File
1229 * [4] = DimIndirect.Index
1230 * .y = DimIndirect.SwizzleX
1232 if (reg
->Dimension
.Indirect
) {
1233 union tgsi_exec_channel index2
;
1234 union tgsi_exec_channel indir_index
;
1235 const uint execmask
= mach
->ExecMask
;
1241 index2
.i
[3] = reg
->DimIndirect
.Index
;
1243 swizzle
= reg
->DimIndirect
.Swizzle
;
1244 fetch_src_file_channel(mach
,
1246 reg
->DimIndirect
.File
,
1252 index2D
.i
[0] += indir_index
.i
[0];
1253 index2D
.i
[1] += indir_index
.i
[1];
1254 index2D
.i
[2] += indir_index
.i
[2];
1255 index2D
.i
[3] += indir_index
.i
[3];
1257 /* for disabled execution channels, zero-out the index to
1258 * avoid using a potential garbage value.
1260 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1261 if ((execmask
& (1 << i
)) == 0) {
1267 /* If by any chance there was a need for a 3D array of register
1268 * files, we would have to check whether Dimension is followed
1269 * by a dimension register and continue the saga.
1278 swizzle
= tgsi_util_get_full_src_register_swizzle( reg
, chan_index
);
1279 fetch_src_file_channel(mach
,
1287 if (reg
->Register
.Absolute
) {
1288 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1289 micro_abs(chan
, chan
);
1291 micro_iabs(chan
, chan
);
1295 if (reg
->Register
.Negate
) {
1296 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1297 micro_neg(chan
, chan
);
1299 micro_ineg(chan
, chan
);
1305 store_dest(struct tgsi_exec_machine
*mach
,
1306 const union tgsi_exec_channel
*chan
,
1307 const struct tgsi_full_dst_register
*reg
,
1308 const struct tgsi_full_instruction
*inst
,
1310 enum tgsi_exec_datatype dst_datatype
)
1313 union tgsi_exec_channel null
;
1314 union tgsi_exec_channel
*dst
;
1315 union tgsi_exec_channel index2D
;
1316 uint execmask
= mach
->ExecMask
;
1317 int offset
= 0; /* indirection offset */
1321 if (0 && dst_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1322 check_inf_or_nan(chan
);
1325 /* There is an extra source register that indirectly subscripts
1326 * a register file. The direct index now becomes an offset
1327 * that is being added to the indirect register.
1331 * ind = Indirect.File
1332 * [2] = Indirect.Index
1333 * .x = Indirect.SwizzleX
1335 if (reg
->Register
.Indirect
) {
1336 union tgsi_exec_channel index
;
1337 union tgsi_exec_channel indir_index
;
1340 /* which address register (always zero for now) */
1344 index
.i
[3] = reg
->Indirect
.Index
;
1346 /* get current value of address register[swizzle] */
1347 swizzle
= reg
->Indirect
.Swizzle
;
1349 /* fetch values from the address/indirection register */
1350 fetch_src_file_channel(mach
,
1358 /* save indirection offset */
1359 offset
= indir_index
.i
[0];
1362 /* There is an extra source register that is a second
1363 * subscript to a register file. Effectively it means that
1364 * the register file is actually a 2D array of registers.
1368 * [3] = Dimension.Index
1370 if (reg
->Register
.Dimension
) {
1374 index2D
.i
[3] = reg
->Dimension
.Index
;
1376 /* Again, the second subscript index can be addressed indirectly
1377 * identically to the first one.
1378 * Nothing stops us from indirectly addressing the indirect register,
1379 * but there is no need for that, so we won't exercise it.
1381 * file[ind[4].y+3][1],
1383 * ind = DimIndirect.File
1384 * [4] = DimIndirect.Index
1385 * .y = DimIndirect.SwizzleX
1387 if (reg
->Dimension
.Indirect
) {
1388 union tgsi_exec_channel index2
;
1389 union tgsi_exec_channel indir_index
;
1390 const uint execmask
= mach
->ExecMask
;
1397 index2
.i
[3] = reg
->DimIndirect
.Index
;
1399 swizzle
= reg
->DimIndirect
.Swizzle
;
1400 fetch_src_file_channel(mach
,
1402 reg
->DimIndirect
.File
,
1408 index2D
.i
[0] += indir_index
.i
[0];
1409 index2D
.i
[1] += indir_index
.i
[1];
1410 index2D
.i
[2] += indir_index
.i
[2];
1411 index2D
.i
[3] += indir_index
.i
[3];
1413 /* for disabled execution channels, zero-out the index to
1414 * avoid using a potential garbage value.
1416 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1417 if ((execmask
& (1 << i
)) == 0) {
1423 /* If by any chance there was a need for a 3D array of register
1424 * files, we would have to check whether Dimension is followed
1425 * by a dimension register and continue the saga.
1434 switch (reg
->Register
.File
) {
1435 case TGSI_FILE_NULL
:
1439 case TGSI_FILE_OUTPUT
:
1440 index
= mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0]
1441 + reg
->Register
.Index
;
1442 dst
= &mach
->Outputs
[offset
+ index
].xyzw
[chan_index
];
1444 debug_printf("NumOutputs = %d, TEMP_O_C/I = %d, redindex = %d\n",
1445 mach
->NumOutputs
, mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0],
1446 reg
->Register
.Index
);
1447 if (TGSI_PROCESSOR_GEOMETRY
== mach
->Processor
) {
1448 debug_printf("STORING OUT[%d] mask(%d), = (", offset
+ index
, execmask
);
1449 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1450 if (execmask
& (1 << i
))
1451 debug_printf("%f, ", chan
->f
[i
]);
1452 debug_printf(")\n");
1457 case TGSI_FILE_TEMPORARY
:
1458 index
= reg
->Register
.Index
;
1459 assert( index
< TGSI_EXEC_NUM_TEMPS
);
1460 dst
= &mach
->Temps
[offset
+ index
].xyzw
[chan_index
];
1463 case TGSI_FILE_ADDRESS
:
1464 index
= reg
->Register
.Index
;
1465 dst
= &mach
->Addrs
[index
].xyzw
[chan_index
];
1468 case TGSI_FILE_PREDICATE
:
1469 index
= reg
->Register
.Index
;
1470 assert(index
< TGSI_EXEC_NUM_PREDS
);
1471 dst
= &mach
->Predicates
[index
].xyzw
[chan_index
];
1479 if (inst
->Instruction
.Predicate
) {
1481 union tgsi_exec_channel
*pred
;
1483 switch (chan_index
) {
1485 swizzle
= inst
->Predicate
.SwizzleX
;
1488 swizzle
= inst
->Predicate
.SwizzleY
;
1491 swizzle
= inst
->Predicate
.SwizzleZ
;
1494 swizzle
= inst
->Predicate
.SwizzleW
;
1501 assert(inst
->Predicate
.Index
== 0);
1503 pred
= &mach
->Predicates
[inst
->Predicate
.Index
].xyzw
[swizzle
];
1505 if (inst
->Predicate
.Negate
) {
1506 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1508 execmask
&= ~(1 << i
);
1512 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1514 execmask
&= ~(1 << i
);
1520 switch (inst
->Instruction
.Saturate
) {
1522 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1523 if (execmask
& (1 << i
))
1524 dst
->i
[i
] = chan
->i
[i
];
1527 case TGSI_SAT_ZERO_ONE
:
1528 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1529 if (execmask
& (1 << i
)) {
1530 if (chan
->f
[i
] < 0.0f
)
1532 else if (chan
->f
[i
] > 1.0f
)
1535 dst
->i
[i
] = chan
->i
[i
];
1539 case TGSI_SAT_MINUS_PLUS_ONE
:
1540 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++)
1541 if (execmask
& (1 << i
)) {
1542 if (chan
->f
[i
] < -1.0f
)
1544 else if (chan
->f
[i
] > 1.0f
)
1547 dst
->i
[i
] = chan
->i
[i
];
1556 #define FETCH(VAL,INDEX,CHAN)\
1557 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT)
1559 #define IFETCH(VAL,INDEX,CHAN)\
1560 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT)
1564 * Execute ARB-style KIL which is predicated by a src register.
1565 * Kill fragment if any of the four values is less than zero.
1568 exec_kill_if(struct tgsi_exec_machine
*mach
,
1569 const struct tgsi_full_instruction
*inst
)
1573 uint kilmask
= 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1574 union tgsi_exec_channel r
[1];
1576 /* This mask stores component bits that were already tested. */
1579 for (chan_index
= 0; chan_index
< 4; chan_index
++)
1584 /* unswizzle channel */
1585 swizzle
= tgsi_util_get_full_src_register_swizzle (
1589 /* check if the component has not been already tested */
1590 if (uniquemask
& (1 << swizzle
))
1592 uniquemask
|= 1 << swizzle
;
1594 FETCH(&r
[0], 0, chan_index
);
1595 for (i
= 0; i
< 4; i
++)
1596 if (r
[0].f
[i
] < 0.0f
)
1600 /* restrict to fragments currently executing */
1601 kilmask
&= mach
->ExecMask
;
1603 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1607 * Unconditional fragment kill/discard.
1610 exec_kill(struct tgsi_exec_machine
*mach
,
1611 const struct tgsi_full_instruction
*inst
)
1613 uint kilmask
; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1615 /* kill fragment for all fragments currently executing */
1616 kilmask
= mach
->ExecMask
;
1617 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1621 emit_vertex(struct tgsi_exec_machine
*mach
)
1623 /* FIXME: check for exec mask correctly
1625 for (i = 0; i < TGSI_QUAD_SIZE; ++i) {
1626 if ((mach->ExecMask & (1 << i)))
1628 if (mach
->ExecMask
) {
1629 if (mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]] >= mach
->MaxOutputVertices
)
1632 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] += mach
->NumOutputs
;
1633 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]]++;
1638 emit_primitive(struct tgsi_exec_machine
*mach
)
1640 unsigned *prim_count
= &mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0];
1641 /* FIXME: check for exec mask correctly
1643 for (i = 0; i < TGSI_QUAD_SIZE; ++i) {
1644 if ((mach->ExecMask & (1 << i)))
1646 if (mach
->ExecMask
) {
1648 debug_assert((*prim_count
* mach
->NumOutputs
) < mach
->MaxGeometryShaderOutputs
);
1649 mach
->Primitives
[*prim_count
] = 0;
1654 conditional_emit_primitive(struct tgsi_exec_machine
*mach
)
1656 if (TGSI_PROCESSOR_GEOMETRY
== mach
->Processor
) {
1658 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]];
1659 if (emitted_verts
) {
1660 emit_primitive(mach
);
1667 * Fetch four texture samples using STR texture coordinates.
1670 fetch_texel( struct tgsi_sampler
*sampler
,
1671 const unsigned sview_idx
,
1672 const unsigned sampler_idx
,
1673 const union tgsi_exec_channel
*s
,
1674 const union tgsi_exec_channel
*t
,
1675 const union tgsi_exec_channel
*p
,
1676 const union tgsi_exec_channel
*c0
,
1677 const union tgsi_exec_channel
*c1
,
1678 float derivs
[3][2][TGSI_QUAD_SIZE
],
1679 const int8_t offset
[3],
1680 enum tgsi_sampler_control control
,
1681 union tgsi_exec_channel
*r
,
1682 union tgsi_exec_channel
*g
,
1683 union tgsi_exec_channel
*b
,
1684 union tgsi_exec_channel
*a
)
1687 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
1689 /* FIXME: handle explicit derivs, offsets */
1690 sampler
->get_samples(sampler
, sview_idx
, sampler_idx
,
1691 s
->f
, t
->f
, p
->f
, c0
->f
, c1
->f
, derivs
, offset
, control
, rgba
);
1693 for (j
= 0; j
< 4; j
++) {
1694 r
->f
[j
] = rgba
[0][j
];
1695 g
->f
[j
] = rgba
[1][j
];
1696 b
->f
[j
] = rgba
[2][j
];
1697 a
->f
[j
] = rgba
[3][j
];
1702 #define TEX_MODIFIER_NONE 0
1703 #define TEX_MODIFIER_PROJECTED 1
1704 #define TEX_MODIFIER_LOD_BIAS 2
1705 #define TEX_MODIFIER_EXPLICIT_LOD 3
1706 #define TEX_MODIFIER_LEVEL_ZERO 4
1710 * Fetch all 3 (for s,t,r coords) texel offsets, put them into int array.
1713 fetch_texel_offsets(struct tgsi_exec_machine
*mach
,
1714 const struct tgsi_full_instruction
*inst
,
1717 if (inst
->Texture
.NumOffsets
== 1) {
1718 union tgsi_exec_channel index
;
1719 union tgsi_exec_channel offset
[3];
1720 index
.i
[0] = index
.i
[1] = index
.i
[2] = index
.i
[3] = inst
->TexOffsets
[0].Index
;
1721 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
1722 inst
->TexOffsets
[0].SwizzleX
, &index
, &ZeroVec
, &offset
[0]);
1723 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
1724 inst
->TexOffsets
[0].SwizzleY
, &index
, &ZeroVec
, &offset
[1]);
1725 fetch_src_file_channel(mach
, 0, inst
->TexOffsets
[0].File
,
1726 inst
->TexOffsets
[0].SwizzleZ
, &index
, &ZeroVec
, &offset
[2]);
1727 offsets
[0] = offset
[0].i
[0];
1728 offsets
[1] = offset
[1].i
[0];
1729 offsets
[2] = offset
[2].i
[0];
1731 assert(inst
->Texture
.NumOffsets
== 0);
1732 offsets
[0] = offsets
[1] = offsets
[2] = 0;
1738 * Fetch dx and dy values for one channel (s, t or r).
1739 * Put dx values into one float array, dy values into another.
1742 fetch_assign_deriv_channel(struct tgsi_exec_machine
*mach
,
1743 const struct tgsi_full_instruction
*inst
,
1746 float derivs
[2][TGSI_QUAD_SIZE
])
1748 union tgsi_exec_channel d
;
1749 FETCH(&d
, regdsrcx
, chan
);
1750 derivs
[0][0] = d
.f
[0];
1751 derivs
[0][1] = d
.f
[1];
1752 derivs
[0][2] = d
.f
[2];
1753 derivs
[0][3] = d
.f
[3];
1754 FETCH(&d
, regdsrcx
+ 1, chan
);
1755 derivs
[1][0] = d
.f
[0];
1756 derivs
[1][1] = d
.f
[1];
1757 derivs
[1][2] = d
.f
[2];
1758 derivs
[1][3] = d
.f
[3];
1763 * execute a texture instruction.
1765 * modifier is used to control the channel routing for the\
1766 * instruction variants like proj, lod, and texture with lod bias.
1767 * sampler indicates which src register the sampler is contained in.
1770 exec_tex(struct tgsi_exec_machine
*mach
,
1771 const struct tgsi_full_instruction
*inst
,
1772 uint modifier
, uint sampler
)
1774 const uint unit
= inst
->Src
[sampler
].Register
.Index
;
1775 const union tgsi_exec_channel
*args
[5], *proj
= NULL
;
1776 union tgsi_exec_channel r
[5];
1777 enum tgsi_sampler_control control
= tgsi_sampler_lod_none
;
1780 int dim
, shadow_ref
, i
;
1782 /* always fetch all 3 offsets, overkill but keeps code simple */
1783 fetch_texel_offsets(mach
, inst
, offsets
);
1785 assert(modifier
!= TEX_MODIFIER_LEVEL_ZERO
);
1786 assert(inst
->Texture
.Texture
!= TGSI_TEXTURE_BUFFER
);
1788 dim
= tgsi_util_get_texture_coord_dim(inst
->Texture
.Texture
, &shadow_ref
);
1791 if (shadow_ref
>= 0)
1792 assert(shadow_ref
>= dim
&& shadow_ref
< Elements(args
));
1794 /* fetch modifier to the last argument */
1795 if (modifier
!= TEX_MODIFIER_NONE
) {
1796 const int last
= Elements(args
) - 1;
1798 /* fetch modifier from src0.w or src1.x */
1800 assert(dim
<= TGSI_CHAN_W
&& shadow_ref
!= TGSI_CHAN_W
);
1801 FETCH(&r
[last
], 0, TGSI_CHAN_W
);
1804 assert(shadow_ref
!= 4);
1805 FETCH(&r
[last
], 1, TGSI_CHAN_X
);
1808 if (modifier
!= TEX_MODIFIER_PROJECTED
) {
1809 args
[last
] = &r
[last
];
1813 args
[last
] = &ZeroVec
;
1816 /* point unused arguments to zero vector */
1817 for (i
= dim
; i
< last
; i
++)
1820 if (modifier
== TEX_MODIFIER_EXPLICIT_LOD
)
1821 control
= tgsi_sampler_lod_explicit
;
1822 else if (modifier
== TEX_MODIFIER_LOD_BIAS
)
1823 control
= tgsi_sampler_lod_bias
;
1826 for (i
= dim
; i
< Elements(args
); i
++)
1830 /* fetch coordinates */
1831 for (i
= 0; i
< dim
; i
++) {
1832 FETCH(&r
[i
], 0, TGSI_CHAN_X
+ i
);
1835 micro_div(&r
[i
], &r
[i
], proj
);
1840 /* fetch reference value */
1841 if (shadow_ref
>= 0) {
1842 FETCH(&r
[shadow_ref
], shadow_ref
/ 4, TGSI_CHAN_X
+ (shadow_ref
% 4));
1845 micro_div(&r
[shadow_ref
], &r
[shadow_ref
], proj
);
1847 args
[shadow_ref
] = &r
[shadow_ref
];
1850 fetch_texel(mach
->Sampler
, unit
, unit
,
1851 args
[0], args
[1], args
[2], args
[3], args
[4],
1852 NULL
, offsets
, control
,
1853 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1856 debug_printf("fetch r: %g %g %g %g\n",
1857 r
[0].f
[0], r
[0].f
[1], r
[0].f
[2], r
[0].f
[3]);
1858 debug_printf("fetch g: %g %g %g %g\n",
1859 r
[1].f
[0], r
[1].f
[1], r
[1].f
[2], r
[1].f
[3]);
1860 debug_printf("fetch b: %g %g %g %g\n",
1861 r
[2].f
[0], r
[2].f
[1], r
[2].f
[2], r
[2].f
[3]);
1862 debug_printf("fetch a: %g %g %g %g\n",
1863 r
[3].f
[0], r
[3].f
[1], r
[3].f
[2], r
[3].f
[3]);
1866 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
1867 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1868 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
1875 exec_txd(struct tgsi_exec_machine
*mach
,
1876 const struct tgsi_full_instruction
*inst
)
1878 const uint unit
= inst
->Src
[3].Register
.Index
;
1879 union tgsi_exec_channel r
[4];
1880 float derivs
[3][2][TGSI_QUAD_SIZE
];
1884 /* always fetch all 3 offsets, overkill but keeps code simple */
1885 fetch_texel_offsets(mach
, inst
, offsets
);
1887 switch (inst
->Texture
.Texture
) {
1888 case TGSI_TEXTURE_1D
:
1889 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1891 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1893 fetch_texel(mach
->Sampler
, unit
, unit
,
1894 &r
[0], &ZeroVec
, &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
1895 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1896 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1899 case TGSI_TEXTURE_SHADOW1D
:
1900 case TGSI_TEXTURE_1D_ARRAY
:
1901 case TGSI_TEXTURE_SHADOW1D_ARRAY
:
1902 /* SHADOW1D/1D_ARRAY would not need Y/Z respectively, but don't bother */
1903 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1904 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
1905 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
1907 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1909 fetch_texel(mach
->Sampler
, unit
, unit
,
1910 &r
[0], &r
[1], &r
[2], &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
1911 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1912 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1915 case TGSI_TEXTURE_2D
:
1916 case TGSI_TEXTURE_RECT
:
1917 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1918 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
1920 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1921 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
1923 fetch_texel(mach
->Sampler
, unit
, unit
,
1924 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
1925 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1926 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1930 case TGSI_TEXTURE_SHADOW2D
:
1931 case TGSI_TEXTURE_SHADOWRECT
:
1932 case TGSI_TEXTURE_2D_ARRAY
:
1933 case TGSI_TEXTURE_SHADOW2D_ARRAY
:
1934 /* only SHADOW2D_ARRAY actually needs W */
1935 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1936 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
1937 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
1938 FETCH(&r
[3], 0, TGSI_CHAN_W
);
1940 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1941 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
1943 fetch_texel(mach
->Sampler
, unit
, unit
,
1944 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
, /* inputs */
1945 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1946 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1949 case TGSI_TEXTURE_3D
:
1950 case TGSI_TEXTURE_CUBE
:
1951 case TGSI_TEXTURE_CUBE_ARRAY
:
1952 case TGSI_TEXTURE_SHADOWCUBE
:
1953 /* only TEXTURE_CUBE_ARRAY and TEXTURE_SHADOWCUBE actually need W */
1954 FETCH(&r
[0], 0, TGSI_CHAN_X
);
1955 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
1956 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
1957 FETCH(&r
[3], 0, TGSI_CHAN_W
);
1959 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_X
, derivs
[0]);
1960 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Y
, derivs
[1]);
1961 fetch_assign_deriv_channel(mach
, inst
, 1, TGSI_CHAN_Z
, derivs
[2]);
1963 fetch_texel(mach
->Sampler
, unit
, unit
,
1964 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
, /* inputs */
1965 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
1966 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1973 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
1974 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1975 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
1982 exec_txf(struct tgsi_exec_machine
*mach
,
1983 const struct tgsi_full_instruction
*inst
)
1985 const uint unit
= inst
->Src
[1].Register
.Index
;
1986 union tgsi_exec_channel r
[4];
1988 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
1993 /* always fetch all 3 offsets, overkill but keeps code simple */
1994 fetch_texel_offsets(mach
, inst
, offsets
);
1996 IFETCH(&r
[3], 0, TGSI_CHAN_W
);
1998 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_SAMPLE_I
) {
1999 target
= mach
->SamplerViews
[unit
].Resource
;
2002 target
= inst
->Texture
.Texture
;
2005 case TGSI_TEXTURE_3D
:
2006 case TGSI_TEXTURE_2D_ARRAY
:
2007 case TGSI_TEXTURE_SHADOW2D_ARRAY
:
2008 case TGSI_TEXTURE_2D_ARRAY_MSAA
:
2009 IFETCH(&r
[2], 0, TGSI_CHAN_Z
);
2011 case TGSI_TEXTURE_2D
:
2012 case TGSI_TEXTURE_RECT
:
2013 case TGSI_TEXTURE_SHADOW1D_ARRAY
:
2014 case TGSI_TEXTURE_SHADOW2D
:
2015 case TGSI_TEXTURE_SHADOWRECT
:
2016 case TGSI_TEXTURE_1D_ARRAY
:
2017 case TGSI_TEXTURE_2D_MSAA
:
2018 IFETCH(&r
[1], 0, TGSI_CHAN_Y
);
2020 case TGSI_TEXTURE_BUFFER
:
2021 case TGSI_TEXTURE_1D
:
2022 case TGSI_TEXTURE_SHADOW1D
:
2023 IFETCH(&r
[0], 0, TGSI_CHAN_X
);
2030 mach
->Sampler
->get_texel(mach
->Sampler
, unit
, r
[0].i
, r
[1].i
, r
[2].i
, r
[3].i
,
2033 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2034 r
[0].f
[j
] = rgba
[0][j
];
2035 r
[1].f
[j
] = rgba
[1][j
];
2036 r
[2].f
[j
] = rgba
[2][j
];
2037 r
[3].f
[j
] = rgba
[3][j
];
2040 if (inst
->Instruction
.Opcode
== TGSI_OPCODE_SAMPLE_I
) {
2041 unsigned char swizzles
[4];
2042 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2043 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2044 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2045 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2047 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2048 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2049 store_dest(mach
, &r
[swizzles
[chan
]],
2050 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2055 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2056 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2057 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2064 exec_txq(struct tgsi_exec_machine
*mach
,
2065 const struct tgsi_full_instruction
*inst
)
2067 const uint unit
= inst
->Src
[1].Register
.Index
;
2069 union tgsi_exec_channel r
[4], src
;
2073 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_INT
);
2075 /* XXX: This interface can't return per-pixel values */
2076 mach
->Sampler
->get_dims(mach
->Sampler
, unit
, src
.i
[0], result
);
2078 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2079 for (j
= 0; j
< 4; j
++) {
2080 r
[j
].i
[i
] = result
[j
];
2084 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2085 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2086 store_dest(mach
, &r
[chan
], &inst
->Dst
[0], inst
, chan
,
2087 TGSI_EXEC_DATA_INT
);
2093 exec_sample(struct tgsi_exec_machine
*mach
,
2094 const struct tgsi_full_instruction
*inst
,
2095 uint modifier
, boolean compare
)
2097 const uint resource_unit
= inst
->Src
[1].Register
.Index
;
2098 const uint sampler_unit
= inst
->Src
[2].Register
.Index
;
2099 union tgsi_exec_channel r
[4], c1
;
2100 const union tgsi_exec_channel
*lod
= &ZeroVec
;
2101 enum tgsi_sampler_control control
= tgsi_sampler_lod_none
;
2103 unsigned char swizzles
[4];
2106 /* always fetch all 3 offsets, overkill but keeps code simple */
2107 fetch_texel_offsets(mach
, inst
, offsets
);
2109 assert(modifier
!= TEX_MODIFIER_PROJECTED
);
2111 if (modifier
!= TEX_MODIFIER_NONE
) {
2112 if (modifier
== TEX_MODIFIER_LOD_BIAS
) {
2113 FETCH(&c1
, 3, TGSI_CHAN_X
);
2115 control
= tgsi_sampler_lod_bias
;
2117 else if (modifier
== TEX_MODIFIER_EXPLICIT_LOD
) {
2118 FETCH(&c1
, 3, TGSI_CHAN_X
);
2120 control
= tgsi_sampler_lod_explicit
;
2123 assert(modifier
== TEX_MODIFIER_LEVEL_ZERO
);
2124 control
= tgsi_sampler_lod_zero
;
2128 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2130 switch (mach
->SamplerViews
[resource_unit
].Resource
) {
2131 case TGSI_TEXTURE_1D
:
2133 FETCH(&r
[2], 3, TGSI_CHAN_X
);
2134 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2135 &r
[0], &ZeroVec
, &r
[2], &ZeroVec
, lod
, /* S, T, P, C, LOD */
2136 NULL
, offsets
, control
,
2137 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2140 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2141 &r
[0], &ZeroVec
, &ZeroVec
, &ZeroVec
, lod
, /* S, T, P, C, LOD */
2142 NULL
, offsets
, control
,
2143 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2147 case TGSI_TEXTURE_1D_ARRAY
:
2148 case TGSI_TEXTURE_2D
:
2149 case TGSI_TEXTURE_RECT
:
2150 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2152 FETCH(&r
[2], 3, TGSI_CHAN_X
);
2153 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2154 &r
[0], &r
[1], &r
[2], &ZeroVec
, lod
, /* S, T, P, C, LOD */
2155 NULL
, offsets
, control
,
2156 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2159 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2160 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, lod
, /* S, T, P, C, LOD */
2161 NULL
, offsets
, control
,
2162 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2166 case TGSI_TEXTURE_2D_ARRAY
:
2167 case TGSI_TEXTURE_3D
:
2168 case TGSI_TEXTURE_CUBE
:
2169 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2170 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2172 FETCH(&r
[3], 3, TGSI_CHAN_X
);
2173 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2174 &r
[0], &r
[1], &r
[2], &r
[3], lod
,
2175 NULL
, offsets
, control
,
2176 &r
[0], &r
[1], &r
[2], &r
[3]);
2179 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2180 &r
[0], &r
[1], &r
[2], &ZeroVec
, lod
,
2181 NULL
, offsets
, control
,
2182 &r
[0], &r
[1], &r
[2], &r
[3]);
2186 case TGSI_TEXTURE_CUBE_ARRAY
:
2187 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2188 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2189 FETCH(&r
[3], 0, TGSI_CHAN_W
);
2191 FETCH(&r
[4], 3, TGSI_CHAN_X
);
2192 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2193 &r
[0], &r
[1], &r
[2], &r
[3], &r
[4],
2194 NULL
, offsets
, control
,
2195 &r
[0], &r
[1], &r
[2], &r
[3]);
2198 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2199 &r
[0], &r
[1], &r
[2], &r
[3], lod
,
2200 NULL
, offsets
, control
,
2201 &r
[0], &r
[1], &r
[2], &r
[3]);
2210 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2211 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2212 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2213 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2215 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2216 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2217 store_dest(mach
, &r
[swizzles
[chan
]],
2218 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2224 exec_sample_d(struct tgsi_exec_machine
*mach
,
2225 const struct tgsi_full_instruction
*inst
)
2227 const uint resource_unit
= inst
->Src
[1].Register
.Index
;
2228 const uint sampler_unit
= inst
->Src
[2].Register
.Index
;
2229 union tgsi_exec_channel r
[4];
2230 float derivs
[3][2][TGSI_QUAD_SIZE
];
2232 unsigned char swizzles
[4];
2235 /* always fetch all 3 offsets, overkill but keeps code simple */
2236 fetch_texel_offsets(mach
, inst
, offsets
);
2238 FETCH(&r
[0], 0, TGSI_CHAN_X
);
2240 switch (mach
->SamplerViews
[resource_unit
].Resource
) {
2241 case TGSI_TEXTURE_1D
:
2242 case TGSI_TEXTURE_1D_ARRAY
:
2243 /* only 1D array actually needs Y */
2244 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2246 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2248 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2249 &r
[0], &r
[1], &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, C, LOD */
2250 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
2251 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
2254 case TGSI_TEXTURE_2D
:
2255 case TGSI_TEXTURE_RECT
:
2256 case TGSI_TEXTURE_2D_ARRAY
:
2257 /* only 2D array actually needs Z */
2258 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2259 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2261 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2262 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Y
, derivs
[1]);
2264 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2265 &r
[0], &r
[1], &r
[2], &ZeroVec
, &ZeroVec
, /* inputs */
2266 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
2267 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
2270 case TGSI_TEXTURE_3D
:
2271 case TGSI_TEXTURE_CUBE
:
2272 case TGSI_TEXTURE_CUBE_ARRAY
:
2273 /* only cube array actually needs W */
2274 FETCH(&r
[1], 0, TGSI_CHAN_Y
);
2275 FETCH(&r
[2], 0, TGSI_CHAN_Z
);
2276 FETCH(&r
[3], 0, TGSI_CHAN_W
);
2278 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_X
, derivs
[0]);
2279 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Y
, derivs
[1]);
2280 fetch_assign_deriv_channel(mach
, inst
, 3, TGSI_CHAN_Z
, derivs
[2]);
2282 fetch_texel(mach
->Sampler
, resource_unit
, sampler_unit
,
2283 &r
[0], &r
[1], &r
[2], &r
[3], &ZeroVec
,
2284 derivs
, offsets
, tgsi_sampler_derivs_explicit
,
2285 &r
[0], &r
[1], &r
[2], &r
[3]);
2292 swizzles
[0] = inst
->Src
[1].Register
.SwizzleX
;
2293 swizzles
[1] = inst
->Src
[1].Register
.SwizzleY
;
2294 swizzles
[2] = inst
->Src
[1].Register
.SwizzleZ
;
2295 swizzles
[3] = inst
->Src
[1].Register
.SwizzleW
;
2297 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2298 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2299 store_dest(mach
, &r
[swizzles
[chan
]],
2300 &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2307 * Evaluate a constant-valued coefficient at the position of the
2312 struct tgsi_exec_machine
*mach
,
2318 for( i
= 0; i
< TGSI_QUAD_SIZE
; i
++ ) {
2319 mach
->Inputs
[attrib
].xyzw
[chan
].f
[i
] = mach
->InterpCoefs
[attrib
].a0
[chan
];
2324 * Evaluate a linear-valued coefficient at the position of the
2329 struct tgsi_exec_machine
*mach
,
2333 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
2334 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
2335 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
2336 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
2337 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
2338 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
;
2339 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = a0
+ dadx
;
2340 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = a0
+ dady
;
2341 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = a0
+ dadx
+ dady
;
2345 * Evaluate a perspective-valued coefficient at the position of the
2349 eval_perspective_coef(
2350 struct tgsi_exec_machine
*mach
,
2354 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
2355 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
2356 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
2357 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
2358 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
2359 const float *w
= mach
->QuadPos
.xyzw
[3].f
;
2360 /* divide by W here */
2361 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
/ w
[0];
2362 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = (a0
+ dadx
) / w
[1];
2363 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = (a0
+ dady
) / w
[2];
2364 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = (a0
+ dadx
+ dady
) / w
[3];
2368 typedef void (* eval_coef_func
)(
2369 struct tgsi_exec_machine
*mach
,
2374 exec_declaration(struct tgsi_exec_machine
*mach
,
2375 const struct tgsi_full_declaration
*decl
)
2377 if (decl
->Declaration
.File
== TGSI_FILE_SAMPLER_VIEW
) {
2378 mach
->SamplerViews
[decl
->Range
.First
] = decl
->SamplerView
;
2382 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
2383 if (decl
->Declaration
.File
== TGSI_FILE_INPUT
) {
2384 uint first
, last
, mask
;
2386 first
= decl
->Range
.First
;
2387 last
= decl
->Range
.Last
;
2388 mask
= decl
->Declaration
.UsageMask
;
2390 /* XXX we could remove this special-case code since
2391 * mach->InterpCoefs[first].a0 should already have the
2392 * front/back-face value. But we should first update the
2393 * ureg code to emit the right UsageMask value (WRITEMASK_X).
2394 * Then, we could remove the tgsi_exec_machine::Face field.
2396 /* XXX make FACE a system value */
2397 if (decl
->Semantic
.Name
== TGSI_SEMANTIC_FACE
) {
2400 assert(decl
->Semantic
.Index
== 0);
2401 assert(first
== last
);
2403 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2404 mach
->Inputs
[first
].xyzw
[0].f
[i
] = mach
->Face
;
2407 eval_coef_func eval
;
2410 switch (decl
->Interp
.Interpolate
) {
2411 case TGSI_INTERPOLATE_CONSTANT
:
2412 eval
= eval_constant_coef
;
2415 case TGSI_INTERPOLATE_LINEAR
:
2416 eval
= eval_linear_coef
;
2419 case TGSI_INTERPOLATE_PERSPECTIVE
:
2420 eval
= eval_perspective_coef
;
2423 case TGSI_INTERPOLATE_COLOR
:
2424 eval
= mach
->flatshade_color
? eval_constant_coef
: eval_perspective_coef
;
2432 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
2433 if (mask
& (1 << j
)) {
2434 for (i
= first
; i
<= last
; i
++) {
2441 if (DEBUG_EXECUTION
) {
2443 for (i
= first
; i
<= last
; ++i
) {
2444 debug_printf("IN[%2u] = ", i
);
2445 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
2449 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
2450 mach
->Inputs
[i
].xyzw
[0].f
[j
], mach
->Inputs
[i
].xyzw
[0].u
[j
],
2451 mach
->Inputs
[i
].xyzw
[1].f
[j
], mach
->Inputs
[i
].xyzw
[1].u
[j
],
2452 mach
->Inputs
[i
].xyzw
[2].f
[j
], mach
->Inputs
[i
].xyzw
[2].u
[j
],
2453 mach
->Inputs
[i
].xyzw
[3].f
[j
], mach
->Inputs
[i
].xyzw
[3].u
[j
]);
2460 if (decl
->Declaration
.File
== TGSI_FILE_SYSTEM_VALUE
) {
2461 mach
->SysSemanticToIndex
[decl
->Declaration
.Semantic
] = decl
->Range
.First
;
2466 typedef void (* micro_op
)(union tgsi_exec_channel
*dst
);
2469 exec_vector(struct tgsi_exec_machine
*mach
,
2470 const struct tgsi_full_instruction
*inst
,
2472 enum tgsi_exec_datatype dst_datatype
)
2476 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2477 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2478 union tgsi_exec_channel dst
;
2481 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2486 typedef void (* micro_unary_op
)(union tgsi_exec_channel
*dst
,
2487 const union tgsi_exec_channel
*src
);
2490 exec_scalar_unary(struct tgsi_exec_machine
*mach
,
2491 const struct tgsi_full_instruction
*inst
,
2493 enum tgsi_exec_datatype dst_datatype
,
2494 enum tgsi_exec_datatype src_datatype
)
2497 union tgsi_exec_channel src
;
2498 union tgsi_exec_channel dst
;
2500 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, src_datatype
);
2502 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2503 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2504 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2510 exec_vector_unary(struct tgsi_exec_machine
*mach
,
2511 const struct tgsi_full_instruction
*inst
,
2513 enum tgsi_exec_datatype dst_datatype
,
2514 enum tgsi_exec_datatype src_datatype
)
2517 struct tgsi_exec_vector dst
;
2519 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2520 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2521 union tgsi_exec_channel src
;
2523 fetch_source(mach
, &src
, &inst
->Src
[0], chan
, src_datatype
);
2524 op(&dst
.xyzw
[chan
], &src
);
2527 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2528 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2529 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2534 typedef void (* micro_binary_op
)(union tgsi_exec_channel
*dst
,
2535 const union tgsi_exec_channel
*src0
,
2536 const union tgsi_exec_channel
*src1
);
2539 exec_scalar_binary(struct tgsi_exec_machine
*mach
,
2540 const struct tgsi_full_instruction
*inst
,
2542 enum tgsi_exec_datatype dst_datatype
,
2543 enum tgsi_exec_datatype src_datatype
)
2546 union tgsi_exec_channel src
[2];
2547 union tgsi_exec_channel dst
;
2549 fetch_source(mach
, &src
[0], &inst
->Src
[0], TGSI_CHAN_X
, src_datatype
);
2550 fetch_source(mach
, &src
[1], &inst
->Src
[1], TGSI_CHAN_X
, src_datatype
);
2551 op(&dst
, &src
[0], &src
[1]);
2552 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2553 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2554 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2560 exec_vector_binary(struct tgsi_exec_machine
*mach
,
2561 const struct tgsi_full_instruction
*inst
,
2563 enum tgsi_exec_datatype dst_datatype
,
2564 enum tgsi_exec_datatype src_datatype
)
2567 struct tgsi_exec_vector dst
;
2569 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2570 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2571 union tgsi_exec_channel src
[2];
2573 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
2574 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
2575 op(&dst
.xyzw
[chan
], &src
[0], &src
[1]);
2578 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2579 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2580 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2585 typedef void (* micro_trinary_op
)(union tgsi_exec_channel
*dst
,
2586 const union tgsi_exec_channel
*src0
,
2587 const union tgsi_exec_channel
*src1
,
2588 const union tgsi_exec_channel
*src2
);
2591 exec_vector_trinary(struct tgsi_exec_machine
*mach
,
2592 const struct tgsi_full_instruction
*inst
,
2593 micro_trinary_op op
,
2594 enum tgsi_exec_datatype dst_datatype
,
2595 enum tgsi_exec_datatype src_datatype
)
2598 struct tgsi_exec_vector dst
;
2600 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2601 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2602 union tgsi_exec_channel src
[3];
2604 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
2605 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
2606 fetch_source(mach
, &src
[2], &inst
->Src
[2], chan
, src_datatype
);
2607 op(&dst
.xyzw
[chan
], &src
[0], &src
[1], &src
[2]);
2610 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2611 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2612 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2617 typedef void (* micro_quaternary_op
)(union tgsi_exec_channel
*dst
,
2618 const union tgsi_exec_channel
*src0
,
2619 const union tgsi_exec_channel
*src1
,
2620 const union tgsi_exec_channel
*src2
,
2621 const union tgsi_exec_channel
*src3
);
2624 exec_vector_quaternary(struct tgsi_exec_machine
*mach
,
2625 const struct tgsi_full_instruction
*inst
,
2626 micro_quaternary_op op
,
2627 enum tgsi_exec_datatype dst_datatype
,
2628 enum tgsi_exec_datatype src_datatype
)
2631 struct tgsi_exec_vector dst
;
2633 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2634 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2635 union tgsi_exec_channel src
[4];
2637 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
2638 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
2639 fetch_source(mach
, &src
[2], &inst
->Src
[2], chan
, src_datatype
);
2640 fetch_source(mach
, &src
[3], &inst
->Src
[3], chan
, src_datatype
);
2641 op(&dst
.xyzw
[chan
], &src
[0], &src
[1], &src
[2], &src
[3]);
2644 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2645 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2646 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
2652 exec_dp3(struct tgsi_exec_machine
*mach
,
2653 const struct tgsi_full_instruction
*inst
)
2656 union tgsi_exec_channel arg
[3];
2658 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2659 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2660 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2662 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_Z
; chan
++) {
2663 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2664 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
2665 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2668 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2669 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2670 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2676 exec_dp4(struct tgsi_exec_machine
*mach
,
2677 const struct tgsi_full_instruction
*inst
)
2680 union tgsi_exec_channel arg
[3];
2682 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2683 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2684 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2686 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_W
; chan
++) {
2687 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2688 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
2689 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2692 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2693 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2694 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2700 exec_dp2a(struct tgsi_exec_machine
*mach
,
2701 const struct tgsi_full_instruction
*inst
)
2704 union tgsi_exec_channel arg
[3];
2706 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2707 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2708 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2710 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2711 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2712 micro_mad(&arg
[0], &arg
[0], &arg
[1], &arg
[2]);
2714 fetch_source(mach
, &arg
[1], &inst
->Src
[2], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2715 micro_add(&arg
[0], &arg
[0], &arg
[1]);
2717 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2718 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2719 store_dest(mach
, &arg
[0], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2725 exec_dph(struct tgsi_exec_machine
*mach
,
2726 const struct tgsi_full_instruction
*inst
)
2729 union tgsi_exec_channel arg
[3];
2731 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2732 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2733 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2735 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2736 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2737 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2739 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2740 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2741 micro_mad(&arg
[0], &arg
[0], &arg
[1], &arg
[2]);
2743 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2744 micro_add(&arg
[0], &arg
[0], &arg
[1]);
2746 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2747 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2748 store_dest(mach
, &arg
[0], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2754 exec_dp2(struct tgsi_exec_machine
*mach
,
2755 const struct tgsi_full_instruction
*inst
)
2758 union tgsi_exec_channel arg
[3];
2760 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2761 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2762 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2764 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2765 fetch_source(mach
, &arg
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2766 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2768 for (chan
= 0; chan
< TGSI_NUM_CHANNELS
; chan
++) {
2769 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2770 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2776 exec_nrm4(struct tgsi_exec_machine
*mach
,
2777 const struct tgsi_full_instruction
*inst
)
2780 union tgsi_exec_channel arg
[4];
2781 union tgsi_exec_channel scale
;
2783 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2784 micro_mul(&scale
, &arg
[0], &arg
[0]);
2786 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_W
; chan
++) {
2787 union tgsi_exec_channel product
;
2789 fetch_source(mach
, &arg
[chan
], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2790 micro_mul(&product
, &arg
[chan
], &arg
[chan
]);
2791 micro_add(&scale
, &scale
, &product
);
2794 micro_rsq(&scale
, &scale
);
2796 for (chan
= TGSI_CHAN_X
; chan
<= TGSI_CHAN_W
; chan
++) {
2797 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2798 micro_mul(&arg
[chan
], &arg
[chan
], &scale
);
2799 store_dest(mach
, &arg
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2805 exec_nrm3(struct tgsi_exec_machine
*mach
,
2806 const struct tgsi_full_instruction
*inst
)
2808 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XYZ
) {
2810 union tgsi_exec_channel arg
[3];
2811 union tgsi_exec_channel scale
;
2813 fetch_source(mach
, &arg
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2814 micro_mul(&scale
, &arg
[0], &arg
[0]);
2816 for (chan
= TGSI_CHAN_Y
; chan
<= TGSI_CHAN_Z
; chan
++) {
2817 union tgsi_exec_channel product
;
2819 fetch_source(mach
, &arg
[chan
], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2820 micro_mul(&product
, &arg
[chan
], &arg
[chan
]);
2821 micro_add(&scale
, &scale
, &product
);
2824 micro_rsq(&scale
, &scale
);
2826 for (chan
= TGSI_CHAN_X
; chan
<= TGSI_CHAN_Z
; chan
++) {
2827 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2828 micro_mul(&arg
[chan
], &arg
[chan
], &scale
);
2829 store_dest(mach
, &arg
[chan
], &inst
->Dst
[0], inst
, chan
, 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_scs(struct tgsi_exec_machine
*mach
,
2841 const struct tgsi_full_instruction
*inst
)
2843 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XY
) {
2844 union tgsi_exec_channel arg
;
2845 union tgsi_exec_channel result
;
2847 fetch_source(mach
, &arg
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2849 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2850 micro_cos(&result
, &arg
);
2851 store_dest(mach
, &result
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2853 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2854 micro_sin(&result
, &arg
);
2855 store_dest(mach
, &result
, &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2858 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2859 store_dest(mach
, &ZeroVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2861 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2862 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2867 exec_x2d(struct tgsi_exec_machine
*mach
,
2868 const struct tgsi_full_instruction
*inst
)
2870 union tgsi_exec_channel r
[4];
2871 union tgsi_exec_channel d
[2];
2873 fetch_source(mach
, &r
[0], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2874 fetch_source(mach
, &r
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2875 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XZ
) {
2876 fetch_source(mach
, &r
[2], &inst
->Src
[2], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2877 micro_mul(&r
[2], &r
[2], &r
[0]);
2878 fetch_source(mach
, &r
[3], &inst
->Src
[2], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2879 micro_mul(&r
[3], &r
[3], &r
[1]);
2880 micro_add(&r
[2], &r
[2], &r
[3]);
2881 fetch_source(mach
, &r
[3], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2882 micro_add(&d
[0], &r
[2], &r
[3]);
2884 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_YW
) {
2885 fetch_source(mach
, &r
[2], &inst
->Src
[2], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2886 micro_mul(&r
[2], &r
[2], &r
[0]);
2887 fetch_source(mach
, &r
[3], &inst
->Src
[2], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2888 micro_mul(&r
[3], &r
[3], &r
[1]);
2889 micro_add(&r
[2], &r
[2], &r
[3]);
2890 fetch_source(mach
, &r
[3], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2891 micro_add(&d
[1], &r
[2], &r
[3]);
2893 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2894 store_dest(mach
, &d
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2896 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2897 store_dest(mach
, &d
[1], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2899 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2900 store_dest(mach
, &d
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2902 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2903 store_dest(mach
, &d
[1], &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2908 exec_rfl(struct tgsi_exec_machine
*mach
,
2909 const struct tgsi_full_instruction
*inst
)
2911 union tgsi_exec_channel r
[9];
2913 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XYZ
) {
2914 /* r0 = dp3(src0, src0) */
2915 fetch_source(mach
, &r
[2], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2916 micro_mul(&r
[0], &r
[2], &r
[2]);
2917 fetch_source(mach
, &r
[4], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2918 micro_mul(&r
[8], &r
[4], &r
[4]);
2919 micro_add(&r
[0], &r
[0], &r
[8]);
2920 fetch_source(mach
, &r
[6], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2921 micro_mul(&r
[8], &r
[6], &r
[6]);
2922 micro_add(&r
[0], &r
[0], &r
[8]);
2924 /* r1 = dp3(src0, src1) */
2925 fetch_source(mach
, &r
[3], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2926 micro_mul(&r
[1], &r
[2], &r
[3]);
2927 fetch_source(mach
, &r
[5], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2928 micro_mul(&r
[8], &r
[4], &r
[5]);
2929 micro_add(&r
[1], &r
[1], &r
[8]);
2930 fetch_source(mach
, &r
[7], &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2931 micro_mul(&r
[8], &r
[6], &r
[7]);
2932 micro_add(&r
[1], &r
[1], &r
[8]);
2934 /* r1 = 2 * r1 / r0 */
2935 micro_add(&r
[1], &r
[1], &r
[1]);
2936 micro_div(&r
[1], &r
[1], &r
[0]);
2938 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2939 micro_mul(&r
[2], &r
[2], &r
[1]);
2940 micro_sub(&r
[2], &r
[2], &r
[3]);
2941 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2943 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2944 micro_mul(&r
[4], &r
[4], &r
[1]);
2945 micro_sub(&r
[4], &r
[4], &r
[5]);
2946 store_dest(mach
, &r
[4], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2948 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2949 micro_mul(&r
[6], &r
[6], &r
[1]);
2950 micro_sub(&r
[6], &r
[6], &r
[7]);
2951 store_dest(mach
, &r
[6], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2954 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2955 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2960 exec_xpd(struct tgsi_exec_machine
*mach
,
2961 const struct tgsi_full_instruction
*inst
)
2963 union tgsi_exec_channel r
[6];
2964 union tgsi_exec_channel d
[3];
2966 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2967 fetch_source(mach
, &r
[1], &inst
->Src
[1], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2969 micro_mul(&r
[2], &r
[0], &r
[1]);
2971 fetch_source(mach
, &r
[3], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2972 fetch_source(mach
, &r
[4], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2974 micro_mul(&r
[5], &r
[3], &r
[4] );
2975 micro_sub(&d
[TGSI_CHAN_X
], &r
[2], &r
[5]);
2977 fetch_source(mach
, &r
[2], &inst
->Src
[1], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2979 micro_mul(&r
[3], &r
[3], &r
[2]);
2981 fetch_source(mach
, &r
[5], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2983 micro_mul(&r
[1], &r
[1], &r
[5]);
2984 micro_sub(&d
[TGSI_CHAN_Y
], &r
[3], &r
[1]);
2986 micro_mul(&r
[5], &r
[5], &r
[4]);
2987 micro_mul(&r
[0], &r
[0], &r
[2]);
2988 micro_sub(&d
[TGSI_CHAN_Z
], &r
[5], &r
[0]);
2990 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
2991 store_dest(mach
, &d
[TGSI_CHAN_X
], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2993 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
2994 store_dest(mach
, &d
[TGSI_CHAN_Y
], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2996 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
2997 store_dest(mach
, &d
[TGSI_CHAN_Z
], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2999 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3000 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3005 exec_dst(struct tgsi_exec_machine
*mach
,
3006 const struct tgsi_full_instruction
*inst
)
3008 union tgsi_exec_channel r
[2];
3009 union tgsi_exec_channel d
[4];
3011 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3012 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3013 fetch_source(mach
, &r
[1], &inst
->Src
[1], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3014 micro_mul(&d
[TGSI_CHAN_Y
], &r
[0], &r
[1]);
3016 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3017 fetch_source(mach
, &d
[TGSI_CHAN_Z
], &inst
->Src
[0], TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3019 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3020 fetch_source(mach
, &d
[TGSI_CHAN_W
], &inst
->Src
[1], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3023 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3024 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3026 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3027 store_dest(mach
, &d
[TGSI_CHAN_Y
], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3029 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3030 store_dest(mach
, &d
[TGSI_CHAN_Z
], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3032 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3033 store_dest(mach
, &d
[TGSI_CHAN_W
], &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3038 exec_log(struct tgsi_exec_machine
*mach
,
3039 const struct tgsi_full_instruction
*inst
)
3041 union tgsi_exec_channel r
[3];
3043 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3044 micro_abs(&r
[2], &r
[0]); /* r2 = abs(r0) */
3045 micro_lg2(&r
[1], &r
[2]); /* r1 = lg2(r2) */
3046 micro_flr(&r
[0], &r
[1]); /* r0 = floor(r1) */
3047 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3048 store_dest(mach
, &r
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3050 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3051 micro_exp2(&r
[0], &r
[0]); /* r0 = 2 ^ r0 */
3052 micro_div(&r
[0], &r
[2], &r
[0]); /* r0 = r2 / r0 */
3053 store_dest(mach
, &r
[0], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3055 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3056 store_dest(mach
, &r
[1], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3058 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3059 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3064 exec_exp(struct tgsi_exec_machine
*mach
,
3065 const struct tgsi_full_instruction
*inst
)
3067 union tgsi_exec_channel r
[3];
3069 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3070 micro_flr(&r
[1], &r
[0]); /* r1 = floor(r0) */
3071 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3072 micro_exp2(&r
[2], &r
[1]); /* r2 = 2 ^ r1 */
3073 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3075 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3076 micro_sub(&r
[2], &r
[0], &r
[1]); /* r2 = r0 - r1 */
3077 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3079 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3080 micro_exp2(&r
[2], &r
[0]); /* r2 = 2 ^ r0 */
3081 store_dest(mach
, &r
[2], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3083 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3084 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3089 exec_lit(struct tgsi_exec_machine
*mach
,
3090 const struct tgsi_full_instruction
*inst
)
3092 union tgsi_exec_channel r
[3];
3093 union tgsi_exec_channel d
[3];
3095 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_YZ
) {
3096 fetch_source(mach
, &r
[0], &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3097 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Z
) {
3098 fetch_source(mach
, &r
[1], &inst
->Src
[0], TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3099 micro_max(&r
[1], &r
[1], &ZeroVec
);
3101 fetch_source(mach
, &r
[2], &inst
->Src
[0], TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3102 micro_min(&r
[2], &r
[2], &P128Vec
);
3103 micro_max(&r
[2], &r
[2], &M128Vec
);
3104 micro_pow(&r
[1], &r
[1], &r
[2]);
3105 micro_lt(&d
[TGSI_CHAN_Z
], &ZeroVec
, &r
[0], &r
[1], &ZeroVec
);
3106 store_dest(mach
, &d
[TGSI_CHAN_Z
], &inst
->Dst
[0], inst
, TGSI_CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
3108 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_Y
) {
3109 micro_max(&d
[TGSI_CHAN_Y
], &r
[0], &ZeroVec
);
3110 store_dest(mach
, &d
[TGSI_CHAN_Y
], &inst
->Dst
[0], inst
, TGSI_CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
3113 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_X
) {
3114 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
3117 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
3118 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, TGSI_CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
3123 exec_break(struct tgsi_exec_machine
*mach
)
3125 if (mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_LOOP
) {
3126 /* turn off loop channels for each enabled exec channel */
3127 mach
->LoopMask
&= ~mach
->ExecMask
;
3128 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3129 UPDATE_EXEC_MASK(mach
);
3131 assert(mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_SWITCH
);
3133 mach
->Switch
.mask
= 0x0;
3135 UPDATE_EXEC_MASK(mach
);
3140 exec_switch(struct tgsi_exec_machine
*mach
,
3141 const struct tgsi_full_instruction
*inst
)
3143 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
3144 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
3146 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
3147 fetch_source(mach
, &mach
->Switch
.selector
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_UINT
);
3148 mach
->Switch
.mask
= 0x0;
3149 mach
->Switch
.defaultMask
= 0x0;
3151 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
3152 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_SWITCH
;
3154 UPDATE_EXEC_MASK(mach
);
3158 exec_case(struct tgsi_exec_machine
*mach
,
3159 const struct tgsi_full_instruction
*inst
)
3161 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
3162 union tgsi_exec_channel src
;
3165 fetch_source(mach
, &src
, &inst
->Src
[0], TGSI_CHAN_X
, TGSI_EXEC_DATA_UINT
);
3167 if (mach
->Switch
.selector
.u
[0] == src
.u
[0]) {
3170 if (mach
->Switch
.selector
.u
[1] == src
.u
[1]) {
3173 if (mach
->Switch
.selector
.u
[2] == src
.u
[2]) {
3176 if (mach
->Switch
.selector
.u
[3] == src
.u
[3]) {
3180 mach
->Switch
.defaultMask
|= mask
;
3182 mach
->Switch
.mask
|= mask
& prevMask
;
3184 UPDATE_EXEC_MASK(mach
);
3187 /* FIXME: this will only work if default is last */
3189 exec_default(struct tgsi_exec_machine
*mach
)
3191 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
3193 mach
->Switch
.mask
|= ~mach
->Switch
.defaultMask
& prevMask
;
3195 UPDATE_EXEC_MASK(mach
);
3199 exec_endswitch(struct tgsi_exec_machine
*mach
)
3201 mach
->Switch
= mach
->SwitchStack
[--mach
->SwitchStackTop
];
3202 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
3204 UPDATE_EXEC_MASK(mach
);
3208 micro_i2f(union tgsi_exec_channel
*dst
,
3209 const union tgsi_exec_channel
*src
)
3211 dst
->f
[0] = (float)src
->i
[0];
3212 dst
->f
[1] = (float)src
->i
[1];
3213 dst
->f
[2] = (float)src
->i
[2];
3214 dst
->f
[3] = (float)src
->i
[3];
3218 micro_not(union tgsi_exec_channel
*dst
,
3219 const union tgsi_exec_channel
*src
)
3221 dst
->u
[0] = ~src
->u
[0];
3222 dst
->u
[1] = ~src
->u
[1];
3223 dst
->u
[2] = ~src
->u
[2];
3224 dst
->u
[3] = ~src
->u
[3];
3228 micro_shl(union tgsi_exec_channel
*dst
,
3229 const union tgsi_exec_channel
*src0
,
3230 const union tgsi_exec_channel
*src1
)
3232 unsigned masked_count
;
3233 masked_count
= src1
->u
[0] & 0x1f;
3234 dst
->u
[0] = src0
->u
[0] << masked_count
;
3235 masked_count
= src1
->u
[1] & 0x1f;
3236 dst
->u
[1] = src0
->u
[1] << masked_count
;
3237 masked_count
= src1
->u
[2] & 0x1f;
3238 dst
->u
[2] = src0
->u
[2] << masked_count
;
3239 masked_count
= src1
->u
[3] & 0x1f;
3240 dst
->u
[3] = src0
->u
[3] << masked_count
;
3244 micro_and(union tgsi_exec_channel
*dst
,
3245 const union tgsi_exec_channel
*src0
,
3246 const union tgsi_exec_channel
*src1
)
3248 dst
->u
[0] = src0
->u
[0] & src1
->u
[0];
3249 dst
->u
[1] = src0
->u
[1] & src1
->u
[1];
3250 dst
->u
[2] = src0
->u
[2] & src1
->u
[2];
3251 dst
->u
[3] = src0
->u
[3] & src1
->u
[3];
3255 micro_or(union tgsi_exec_channel
*dst
,
3256 const union tgsi_exec_channel
*src0
,
3257 const union tgsi_exec_channel
*src1
)
3259 dst
->u
[0] = src0
->u
[0] | src1
->u
[0];
3260 dst
->u
[1] = src0
->u
[1] | src1
->u
[1];
3261 dst
->u
[2] = src0
->u
[2] | src1
->u
[2];
3262 dst
->u
[3] = src0
->u
[3] | src1
->u
[3];
3266 micro_xor(union tgsi_exec_channel
*dst
,
3267 const union tgsi_exec_channel
*src0
,
3268 const union tgsi_exec_channel
*src1
)
3270 dst
->u
[0] = src0
->u
[0] ^ src1
->u
[0];
3271 dst
->u
[1] = src0
->u
[1] ^ src1
->u
[1];
3272 dst
->u
[2] = src0
->u
[2] ^ src1
->u
[2];
3273 dst
->u
[3] = src0
->u
[3] ^ src1
->u
[3];
3277 micro_mod(union tgsi_exec_channel
*dst
,
3278 const union tgsi_exec_channel
*src0
,
3279 const union tgsi_exec_channel
*src1
)
3281 dst
->i
[0] = src0
->i
[0] % src1
->i
[0];
3282 dst
->i
[1] = src0
->i
[1] % src1
->i
[1];
3283 dst
->i
[2] = src0
->i
[2] % src1
->i
[2];
3284 dst
->i
[3] = src0
->i
[3] % src1
->i
[3];
3288 micro_f2i(union tgsi_exec_channel
*dst
,
3289 const union tgsi_exec_channel
*src
)
3291 dst
->i
[0] = (int)src
->f
[0];
3292 dst
->i
[1] = (int)src
->f
[1];
3293 dst
->i
[2] = (int)src
->f
[2];
3294 dst
->i
[3] = (int)src
->f
[3];
3298 micro_fseq(union tgsi_exec_channel
*dst
,
3299 const union tgsi_exec_channel
*src0
,
3300 const union tgsi_exec_channel
*src1
)
3302 dst
->u
[0] = src0
->f
[0] == src1
->f
[0] ? ~0 : 0;
3303 dst
->u
[1] = src0
->f
[1] == src1
->f
[1] ? ~0 : 0;
3304 dst
->u
[2] = src0
->f
[2] == src1
->f
[2] ? ~0 : 0;
3305 dst
->u
[3] = src0
->f
[3] == src1
->f
[3] ? ~0 : 0;
3309 micro_fsge(union tgsi_exec_channel
*dst
,
3310 const union tgsi_exec_channel
*src0
,
3311 const union tgsi_exec_channel
*src1
)
3313 dst
->u
[0] = src0
->f
[0] >= src1
->f
[0] ? ~0 : 0;
3314 dst
->u
[1] = src0
->f
[1] >= src1
->f
[1] ? ~0 : 0;
3315 dst
->u
[2] = src0
->f
[2] >= src1
->f
[2] ? ~0 : 0;
3316 dst
->u
[3] = src0
->f
[3] >= src1
->f
[3] ? ~0 : 0;
3320 micro_fslt(union tgsi_exec_channel
*dst
,
3321 const union tgsi_exec_channel
*src0
,
3322 const union tgsi_exec_channel
*src1
)
3324 dst
->u
[0] = src0
->f
[0] < src1
->f
[0] ? ~0 : 0;
3325 dst
->u
[1] = src0
->f
[1] < src1
->f
[1] ? ~0 : 0;
3326 dst
->u
[2] = src0
->f
[2] < src1
->f
[2] ? ~0 : 0;
3327 dst
->u
[3] = src0
->f
[3] < src1
->f
[3] ? ~0 : 0;
3331 micro_fsne(union tgsi_exec_channel
*dst
,
3332 const union tgsi_exec_channel
*src0
,
3333 const union tgsi_exec_channel
*src1
)
3335 dst
->u
[0] = src0
->f
[0] != src1
->f
[0] ? ~0 : 0;
3336 dst
->u
[1] = src0
->f
[1] != src1
->f
[1] ? ~0 : 0;
3337 dst
->u
[2] = src0
->f
[2] != src1
->f
[2] ? ~0 : 0;
3338 dst
->u
[3] = src0
->f
[3] != src1
->f
[3] ? ~0 : 0;
3342 micro_idiv(union tgsi_exec_channel
*dst
,
3343 const union tgsi_exec_channel
*src0
,
3344 const union tgsi_exec_channel
*src1
)
3346 dst
->i
[0] = src1
->i
[0] ? src0
->i
[0] / src1
->i
[0] : 0;
3347 dst
->i
[1] = src1
->i
[1] ? src0
->i
[1] / src1
->i
[1] : 0;
3348 dst
->i
[2] = src1
->i
[2] ? src0
->i
[2] / src1
->i
[2] : 0;
3349 dst
->i
[3] = src1
->i
[3] ? src0
->i
[3] / src1
->i
[3] : 0;
3353 micro_imax(union tgsi_exec_channel
*dst
,
3354 const union tgsi_exec_channel
*src0
,
3355 const union tgsi_exec_channel
*src1
)
3357 dst
->i
[0] = src0
->i
[0] > src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
3358 dst
->i
[1] = src0
->i
[1] > src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
3359 dst
->i
[2] = src0
->i
[2] > src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
3360 dst
->i
[3] = src0
->i
[3] > src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
3364 micro_imin(union tgsi_exec_channel
*dst
,
3365 const union tgsi_exec_channel
*src0
,
3366 const union tgsi_exec_channel
*src1
)
3368 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
3369 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
3370 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
3371 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
3375 micro_isge(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_ishr(union tgsi_exec_channel
*dst
,
3387 const union tgsi_exec_channel
*src0
,
3388 const union tgsi_exec_channel
*src1
)
3390 unsigned masked_count
;
3391 masked_count
= src1
->i
[0] & 0x1f;
3392 dst
->i
[0] = src0
->i
[0] >> masked_count
;
3393 masked_count
= src1
->i
[1] & 0x1f;
3394 dst
->i
[1] = src0
->i
[1] >> masked_count
;
3395 masked_count
= src1
->i
[2] & 0x1f;
3396 dst
->i
[2] = src0
->i
[2] >> masked_count
;
3397 masked_count
= src1
->i
[3] & 0x1f;
3398 dst
->i
[3] = src0
->i
[3] >> masked_count
;
3402 micro_islt(union tgsi_exec_channel
*dst
,
3403 const union tgsi_exec_channel
*src0
,
3404 const union tgsi_exec_channel
*src1
)
3406 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? -1 : 0;
3407 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? -1 : 0;
3408 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? -1 : 0;
3409 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? -1 : 0;
3413 micro_f2u(union tgsi_exec_channel
*dst
,
3414 const union tgsi_exec_channel
*src
)
3416 dst
->u
[0] = (uint
)src
->f
[0];
3417 dst
->u
[1] = (uint
)src
->f
[1];
3418 dst
->u
[2] = (uint
)src
->f
[2];
3419 dst
->u
[3] = (uint
)src
->f
[3];
3423 micro_u2f(union tgsi_exec_channel
*dst
,
3424 const union tgsi_exec_channel
*src
)
3426 dst
->f
[0] = (float)src
->u
[0];
3427 dst
->f
[1] = (float)src
->u
[1];
3428 dst
->f
[2] = (float)src
->u
[2];
3429 dst
->f
[3] = (float)src
->u
[3];
3433 micro_uadd(union tgsi_exec_channel
*dst
,
3434 const union tgsi_exec_channel
*src0
,
3435 const union tgsi_exec_channel
*src1
)
3437 dst
->u
[0] = src0
->u
[0] + src1
->u
[0];
3438 dst
->u
[1] = src0
->u
[1] + src1
->u
[1];
3439 dst
->u
[2] = src0
->u
[2] + src1
->u
[2];
3440 dst
->u
[3] = src0
->u
[3] + src1
->u
[3];
3444 micro_udiv(union tgsi_exec_channel
*dst
,
3445 const union tgsi_exec_channel
*src0
,
3446 const union tgsi_exec_channel
*src1
)
3448 dst
->u
[0] = src1
->u
[0] ? src0
->u
[0] / src1
->u
[0] : ~0u;
3449 dst
->u
[1] = src1
->u
[1] ? src0
->u
[1] / src1
->u
[1] : ~0u;
3450 dst
->u
[2] = src1
->u
[2] ? src0
->u
[2] / src1
->u
[2] : ~0u;
3451 dst
->u
[3] = src1
->u
[3] ? src0
->u
[3] / src1
->u
[3] : ~0u;
3455 micro_umad(union tgsi_exec_channel
*dst
,
3456 const union tgsi_exec_channel
*src0
,
3457 const union tgsi_exec_channel
*src1
,
3458 const union tgsi_exec_channel
*src2
)
3460 dst
->u
[0] = src0
->u
[0] * src1
->u
[0] + src2
->u
[0];
3461 dst
->u
[1] = src0
->u
[1] * src1
->u
[1] + src2
->u
[1];
3462 dst
->u
[2] = src0
->u
[2] * src1
->u
[2] + src2
->u
[2];
3463 dst
->u
[3] = src0
->u
[3] * src1
->u
[3] + src2
->u
[3];
3467 micro_umax(union tgsi_exec_channel
*dst
,
3468 const union tgsi_exec_channel
*src0
,
3469 const union tgsi_exec_channel
*src1
)
3471 dst
->u
[0] = src0
->u
[0] > src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
3472 dst
->u
[1] = src0
->u
[1] > src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
3473 dst
->u
[2] = src0
->u
[2] > src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
3474 dst
->u
[3] = src0
->u
[3] > src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
3478 micro_umin(union tgsi_exec_channel
*dst
,
3479 const union tgsi_exec_channel
*src0
,
3480 const union tgsi_exec_channel
*src1
)
3482 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
3483 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
3484 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
3485 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
3489 micro_umod(union tgsi_exec_channel
*dst
,
3490 const union tgsi_exec_channel
*src0
,
3491 const union tgsi_exec_channel
*src1
)
3493 dst
->u
[0] = src1
->u
[0] ? src0
->u
[0] % src1
->u
[0] : ~0u;
3494 dst
->u
[1] = src1
->u
[1] ? src0
->u
[1] % src1
->u
[1] : ~0u;
3495 dst
->u
[2] = src1
->u
[2] ? src0
->u
[2] % src1
->u
[2] : ~0u;
3496 dst
->u
[3] = src1
->u
[3] ? src0
->u
[3] % src1
->u
[3] : ~0u;
3500 micro_umul(union tgsi_exec_channel
*dst
,
3501 const union tgsi_exec_channel
*src0
,
3502 const union tgsi_exec_channel
*src1
)
3504 dst
->u
[0] = src0
->u
[0] * src1
->u
[0];
3505 dst
->u
[1] = src0
->u
[1] * src1
->u
[1];
3506 dst
->u
[2] = src0
->u
[2] * src1
->u
[2];
3507 dst
->u
[3] = src0
->u
[3] * src1
->u
[3];
3511 micro_imul_hi(union tgsi_exec_channel
*dst
,
3512 const union tgsi_exec_channel
*src0
,
3513 const union tgsi_exec_channel
*src1
)
3515 #define I64M(x, y) ((((int64_t)x) * ((int64_t)y)) >> 32)
3516 dst
->i
[0] = I64M(src0
->i
[0], src1
->i
[0]);
3517 dst
->i
[1] = I64M(src0
->i
[1], src1
->i
[1]);
3518 dst
->i
[2] = I64M(src0
->i
[2], src1
->i
[2]);
3519 dst
->i
[3] = I64M(src0
->i
[3], src1
->i
[3]);
3524 micro_umul_hi(union tgsi_exec_channel
*dst
,
3525 const union tgsi_exec_channel
*src0
,
3526 const union tgsi_exec_channel
*src1
)
3528 #define U64M(x, y) ((((uint64_t)x) * ((uint64_t)y)) >> 32)
3529 dst
->u
[0] = U64M(src0
->u
[0], src1
->u
[0]);
3530 dst
->u
[1] = U64M(src0
->u
[1], src1
->u
[1]);
3531 dst
->u
[2] = U64M(src0
->u
[2], src1
->u
[2]);
3532 dst
->u
[3] = U64M(src0
->u
[3], src1
->u
[3]);
3537 micro_useq(union tgsi_exec_channel
*dst
,
3538 const union tgsi_exec_channel
*src0
,
3539 const union tgsi_exec_channel
*src1
)
3541 dst
->u
[0] = src0
->u
[0] == src1
->u
[0] ? ~0 : 0;
3542 dst
->u
[1] = src0
->u
[1] == src1
->u
[1] ? ~0 : 0;
3543 dst
->u
[2] = src0
->u
[2] == src1
->u
[2] ? ~0 : 0;
3544 dst
->u
[3] = src0
->u
[3] == src1
->u
[3] ? ~0 : 0;
3548 micro_usge(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_ushr(union tgsi_exec_channel
*dst
,
3560 const union tgsi_exec_channel
*src0
,
3561 const union tgsi_exec_channel
*src1
)
3563 unsigned masked_count
;
3564 masked_count
= src1
->u
[0] & 0x1f;
3565 dst
->u
[0] = src0
->u
[0] >> masked_count
;
3566 masked_count
= src1
->u
[1] & 0x1f;
3567 dst
->u
[1] = src0
->u
[1] >> masked_count
;
3568 masked_count
= src1
->u
[2] & 0x1f;
3569 dst
->u
[2] = src0
->u
[2] >> masked_count
;
3570 masked_count
= src1
->u
[3] & 0x1f;
3571 dst
->u
[3] = src0
->u
[3] >> masked_count
;
3575 micro_uslt(union tgsi_exec_channel
*dst
,
3576 const union tgsi_exec_channel
*src0
,
3577 const union tgsi_exec_channel
*src1
)
3579 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? ~0 : 0;
3580 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? ~0 : 0;
3581 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? ~0 : 0;
3582 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? ~0 : 0;
3586 micro_usne(union tgsi_exec_channel
*dst
,
3587 const union tgsi_exec_channel
*src0
,
3588 const union tgsi_exec_channel
*src1
)
3590 dst
->u
[0] = src0
->u
[0] != src1
->u
[0] ? ~0 : 0;
3591 dst
->u
[1] = src0
->u
[1] != src1
->u
[1] ? ~0 : 0;
3592 dst
->u
[2] = src0
->u
[2] != src1
->u
[2] ? ~0 : 0;
3593 dst
->u
[3] = src0
->u
[3] != src1
->u
[3] ? ~0 : 0;
3597 micro_uarl(union tgsi_exec_channel
*dst
,
3598 const union tgsi_exec_channel
*src
)
3600 dst
->i
[0] = src
->u
[0];
3601 dst
->i
[1] = src
->u
[1];
3602 dst
->i
[2] = src
->u
[2];
3603 dst
->i
[3] = src
->u
[3];
3607 micro_ucmp(union tgsi_exec_channel
*dst
,
3608 const union tgsi_exec_channel
*src0
,
3609 const union tgsi_exec_channel
*src1
,
3610 const union tgsi_exec_channel
*src2
)
3612 dst
->u
[0] = src0
->u
[0] ? src1
->u
[0] : src2
->u
[0];
3613 dst
->u
[1] = src0
->u
[1] ? src1
->u
[1] : src2
->u
[1];
3614 dst
->u
[2] = src0
->u
[2] ? src1
->u
[2] : src2
->u
[2];
3615 dst
->u
[3] = src0
->u
[3] ? src1
->u
[3] : src2
->u
[3];
3619 * Signed bitfield extract (i.e. sign-extend the extracted bits)
3622 micro_ibfe(union tgsi_exec_channel
*dst
,
3623 const union tgsi_exec_channel
*src0
,
3624 const union tgsi_exec_channel
*src1
,
3625 const union tgsi_exec_channel
*src2
)
3628 for (i
= 0; i
< 4; i
++) {
3629 int width
= src2
->i
[i
] & 0x1f;
3630 int offset
= src1
->i
[i
] & 0x1f;
3633 else if (width
+ offset
< 32)
3634 dst
->i
[i
] = (src0
->i
[i
] << (32 - width
- offset
)) >> (32 - width
);
3636 dst
->i
[i
] = src0
->i
[i
] >> offset
;
3641 * Unsigned bitfield extract
3644 micro_ubfe(union tgsi_exec_channel
*dst
,
3645 const union tgsi_exec_channel
*src0
,
3646 const union tgsi_exec_channel
*src1
,
3647 const union tgsi_exec_channel
*src2
)
3650 for (i
= 0; i
< 4; i
++) {
3651 int width
= src2
->u
[i
] & 0x1f;
3652 int offset
= src1
->u
[i
] & 0x1f;
3655 else if (width
+ offset
< 32)
3656 dst
->u
[i
] = (src0
->u
[i
] << (32 - width
- offset
)) >> (32 - width
);
3658 dst
->u
[i
] = src0
->u
[i
] >> offset
;
3663 * Bitfield insert: copy low bits from src1 into a region of src0.
3666 micro_bfi(union tgsi_exec_channel
*dst
,
3667 const union tgsi_exec_channel
*src0
,
3668 const union tgsi_exec_channel
*src1
,
3669 const union tgsi_exec_channel
*src2
,
3670 const union tgsi_exec_channel
*src3
)
3673 for (i
= 0; i
< 4; i
++) {
3674 int width
= src3
->u
[i
] & 0x1f;
3675 int offset
= src2
->u
[i
] & 0x1f;
3676 int bitmask
= ((1 << width
) - 1) << offset
;
3677 dst
->u
[i
] = ((src1
->u
[i
] << offset
) & bitmask
) | (src0
->u
[i
] & ~bitmask
);
3682 micro_brev(union tgsi_exec_channel
*dst
,
3683 const union tgsi_exec_channel
*src
)
3685 dst
->u
[0] = util_bitreverse(src
->u
[0]);
3686 dst
->u
[1] = util_bitreverse(src
->u
[1]);
3687 dst
->u
[2] = util_bitreverse(src
->u
[2]);
3688 dst
->u
[3] = util_bitreverse(src
->u
[3]);
3692 micro_popc(union tgsi_exec_channel
*dst
,
3693 const union tgsi_exec_channel
*src
)
3695 dst
->u
[0] = util_bitcount(src
->u
[0]);
3696 dst
->u
[1] = util_bitcount(src
->u
[1]);
3697 dst
->u
[2] = util_bitcount(src
->u
[2]);
3698 dst
->u
[3] = util_bitcount(src
->u
[3]);
3702 micro_lsb(union tgsi_exec_channel
*dst
,
3703 const union tgsi_exec_channel
*src
)
3705 dst
->i
[0] = ffs(src
->u
[0]) - 1;
3706 dst
->i
[1] = ffs(src
->u
[1]) - 1;
3707 dst
->i
[2] = ffs(src
->u
[2]) - 1;
3708 dst
->i
[3] = ffs(src
->u
[3]) - 1;
3712 micro_imsb(union tgsi_exec_channel
*dst
,
3713 const union tgsi_exec_channel
*src
)
3715 dst
->i
[0] = util_last_bit_signed(src
->i
[0]) - 1;
3716 dst
->i
[1] = util_last_bit_signed(src
->i
[1]) - 1;
3717 dst
->i
[2] = util_last_bit_signed(src
->i
[2]) - 1;
3718 dst
->i
[3] = util_last_bit_signed(src
->i
[3]) - 1;
3722 micro_umsb(union tgsi_exec_channel
*dst
,
3723 const union tgsi_exec_channel
*src
)
3725 dst
->i
[0] = util_last_bit(src
->u
[0]) - 1;
3726 dst
->i
[1] = util_last_bit(src
->u
[1]) - 1;
3727 dst
->i
[2] = util_last_bit(src
->u
[2]) - 1;
3728 dst
->i
[3] = util_last_bit(src
->u
[3]) - 1;
3733 struct tgsi_exec_machine
*mach
,
3734 const struct tgsi_full_instruction
*inst
,
3737 union tgsi_exec_channel r
[10];
3741 switch (inst
->Instruction
.Opcode
) {
3742 case TGSI_OPCODE_ARL
:
3743 exec_vector_unary(mach
, inst
, micro_arl
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
3746 case TGSI_OPCODE_MOV
:
3747 exec_vector_unary(mach
, inst
, micro_mov
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
3750 case TGSI_OPCODE_LIT
:
3751 exec_lit(mach
, inst
);
3754 case TGSI_OPCODE_RCP
:
3755 exec_scalar_unary(mach
, inst
, micro_rcp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3758 case TGSI_OPCODE_RSQ
:
3759 exec_scalar_unary(mach
, inst
, micro_rsq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3762 case TGSI_OPCODE_EXP
:
3763 exec_exp(mach
, inst
);
3766 case TGSI_OPCODE_LOG
:
3767 exec_log(mach
, inst
);
3770 case TGSI_OPCODE_MUL
:
3771 exec_vector_binary(mach
, inst
, micro_mul
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3774 case TGSI_OPCODE_ADD
:
3775 exec_vector_binary(mach
, inst
, micro_add
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3778 case TGSI_OPCODE_DP3
:
3779 exec_dp3(mach
, inst
);
3782 case TGSI_OPCODE_DP4
:
3783 exec_dp4(mach
, inst
);
3786 case TGSI_OPCODE_DST
:
3787 exec_dst(mach
, inst
);
3790 case TGSI_OPCODE_MIN
:
3791 exec_vector_binary(mach
, inst
, micro_min
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3794 case TGSI_OPCODE_MAX
:
3795 exec_vector_binary(mach
, inst
, micro_max
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3798 case TGSI_OPCODE_SLT
:
3799 exec_vector_binary(mach
, inst
, micro_slt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3802 case TGSI_OPCODE_SGE
:
3803 exec_vector_binary(mach
, inst
, micro_sge
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3806 case TGSI_OPCODE_MAD
:
3807 exec_vector_trinary(mach
, inst
, micro_mad
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3810 case TGSI_OPCODE_SUB
:
3811 exec_vector_binary(mach
, inst
, micro_sub
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3814 case TGSI_OPCODE_LRP
:
3815 exec_vector_trinary(mach
, inst
, micro_lrp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3818 case TGSI_OPCODE_CND
:
3819 exec_vector_trinary(mach
, inst
, micro_cnd
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3822 case TGSI_OPCODE_SQRT
:
3823 exec_scalar_unary(mach
, inst
, micro_sqrt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3826 case TGSI_OPCODE_DP2A
:
3827 exec_dp2a(mach
, inst
);
3830 case TGSI_OPCODE_FRC
:
3831 exec_vector_unary(mach
, inst
, micro_frc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3834 case TGSI_OPCODE_CLAMP
:
3835 exec_vector_trinary(mach
, inst
, micro_clamp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3838 case TGSI_OPCODE_FLR
:
3839 exec_vector_unary(mach
, inst
, micro_flr
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3842 case TGSI_OPCODE_ROUND
:
3843 exec_vector_unary(mach
, inst
, micro_rnd
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3846 case TGSI_OPCODE_EX2
:
3847 exec_scalar_unary(mach
, inst
, micro_exp2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3850 case TGSI_OPCODE_LG2
:
3851 exec_scalar_unary(mach
, inst
, micro_lg2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3854 case TGSI_OPCODE_POW
:
3855 exec_scalar_binary(mach
, inst
, micro_pow
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3858 case TGSI_OPCODE_XPD
:
3859 exec_xpd(mach
, inst
);
3862 case TGSI_OPCODE_ABS
:
3863 exec_vector_unary(mach
, inst
, micro_abs
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3866 case TGSI_OPCODE_RCC
:
3867 exec_scalar_unary(mach
, inst
, micro_rcc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3870 case TGSI_OPCODE_DPH
:
3871 exec_dph(mach
, inst
);
3874 case TGSI_OPCODE_COS
:
3875 exec_scalar_unary(mach
, inst
, micro_cos
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3878 case TGSI_OPCODE_DDX
:
3879 exec_vector_unary(mach
, inst
, micro_ddx
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3882 case TGSI_OPCODE_DDY
:
3883 exec_vector_unary(mach
, inst
, micro_ddy
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3886 case TGSI_OPCODE_KILL
:
3887 exec_kill (mach
, inst
);
3890 case TGSI_OPCODE_KILL_IF
:
3891 exec_kill_if (mach
, inst
);
3894 case TGSI_OPCODE_PK2H
:
3898 case TGSI_OPCODE_PK2US
:
3902 case TGSI_OPCODE_PK4B
:
3906 case TGSI_OPCODE_PK4UB
:
3910 case TGSI_OPCODE_RFL
:
3911 exec_rfl(mach
, inst
);
3914 case TGSI_OPCODE_SEQ
:
3915 exec_vector_binary(mach
, inst
, micro_seq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3918 case TGSI_OPCODE_SFL
:
3919 exec_vector(mach
, inst
, micro_sfl
, TGSI_EXEC_DATA_FLOAT
);
3922 case TGSI_OPCODE_SGT
:
3923 exec_vector_binary(mach
, inst
, micro_sgt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3926 case TGSI_OPCODE_SIN
:
3927 exec_scalar_unary(mach
, inst
, micro_sin
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3930 case TGSI_OPCODE_SLE
:
3931 exec_vector_binary(mach
, inst
, micro_sle
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3934 case TGSI_OPCODE_SNE
:
3935 exec_vector_binary(mach
, inst
, micro_sne
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3938 case TGSI_OPCODE_STR
:
3939 exec_vector(mach
, inst
, micro_str
, TGSI_EXEC_DATA_FLOAT
);
3942 case TGSI_OPCODE_TEX
:
3943 /* simple texture lookup */
3944 /* src[0] = texcoord */
3945 /* src[1] = sampler unit */
3946 exec_tex(mach
, inst
, TEX_MODIFIER_NONE
, 1);
3949 case TGSI_OPCODE_TXB
:
3950 /* Texture lookup with lod bias */
3951 /* src[0] = texcoord (src[0].w = LOD bias) */
3952 /* src[1] = sampler unit */
3953 exec_tex(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, 1);
3956 case TGSI_OPCODE_TXD
:
3957 /* Texture lookup with explict partial derivatives */
3958 /* src[0] = texcoord */
3959 /* src[1] = d[strq]/dx */
3960 /* src[2] = d[strq]/dy */
3961 /* src[3] = sampler unit */
3962 exec_txd(mach
, inst
);
3965 case TGSI_OPCODE_TXL
:
3966 /* Texture lookup with explit LOD */
3967 /* src[0] = texcoord (src[0].w = LOD) */
3968 /* src[1] = sampler unit */
3969 exec_tex(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, 1);
3972 case TGSI_OPCODE_TXP
:
3973 /* Texture lookup with projection */
3974 /* src[0] = texcoord (src[0].w = projection) */
3975 /* src[1] = sampler unit */
3976 exec_tex(mach
, inst
, TEX_MODIFIER_PROJECTED
, 1);
3979 case TGSI_OPCODE_UP2H
:
3983 case TGSI_OPCODE_UP2US
:
3987 case TGSI_OPCODE_UP4B
:
3991 case TGSI_OPCODE_UP4UB
:
3995 case TGSI_OPCODE_X2D
:
3996 exec_x2d(mach
, inst
);
3999 case TGSI_OPCODE_ARA
:
4003 case TGSI_OPCODE_ARR
:
4004 exec_vector_unary(mach
, inst
, micro_arr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
4007 case TGSI_OPCODE_BRA
:
4011 case TGSI_OPCODE_CAL
:
4012 /* skip the call if no execution channels are enabled */
4013 if (mach
->ExecMask
) {
4016 /* First, record the depths of the execution stacks.
4017 * This is important for deeply nested/looped return statements.
4018 * We have to unwind the stacks by the correct amount. For a
4019 * real code generator, we could determine the number of entries
4020 * to pop off each stack with simple static analysis and avoid
4021 * implementing this data structure at run time.
4023 mach
->CallStack
[mach
->CallStackTop
].CondStackTop
= mach
->CondStackTop
;
4024 mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
= mach
->LoopStackTop
;
4025 mach
->CallStack
[mach
->CallStackTop
].ContStackTop
= mach
->ContStackTop
;
4026 mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
= mach
->SwitchStackTop
;
4027 mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
= mach
->BreakStackTop
;
4028 /* note that PC was already incremented above */
4029 mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
= *pc
;
4031 mach
->CallStackTop
++;
4033 /* Second, push the Cond, Loop, Cont, Func stacks */
4034 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
4035 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
4036 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
4037 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
4038 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
4039 assert(mach
->FuncStackTop
< TGSI_EXEC_MAX_CALL_NESTING
);
4041 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
4042 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
4043 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
4044 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
4045 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
4046 mach
->FuncStack
[mach
->FuncStackTop
++] = mach
->FuncMask
;
4048 /* Finally, jump to the subroutine */
4049 *pc
= inst
->Label
.Label
;
4053 case TGSI_OPCODE_RET
:
4054 mach
->FuncMask
&= ~mach
->ExecMask
;
4055 UPDATE_EXEC_MASK(mach
);
4057 if (mach
->FuncMask
== 0x0) {
4058 /* really return now (otherwise, keep executing */
4060 if (mach
->CallStackTop
== 0) {
4061 /* returning from main() */
4062 mach
->CondStackTop
= 0;
4063 mach
->LoopStackTop
= 0;
4068 assert(mach
->CallStackTop
> 0);
4069 mach
->CallStackTop
--;
4071 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
4072 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
4074 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
4075 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
4077 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
4078 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
4080 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
4081 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
4083 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
4084 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
4086 assert(mach
->FuncStackTop
> 0);
4087 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
4089 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
4091 UPDATE_EXEC_MASK(mach
);
4095 case TGSI_OPCODE_SSG
:
4096 exec_vector_unary(mach
, inst
, micro_sgn
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
4099 case TGSI_OPCODE_CMP
:
4100 exec_vector_trinary(mach
, inst
, micro_cmp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
4103 case TGSI_OPCODE_SCS
:
4104 exec_scs(mach
, inst
);
4107 case TGSI_OPCODE_NRM
:
4108 exec_nrm3(mach
, inst
);
4111 case TGSI_OPCODE_NRM4
:
4112 exec_nrm4(mach
, inst
);
4115 case TGSI_OPCODE_DIV
:
4116 exec_vector_binary(mach
, inst
, micro_div
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
4119 case TGSI_OPCODE_DP2
:
4120 exec_dp2(mach
, inst
);
4123 case TGSI_OPCODE_IF
:
4125 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
4126 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
4127 FETCH( &r
[0], 0, TGSI_CHAN_X
);
4128 /* update CondMask */
4130 mach
->CondMask
&= ~0x1;
4133 mach
->CondMask
&= ~0x2;
4136 mach
->CondMask
&= ~0x4;
4139 mach
->CondMask
&= ~0x8;
4141 UPDATE_EXEC_MASK(mach
);
4142 /* Todo: If CondMask==0, jump to ELSE */
4145 case TGSI_OPCODE_UIF
:
4147 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
4148 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
4149 IFETCH( &r
[0], 0, TGSI_CHAN_X
);
4150 /* update CondMask */
4152 mach
->CondMask
&= ~0x1;
4155 mach
->CondMask
&= ~0x2;
4158 mach
->CondMask
&= ~0x4;
4161 mach
->CondMask
&= ~0x8;
4163 UPDATE_EXEC_MASK(mach
);
4164 /* Todo: If CondMask==0, jump to ELSE */
4167 case TGSI_OPCODE_ELSE
:
4168 /* invert CondMask wrt previous mask */
4171 assert(mach
->CondStackTop
> 0);
4172 prevMask
= mach
->CondStack
[mach
->CondStackTop
- 1];
4173 mach
->CondMask
= ~mach
->CondMask
& prevMask
;
4174 UPDATE_EXEC_MASK(mach
);
4175 /* Todo: If CondMask==0, jump to ENDIF */
4179 case TGSI_OPCODE_ENDIF
:
4181 assert(mach
->CondStackTop
> 0);
4182 mach
->CondMask
= mach
->CondStack
[--mach
->CondStackTop
];
4183 UPDATE_EXEC_MASK(mach
);
4186 case TGSI_OPCODE_END
:
4187 /* make sure we end primitives which haven't
4188 * been explicitly emitted */
4189 conditional_emit_primitive(mach
);
4190 /* halt execution */
4194 case TGSI_OPCODE_PUSHA
:
4198 case TGSI_OPCODE_POPA
:
4202 case TGSI_OPCODE_CEIL
:
4203 exec_vector_unary(mach
, inst
, micro_ceil
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
4206 case TGSI_OPCODE_I2F
:
4207 exec_vector_unary(mach
, inst
, micro_i2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_INT
);
4210 case TGSI_OPCODE_NOT
:
4211 exec_vector_unary(mach
, inst
, micro_not
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4214 case TGSI_OPCODE_TRUNC
:
4215 exec_vector_unary(mach
, inst
, micro_trunc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
4218 case TGSI_OPCODE_SHL
:
4219 exec_vector_binary(mach
, inst
, micro_shl
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4222 case TGSI_OPCODE_AND
:
4223 exec_vector_binary(mach
, inst
, micro_and
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4226 case TGSI_OPCODE_OR
:
4227 exec_vector_binary(mach
, inst
, micro_or
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4230 case TGSI_OPCODE_MOD
:
4231 exec_vector_binary(mach
, inst
, micro_mod
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4234 case TGSI_OPCODE_XOR
:
4235 exec_vector_binary(mach
, inst
, micro_xor
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4238 case TGSI_OPCODE_SAD
:
4242 case TGSI_OPCODE_TXF
:
4243 exec_txf(mach
, inst
);
4246 case TGSI_OPCODE_TXQ
:
4247 exec_txq(mach
, inst
);
4250 case TGSI_OPCODE_EMIT
:
4254 case TGSI_OPCODE_ENDPRIM
:
4255 emit_primitive(mach
);
4258 case TGSI_OPCODE_BGNLOOP
:
4259 /* push LoopMask and ContMasks */
4260 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
4261 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
4262 assert(mach
->LoopLabelStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
4263 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
4265 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
4266 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
4267 mach
->LoopLabelStack
[mach
->LoopLabelStackTop
++] = *pc
- 1;
4268 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
4269 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_LOOP
;
4272 case TGSI_OPCODE_ENDLOOP
:
4273 /* Restore ContMask, but don't pop */
4274 assert(mach
->ContStackTop
> 0);
4275 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
- 1];
4276 UPDATE_EXEC_MASK(mach
);
4277 if (mach
->ExecMask
) {
4278 /* repeat loop: jump to instruction just past BGNLOOP */
4279 assert(mach
->LoopLabelStackTop
> 0);
4280 *pc
= mach
->LoopLabelStack
[mach
->LoopLabelStackTop
- 1] + 1;
4283 /* exit loop: pop LoopMask */
4284 assert(mach
->LoopStackTop
> 0);
4285 mach
->LoopMask
= mach
->LoopStack
[--mach
->LoopStackTop
];
4287 assert(mach
->ContStackTop
> 0);
4288 mach
->ContMask
= mach
->ContStack
[--mach
->ContStackTop
];
4289 assert(mach
->LoopLabelStackTop
> 0);
4290 --mach
->LoopLabelStackTop
;
4292 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
4294 UPDATE_EXEC_MASK(mach
);
4297 case TGSI_OPCODE_BRK
:
4301 case TGSI_OPCODE_CONT
:
4302 /* turn off cont channels for each enabled exec channel */
4303 mach
->ContMask
&= ~mach
->ExecMask
;
4304 /* Todo: if mach->LoopMask == 0, jump to end of loop */
4305 UPDATE_EXEC_MASK(mach
);
4308 case TGSI_OPCODE_BGNSUB
:
4312 case TGSI_OPCODE_ENDSUB
:
4314 * XXX: This really should be a no-op. We should never reach this opcode.
4317 assert(mach
->CallStackTop
> 0);
4318 mach
->CallStackTop
--;
4320 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
4321 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
4323 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
4324 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
4326 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
4327 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
4329 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
4330 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
4332 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
4333 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
4335 assert(mach
->FuncStackTop
> 0);
4336 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
4338 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
4340 UPDATE_EXEC_MASK(mach
);
4343 case TGSI_OPCODE_NOP
:
4346 case TGSI_OPCODE_BREAKC
:
4347 IFETCH(&r
[0], 0, TGSI_CHAN_X
);
4348 /* update CondMask */
4349 if (r
[0].u
[0] && (mach
->ExecMask
& 0x1)) {
4350 mach
->LoopMask
&= ~0x1;
4352 if (r
[0].u
[1] && (mach
->ExecMask
& 0x2)) {
4353 mach
->LoopMask
&= ~0x2;
4355 if (r
[0].u
[2] && (mach
->ExecMask
& 0x4)) {
4356 mach
->LoopMask
&= ~0x4;
4358 if (r
[0].u
[3] && (mach
->ExecMask
& 0x8)) {
4359 mach
->LoopMask
&= ~0x8;
4361 /* Todo: if mach->LoopMask == 0, jump to end of loop */
4362 UPDATE_EXEC_MASK(mach
);
4365 case TGSI_OPCODE_F2I
:
4366 exec_vector_unary(mach
, inst
, micro_f2i
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
4369 case TGSI_OPCODE_FSEQ
:
4370 exec_vector_binary(mach
, inst
, micro_fseq
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4373 case TGSI_OPCODE_FSGE
:
4374 exec_vector_binary(mach
, inst
, micro_fsge
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4377 case TGSI_OPCODE_FSLT
:
4378 exec_vector_binary(mach
, inst
, micro_fslt
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4381 case TGSI_OPCODE_FSNE
:
4382 exec_vector_binary(mach
, inst
, micro_fsne
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4385 case TGSI_OPCODE_IDIV
:
4386 exec_vector_binary(mach
, inst
, micro_idiv
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4389 case TGSI_OPCODE_IMAX
:
4390 exec_vector_binary(mach
, inst
, micro_imax
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4393 case TGSI_OPCODE_IMIN
:
4394 exec_vector_binary(mach
, inst
, micro_imin
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4397 case TGSI_OPCODE_INEG
:
4398 exec_vector_unary(mach
, inst
, micro_ineg
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4401 case TGSI_OPCODE_ISGE
:
4402 exec_vector_binary(mach
, inst
, micro_isge
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4405 case TGSI_OPCODE_ISHR
:
4406 exec_vector_binary(mach
, inst
, micro_ishr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4409 case TGSI_OPCODE_ISLT
:
4410 exec_vector_binary(mach
, inst
, micro_islt
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4413 case TGSI_OPCODE_F2U
:
4414 exec_vector_unary(mach
, inst
, micro_f2u
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
4417 case TGSI_OPCODE_U2F
:
4418 exec_vector_unary(mach
, inst
, micro_u2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_UINT
);
4421 case TGSI_OPCODE_UADD
:
4422 exec_vector_binary(mach
, inst
, micro_uadd
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4425 case TGSI_OPCODE_UDIV
:
4426 exec_vector_binary(mach
, inst
, micro_udiv
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4429 case TGSI_OPCODE_UMAD
:
4430 exec_vector_trinary(mach
, inst
, micro_umad
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4433 case TGSI_OPCODE_UMAX
:
4434 exec_vector_binary(mach
, inst
, micro_umax
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4437 case TGSI_OPCODE_UMIN
:
4438 exec_vector_binary(mach
, inst
, micro_umin
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4441 case TGSI_OPCODE_UMOD
:
4442 exec_vector_binary(mach
, inst
, micro_umod
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4445 case TGSI_OPCODE_UMUL
:
4446 exec_vector_binary(mach
, inst
, micro_umul
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4449 case TGSI_OPCODE_IMUL_HI
:
4450 exec_vector_binary(mach
, inst
, micro_imul_hi
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4453 case TGSI_OPCODE_UMUL_HI
:
4454 exec_vector_binary(mach
, inst
, micro_umul_hi
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4457 case TGSI_OPCODE_USEQ
:
4458 exec_vector_binary(mach
, inst
, micro_useq
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4461 case TGSI_OPCODE_USGE
:
4462 exec_vector_binary(mach
, inst
, micro_usge
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4465 case TGSI_OPCODE_USHR
:
4466 exec_vector_binary(mach
, inst
, micro_ushr
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4469 case TGSI_OPCODE_USLT
:
4470 exec_vector_binary(mach
, inst
, micro_uslt
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4473 case TGSI_OPCODE_USNE
:
4474 exec_vector_binary(mach
, inst
, micro_usne
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4477 case TGSI_OPCODE_SWITCH
:
4478 exec_switch(mach
, inst
);
4481 case TGSI_OPCODE_CASE
:
4482 exec_case(mach
, inst
);
4485 case TGSI_OPCODE_DEFAULT
:
4489 case TGSI_OPCODE_ENDSWITCH
:
4490 exec_endswitch(mach
);
4493 case TGSI_OPCODE_SAMPLE_I
:
4494 exec_txf(mach
, inst
);
4497 case TGSI_OPCODE_SAMPLE_I_MS
:
4501 case TGSI_OPCODE_SAMPLE
:
4502 exec_sample(mach
, inst
, TEX_MODIFIER_NONE
, FALSE
);
4505 case TGSI_OPCODE_SAMPLE_B
:
4506 exec_sample(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, FALSE
);
4509 case TGSI_OPCODE_SAMPLE_C
:
4510 exec_sample(mach
, inst
, TEX_MODIFIER_NONE
, TRUE
);
4513 case TGSI_OPCODE_SAMPLE_C_LZ
:
4514 exec_sample(mach
, inst
, TEX_MODIFIER_LEVEL_ZERO
, TRUE
);
4517 case TGSI_OPCODE_SAMPLE_D
:
4518 exec_sample_d(mach
, inst
);
4521 case TGSI_OPCODE_SAMPLE_L
:
4522 exec_sample(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, FALSE
);
4525 case TGSI_OPCODE_GATHER4
:
4529 case TGSI_OPCODE_SVIEWINFO
:
4530 exec_txq(mach
, inst
);
4533 case TGSI_OPCODE_SAMPLE_POS
:
4537 case TGSI_OPCODE_SAMPLE_INFO
:
4541 case TGSI_OPCODE_UARL
:
4542 exec_vector_unary(mach
, inst
, micro_uarl
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_UINT
);
4545 case TGSI_OPCODE_UCMP
:
4546 exec_vector_trinary(mach
, inst
, micro_ucmp
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4549 case TGSI_OPCODE_IABS
:
4550 exec_vector_unary(mach
, inst
, micro_iabs
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4553 case TGSI_OPCODE_ISSG
:
4554 exec_vector_unary(mach
, inst
, micro_isgn
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4557 case TGSI_OPCODE_TEX2
:
4558 /* simple texture lookup */
4559 /* src[0] = texcoord */
4560 /* src[1] = compare */
4561 /* src[2] = sampler unit */
4562 exec_tex(mach
, inst
, TEX_MODIFIER_NONE
, 2);
4564 case TGSI_OPCODE_TXB2
:
4565 /* simple texture lookup */
4566 /* src[0] = texcoord */
4568 /* src[2] = sampler unit */
4569 exec_tex(mach
, inst
, TEX_MODIFIER_LOD_BIAS
, 2);
4571 case TGSI_OPCODE_TXL2
:
4572 /* simple texture lookup */
4573 /* src[0] = texcoord */
4575 /* src[2] = sampler unit */
4576 exec_tex(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
, 2);
4579 case TGSI_OPCODE_IBFE
:
4580 exec_vector_trinary(mach
, inst
, micro_ibfe
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4582 case TGSI_OPCODE_UBFE
:
4583 exec_vector_trinary(mach
, inst
, micro_ubfe
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4585 case TGSI_OPCODE_BFI
:
4586 exec_vector_quaternary(mach
, inst
, micro_bfi
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4588 case TGSI_OPCODE_BREV
:
4589 exec_vector_unary(mach
, inst
, micro_brev
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4591 case TGSI_OPCODE_POPC
:
4592 exec_vector_unary(mach
, inst
, micro_popc
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
4594 case TGSI_OPCODE_LSB
:
4595 exec_vector_unary(mach
, inst
, micro_lsb
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_UINT
);
4597 case TGSI_OPCODE_IMSB
:
4598 exec_vector_unary(mach
, inst
, micro_imsb
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
4600 case TGSI_OPCODE_UMSB
:
4601 exec_vector_unary(mach
, inst
, micro_umsb
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_UINT
);
4610 * Run TGSI interpreter.
4611 * \return bitmask of "alive" quad components
4614 tgsi_exec_machine_run( struct tgsi_exec_machine
*mach
)
4618 uint default_mask
= 0xf;
4620 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] = 0;
4621 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] = 0;
4623 if( mach
->Processor
== TGSI_PROCESSOR_GEOMETRY
) {
4624 mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0] = 0;
4625 mach
->Primitives
[0] = 0;
4626 /* GS runs on a single primitive for now */
4630 mach
->CondMask
= default_mask
;
4631 mach
->LoopMask
= default_mask
;
4632 mach
->ContMask
= default_mask
;
4633 mach
->FuncMask
= default_mask
;
4634 mach
->ExecMask
= default_mask
;
4636 mach
->Switch
.mask
= default_mask
;
4638 assert(mach
->CondStackTop
== 0);
4639 assert(mach
->LoopStackTop
== 0);
4640 assert(mach
->ContStackTop
== 0);
4641 assert(mach
->SwitchStackTop
== 0);
4642 assert(mach
->BreakStackTop
== 0);
4643 assert(mach
->CallStackTop
== 0);
4646 /* execute declarations (interpolants) */
4647 for (i
= 0; i
< mach
->NumDeclarations
; i
++) {
4648 exec_declaration( mach
, mach
->Declarations
+i
);
4653 struct tgsi_exec_vector temps
[TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
];
4654 struct tgsi_exec_vector outputs
[PIPE_MAX_ATTRIBS
];
4657 memset(mach
->Temps
, 0, sizeof(temps
));
4658 memset(mach
->Outputs
, 0, sizeof(outputs
));
4659 memset(temps
, 0, sizeof(temps
));
4660 memset(outputs
, 0, sizeof(outputs
));
4663 /* execute instructions, until pc is set to -1 */
4669 tgsi_dump_instruction(&mach
->Instructions
[pc
], inst
++);
4672 assert(pc
< (int) mach
->NumInstructions
);
4673 exec_instruction(mach
, mach
->Instructions
+ pc
, &pc
);
4676 for (i
= 0; i
< TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
; i
++) {
4677 if (memcmp(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]))) {
4680 memcpy(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]));
4681 debug_printf("TEMP[%2u] = ", i
);
4682 for (j
= 0; j
< 4; j
++) {
4686 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
4687 temps
[i
].xyzw
[0].f
[j
], temps
[i
].xyzw
[0].u
[j
],
4688 temps
[i
].xyzw
[1].f
[j
], temps
[i
].xyzw
[1].u
[j
],
4689 temps
[i
].xyzw
[2].f
[j
], temps
[i
].xyzw
[2].u
[j
],
4690 temps
[i
].xyzw
[3].f
[j
], temps
[i
].xyzw
[3].u
[j
]);
4694 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
4695 if (memcmp(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]))) {
4698 memcpy(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]));
4699 debug_printf("OUT[%2u] = ", i
);
4700 for (j
= 0; j
< 4; j
++) {
4704 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
4705 outputs
[i
].xyzw
[0].f
[j
], outputs
[i
].xyzw
[0].u
[j
],
4706 outputs
[i
].xyzw
[1].f
[j
], outputs
[i
].xyzw
[1].u
[j
],
4707 outputs
[i
].xyzw
[2].f
[j
], outputs
[i
].xyzw
[2].u
[j
],
4708 outputs
[i
].xyzw
[3].f
[j
], outputs
[i
].xyzw
[3].u
[j
]);
4717 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
4718 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
4720 * Scale back depth component.
4722 for (i
= 0; i
< 4; i
++)
4723 mach
->Outputs
[0].xyzw
[2].f
[i
] *= ctx
->DrawBuffer
->_DepthMaxF
;
4727 /* Strictly speaking, these assertions aren't really needed but they
4728 * can potentially catch some bugs in the control flow code.
4730 assert(mach
->CondStackTop
== 0);
4731 assert(mach
->LoopStackTop
== 0);
4732 assert(mach
->ContStackTop
== 0);
4733 assert(mach
->SwitchStackTop
== 0);
4734 assert(mach
->BreakStackTop
== 0);
4735 assert(mach
->CallStackTop
== 0);
4737 return ~mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];