1 /**************************************************************************
3 * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2009-2010 VMware, Inc. All rights Reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
30 * TGSI interpreter/executor.
32 * Flow control information:
34 * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
35 * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
36 * care since a condition may be true for some quad components but false
37 * for other components.
39 * We basically execute all statements (even if they're in the part of
40 * an IF/ELSE clause that's "not taken") and use a special mask to
41 * control writing to destination registers. This is the ExecMask.
44 * The ExecMask is computed from three other masks (CondMask, LoopMask and
45 * ContMask) which are controlled by the flow control instructions (namely:
46 * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
54 #include "pipe/p_compiler.h"
55 #include "pipe/p_state.h"
56 #include "pipe/p_shader_tokens.h"
57 #include "tgsi/tgsi_dump.h"
58 #include "tgsi/tgsi_parse.h"
59 #include "tgsi/tgsi_util.h"
60 #include "tgsi_exec.h"
61 #include "util/u_memory.h"
62 #include "util/u_math.h"
67 #define TILE_TOP_LEFT 0
68 #define TILE_TOP_RIGHT 1
69 #define TILE_BOTTOM_LEFT 2
70 #define TILE_BOTTOM_RIGHT 3
73 micro_abs(union tgsi_exec_channel
*dst
,
74 const union tgsi_exec_channel
*src
)
76 dst
->f
[0] = fabsf(src
->f
[0]);
77 dst
->f
[1] = fabsf(src
->f
[1]);
78 dst
->f
[2] = fabsf(src
->f
[2]);
79 dst
->f
[3] = fabsf(src
->f
[3]);
83 micro_arl(union tgsi_exec_channel
*dst
,
84 const union tgsi_exec_channel
*src
)
86 dst
->i
[0] = (int)floorf(src
->f
[0]);
87 dst
->i
[1] = (int)floorf(src
->f
[1]);
88 dst
->i
[2] = (int)floorf(src
->f
[2]);
89 dst
->i
[3] = (int)floorf(src
->f
[3]);
93 micro_arr(union tgsi_exec_channel
*dst
,
94 const union tgsi_exec_channel
*src
)
96 dst
->i
[0] = (int)floorf(src
->f
[0] + 0.5f
);
97 dst
->i
[1] = (int)floorf(src
->f
[1] + 0.5f
);
98 dst
->i
[2] = (int)floorf(src
->f
[2] + 0.5f
);
99 dst
->i
[3] = (int)floorf(src
->f
[3] + 0.5f
);
103 micro_ceil(union tgsi_exec_channel
*dst
,
104 const union tgsi_exec_channel
*src
)
106 dst
->f
[0] = ceilf(src
->f
[0]);
107 dst
->f
[1] = ceilf(src
->f
[1]);
108 dst
->f
[2] = ceilf(src
->f
[2]);
109 dst
->f
[3] = ceilf(src
->f
[3]);
113 micro_clamp(union tgsi_exec_channel
*dst
,
114 const union tgsi_exec_channel
*src0
,
115 const union tgsi_exec_channel
*src1
,
116 const union tgsi_exec_channel
*src2
)
118 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src1
->f
[0] : src0
->f
[0] > src2
->f
[0] ? src2
->f
[0] : src0
->f
[0];
119 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src1
->f
[1] : src0
->f
[1] > src2
->f
[1] ? src2
->f
[1] : src0
->f
[1];
120 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src1
->f
[2] : src0
->f
[2] > src2
->f
[2] ? src2
->f
[2] : src0
->f
[2];
121 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src1
->f
[3] : src0
->f
[3] > src2
->f
[3] ? src2
->f
[3] : src0
->f
[3];
125 micro_cmp(union tgsi_exec_channel
*dst
,
126 const union tgsi_exec_channel
*src0
,
127 const union tgsi_exec_channel
*src1
,
128 const union tgsi_exec_channel
*src2
)
130 dst
->f
[0] = src0
->f
[0] < 0.0f
? src1
->f
[0] : src2
->f
[0];
131 dst
->f
[1] = src0
->f
[1] < 0.0f
? src1
->f
[1] : src2
->f
[1];
132 dst
->f
[2] = src0
->f
[2] < 0.0f
? src1
->f
[2] : src2
->f
[2];
133 dst
->f
[3] = src0
->f
[3] < 0.0f
? src1
->f
[3] : src2
->f
[3];
137 micro_cnd(union tgsi_exec_channel
*dst
,
138 const union tgsi_exec_channel
*src0
,
139 const union tgsi_exec_channel
*src1
,
140 const union tgsi_exec_channel
*src2
)
142 dst
->f
[0] = src2
->f
[0] > 0.5f
? src0
->f
[0] : src1
->f
[0];
143 dst
->f
[1] = src2
->f
[1] > 0.5f
? src0
->f
[1] : src1
->f
[1];
144 dst
->f
[2] = src2
->f
[2] > 0.5f
? src0
->f
[2] : src1
->f
[2];
145 dst
->f
[3] = src2
->f
[3] > 0.5f
? src0
->f
[3] : src1
->f
[3];
149 micro_cos(union tgsi_exec_channel
*dst
,
150 const union tgsi_exec_channel
*src
)
152 dst
->f
[0] = cosf(src
->f
[0]);
153 dst
->f
[1] = cosf(src
->f
[1]);
154 dst
->f
[2] = cosf(src
->f
[2]);
155 dst
->f
[3] = cosf(src
->f
[3]);
159 micro_ddx(union tgsi_exec_channel
*dst
,
160 const union tgsi_exec_channel
*src
)
165 dst
->f
[3] = src
->f
[TILE_BOTTOM_RIGHT
] - src
->f
[TILE_BOTTOM_LEFT
];
169 micro_ddy(union tgsi_exec_channel
*dst
,
170 const union tgsi_exec_channel
*src
)
175 dst
->f
[3] = src
->f
[TILE_BOTTOM_LEFT
] - src
->f
[TILE_TOP_LEFT
];
179 micro_exp2(union tgsi_exec_channel
*dst
,
180 const union tgsi_exec_channel
*src
)
183 dst
->f
[0] = util_fast_exp2(src
->f
[0]);
184 dst
->f
[1] = util_fast_exp2(src
->f
[1]);
185 dst
->f
[2] = util_fast_exp2(src
->f
[2]);
186 dst
->f
[3] = util_fast_exp2(src
->f
[3]);
189 /* Inf is okay for this instruction, so clamp it to silence assertions. */
191 union tgsi_exec_channel clamped
;
193 for (i
= 0; i
< 4; i
++) {
194 if (src
->f
[i
] > 127.99999f
) {
195 clamped
.f
[i
] = 127.99999f
;
196 } else if (src
->f
[i
] < -126.99999f
) {
197 clamped
.f
[i
] = -126.99999f
;
199 clamped
.f
[i
] = src
->f
[i
];
205 dst
->f
[0] = powf(2.0f
, src
->f
[0]);
206 dst
->f
[1] = powf(2.0f
, src
->f
[1]);
207 dst
->f
[2] = powf(2.0f
, src
->f
[2]);
208 dst
->f
[3] = powf(2.0f
, src
->f
[3]);
209 #endif /* FAST_MATH */
213 micro_flr(union tgsi_exec_channel
*dst
,
214 const union tgsi_exec_channel
*src
)
216 dst
->f
[0] = floorf(src
->f
[0]);
217 dst
->f
[1] = floorf(src
->f
[1]);
218 dst
->f
[2] = floorf(src
->f
[2]);
219 dst
->f
[3] = floorf(src
->f
[3]);
223 micro_frc(union tgsi_exec_channel
*dst
,
224 const union tgsi_exec_channel
*src
)
226 dst
->f
[0] = src
->f
[0] - floorf(src
->f
[0]);
227 dst
->f
[1] = src
->f
[1] - floorf(src
->f
[1]);
228 dst
->f
[2] = src
->f
[2] - floorf(src
->f
[2]);
229 dst
->f
[3] = src
->f
[3] - floorf(src
->f
[3]);
233 micro_iabs(union tgsi_exec_channel
*dst
,
234 const union tgsi_exec_channel
*src
)
236 dst
->i
[0] = src
->i
[0] >= 0 ? src
->i
[0] : -src
->i
[0];
237 dst
->i
[1] = src
->i
[1] >= 0 ? src
->i
[1] : -src
->i
[1];
238 dst
->i
[2] = src
->i
[2] >= 0 ? src
->i
[2] : -src
->i
[2];
239 dst
->i
[3] = src
->i
[3] >= 0 ? src
->i
[3] : -src
->i
[3];
243 micro_ineg(union tgsi_exec_channel
*dst
,
244 const union tgsi_exec_channel
*src
)
246 dst
->i
[0] = -src
->i
[0];
247 dst
->i
[1] = -src
->i
[1];
248 dst
->i
[2] = -src
->i
[2];
249 dst
->i
[3] = -src
->i
[3];
253 micro_lg2(union tgsi_exec_channel
*dst
,
254 const union tgsi_exec_channel
*src
)
257 dst
->f
[0] = util_fast_log2(src
->f
[0]);
258 dst
->f
[1] = util_fast_log2(src
->f
[1]);
259 dst
->f
[2] = util_fast_log2(src
->f
[2]);
260 dst
->f
[3] = util_fast_log2(src
->f
[3]);
262 dst
->f
[0] = logf(src
->f
[0]) * 1.442695f
;
263 dst
->f
[1] = logf(src
->f
[1]) * 1.442695f
;
264 dst
->f
[2] = logf(src
->f
[2]) * 1.442695f
;
265 dst
->f
[3] = logf(src
->f
[3]) * 1.442695f
;
270 micro_lrp(union tgsi_exec_channel
*dst
,
271 const union tgsi_exec_channel
*src0
,
272 const union tgsi_exec_channel
*src1
,
273 const union tgsi_exec_channel
*src2
)
275 dst
->f
[0] = src0
->f
[0] * (src1
->f
[0] - src2
->f
[0]) + src2
->f
[0];
276 dst
->f
[1] = src0
->f
[1] * (src1
->f
[1] - src2
->f
[1]) + src2
->f
[1];
277 dst
->f
[2] = src0
->f
[2] * (src1
->f
[2] - src2
->f
[2]) + src2
->f
[2];
278 dst
->f
[3] = src0
->f
[3] * (src1
->f
[3] - src2
->f
[3]) + src2
->f
[3];
282 micro_mad(union tgsi_exec_channel
*dst
,
283 const union tgsi_exec_channel
*src0
,
284 const union tgsi_exec_channel
*src1
,
285 const union tgsi_exec_channel
*src2
)
287 dst
->f
[0] = src0
->f
[0] * src1
->f
[0] + src2
->f
[0];
288 dst
->f
[1] = src0
->f
[1] * src1
->f
[1] + src2
->f
[1];
289 dst
->f
[2] = src0
->f
[2] * src1
->f
[2] + src2
->f
[2];
290 dst
->f
[3] = src0
->f
[3] * src1
->f
[3] + src2
->f
[3];
294 micro_mov(union tgsi_exec_channel
*dst
,
295 const union tgsi_exec_channel
*src
)
297 dst
->u
[0] = src
->u
[0];
298 dst
->u
[1] = src
->u
[1];
299 dst
->u
[2] = src
->u
[2];
300 dst
->u
[3] = src
->u
[3];
304 micro_rcp(union tgsi_exec_channel
*dst
,
305 const union tgsi_exec_channel
*src
)
307 #if 0 /* for debugging */
308 assert(src
->f
[0] != 0.0f
);
309 assert(src
->f
[1] != 0.0f
);
310 assert(src
->f
[2] != 0.0f
);
311 assert(src
->f
[3] != 0.0f
);
313 dst
->f
[0] = 1.0f
/ src
->f
[0];
314 dst
->f
[1] = 1.0f
/ src
->f
[1];
315 dst
->f
[2] = 1.0f
/ src
->f
[2];
316 dst
->f
[3] = 1.0f
/ src
->f
[3];
320 micro_rnd(union tgsi_exec_channel
*dst
,
321 const union tgsi_exec_channel
*src
)
323 dst
->f
[0] = floorf(src
->f
[0] + 0.5f
);
324 dst
->f
[1] = floorf(src
->f
[1] + 0.5f
);
325 dst
->f
[2] = floorf(src
->f
[2] + 0.5f
);
326 dst
->f
[3] = floorf(src
->f
[3] + 0.5f
);
330 micro_rsq(union tgsi_exec_channel
*dst
,
331 const union tgsi_exec_channel
*src
)
333 #if 0 /* for debugging */
334 assert(src
->f
[0] != 0.0f
);
335 assert(src
->f
[1] != 0.0f
);
336 assert(src
->f
[2] != 0.0f
);
337 assert(src
->f
[3] != 0.0f
);
339 dst
->f
[0] = 1.0f
/ sqrtf(fabsf(src
->f
[0]));
340 dst
->f
[1] = 1.0f
/ sqrtf(fabsf(src
->f
[1]));
341 dst
->f
[2] = 1.0f
/ sqrtf(fabsf(src
->f
[2]));
342 dst
->f
[3] = 1.0f
/ sqrtf(fabsf(src
->f
[3]));
346 micro_seq(union tgsi_exec_channel
*dst
,
347 const union tgsi_exec_channel
*src0
,
348 const union tgsi_exec_channel
*src1
)
350 dst
->f
[0] = src0
->f
[0] == src1
->f
[0] ? 1.0f
: 0.0f
;
351 dst
->f
[1] = src0
->f
[1] == src1
->f
[1] ? 1.0f
: 0.0f
;
352 dst
->f
[2] = src0
->f
[2] == src1
->f
[2] ? 1.0f
: 0.0f
;
353 dst
->f
[3] = src0
->f
[3] == src1
->f
[3] ? 1.0f
: 0.0f
;
357 micro_sge(union tgsi_exec_channel
*dst
,
358 const union tgsi_exec_channel
*src0
,
359 const union tgsi_exec_channel
*src1
)
361 dst
->f
[0] = src0
->f
[0] >= src1
->f
[0] ? 1.0f
: 0.0f
;
362 dst
->f
[1] = src0
->f
[1] >= src1
->f
[1] ? 1.0f
: 0.0f
;
363 dst
->f
[2] = src0
->f
[2] >= src1
->f
[2] ? 1.0f
: 0.0f
;
364 dst
->f
[3] = src0
->f
[3] >= src1
->f
[3] ? 1.0f
: 0.0f
;
368 micro_sgn(union tgsi_exec_channel
*dst
,
369 const union tgsi_exec_channel
*src
)
371 dst
->f
[0] = src
->f
[0] < 0.0f
? -1.0f
: src
->f
[0] > 0.0f
? 1.0f
: 0.0f
;
372 dst
->f
[1] = src
->f
[1] < 0.0f
? -1.0f
: src
->f
[1] > 0.0f
? 1.0f
: 0.0f
;
373 dst
->f
[2] = src
->f
[2] < 0.0f
? -1.0f
: src
->f
[2] > 0.0f
? 1.0f
: 0.0f
;
374 dst
->f
[3] = src
->f
[3] < 0.0f
? -1.0f
: src
->f
[3] > 0.0f
? 1.0f
: 0.0f
;
378 micro_sgt(union tgsi_exec_channel
*dst
,
379 const union tgsi_exec_channel
*src0
,
380 const union tgsi_exec_channel
*src1
)
382 dst
->f
[0] = src0
->f
[0] > src1
->f
[0] ? 1.0f
: 0.0f
;
383 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? 1.0f
: 0.0f
;
384 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? 1.0f
: 0.0f
;
385 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? 1.0f
: 0.0f
;
389 micro_sin(union tgsi_exec_channel
*dst
,
390 const union tgsi_exec_channel
*src
)
392 dst
->f
[0] = sinf(src
->f
[0]);
393 dst
->f
[1] = sinf(src
->f
[1]);
394 dst
->f
[2] = sinf(src
->f
[2]);
395 dst
->f
[3] = sinf(src
->f
[3]);
399 micro_sle(union tgsi_exec_channel
*dst
,
400 const union tgsi_exec_channel
*src0
,
401 const union tgsi_exec_channel
*src1
)
403 dst
->f
[0] = src0
->f
[0] <= src1
->f
[0] ? 1.0f
: 0.0f
;
404 dst
->f
[1] = src0
->f
[1] <= src1
->f
[1] ? 1.0f
: 0.0f
;
405 dst
->f
[2] = src0
->f
[2] <= src1
->f
[2] ? 1.0f
: 0.0f
;
406 dst
->f
[3] = src0
->f
[3] <= src1
->f
[3] ? 1.0f
: 0.0f
;
410 micro_slt(union tgsi_exec_channel
*dst
,
411 const union tgsi_exec_channel
*src0
,
412 const union tgsi_exec_channel
*src1
)
414 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? 1.0f
: 0.0f
;
415 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? 1.0f
: 0.0f
;
416 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? 1.0f
: 0.0f
;
417 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? 1.0f
: 0.0f
;
421 micro_sne(union tgsi_exec_channel
*dst
,
422 const union tgsi_exec_channel
*src0
,
423 const union tgsi_exec_channel
*src1
)
425 dst
->f
[0] = src0
->f
[0] != src1
->f
[0] ? 1.0f
: 0.0f
;
426 dst
->f
[1] = src0
->f
[1] != src1
->f
[1] ? 1.0f
: 0.0f
;
427 dst
->f
[2] = src0
->f
[2] != src1
->f
[2] ? 1.0f
: 0.0f
;
428 dst
->f
[3] = src0
->f
[3] != src1
->f
[3] ? 1.0f
: 0.0f
;
432 micro_trunc(union tgsi_exec_channel
*dst
,
433 const union tgsi_exec_channel
*src
)
435 dst
->f
[0] = (float)(int)src
->f
[0];
436 dst
->f
[1] = (float)(int)src
->f
[1];
437 dst
->f
[2] = (float)(int)src
->f
[2];
438 dst
->f
[3] = (float)(int)src
->f
[3];
447 enum tgsi_exec_datatype
{
448 TGSI_EXEC_DATA_FLOAT
,
454 * Shorthand locations of various utility registers (_I = Index, _C = Channel)
456 #define TEMP_0_I TGSI_EXEC_TEMP_00000000_I
457 #define TEMP_0_C TGSI_EXEC_TEMP_00000000_C
458 #define TEMP_7F_I TGSI_EXEC_TEMP_7FFFFFFF_I
459 #define TEMP_7F_C TGSI_EXEC_TEMP_7FFFFFFF_C
460 #define TEMP_80_I TGSI_EXEC_TEMP_80000000_I
461 #define TEMP_80_C TGSI_EXEC_TEMP_80000000_C
462 #define TEMP_FF_I TGSI_EXEC_TEMP_FFFFFFFF_I
463 #define TEMP_FF_C TGSI_EXEC_TEMP_FFFFFFFF_C
464 #define TEMP_1_I TGSI_EXEC_TEMP_ONE_I
465 #define TEMP_1_C TGSI_EXEC_TEMP_ONE_C
466 #define TEMP_2_I TGSI_EXEC_TEMP_TWO_I
467 #define TEMP_2_C TGSI_EXEC_TEMP_TWO_C
468 #define TEMP_128_I TGSI_EXEC_TEMP_128_I
469 #define TEMP_128_C TGSI_EXEC_TEMP_128_C
470 #define TEMP_M128_I TGSI_EXEC_TEMP_MINUS_128_I
471 #define TEMP_M128_C TGSI_EXEC_TEMP_MINUS_128_C
472 #define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I
473 #define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C
474 #define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I
475 #define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C
476 #define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I
477 #define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C
478 #define TEMP_CC_I TGSI_EXEC_TEMP_CC_I
479 #define TEMP_CC_C TGSI_EXEC_TEMP_CC_C
480 #define TEMP_3_I TGSI_EXEC_TEMP_THREE_I
481 #define TEMP_3_C TGSI_EXEC_TEMP_THREE_C
482 #define TEMP_HALF_I TGSI_EXEC_TEMP_HALF_I
483 #define TEMP_HALF_C TGSI_EXEC_TEMP_HALF_C
484 #define TEMP_R0 TGSI_EXEC_TEMP_R0
485 #define TEMP_P0 TGSI_EXEC_TEMP_P0
487 #define IS_CHANNEL_ENABLED(INST, CHAN)\
488 ((INST).Dst[0].Register.WriteMask & (1 << (CHAN)))
490 #define IS_CHANNEL_ENABLED2(INST, CHAN)\
491 ((INST).Dst[1].Register.WriteMask & (1 << (CHAN)))
493 #define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\
494 for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
495 if (IS_CHANNEL_ENABLED( INST, CHAN ))
497 #define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\
498 for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
499 if (IS_CHANNEL_ENABLED2( INST, CHAN ))
502 /** The execution mask depends on the conditional mask and the loop mask */
503 #define UPDATE_EXEC_MASK(MACH) \
504 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask
507 static const union tgsi_exec_channel ZeroVec
=
508 { { 0.0, 0.0, 0.0, 0.0 } };
510 static const union tgsi_exec_channel OneVec
= {
511 {1.0f
, 1.0f
, 1.0f
, 1.0f
}
516 * Assert that none of the float values in 'chan' are infinite or NaN.
517 * NaN and Inf may occur normally during program execution and should
518 * not lead to crashes, etc. But when debugging, it's helpful to catch
522 check_inf_or_nan(const union tgsi_exec_channel
*chan
)
524 assert(!util_is_inf_or_nan((chan
)->f
[0]));
525 assert(!util_is_inf_or_nan((chan
)->f
[1]));
526 assert(!util_is_inf_or_nan((chan
)->f
[2]));
527 assert(!util_is_inf_or_nan((chan
)->f
[3]));
533 print_chan(const char *msg
, const union tgsi_exec_channel
*chan
)
535 debug_printf("%s = {%f, %f, %f, %f}\n",
536 msg
, chan
->f
[0], chan
->f
[1], chan
->f
[2], chan
->f
[3]);
543 print_temp(const struct tgsi_exec_machine
*mach
, uint index
)
545 const struct tgsi_exec_vector
*tmp
= &mach
->Temps
[index
];
547 debug_printf("Temp[%u] =\n", index
);
548 for (i
= 0; i
< 4; i
++) {
549 debug_printf(" %c: { %f, %f, %f, %f }\n",
561 * Check if there's a potential src/dst register data dependency when
562 * using SOA execution.
565 * This would expand into:
570 * The second instruction will have the wrong value for t0 if executed as-is.
573 tgsi_check_soa_dependencies(const struct tgsi_full_instruction
*inst
)
577 uint writemask
= inst
->Dst
[0].Register
.WriteMask
;
578 if (writemask
== TGSI_WRITEMASK_X
||
579 writemask
== TGSI_WRITEMASK_Y
||
580 writemask
== TGSI_WRITEMASK_Z
||
581 writemask
== TGSI_WRITEMASK_W
||
582 writemask
== TGSI_WRITEMASK_NONE
) {
583 /* no chance of data dependency */
587 /* loop over src regs */
588 for (i
= 0; i
< inst
->Instruction
.NumSrcRegs
; i
++) {
589 if ((inst
->Src
[i
].Register
.File
==
590 inst
->Dst
[0].Register
.File
) &&
591 (inst
->Src
[i
].Register
.Index
==
592 inst
->Dst
[0].Register
.Index
)) {
593 /* loop over dest channels */
594 uint channelsWritten
= 0x0;
595 FOR_EACH_ENABLED_CHANNEL(*inst
, chan
) {
596 /* check if we're reading a channel that's been written */
597 uint swizzle
= tgsi_util_get_full_src_register_swizzle(&inst
->Src
[i
], chan
);
598 if (channelsWritten
& (1 << swizzle
)) {
602 channelsWritten
|= (1 << chan
);
611 * Initialize machine state by expanding tokens to full instructions,
612 * allocating temporary storage, setting up constants, etc.
613 * After this, we can call tgsi_exec_machine_run() many times.
616 tgsi_exec_machine_bind_shader(
617 struct tgsi_exec_machine
*mach
,
618 const struct tgsi_token
*tokens
,
620 struct tgsi_sampler
**samplers
)
623 struct tgsi_parse_context parse
;
624 struct tgsi_exec_labels
*labels
= &mach
->Labels
;
625 struct tgsi_full_instruction
*instructions
;
626 struct tgsi_full_declaration
*declarations
;
627 uint maxInstructions
= 10, numInstructions
= 0;
628 uint maxDeclarations
= 10, numDeclarations
= 0;
632 tgsi_dump(tokens
, 0);
637 mach
->Tokens
= tokens
;
638 mach
->Samplers
= samplers
;
640 k
= tgsi_parse_init (&parse
, mach
->Tokens
);
641 if (k
!= TGSI_PARSE_OK
) {
642 debug_printf( "Problem parsing!\n" );
646 mach
->Processor
= parse
.FullHeader
.Processor
.Processor
;
650 declarations
= (struct tgsi_full_declaration
*)
651 MALLOC( maxDeclarations
* sizeof(struct tgsi_full_declaration
) );
657 instructions
= (struct tgsi_full_instruction
*)
658 MALLOC( maxInstructions
* sizeof(struct tgsi_full_instruction
) );
661 FREE( declarations
);
665 while( !tgsi_parse_end_of_tokens( &parse
) ) {
666 uint pointer
= parse
.Position
;
669 tgsi_parse_token( &parse
);
670 switch( parse
.FullToken
.Token
.Type
) {
671 case TGSI_TOKEN_TYPE_DECLARATION
:
672 /* save expanded declaration */
673 if (numDeclarations
== maxDeclarations
) {
674 declarations
= REALLOC(declarations
,
676 * sizeof(struct tgsi_full_declaration
),
677 (maxDeclarations
+ 10)
678 * sizeof(struct tgsi_full_declaration
));
679 maxDeclarations
+= 10;
681 if (parse
.FullToken
.FullDeclaration
.Declaration
.File
== TGSI_FILE_OUTPUT
) {
683 for (reg
= parse
.FullToken
.FullDeclaration
.Range
.First
;
684 reg
<= parse
.FullToken
.FullDeclaration
.Range
.Last
;
689 memcpy(declarations
+ numDeclarations
,
690 &parse
.FullToken
.FullDeclaration
,
691 sizeof(declarations
[0]));
695 case TGSI_TOKEN_TYPE_IMMEDIATE
:
697 uint size
= parse
.FullToken
.FullImmediate
.Immediate
.NrTokens
- 1;
699 assert( mach
->ImmLimit
+ 1 <= TGSI_EXEC_NUM_IMMEDIATES
);
701 for( i
= 0; i
< size
; i
++ ) {
702 mach
->Imms
[mach
->ImmLimit
][i
] =
703 parse
.FullToken
.FullImmediate
.u
[i
].Float
;
709 case TGSI_TOKEN_TYPE_INSTRUCTION
:
710 assert( labels
->count
< MAX_LABELS
);
712 labels
->labels
[labels
->count
][0] = instno
;
713 labels
->labels
[labels
->count
][1] = pointer
;
716 /* save expanded instruction */
717 if (numInstructions
== maxInstructions
) {
718 instructions
= REALLOC(instructions
,
720 * sizeof(struct tgsi_full_instruction
),
721 (maxInstructions
+ 10)
722 * sizeof(struct tgsi_full_instruction
));
723 maxInstructions
+= 10;
726 memcpy(instructions
+ numInstructions
,
727 &parse
.FullToken
.FullInstruction
,
728 sizeof(instructions
[0]));
733 case TGSI_TOKEN_TYPE_PROPERTY
:
740 tgsi_parse_free (&parse
);
742 if (mach
->Declarations
) {
743 FREE( mach
->Declarations
);
745 mach
->Declarations
= declarations
;
746 mach
->NumDeclarations
= numDeclarations
;
748 if (mach
->Instructions
) {
749 FREE( mach
->Instructions
);
751 mach
->Instructions
= instructions
;
752 mach
->NumInstructions
= numInstructions
;
756 struct tgsi_exec_machine
*
757 tgsi_exec_machine_create( void )
759 struct tgsi_exec_machine
*mach
;
762 mach
= align_malloc( sizeof *mach
, 16 );
766 memset(mach
, 0, sizeof(*mach
));
768 mach
->Addrs
= &mach
->Temps
[TGSI_EXEC_TEMP_ADDR
];
769 mach
->MaxGeometryShaderOutputs
= TGSI_MAX_TOTAL_VERTICES
;
770 mach
->Predicates
= &mach
->Temps
[TGSI_EXEC_TEMP_P0
];
772 /* Setup constants. */
773 for( i
= 0; i
< 4; i
++ ) {
774 mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
].u
[i
] = 0x00000000;
775 mach
->Temps
[TEMP_7F_I
].xyzw
[TEMP_7F_C
].u
[i
] = 0x7FFFFFFF;
776 mach
->Temps
[TEMP_80_I
].xyzw
[TEMP_80_C
].u
[i
] = 0x80000000;
777 mach
->Temps
[TEMP_FF_I
].xyzw
[TEMP_FF_C
].u
[i
] = 0xFFFFFFFF;
778 mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
].f
[i
] = 1.0f
;
779 mach
->Temps
[TEMP_2_I
].xyzw
[TEMP_2_C
].f
[i
] = 2.0f
;
780 mach
->Temps
[TEMP_128_I
].xyzw
[TEMP_128_C
].f
[i
] = 128.0f
;
781 mach
->Temps
[TEMP_M128_I
].xyzw
[TEMP_M128_C
].f
[i
] = -128.0f
;
782 mach
->Temps
[TEMP_3_I
].xyzw
[TEMP_3_C
].f
[i
] = 3.0f
;
783 mach
->Temps
[TEMP_HALF_I
].xyzw
[TEMP_HALF_C
].f
[i
] = 0.5f
;
787 /* silence warnings */
801 tgsi_exec_machine_destroy(struct tgsi_exec_machine
*mach
)
804 FREE(mach
->Instructions
);
805 FREE(mach
->Declarations
);
812 micro_add(union tgsi_exec_channel
*dst
,
813 const union tgsi_exec_channel
*src0
,
814 const union tgsi_exec_channel
*src1
)
816 dst
->f
[0] = src0
->f
[0] + src1
->f
[0];
817 dst
->f
[1] = src0
->f
[1] + src1
->f
[1];
818 dst
->f
[2] = src0
->f
[2] + src1
->f
[2];
819 dst
->f
[3] = src0
->f
[3] + src1
->f
[3];
824 union tgsi_exec_channel
*dst
,
825 const union tgsi_exec_channel
*src0
,
826 const union tgsi_exec_channel
*src1
)
828 if (src1
->f
[0] != 0) {
829 dst
->f
[0] = src0
->f
[0] / src1
->f
[0];
831 if (src1
->f
[1] != 0) {
832 dst
->f
[1] = src0
->f
[1] / src1
->f
[1];
834 if (src1
->f
[2] != 0) {
835 dst
->f
[2] = src0
->f
[2] / src1
->f
[2];
837 if (src1
->f
[3] != 0) {
838 dst
->f
[3] = src0
->f
[3] / src1
->f
[3];
843 micro_float_clamp(union tgsi_exec_channel
*dst
,
844 const union tgsi_exec_channel
*src
)
848 for (i
= 0; i
< 4; i
++) {
849 if (src
->f
[i
] > 0.0f
) {
850 if (src
->f
[i
] > 1.884467e+019f
)
851 dst
->f
[i
] = 1.884467e+019f
;
852 else if (src
->f
[i
] < 5.42101e-020f
)
853 dst
->f
[i
] = 5.42101e-020f
;
855 dst
->f
[i
] = src
->f
[i
];
858 if (src
->f
[i
] < -1.884467e+019f
)
859 dst
->f
[i
] = -1.884467e+019f
;
860 else if (src
->f
[i
] > -5.42101e-020f
)
861 dst
->f
[i
] = -5.42101e-020f
;
863 dst
->f
[i
] = src
->f
[i
];
870 union tgsi_exec_channel
*dst
,
871 const union tgsi_exec_channel
*src0
,
872 const union tgsi_exec_channel
*src1
,
873 const union tgsi_exec_channel
*src2
,
874 const union tgsi_exec_channel
*src3
)
876 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src2
->f
[0] : src3
->f
[0];
877 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src2
->f
[1] : src3
->f
[1];
878 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src2
->f
[2] : src3
->f
[2];
879 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src2
->f
[3] : src3
->f
[3];
883 micro_max(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] ? src0
->f
[0] : src1
->f
[0];
888 dst
->f
[1] = src0
->f
[1] > src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
889 dst
->f
[2] = src0
->f
[2] > src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
890 dst
->f
[3] = src0
->f
[3] > src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
894 micro_min(union tgsi_exec_channel
*dst
,
895 const union tgsi_exec_channel
*src0
,
896 const union tgsi_exec_channel
*src1
)
898 dst
->f
[0] = src0
->f
[0] < src1
->f
[0] ? src0
->f
[0] : src1
->f
[0];
899 dst
->f
[1] = src0
->f
[1] < src1
->f
[1] ? src0
->f
[1] : src1
->f
[1];
900 dst
->f
[2] = src0
->f
[2] < src1
->f
[2] ? src0
->f
[2] : src1
->f
[2];
901 dst
->f
[3] = src0
->f
[3] < src1
->f
[3] ? src0
->f
[3] : src1
->f
[3];
905 micro_mul(union tgsi_exec_channel
*dst
,
906 const union tgsi_exec_channel
*src0
,
907 const union tgsi_exec_channel
*src1
)
909 dst
->f
[0] = src0
->f
[0] * src1
->f
[0];
910 dst
->f
[1] = src0
->f
[1] * src1
->f
[1];
911 dst
->f
[2] = src0
->f
[2] * src1
->f
[2];
912 dst
->f
[3] = src0
->f
[3] * src1
->f
[3];
918 union tgsi_exec_channel
*dst0
,
919 union tgsi_exec_channel
*dst1
,
920 const union tgsi_exec_channel
*src0
,
921 const union tgsi_exec_channel
*src1
)
923 dst1
->i
[0] = src0
->i
[0] * src1
->i
[0];
924 dst1
->i
[1] = src0
->i
[1] * src1
->i
[1];
925 dst1
->i
[2] = src0
->i
[2] * src1
->i
[2];
926 dst1
->i
[3] = src0
->i
[3] * src1
->i
[3];
937 union tgsi_exec_channel
*dst0
,
938 union tgsi_exec_channel
*dst1
,
939 const union tgsi_exec_channel
*src0
,
940 const union tgsi_exec_channel
*src1
)
942 dst1
->u
[0] = src0
->u
[0] * src1
->u
[0];
943 dst1
->u
[1] = src0
->u
[1] * src1
->u
[1];
944 dst1
->u
[2] = src0
->u
[2] * src1
->u
[2];
945 dst1
->u
[3] = src0
->u
[3] * src1
->u
[3];
957 union tgsi_exec_channel
*dst
,
958 const union tgsi_exec_channel
*src0
,
959 const union tgsi_exec_channel
*src1
,
960 const union tgsi_exec_channel
*src2
)
962 dst
->u
[0] = src0
->u
[0] ? src1
->u
[0] : src2
->u
[0];
963 dst
->u
[1] = src0
->u
[1] ? src1
->u
[1] : src2
->u
[1];
964 dst
->u
[2] = src0
->u
[2] ? src1
->u
[2] : src2
->u
[2];
965 dst
->u
[3] = src0
->u
[3] ? src1
->u
[3] : src2
->u
[3];
971 union tgsi_exec_channel
*dst
,
972 const union tgsi_exec_channel
*src
)
974 dst
->f
[0] = -src
->f
[0];
975 dst
->f
[1] = -src
->f
[1];
976 dst
->f
[2] = -src
->f
[2];
977 dst
->f
[3] = -src
->f
[3];
982 union tgsi_exec_channel
*dst
,
983 const union tgsi_exec_channel
*src0
,
984 const union tgsi_exec_channel
*src1
)
987 dst
->f
[0] = util_fast_pow( src0
->f
[0], src1
->f
[0] );
988 dst
->f
[1] = util_fast_pow( src0
->f
[1], src1
->f
[1] );
989 dst
->f
[2] = util_fast_pow( src0
->f
[2], src1
->f
[2] );
990 dst
->f
[3] = util_fast_pow( src0
->f
[3], src1
->f
[3] );
992 dst
->f
[0] = powf( src0
->f
[0], src1
->f
[0] );
993 dst
->f
[1] = powf( src0
->f
[1], src1
->f
[1] );
994 dst
->f
[2] = powf( src0
->f
[2], src1
->f
[2] );
995 dst
->f
[3] = powf( src0
->f
[3], src1
->f
[3] );
1000 micro_sub(union tgsi_exec_channel
*dst
,
1001 const union tgsi_exec_channel
*src0
,
1002 const union tgsi_exec_channel
*src1
)
1004 dst
->f
[0] = src0
->f
[0] - src1
->f
[0];
1005 dst
->f
[1] = src0
->f
[1] - src1
->f
[1];
1006 dst
->f
[2] = src0
->f
[2] - src1
->f
[2];
1007 dst
->f
[3] = src0
->f
[3] - src1
->f
[3];
1011 fetch_src_file_channel(const struct tgsi_exec_machine
*mach
,
1014 const union tgsi_exec_channel
*index
,
1015 const union tgsi_exec_channel
*index2D
,
1016 union tgsi_exec_channel
*chan
)
1021 case TGSI_FILE_CONSTANT
:
1022 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1023 assert(index2D
->i
[i
] >= 0 && index2D
->i
[i
] < PIPE_MAX_CONSTANT_BUFFERS
);
1024 assert(mach
->Consts
[index2D
->i
[i
]]);
1026 if (index
->i
[i
] < 0) {
1029 const uint
*p
= (const uint
*)mach
->Consts
[index2D
->i
[i
]];
1031 chan
->u
[i
] = p
[index
->i
[i
] * 4 + swizzle
];
1036 case TGSI_FILE_INPUT
:
1037 case TGSI_FILE_SYSTEM_VALUE
:
1038 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1039 /* XXX: 2D indexing */
1040 chan
->u
[i
] = mach
->Inputs
[index2D
->i
[i
] * TGSI_EXEC_MAX_INPUT_ATTRIBS
+ index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1044 case TGSI_FILE_TEMPORARY
:
1045 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1046 assert(index
->i
[i
] < TGSI_EXEC_NUM_TEMPS
);
1047 assert(index2D
->i
[i
] == 0);
1049 chan
->u
[i
] = mach
->Temps
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1053 case TGSI_FILE_IMMEDIATE
:
1054 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1055 assert(index
->i
[i
] >= 0 && index
->i
[i
] < (int)mach
->ImmLimit
);
1056 assert(index2D
->i
[i
] == 0);
1058 chan
->f
[i
] = mach
->Imms
[index
->i
[i
]][swizzle
];
1062 case TGSI_FILE_ADDRESS
:
1063 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1064 assert(index
->i
[i
] >= 0);
1065 assert(index2D
->i
[i
] == 0);
1067 chan
->u
[i
] = mach
->Addrs
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1071 case TGSI_FILE_PREDICATE
:
1072 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1073 assert(index
->i
[i
] >= 0 && index
->i
[i
] < TGSI_EXEC_NUM_PREDS
);
1074 assert(index2D
->i
[i
] == 0);
1076 chan
->u
[i
] = mach
->Predicates
[0].xyzw
[swizzle
].u
[i
];
1080 case TGSI_FILE_OUTPUT
:
1081 /* vertex/fragment output vars can be read too */
1082 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1083 assert(index
->i
[i
] >= 0);
1084 assert(index2D
->i
[i
] == 0);
1086 chan
->u
[i
] = mach
->Outputs
[index
->i
[i
]].xyzw
[swizzle
].u
[i
];
1092 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1099 fetch_source(const struct tgsi_exec_machine
*mach
,
1100 union tgsi_exec_channel
*chan
,
1101 const struct tgsi_full_src_register
*reg
,
1102 const uint chan_index
,
1103 enum tgsi_exec_datatype src_datatype
)
1105 union tgsi_exec_channel index
;
1106 union tgsi_exec_channel index2D
;
1109 /* We start with a direct index into a register file.
1113 * file = Register.File
1114 * [1] = Register.Index
1119 index
.i
[3] = reg
->Register
.Index
;
1121 /* There is an extra source register that indirectly subscripts
1122 * a register file. The direct index now becomes an offset
1123 * that is being added to the indirect register.
1127 * ind = Indirect.File
1128 * [2] = Indirect.Index
1129 * .x = Indirect.SwizzleX
1131 if (reg
->Register
.Indirect
) {
1132 union tgsi_exec_channel index2
;
1133 union tgsi_exec_channel indir_index
;
1134 const uint execmask
= mach
->ExecMask
;
1137 /* which address register (always zero now) */
1141 index2
.i
[3] = reg
->Indirect
.Index
;
1143 /* get current value of address register[swizzle] */
1144 swizzle
= tgsi_util_get_src_register_swizzle( ®
->Indirect
, CHAN_X
);
1145 fetch_src_file_channel(mach
,
1152 /* add value of address register to the offset */
1153 index
.i
[0] += indir_index
.i
[0];
1154 index
.i
[1] += indir_index
.i
[1];
1155 index
.i
[2] += indir_index
.i
[2];
1156 index
.i
[3] += indir_index
.i
[3];
1158 /* for disabled execution channels, zero-out the index to
1159 * avoid using a potential garbage value.
1161 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1162 if ((execmask
& (1 << i
)) == 0)
1167 /* There is an extra source register that is a second
1168 * subscript to a register file. Effectively it means that
1169 * the register file is actually a 2D array of registers.
1173 * [3] = Dimension.Index
1175 if (reg
->Register
.Dimension
) {
1179 index2D
.i
[3] = reg
->Dimension
.Index
;
1181 /* Again, the second subscript index can be addressed indirectly
1182 * identically to the first one.
1183 * Nothing stops us from indirectly addressing the indirect register,
1184 * but there is no need for that, so we won't exercise it.
1186 * file[ind[4].y+3][1],
1188 * ind = DimIndirect.File
1189 * [4] = DimIndirect.Index
1190 * .y = DimIndirect.SwizzleX
1192 if (reg
->Dimension
.Indirect
) {
1193 union tgsi_exec_channel index2
;
1194 union tgsi_exec_channel indir_index
;
1195 const uint execmask
= mach
->ExecMask
;
1201 index2
.i
[3] = reg
->DimIndirect
.Index
;
1203 swizzle
= tgsi_util_get_src_register_swizzle( ®
->DimIndirect
, CHAN_X
);
1204 fetch_src_file_channel(mach
,
1205 reg
->DimIndirect
.File
,
1211 index2D
.i
[0] += indir_index
.i
[0];
1212 index2D
.i
[1] += indir_index
.i
[1];
1213 index2D
.i
[2] += indir_index
.i
[2];
1214 index2D
.i
[3] += indir_index
.i
[3];
1216 /* for disabled execution channels, zero-out the index to
1217 * avoid using a potential garbage value.
1219 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1220 if ((execmask
& (1 << i
)) == 0) {
1226 /* If by any chance there was a need for a 3D array of register
1227 * files, we would have to check whether Dimension is followed
1228 * by a dimension register and continue the saga.
1237 swizzle
= tgsi_util_get_full_src_register_swizzle( reg
, chan_index
);
1238 fetch_src_file_channel(mach
,
1245 if (reg
->Register
.Absolute
) {
1246 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1247 micro_abs(chan
, chan
);
1249 micro_iabs(chan
, chan
);
1253 if (reg
->Register
.Negate
) {
1254 if (src_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1255 micro_neg(chan
, chan
);
1257 micro_ineg(chan
, chan
);
1263 store_dest(struct tgsi_exec_machine
*mach
,
1264 const union tgsi_exec_channel
*chan
,
1265 const struct tgsi_full_dst_register
*reg
,
1266 const struct tgsi_full_instruction
*inst
,
1268 enum tgsi_exec_datatype dst_datatype
)
1271 union tgsi_exec_channel null
;
1272 union tgsi_exec_channel
*dst
;
1273 uint execmask
= mach
->ExecMask
;
1274 int offset
= 0; /* indirection offset */
1278 if (0 && dst_datatype
== TGSI_EXEC_DATA_FLOAT
) {
1279 check_inf_or_nan(chan
);
1282 /* There is an extra source register that indirectly subscripts
1283 * a register file. The direct index now becomes an offset
1284 * that is being added to the indirect register.
1288 * ind = Indirect.File
1289 * [2] = Indirect.Index
1290 * .x = Indirect.SwizzleX
1292 if (reg
->Register
.Indirect
) {
1293 union tgsi_exec_channel index
;
1294 union tgsi_exec_channel indir_index
;
1297 /* which address register (always zero for now) */
1301 index
.i
[3] = reg
->Indirect
.Index
;
1303 /* get current value of address register[swizzle] */
1304 swizzle
= tgsi_util_get_src_register_swizzle( ®
->Indirect
, CHAN_X
);
1306 /* fetch values from the address/indirection register */
1307 fetch_src_file_channel(mach
,
1314 /* save indirection offset */
1315 offset
= indir_index
.i
[0];
1318 switch (reg
->Register
.File
) {
1319 case TGSI_FILE_NULL
:
1323 case TGSI_FILE_OUTPUT
:
1324 index
= mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0]
1325 + reg
->Register
.Index
;
1326 dst
= &mach
->Outputs
[offset
+ index
].xyzw
[chan_index
];
1328 if (TGSI_PROCESSOR_GEOMETRY
== mach
->Processor
) {
1329 fprintf(stderr
, "STORING OUT[%d] mask(%d), = (", offset
+ index
, execmask
);
1330 for (i
= 0; i
< QUAD_SIZE
; i
++)
1331 if (execmask
& (1 << i
))
1332 fprintf(stderr
, "%f, ", chan
->f
[i
]);
1333 fprintf(stderr
, ")\n");
1338 case TGSI_FILE_TEMPORARY
:
1339 index
= reg
->Register
.Index
;
1340 assert( index
< TGSI_EXEC_NUM_TEMPS
);
1341 dst
= &mach
->Temps
[offset
+ index
].xyzw
[chan_index
];
1344 case TGSI_FILE_ADDRESS
:
1345 index
= reg
->Register
.Index
;
1346 dst
= &mach
->Addrs
[index
].xyzw
[chan_index
];
1349 case TGSI_FILE_PREDICATE
:
1350 index
= reg
->Register
.Index
;
1351 assert(index
< TGSI_EXEC_NUM_PREDS
);
1352 dst
= &mach
->Predicates
[index
].xyzw
[chan_index
];
1360 if (inst
->Instruction
.Predicate
) {
1362 union tgsi_exec_channel
*pred
;
1364 switch (chan_index
) {
1366 swizzle
= inst
->Predicate
.SwizzleX
;
1369 swizzle
= inst
->Predicate
.SwizzleY
;
1372 swizzle
= inst
->Predicate
.SwizzleZ
;
1375 swizzle
= inst
->Predicate
.SwizzleW
;
1382 assert(inst
->Predicate
.Index
== 0);
1384 pred
= &mach
->Predicates
[inst
->Predicate
.Index
].xyzw
[swizzle
];
1386 if (inst
->Predicate
.Negate
) {
1387 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1389 execmask
&= ~(1 << i
);
1393 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1395 execmask
&= ~(1 << i
);
1401 switch (inst
->Instruction
.Saturate
) {
1403 for (i
= 0; i
< QUAD_SIZE
; i
++)
1404 if (execmask
& (1 << i
))
1405 dst
->i
[i
] = chan
->i
[i
];
1408 case TGSI_SAT_ZERO_ONE
:
1409 for (i
= 0; i
< QUAD_SIZE
; i
++)
1410 if (execmask
& (1 << i
)) {
1411 if (chan
->f
[i
] < 0.0f
)
1413 else if (chan
->f
[i
] > 1.0f
)
1416 dst
->i
[i
] = chan
->i
[i
];
1420 case TGSI_SAT_MINUS_PLUS_ONE
:
1421 for (i
= 0; i
< QUAD_SIZE
; i
++)
1422 if (execmask
& (1 << i
)) {
1423 if (chan
->f
[i
] < -1.0f
)
1425 else if (chan
->f
[i
] > 1.0f
)
1428 dst
->i
[i
] = chan
->i
[i
];
1437 #define FETCH(VAL,INDEX,CHAN)\
1438 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT)
1440 #define STORE(VAL,INDEX,CHAN)\
1441 store_dest(mach, VAL, &inst->Dst[INDEX], inst, CHAN, TGSI_EXEC_DATA_FLOAT)
1445 * Execute ARB-style KIL which is predicated by a src register.
1446 * Kill fragment if any of the four values is less than zero.
1449 exec_kil(struct tgsi_exec_machine
*mach
,
1450 const struct tgsi_full_instruction
*inst
)
1454 uint kilmask
= 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1455 union tgsi_exec_channel r
[1];
1457 /* This mask stores component bits that were already tested. */
1460 for (chan_index
= 0; chan_index
< 4; chan_index
++)
1465 /* unswizzle channel */
1466 swizzle
= tgsi_util_get_full_src_register_swizzle (
1470 /* check if the component has not been already tested */
1471 if (uniquemask
& (1 << swizzle
))
1473 uniquemask
|= 1 << swizzle
;
1475 FETCH(&r
[0], 0, chan_index
);
1476 for (i
= 0; i
< 4; i
++)
1477 if (r
[0].f
[i
] < 0.0f
)
1481 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1485 * Execute NVIDIA-style KIL which is predicated by a condition code.
1486 * Kill fragment if the condition code is TRUE.
1489 exec_kilp(struct tgsi_exec_machine
*mach
,
1490 const struct tgsi_full_instruction
*inst
)
1492 uint kilmask
; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1494 /* "unconditional" kil */
1495 kilmask
= mach
->ExecMask
;
1496 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] |= kilmask
;
1500 emit_vertex(struct tgsi_exec_machine
*mach
)
1502 /* FIXME: check for exec mask correctly
1504 for (i = 0; i < QUAD_SIZE; ++i) {
1505 if ((mach->ExecMask & (1 << i)))
1507 if (mach
->ExecMask
) {
1508 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] += mach
->NumOutputs
;
1509 mach
->Primitives
[mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0]]++;
1514 emit_primitive(struct tgsi_exec_machine
*mach
)
1516 unsigned *prim_count
= &mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0];
1517 /* FIXME: check for exec mask correctly
1519 for (i = 0; i < QUAD_SIZE; ++i) {
1520 if ((mach->ExecMask & (1 << i)))
1522 if (mach
->ExecMask
) {
1524 debug_assert((*prim_count
* mach
->NumOutputs
) < mach
->MaxGeometryShaderOutputs
);
1525 mach
->Primitives
[*prim_count
] = 0;
1530 * Fetch four texture samples using STR texture coordinates.
1533 fetch_texel( struct tgsi_sampler
*sampler
,
1534 const union tgsi_exec_channel
*s
,
1535 const union tgsi_exec_channel
*t
,
1536 const union tgsi_exec_channel
*p
,
1537 const union tgsi_exec_channel
*c0
,
1538 enum tgsi_sampler_control control
,
1539 union tgsi_exec_channel
*r
,
1540 union tgsi_exec_channel
*g
,
1541 union tgsi_exec_channel
*b
,
1542 union tgsi_exec_channel
*a
)
1545 float rgba
[NUM_CHANNELS
][QUAD_SIZE
];
1547 sampler
->get_samples(sampler
, s
->f
, t
->f
, p
->f
, c0
->f
, control
, rgba
);
1549 for (j
= 0; j
< 4; j
++) {
1550 r
->f
[j
] = rgba
[0][j
];
1551 g
->f
[j
] = rgba
[1][j
];
1552 b
->f
[j
] = rgba
[2][j
];
1553 a
->f
[j
] = rgba
[3][j
];
1558 #define TEX_MODIFIER_NONE 0
1559 #define TEX_MODIFIER_PROJECTED 1
1560 #define TEX_MODIFIER_LOD_BIAS 2
1561 #define TEX_MODIFIER_EXPLICIT_LOD 3
1565 exec_tex(struct tgsi_exec_machine
*mach
,
1566 const struct tgsi_full_instruction
*inst
,
1569 const uint unit
= inst
->Src
[1].Register
.Index
;
1570 union tgsi_exec_channel r
[4];
1571 const union tgsi_exec_channel
*lod
= &ZeroVec
;
1572 enum tgsi_sampler_control control
;
1575 if (modifier
!= TEX_MODIFIER_NONE
) {
1576 FETCH(&r
[3], 0, CHAN_W
);
1577 if (modifier
!= TEX_MODIFIER_PROJECTED
) {
1582 if (modifier
== TEX_MODIFIER_EXPLICIT_LOD
) {
1583 control
= tgsi_sampler_lod_explicit
;
1585 control
= tgsi_sampler_lod_bias
;
1588 switch (inst
->Texture
.Texture
) {
1589 case TGSI_TEXTURE_1D
:
1590 case TGSI_TEXTURE_SHADOW1D
:
1591 FETCH(&r
[0], 0, CHAN_X
);
1593 if (modifier
== TEX_MODIFIER_PROJECTED
) {
1594 micro_div(&r
[0], &r
[0], &r
[3]);
1597 fetch_texel(mach
->Samplers
[unit
],
1598 &r
[0], &ZeroVec
, &ZeroVec
, lod
, /* S, T, P, LOD */
1600 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1603 case TGSI_TEXTURE_2D
:
1604 case TGSI_TEXTURE_RECT
:
1605 case TGSI_TEXTURE_SHADOW2D
:
1606 case TGSI_TEXTURE_SHADOWRECT
:
1607 FETCH(&r
[0], 0, CHAN_X
);
1608 FETCH(&r
[1], 0, CHAN_Y
);
1609 FETCH(&r
[2], 0, CHAN_Z
);
1611 if (modifier
== TEX_MODIFIER_PROJECTED
) {
1612 micro_div(&r
[0], &r
[0], &r
[3]);
1613 micro_div(&r
[1], &r
[1], &r
[3]);
1614 micro_div(&r
[2], &r
[2], &r
[3]);
1617 fetch_texel(mach
->Samplers
[unit
],
1618 &r
[0], &r
[1], &r
[2], lod
, /* S, T, P, LOD */
1620 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1623 case TGSI_TEXTURE_3D
:
1624 case TGSI_TEXTURE_CUBE
:
1625 FETCH(&r
[0], 0, CHAN_X
);
1626 FETCH(&r
[1], 0, CHAN_Y
);
1627 FETCH(&r
[2], 0, CHAN_Z
);
1629 if (modifier
== TEX_MODIFIER_PROJECTED
) {
1630 micro_div(&r
[0], &r
[0], &r
[3]);
1631 micro_div(&r
[1], &r
[1], &r
[3]);
1632 micro_div(&r
[2], &r
[2], &r
[3]);
1635 fetch_texel(mach
->Samplers
[unit
],
1636 &r
[0], &r
[1], &r
[2], lod
,
1638 &r
[0], &r
[1], &r
[2], &r
[3]);
1645 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
1646 STORE(&r
[chan_index
], 0, chan_index
);
1651 exec_txd(struct tgsi_exec_machine
*mach
,
1652 const struct tgsi_full_instruction
*inst
)
1654 const uint unit
= inst
->Src
[3].Register
.Index
;
1655 union tgsi_exec_channel r
[4];
1659 * XXX: This is fake TXD -- the derivatives are not taken into account, yet.
1662 switch (inst
->Texture
.Texture
) {
1663 case TGSI_TEXTURE_1D
:
1664 case TGSI_TEXTURE_SHADOW1D
:
1666 FETCH(&r
[0], 0, CHAN_X
);
1668 fetch_texel(mach
->Samplers
[unit
],
1669 &r
[0], &ZeroVec
, &ZeroVec
, &ZeroVec
, /* S, T, P, BIAS */
1670 tgsi_sampler_lod_bias
,
1671 &r
[0], &r
[1], &r
[2], &r
[3]); /* R, G, B, A */
1674 case TGSI_TEXTURE_2D
:
1675 case TGSI_TEXTURE_RECT
:
1676 case TGSI_TEXTURE_SHADOW2D
:
1677 case TGSI_TEXTURE_SHADOWRECT
:
1679 FETCH(&r
[0], 0, CHAN_X
);
1680 FETCH(&r
[1], 0, CHAN_Y
);
1681 FETCH(&r
[2], 0, CHAN_Z
);
1683 fetch_texel(mach
->Samplers
[unit
],
1684 &r
[0], &r
[1], &r
[2], &ZeroVec
, /* inputs */
1685 tgsi_sampler_lod_bias
,
1686 &r
[0], &r
[1], &r
[2], &r
[3]); /* outputs */
1689 case TGSI_TEXTURE_3D
:
1690 case TGSI_TEXTURE_CUBE
:
1692 FETCH(&r
[0], 0, CHAN_X
);
1693 FETCH(&r
[1], 0, CHAN_Y
);
1694 FETCH(&r
[2], 0, CHAN_Z
);
1696 fetch_texel(mach
->Samplers
[unit
],
1697 &r
[0], &r
[1], &r
[2], &ZeroVec
,
1698 tgsi_sampler_lod_bias
,
1699 &r
[0], &r
[1], &r
[2], &r
[3]);
1706 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
1707 STORE(&r
[chan_index
], 0, chan_index
);
1713 * Evaluate a constant-valued coefficient at the position of the
1718 struct tgsi_exec_machine
*mach
,
1724 for( i
= 0; i
< QUAD_SIZE
; i
++ ) {
1725 mach
->Inputs
[attrib
].xyzw
[chan
].f
[i
] = mach
->InterpCoefs
[attrib
].a0
[chan
];
1730 * Evaluate a linear-valued coefficient at the position of the
1735 struct tgsi_exec_machine
*mach
,
1739 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
1740 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
1741 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
1742 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
1743 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
1744 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
;
1745 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = a0
+ dadx
;
1746 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = a0
+ dady
;
1747 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = a0
+ dadx
+ dady
;
1751 * Evaluate a perspective-valued coefficient at the position of the
1755 eval_perspective_coef(
1756 struct tgsi_exec_machine
*mach
,
1760 const float x
= mach
->QuadPos
.xyzw
[0].f
[0];
1761 const float y
= mach
->QuadPos
.xyzw
[1].f
[0];
1762 const float dadx
= mach
->InterpCoefs
[attrib
].dadx
[chan
];
1763 const float dady
= mach
->InterpCoefs
[attrib
].dady
[chan
];
1764 const float a0
= mach
->InterpCoefs
[attrib
].a0
[chan
] + dadx
* x
+ dady
* y
;
1765 const float *w
= mach
->QuadPos
.xyzw
[3].f
;
1766 /* divide by W here */
1767 mach
->Inputs
[attrib
].xyzw
[chan
].f
[0] = a0
/ w
[0];
1768 mach
->Inputs
[attrib
].xyzw
[chan
].f
[1] = (a0
+ dadx
) / w
[1];
1769 mach
->Inputs
[attrib
].xyzw
[chan
].f
[2] = (a0
+ dady
) / w
[2];
1770 mach
->Inputs
[attrib
].xyzw
[chan
].f
[3] = (a0
+ dadx
+ dady
) / w
[3];
1774 typedef void (* eval_coef_func
)(
1775 struct tgsi_exec_machine
*mach
,
1780 exec_declaration(struct tgsi_exec_machine
*mach
,
1781 const struct tgsi_full_declaration
*decl
)
1783 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
1784 if (decl
->Declaration
.File
== TGSI_FILE_INPUT
||
1785 decl
->Declaration
.File
== TGSI_FILE_SYSTEM_VALUE
) {
1786 uint first
, last
, mask
;
1788 first
= decl
->Range
.First
;
1789 last
= decl
->Range
.Last
;
1790 mask
= decl
->Declaration
.UsageMask
;
1792 /* XXX we could remove this special-case code since
1793 * mach->InterpCoefs[first].a0 should already have the
1794 * front/back-face value. But we should first update the
1795 * ureg code to emit the right UsageMask value (WRITEMASK_X).
1796 * Then, we could remove the tgsi_exec_machine::Face field.
1798 if (decl
->Semantic
.Name
== TGSI_SEMANTIC_FACE
) {
1801 assert(decl
->Semantic
.Index
== 0);
1802 assert(first
== last
);
1804 for (i
= 0; i
< QUAD_SIZE
; i
++) {
1805 mach
->Inputs
[first
].xyzw
[0].f
[i
] = mach
->Face
;
1808 eval_coef_func eval
;
1811 switch (decl
->Declaration
.Interpolate
) {
1812 case TGSI_INTERPOLATE_CONSTANT
:
1813 eval
= eval_constant_coef
;
1816 case TGSI_INTERPOLATE_LINEAR
:
1817 eval
= eval_linear_coef
;
1820 case TGSI_INTERPOLATE_PERSPECTIVE
:
1821 eval
= eval_perspective_coef
;
1829 for (j
= 0; j
< NUM_CHANNELS
; j
++) {
1830 if (mask
& (1 << j
)) {
1831 for (i
= first
; i
<= last
; i
++) {
1841 typedef void (* micro_unary_op
)(union tgsi_exec_channel
*dst
,
1842 const union tgsi_exec_channel
*src
);
1845 exec_scalar_unary(struct tgsi_exec_machine
*mach
,
1846 const struct tgsi_full_instruction
*inst
,
1848 enum tgsi_exec_datatype dst_datatype
,
1849 enum tgsi_exec_datatype src_datatype
)
1852 union tgsi_exec_channel src
;
1853 union tgsi_exec_channel dst
;
1855 fetch_source(mach
, &src
, &inst
->Src
[0], CHAN_X
, src_datatype
);
1857 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1858 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1859 store_dest(mach
, &dst
, &inst
->Dst
[0], inst
, chan
, dst_datatype
);
1865 exec_vector_unary(struct tgsi_exec_machine
*mach
,
1866 const struct tgsi_full_instruction
*inst
,
1868 enum tgsi_exec_datatype dst_datatype
,
1869 enum tgsi_exec_datatype src_datatype
)
1872 struct tgsi_exec_vector dst
;
1874 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1875 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1876 union tgsi_exec_channel src
;
1878 fetch_source(mach
, &src
, &inst
->Src
[0], chan
, src_datatype
);
1879 op(&dst
.xyzw
[chan
], &src
);
1882 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1883 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1884 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
1889 typedef void (* micro_binary_op
)(union tgsi_exec_channel
*dst
,
1890 const union tgsi_exec_channel
*src0
,
1891 const union tgsi_exec_channel
*src1
);
1894 exec_vector_binary(struct tgsi_exec_machine
*mach
,
1895 const struct tgsi_full_instruction
*inst
,
1897 enum tgsi_exec_datatype dst_datatype
,
1898 enum tgsi_exec_datatype src_datatype
)
1901 struct tgsi_exec_vector dst
;
1903 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1904 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1905 union tgsi_exec_channel src
[2];
1907 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
1908 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
1909 op(&dst
.xyzw
[chan
], &src
[0], &src
[1]);
1912 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1913 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1914 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
1919 typedef void (* micro_trinary_op
)(union tgsi_exec_channel
*dst
,
1920 const union tgsi_exec_channel
*src0
,
1921 const union tgsi_exec_channel
*src1
,
1922 const union tgsi_exec_channel
*src2
);
1925 exec_vector_trinary(struct tgsi_exec_machine
*mach
,
1926 const struct tgsi_full_instruction
*inst
,
1927 micro_trinary_op op
,
1928 enum tgsi_exec_datatype dst_datatype
,
1929 enum tgsi_exec_datatype src_datatype
)
1932 struct tgsi_exec_vector dst
;
1934 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1935 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1936 union tgsi_exec_channel src
[3];
1938 fetch_source(mach
, &src
[0], &inst
->Src
[0], chan
, src_datatype
);
1939 fetch_source(mach
, &src
[1], &inst
->Src
[1], chan
, src_datatype
);
1940 fetch_source(mach
, &src
[2], &inst
->Src
[2], chan
, src_datatype
);
1941 op(&dst
.xyzw
[chan
], &src
[0], &src
[1], &src
[2]);
1944 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1945 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1946 store_dest(mach
, &dst
.xyzw
[chan
], &inst
->Dst
[0], inst
, chan
, dst_datatype
);
1952 exec_dp3(struct tgsi_exec_machine
*mach
,
1953 const struct tgsi_full_instruction
*inst
)
1956 union tgsi_exec_channel arg
[3];
1958 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1959 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1960 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
1962 for (chan
= CHAN_Y
; chan
<= CHAN_Z
; chan
++) {
1963 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
1964 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
1965 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
1968 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1969 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1970 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
1976 exec_dp4(struct tgsi_exec_machine
*mach
,
1977 const struct tgsi_full_instruction
*inst
)
1980 union tgsi_exec_channel arg
[3];
1982 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1983 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
1984 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
1986 for (chan
= CHAN_Y
; chan
<= CHAN_W
; chan
++) {
1987 fetch_source(mach
, &arg
[0], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
1988 fetch_source(mach
, &arg
[1], &inst
->Src
[1], chan
, TGSI_EXEC_DATA_FLOAT
);
1989 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
1992 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
1993 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
1994 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2000 exec_dp2a(struct tgsi_exec_machine
*mach
,
2001 const struct tgsi_full_instruction
*inst
)
2004 union tgsi_exec_channel arg
[3];
2006 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2007 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2008 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2010 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2011 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2012 micro_mad(&arg
[0], &arg
[0], &arg
[1], &arg
[2]);
2014 fetch_source(mach
, &arg
[1], &inst
->Src
[2], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2015 micro_add(&arg
[0], &arg
[0], &arg
[1]);
2017 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
2018 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2019 store_dest(mach
, &arg
[0], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2025 exec_dph(struct tgsi_exec_machine
*mach
,
2026 const struct tgsi_full_instruction
*inst
)
2029 union tgsi_exec_channel arg
[3];
2031 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2032 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2033 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2035 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2036 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2037 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2039 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2040 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_Z
, TGSI_EXEC_DATA_FLOAT
);
2041 micro_mad(&arg
[0], &arg
[0], &arg
[1], &arg
[2]);
2043 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2044 micro_add(&arg
[0], &arg
[0], &arg
[1]);
2046 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
2047 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2048 store_dest(mach
, &arg
[0], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2054 exec_dp2(struct tgsi_exec_machine
*mach
,
2055 const struct tgsi_full_instruction
*inst
)
2058 union tgsi_exec_channel arg
[3];
2060 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2061 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2062 micro_mul(&arg
[2], &arg
[0], &arg
[1]);
2064 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2065 fetch_source(mach
, &arg
[1], &inst
->Src
[1], CHAN_Y
, TGSI_EXEC_DATA_FLOAT
);
2066 micro_mad(&arg
[2], &arg
[0], &arg
[1], &arg
[2]);
2068 for (chan
= 0; chan
< NUM_CHANNELS
; chan
++) {
2069 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2070 store_dest(mach
, &arg
[2], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2076 exec_nrm4(struct tgsi_exec_machine
*mach
,
2077 const struct tgsi_full_instruction
*inst
)
2080 union tgsi_exec_channel arg
[4];
2081 union tgsi_exec_channel scale
;
2083 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2084 micro_mul(&scale
, &arg
[0], &arg
[0]);
2086 for (chan
= CHAN_Y
; chan
<= CHAN_W
; chan
++) {
2087 union tgsi_exec_channel product
;
2089 fetch_source(mach
, &arg
[chan
], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2090 micro_mul(&product
, &arg
[chan
], &arg
[chan
]);
2091 micro_add(&scale
, &scale
, &product
);
2094 micro_rsq(&scale
, &scale
);
2096 for (chan
= CHAN_X
; chan
<= CHAN_W
; chan
++) {
2097 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2098 micro_mul(&arg
[chan
], &arg
[chan
], &scale
);
2099 store_dest(mach
, &arg
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2105 exec_nrm3(struct tgsi_exec_machine
*mach
,
2106 const struct tgsi_full_instruction
*inst
)
2108 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_XYZ
) {
2110 union tgsi_exec_channel arg
[3];
2111 union tgsi_exec_channel scale
;
2113 fetch_source(mach
, &arg
[0], &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_FLOAT
);
2114 micro_mul(&scale
, &arg
[0], &arg
[0]);
2116 for (chan
= CHAN_Y
; chan
<= CHAN_Z
; chan
++) {
2117 union tgsi_exec_channel product
;
2119 fetch_source(mach
, &arg
[chan
], &inst
->Src
[0], chan
, TGSI_EXEC_DATA_FLOAT
);
2120 micro_mul(&product
, &arg
[chan
], &arg
[chan
]);
2121 micro_add(&scale
, &scale
, &product
);
2124 micro_rsq(&scale
, &scale
);
2126 for (chan
= CHAN_X
; chan
<= CHAN_Z
; chan
++) {
2127 if (inst
->Dst
[0].Register
.WriteMask
& (1 << chan
)) {
2128 micro_mul(&arg
[chan
], &arg
[chan
], &scale
);
2129 store_dest(mach
, &arg
[chan
], &inst
->Dst
[0], inst
, chan
, TGSI_EXEC_DATA_FLOAT
);
2134 if (inst
->Dst
[0].Register
.WriteMask
& TGSI_WRITEMASK_W
) {
2135 store_dest(mach
, &OneVec
, &inst
->Dst
[0], inst
, CHAN_W
, TGSI_EXEC_DATA_FLOAT
);
2140 exec_break(struct tgsi_exec_machine
*mach
)
2142 if (mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_LOOP
) {
2143 /* turn off loop channels for each enabled exec channel */
2144 mach
->LoopMask
&= ~mach
->ExecMask
;
2145 /* Todo: if mach->LoopMask == 0, jump to end of loop */
2146 UPDATE_EXEC_MASK(mach
);
2148 assert(mach
->BreakType
== TGSI_EXEC_BREAK_INSIDE_SWITCH
);
2150 mach
->Switch
.mask
= 0x0;
2152 UPDATE_EXEC_MASK(mach
);
2157 exec_switch(struct tgsi_exec_machine
*mach
,
2158 const struct tgsi_full_instruction
*inst
)
2160 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
2161 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
2163 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
2164 fetch_source(mach
, &mach
->Switch
.selector
, &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_UINT
);
2165 mach
->Switch
.mask
= 0x0;
2166 mach
->Switch
.defaultMask
= 0x0;
2168 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
2169 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_SWITCH
;
2171 UPDATE_EXEC_MASK(mach
);
2175 exec_case(struct tgsi_exec_machine
*mach
,
2176 const struct tgsi_full_instruction
*inst
)
2178 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
2179 union tgsi_exec_channel src
;
2182 fetch_source(mach
, &src
, &inst
->Src
[0], CHAN_X
, TGSI_EXEC_DATA_UINT
);
2184 if (mach
->Switch
.selector
.u
[0] == src
.u
[0]) {
2187 if (mach
->Switch
.selector
.u
[1] == src
.u
[1]) {
2190 if (mach
->Switch
.selector
.u
[2] == src
.u
[2]) {
2193 if (mach
->Switch
.selector
.u
[3] == src
.u
[3]) {
2197 mach
->Switch
.defaultMask
|= mask
;
2199 mach
->Switch
.mask
|= mask
& prevMask
;
2201 UPDATE_EXEC_MASK(mach
);
2205 exec_default(struct tgsi_exec_machine
*mach
)
2207 uint prevMask
= mach
->SwitchStack
[mach
->SwitchStackTop
- 1].mask
;
2209 mach
->Switch
.mask
|= ~mach
->Switch
.defaultMask
& prevMask
;
2211 UPDATE_EXEC_MASK(mach
);
2215 exec_endswitch(struct tgsi_exec_machine
*mach
)
2217 mach
->Switch
= mach
->SwitchStack
[--mach
->SwitchStackTop
];
2218 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
2220 UPDATE_EXEC_MASK(mach
);
2224 micro_i2f(union tgsi_exec_channel
*dst
,
2225 const union tgsi_exec_channel
*src
)
2227 dst
->f
[0] = (float)src
->i
[0];
2228 dst
->f
[1] = (float)src
->i
[1];
2229 dst
->f
[2] = (float)src
->i
[2];
2230 dst
->f
[3] = (float)src
->i
[3];
2234 micro_not(union tgsi_exec_channel
*dst
,
2235 const union tgsi_exec_channel
*src
)
2237 dst
->u
[0] = ~src
->u
[0];
2238 dst
->u
[1] = ~src
->u
[1];
2239 dst
->u
[2] = ~src
->u
[2];
2240 dst
->u
[3] = ~src
->u
[3];
2244 micro_shl(union tgsi_exec_channel
*dst
,
2245 const union tgsi_exec_channel
*src0
,
2246 const union tgsi_exec_channel
*src1
)
2248 dst
->u
[0] = src0
->u
[0] << src1
->u
[0];
2249 dst
->u
[1] = src0
->u
[1] << src1
->u
[1];
2250 dst
->u
[2] = src0
->u
[2] << src1
->u
[2];
2251 dst
->u
[3] = src0
->u
[3] << src1
->u
[3];
2255 micro_and(union tgsi_exec_channel
*dst
,
2256 const union tgsi_exec_channel
*src0
,
2257 const union tgsi_exec_channel
*src1
)
2259 dst
->u
[0] = src0
->u
[0] & src1
->u
[0];
2260 dst
->u
[1] = src0
->u
[1] & src1
->u
[1];
2261 dst
->u
[2] = src0
->u
[2] & src1
->u
[2];
2262 dst
->u
[3] = src0
->u
[3] & src1
->u
[3];
2266 micro_or(union tgsi_exec_channel
*dst
,
2267 const union tgsi_exec_channel
*src0
,
2268 const union tgsi_exec_channel
*src1
)
2270 dst
->u
[0] = src0
->u
[0] | src1
->u
[0];
2271 dst
->u
[1] = src0
->u
[1] | src1
->u
[1];
2272 dst
->u
[2] = src0
->u
[2] | src1
->u
[2];
2273 dst
->u
[3] = src0
->u
[3] | src1
->u
[3];
2277 micro_xor(union tgsi_exec_channel
*dst
,
2278 const union tgsi_exec_channel
*src0
,
2279 const union tgsi_exec_channel
*src1
)
2281 dst
->u
[0] = src0
->u
[0] ^ src1
->u
[0];
2282 dst
->u
[1] = src0
->u
[1] ^ src1
->u
[1];
2283 dst
->u
[2] = src0
->u
[2] ^ src1
->u
[2];
2284 dst
->u
[3] = src0
->u
[3] ^ src1
->u
[3];
2288 micro_f2i(union tgsi_exec_channel
*dst
,
2289 const union tgsi_exec_channel
*src
)
2291 dst
->i
[0] = (int)src
->f
[0];
2292 dst
->i
[1] = (int)src
->f
[1];
2293 dst
->i
[2] = (int)src
->f
[2];
2294 dst
->i
[3] = (int)src
->f
[3];
2298 micro_idiv(union tgsi_exec_channel
*dst
,
2299 const union tgsi_exec_channel
*src0
,
2300 const union tgsi_exec_channel
*src1
)
2302 dst
->i
[0] = src0
->i
[0] / src1
->i
[0];
2303 dst
->i
[1] = src0
->i
[1] / src1
->i
[1];
2304 dst
->i
[2] = src0
->i
[2] / src1
->i
[2];
2305 dst
->i
[3] = src0
->i
[3] / src1
->i
[3];
2309 micro_imax(union tgsi_exec_channel
*dst
,
2310 const union tgsi_exec_channel
*src0
,
2311 const union tgsi_exec_channel
*src1
)
2313 dst
->i
[0] = src0
->i
[0] > src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
2314 dst
->i
[1] = src0
->i
[1] > src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
2315 dst
->i
[2] = src0
->i
[2] > src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
2316 dst
->i
[3] = src0
->i
[3] > src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
2320 micro_imin(union tgsi_exec_channel
*dst
,
2321 const union tgsi_exec_channel
*src0
,
2322 const union tgsi_exec_channel
*src1
)
2324 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? src0
->i
[0] : src1
->i
[0];
2325 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? src0
->i
[1] : src1
->i
[1];
2326 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? src0
->i
[2] : src1
->i
[2];
2327 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? src0
->i
[3] : src1
->i
[3];
2331 micro_isge(union tgsi_exec_channel
*dst
,
2332 const union tgsi_exec_channel
*src0
,
2333 const union tgsi_exec_channel
*src1
)
2335 dst
->i
[0] = src0
->i
[0] >= src1
->i
[0] ? -1 : 0;
2336 dst
->i
[1] = src0
->i
[1] >= src1
->i
[1] ? -1 : 0;
2337 dst
->i
[2] = src0
->i
[2] >= src1
->i
[2] ? -1 : 0;
2338 dst
->i
[3] = src0
->i
[3] >= src1
->i
[3] ? -1 : 0;
2342 micro_ishr(union tgsi_exec_channel
*dst
,
2343 const union tgsi_exec_channel
*src0
,
2344 const union tgsi_exec_channel
*src1
)
2346 dst
->i
[0] = src0
->i
[0] >> src1
->i
[0];
2347 dst
->i
[1] = src0
->i
[1] >> src1
->i
[1];
2348 dst
->i
[2] = src0
->i
[2] >> src1
->i
[2];
2349 dst
->i
[3] = src0
->i
[3] >> src1
->i
[3];
2353 micro_islt(union tgsi_exec_channel
*dst
,
2354 const union tgsi_exec_channel
*src0
,
2355 const union tgsi_exec_channel
*src1
)
2357 dst
->i
[0] = src0
->i
[0] < src1
->i
[0] ? -1 : 0;
2358 dst
->i
[1] = src0
->i
[1] < src1
->i
[1] ? -1 : 0;
2359 dst
->i
[2] = src0
->i
[2] < src1
->i
[2] ? -1 : 0;
2360 dst
->i
[3] = src0
->i
[3] < src1
->i
[3] ? -1 : 0;
2364 micro_f2u(union tgsi_exec_channel
*dst
,
2365 const union tgsi_exec_channel
*src
)
2367 dst
->u
[0] = (uint
)src
->f
[0];
2368 dst
->u
[1] = (uint
)src
->f
[1];
2369 dst
->u
[2] = (uint
)src
->f
[2];
2370 dst
->u
[3] = (uint
)src
->f
[3];
2374 micro_u2f(union tgsi_exec_channel
*dst
,
2375 const union tgsi_exec_channel
*src
)
2377 dst
->f
[0] = (float)src
->u
[0];
2378 dst
->f
[1] = (float)src
->u
[1];
2379 dst
->f
[2] = (float)src
->u
[2];
2380 dst
->f
[3] = (float)src
->u
[3];
2384 micro_uadd(union tgsi_exec_channel
*dst
,
2385 const union tgsi_exec_channel
*src0
,
2386 const union tgsi_exec_channel
*src1
)
2388 dst
->u
[0] = src0
->u
[0] + src1
->u
[0];
2389 dst
->u
[1] = src0
->u
[1] + src1
->u
[1];
2390 dst
->u
[2] = src0
->u
[2] + src1
->u
[2];
2391 dst
->u
[3] = src0
->u
[3] + src1
->u
[3];
2395 micro_udiv(union tgsi_exec_channel
*dst
,
2396 const union tgsi_exec_channel
*src0
,
2397 const union tgsi_exec_channel
*src1
)
2399 dst
->u
[0] = src0
->u
[0] / src1
->u
[0];
2400 dst
->u
[1] = src0
->u
[1] / src1
->u
[1];
2401 dst
->u
[2] = src0
->u
[2] / src1
->u
[2];
2402 dst
->u
[3] = src0
->u
[3] / src1
->u
[3];
2406 micro_umad(union tgsi_exec_channel
*dst
,
2407 const union tgsi_exec_channel
*src0
,
2408 const union tgsi_exec_channel
*src1
,
2409 const union tgsi_exec_channel
*src2
)
2411 dst
->u
[0] = src0
->u
[0] * src1
->u
[0] + src2
->u
[0];
2412 dst
->u
[1] = src0
->u
[1] * src1
->u
[1] + src2
->u
[1];
2413 dst
->u
[2] = src0
->u
[2] * src1
->u
[2] + src2
->u
[2];
2414 dst
->u
[3] = src0
->u
[3] * src1
->u
[3] + src2
->u
[3];
2418 micro_umax(union tgsi_exec_channel
*dst
,
2419 const union tgsi_exec_channel
*src0
,
2420 const union tgsi_exec_channel
*src1
)
2422 dst
->u
[0] = src0
->u
[0] > src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
2423 dst
->u
[1] = src0
->u
[1] > src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
2424 dst
->u
[2] = src0
->u
[2] > src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
2425 dst
->u
[3] = src0
->u
[3] > src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
2429 micro_umin(union tgsi_exec_channel
*dst
,
2430 const union tgsi_exec_channel
*src0
,
2431 const union tgsi_exec_channel
*src1
)
2433 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? src0
->u
[0] : src1
->u
[0];
2434 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? src0
->u
[1] : src1
->u
[1];
2435 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? src0
->u
[2] : src1
->u
[2];
2436 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? src0
->u
[3] : src1
->u
[3];
2440 micro_umod(union tgsi_exec_channel
*dst
,
2441 const union tgsi_exec_channel
*src0
,
2442 const union tgsi_exec_channel
*src1
)
2444 dst
->u
[0] = src0
->u
[0] % src1
->u
[0];
2445 dst
->u
[1] = src0
->u
[1] % src1
->u
[1];
2446 dst
->u
[2] = src0
->u
[2] % src1
->u
[2];
2447 dst
->u
[3] = src0
->u
[3] % src1
->u
[3];
2451 micro_umul(union tgsi_exec_channel
*dst
,
2452 const union tgsi_exec_channel
*src0
,
2453 const union tgsi_exec_channel
*src1
)
2455 dst
->u
[0] = src0
->u
[0] * src1
->u
[0];
2456 dst
->u
[1] = src0
->u
[1] * src1
->u
[1];
2457 dst
->u
[2] = src0
->u
[2] * src1
->u
[2];
2458 dst
->u
[3] = src0
->u
[3] * src1
->u
[3];
2462 micro_useq(union tgsi_exec_channel
*dst
,
2463 const union tgsi_exec_channel
*src0
,
2464 const union tgsi_exec_channel
*src1
)
2466 dst
->u
[0] = src0
->u
[0] == src1
->u
[0] ? ~0 : 0;
2467 dst
->u
[1] = src0
->u
[1] == src1
->u
[1] ? ~0 : 0;
2468 dst
->u
[2] = src0
->u
[2] == src1
->u
[2] ? ~0 : 0;
2469 dst
->u
[3] = src0
->u
[3] == src1
->u
[3] ? ~0 : 0;
2473 micro_usge(union tgsi_exec_channel
*dst
,
2474 const union tgsi_exec_channel
*src0
,
2475 const union tgsi_exec_channel
*src1
)
2477 dst
->u
[0] = src0
->u
[0] >= src1
->u
[0] ? ~0 : 0;
2478 dst
->u
[1] = src0
->u
[1] >= src1
->u
[1] ? ~0 : 0;
2479 dst
->u
[2] = src0
->u
[2] >= src1
->u
[2] ? ~0 : 0;
2480 dst
->u
[3] = src0
->u
[3] >= src1
->u
[3] ? ~0 : 0;
2484 micro_ushr(union tgsi_exec_channel
*dst
,
2485 const union tgsi_exec_channel
*src0
,
2486 const union tgsi_exec_channel
*src1
)
2488 dst
->u
[0] = src0
->u
[0] >> src1
->u
[0];
2489 dst
->u
[1] = src0
->u
[1] >> src1
->u
[1];
2490 dst
->u
[2] = src0
->u
[2] >> src1
->u
[2];
2491 dst
->u
[3] = src0
->u
[3] >> src1
->u
[3];
2495 micro_uslt(union tgsi_exec_channel
*dst
,
2496 const union tgsi_exec_channel
*src0
,
2497 const union tgsi_exec_channel
*src1
)
2499 dst
->u
[0] = src0
->u
[0] < src1
->u
[0] ? ~0 : 0;
2500 dst
->u
[1] = src0
->u
[1] < src1
->u
[1] ? ~0 : 0;
2501 dst
->u
[2] = src0
->u
[2] < src1
->u
[2] ? ~0 : 0;
2502 dst
->u
[3] = src0
->u
[3] < src1
->u
[3] ? ~0 : 0;
2506 micro_usne(union tgsi_exec_channel
*dst
,
2507 const union tgsi_exec_channel
*src0
,
2508 const union tgsi_exec_channel
*src1
)
2510 dst
->u
[0] = src0
->u
[0] != src1
->u
[0] ? ~0 : 0;
2511 dst
->u
[1] = src0
->u
[1] != src1
->u
[1] ? ~0 : 0;
2512 dst
->u
[2] = src0
->u
[2] != src1
->u
[2] ? ~0 : 0;
2513 dst
->u
[3] = src0
->u
[3] != src1
->u
[3] ? ~0 : 0;
2518 struct tgsi_exec_machine
*mach
,
2519 const struct tgsi_full_instruction
*inst
,
2523 union tgsi_exec_channel r
[10];
2524 union tgsi_exec_channel d
[8];
2528 switch (inst
->Instruction
.Opcode
) {
2529 case TGSI_OPCODE_ARL
:
2530 exec_vector_unary(mach
, inst
, micro_arl
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
2533 case TGSI_OPCODE_MOV
:
2534 exec_vector_unary(mach
, inst
, micro_mov
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
2537 case TGSI_OPCODE_LIT
:
2538 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
) || IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2539 FETCH( &r
[0], 0, CHAN_X
);
2540 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2541 micro_max(&d
[CHAN_Y
], &r
[0], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
]);
2544 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2545 FETCH( &r
[1], 0, CHAN_Y
);
2546 micro_max( &r
[1], &r
[1], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
] );
2548 FETCH( &r
[2], 0, CHAN_W
);
2549 micro_min( &r
[2], &r
[2], &mach
->Temps
[TEMP_128_I
].xyzw
[TEMP_128_C
] );
2550 micro_max( &r
[2], &r
[2], &mach
->Temps
[TEMP_M128_I
].xyzw
[TEMP_M128_C
] );
2551 micro_pow( &r
[1], &r
[1], &r
[2] );
2552 micro_lt(&d
[CHAN_Z
], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], &r
[0], &r
[1], &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
]);
2555 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2556 STORE(&d
[CHAN_Y
], 0, CHAN_Y
);
2558 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2559 STORE(&d
[CHAN_Z
], 0, CHAN_Z
);
2562 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
2563 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_X
);
2565 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2566 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2570 case TGSI_OPCODE_RCP
:
2571 exec_scalar_unary(mach
, inst
, micro_rcp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2574 case TGSI_OPCODE_RSQ
:
2575 exec_scalar_unary(mach
, inst
, micro_rsq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2578 case TGSI_OPCODE_EXP
:
2579 FETCH( &r
[0], 0, CHAN_X
);
2580 micro_flr( &r
[1], &r
[0] ); /* r1 = floor(r0) */
2581 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
2582 micro_exp2( &r
[2], &r
[1] ); /* r2 = 2 ^ r1 */
2583 STORE( &r
[2], 0, CHAN_X
); /* store r2 */
2585 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2586 micro_sub( &r
[2], &r
[0], &r
[1] ); /* r2 = r0 - r1 */
2587 STORE( &r
[2], 0, CHAN_Y
); /* store r2 */
2589 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2590 micro_exp2( &r
[2], &r
[0] ); /* r2 = 2 ^ r0 */
2591 STORE( &r
[2], 0, CHAN_Z
); /* store r2 */
2593 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2594 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2598 case TGSI_OPCODE_LOG
:
2599 FETCH( &r
[0], 0, CHAN_X
);
2600 micro_abs( &r
[2], &r
[0] ); /* r2 = abs(r0) */
2601 micro_lg2( &r
[1], &r
[2] ); /* r1 = lg2(r2) */
2602 micro_flr( &r
[0], &r
[1] ); /* r0 = floor(r1) */
2603 if (IS_CHANNEL_ENABLED( *inst
, CHAN_X
)) {
2604 STORE( &r
[0], 0, CHAN_X
);
2606 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2607 micro_exp2( &r
[0], &r
[0] ); /* r0 = 2 ^ r0 */
2608 micro_div( &r
[0], &r
[2], &r
[0] ); /* r0 = r2 / r0 */
2609 STORE( &r
[0], 0, CHAN_Y
);
2611 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2612 STORE( &r
[1], 0, CHAN_Z
);
2614 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2615 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2619 case TGSI_OPCODE_MUL
:
2620 exec_vector_binary(mach
, inst
, micro_mul
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2623 case TGSI_OPCODE_ADD
:
2624 exec_vector_binary(mach
, inst
, micro_add
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2627 case TGSI_OPCODE_DP3
:
2628 exec_dp3(mach
, inst
);
2631 case TGSI_OPCODE_DP4
:
2632 exec_dp4(mach
, inst
);
2635 case TGSI_OPCODE_DST
:
2636 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Y
)) {
2637 FETCH( &r
[0], 0, CHAN_Y
);
2638 FETCH( &r
[1], 1, CHAN_Y
);
2639 micro_mul(&d
[CHAN_Y
], &r
[0], &r
[1]);
2641 if (IS_CHANNEL_ENABLED( *inst
, CHAN_Z
)) {
2642 FETCH(&d
[CHAN_Z
], 0, CHAN_Z
);
2644 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2645 FETCH(&d
[CHAN_W
], 1, CHAN_W
);
2648 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
2649 STORE(&mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_X
);
2651 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2652 STORE(&d
[CHAN_Y
], 0, CHAN_Y
);
2654 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2655 STORE(&d
[CHAN_Z
], 0, CHAN_Z
);
2657 if (IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
2658 STORE(&d
[CHAN_W
], 0, CHAN_W
);
2662 case TGSI_OPCODE_MIN
:
2663 exec_vector_binary(mach
, inst
, micro_min
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2666 case TGSI_OPCODE_MAX
:
2667 exec_vector_binary(mach
, inst
, micro_max
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2670 case TGSI_OPCODE_SLT
:
2671 exec_vector_binary(mach
, inst
, micro_slt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2674 case TGSI_OPCODE_SGE
:
2675 exec_vector_binary(mach
, inst
, micro_sge
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2678 case TGSI_OPCODE_MAD
:
2679 exec_vector_trinary(mach
, inst
, micro_mad
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2682 case TGSI_OPCODE_SUB
:
2683 exec_vector_binary(mach
, inst
, micro_sub
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2686 case TGSI_OPCODE_LRP
:
2687 exec_vector_trinary(mach
, inst
, micro_lrp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2690 case TGSI_OPCODE_CND
:
2691 exec_vector_trinary(mach
, inst
, micro_cnd
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2694 case TGSI_OPCODE_DP2A
:
2695 exec_dp2a(mach
, inst
);
2698 case TGSI_OPCODE_FRC
:
2699 exec_vector_unary(mach
, inst
, micro_frc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2702 case TGSI_OPCODE_CLAMP
:
2703 exec_vector_trinary(mach
, inst
, micro_clamp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2706 case TGSI_OPCODE_FLR
:
2707 exec_vector_unary(mach
, inst
, micro_flr
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2710 case TGSI_OPCODE_ROUND
:
2711 exec_vector_unary(mach
, inst
, micro_rnd
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2714 case TGSI_OPCODE_EX2
:
2715 exec_scalar_unary(mach
, inst
, micro_exp2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2718 case TGSI_OPCODE_LG2
:
2719 exec_scalar_unary(mach
, inst
, micro_lg2
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2722 case TGSI_OPCODE_POW
:
2723 FETCH(&r
[0], 0, CHAN_X
);
2724 FETCH(&r
[1], 1, CHAN_X
);
2726 micro_pow( &r
[0], &r
[0], &r
[1] );
2728 FOR_EACH_ENABLED_CHANNEL( *inst
, chan_index
) {
2729 STORE( &r
[0], 0, chan_index
);
2733 case TGSI_OPCODE_XPD
:
2734 FETCH(&r
[0], 0, CHAN_Y
);
2735 FETCH(&r
[1], 1, CHAN_Z
);
2737 micro_mul( &r
[2], &r
[0], &r
[1] );
2739 FETCH(&r
[3], 0, CHAN_Z
);
2740 FETCH(&r
[4], 1, CHAN_Y
);
2742 micro_mul( &r
[5], &r
[3], &r
[4] );
2743 micro_sub(&d
[CHAN_X
], &r
[2], &r
[5]);
2745 FETCH(&r
[2], 1, CHAN_X
);
2747 micro_mul( &r
[3], &r
[3], &r
[2] );
2749 FETCH(&r
[5], 0, CHAN_X
);
2751 micro_mul( &r
[1], &r
[1], &r
[5] );
2752 micro_sub(&d
[CHAN_Y
], &r
[3], &r
[1]);
2754 micro_mul( &r
[5], &r
[5], &r
[4] );
2755 micro_mul( &r
[0], &r
[0], &r
[2] );
2756 micro_sub(&d
[CHAN_Z
], &r
[5], &r
[0]);
2758 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
2759 STORE(&d
[CHAN_X
], 0, CHAN_X
);
2761 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2762 STORE(&d
[CHAN_Y
], 0, CHAN_Y
);
2764 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2765 STORE(&d
[CHAN_Z
], 0, CHAN_Z
);
2767 if (IS_CHANNEL_ENABLED( *inst
, CHAN_W
)) {
2768 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2772 case TGSI_OPCODE_ABS
:
2773 exec_vector_unary(mach
, inst
, micro_abs
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2776 case TGSI_OPCODE_RCC
:
2777 FETCH(&r
[0], 0, CHAN_X
);
2778 micro_div(&r
[0], &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], &r
[0]);
2779 micro_float_clamp(&r
[0], &r
[0]);
2780 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2781 STORE(&r
[0], 0, chan_index
);
2785 case TGSI_OPCODE_DPH
:
2786 exec_dph(mach
, inst
);
2789 case TGSI_OPCODE_COS
:
2790 exec_scalar_unary(mach
, inst
, micro_cos
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2793 case TGSI_OPCODE_DDX
:
2794 exec_vector_unary(mach
, inst
, micro_ddx
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2797 case TGSI_OPCODE_DDY
:
2798 exec_vector_unary(mach
, inst
, micro_ddy
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2801 case TGSI_OPCODE_KILP
:
2802 exec_kilp (mach
, inst
);
2805 case TGSI_OPCODE_KIL
:
2806 exec_kil (mach
, inst
);
2809 case TGSI_OPCODE_PK2H
:
2813 case TGSI_OPCODE_PK2US
:
2817 case TGSI_OPCODE_PK4B
:
2821 case TGSI_OPCODE_PK4UB
:
2825 case TGSI_OPCODE_RFL
:
2826 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
) ||
2827 IS_CHANNEL_ENABLED(*inst
, CHAN_Y
) ||
2828 IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2829 /* r0 = dp3(src0, src0) */
2830 FETCH(&r
[2], 0, CHAN_X
);
2831 micro_mul(&r
[0], &r
[2], &r
[2]);
2832 FETCH(&r
[4], 0, CHAN_Y
);
2833 micro_mul(&r
[8], &r
[4], &r
[4]);
2834 micro_add(&r
[0], &r
[0], &r
[8]);
2835 FETCH(&r
[6], 0, CHAN_Z
);
2836 micro_mul(&r
[8], &r
[6], &r
[6]);
2837 micro_add(&r
[0], &r
[0], &r
[8]);
2839 /* r1 = dp3(src0, src1) */
2840 FETCH(&r
[3], 1, CHAN_X
);
2841 micro_mul(&r
[1], &r
[2], &r
[3]);
2842 FETCH(&r
[5], 1, CHAN_Y
);
2843 micro_mul(&r
[8], &r
[4], &r
[5]);
2844 micro_add(&r
[1], &r
[1], &r
[8]);
2845 FETCH(&r
[7], 1, CHAN_Z
);
2846 micro_mul(&r
[8], &r
[6], &r
[7]);
2847 micro_add(&r
[1], &r
[1], &r
[8]);
2849 /* r1 = 2 * r1 / r0 */
2850 micro_add(&r
[1], &r
[1], &r
[1]);
2851 micro_div(&r
[1], &r
[1], &r
[0]);
2853 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
2854 micro_mul(&r
[2], &r
[2], &r
[1]);
2855 micro_sub(&r
[2], &r
[2], &r
[3]);
2856 STORE(&r
[2], 0, CHAN_X
);
2858 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2859 micro_mul(&r
[4], &r
[4], &r
[1]);
2860 micro_sub(&r
[4], &r
[4], &r
[5]);
2861 STORE(&r
[4], 0, CHAN_Y
);
2863 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2864 micro_mul(&r
[6], &r
[6], &r
[1]);
2865 micro_sub(&r
[6], &r
[6], &r
[7]);
2866 STORE(&r
[6], 0, CHAN_Z
);
2869 if (IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
2870 STORE(&mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
2874 case TGSI_OPCODE_SEQ
:
2875 exec_vector_binary(mach
, inst
, micro_seq
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2878 case TGSI_OPCODE_SFL
:
2879 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2880 STORE(&mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], 0, chan_index
);
2884 case TGSI_OPCODE_SGT
:
2885 exec_vector_binary(mach
, inst
, micro_sgt
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2888 case TGSI_OPCODE_SIN
:
2889 exec_scalar_unary(mach
, inst
, micro_sin
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2892 case TGSI_OPCODE_SLE
:
2893 exec_vector_binary(mach
, inst
, micro_sle
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2896 case TGSI_OPCODE_SNE
:
2897 exec_vector_binary(mach
, inst
, micro_sne
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
2900 case TGSI_OPCODE_STR
:
2901 FOR_EACH_ENABLED_CHANNEL(*inst
, chan_index
) {
2902 STORE(&mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, chan_index
);
2906 case TGSI_OPCODE_TEX
:
2907 /* simple texture lookup */
2908 /* src[0] = texcoord */
2909 /* src[1] = sampler unit */
2910 exec_tex(mach
, inst
, TEX_MODIFIER_NONE
);
2913 case TGSI_OPCODE_TXB
:
2914 /* Texture lookup with lod bias */
2915 /* src[0] = texcoord (src[0].w = LOD bias) */
2916 /* src[1] = sampler unit */
2917 exec_tex(mach
, inst
, TEX_MODIFIER_LOD_BIAS
);
2920 case TGSI_OPCODE_TXD
:
2921 /* Texture lookup with explict partial derivatives */
2922 /* src[0] = texcoord */
2923 /* src[1] = d[strq]/dx */
2924 /* src[2] = d[strq]/dy */
2925 /* src[3] = sampler unit */
2926 exec_txd(mach
, inst
);
2929 case TGSI_OPCODE_TXL
:
2930 /* Texture lookup with explit LOD */
2931 /* src[0] = texcoord (src[0].w = LOD) */
2932 /* src[1] = sampler unit */
2933 exec_tex(mach
, inst
, TEX_MODIFIER_EXPLICIT_LOD
);
2936 case TGSI_OPCODE_TXP
:
2937 /* Texture lookup with projection */
2938 /* src[0] = texcoord (src[0].w = projection) */
2939 /* src[1] = sampler unit */
2940 exec_tex(mach
, inst
, TEX_MODIFIER_PROJECTED
);
2943 case TGSI_OPCODE_UP2H
:
2947 case TGSI_OPCODE_UP2US
:
2951 case TGSI_OPCODE_UP4B
:
2955 case TGSI_OPCODE_UP4UB
:
2959 case TGSI_OPCODE_X2D
:
2960 FETCH(&r
[0], 1, CHAN_X
);
2961 FETCH(&r
[1], 1, CHAN_Y
);
2962 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
) ||
2963 IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2964 FETCH(&r
[2], 2, CHAN_X
);
2965 micro_mul(&r
[2], &r
[2], &r
[0]);
2966 FETCH(&r
[3], 2, CHAN_Y
);
2967 micro_mul(&r
[3], &r
[3], &r
[1]);
2968 micro_add(&r
[2], &r
[2], &r
[3]);
2969 FETCH(&r
[3], 0, CHAN_X
);
2970 micro_add(&d
[CHAN_X
], &r
[2], &r
[3]);
2973 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
) ||
2974 IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
2975 FETCH(&r
[2], 2, CHAN_Z
);
2976 micro_mul(&r
[2], &r
[2], &r
[0]);
2977 FETCH(&r
[3], 2, CHAN_W
);
2978 micro_mul(&r
[3], &r
[3], &r
[1]);
2979 micro_add(&r
[2], &r
[2], &r
[3]);
2980 FETCH(&r
[3], 0, CHAN_Y
);
2981 micro_add(&d
[CHAN_Y
], &r
[2], &r
[3]);
2984 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
2985 STORE(&d
[CHAN_X
], 0, CHAN_X
);
2987 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
2988 STORE(&d
[CHAN_Y
], 0, CHAN_Y
);
2990 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Z
)) {
2991 STORE(&d
[CHAN_X
], 0, CHAN_Z
);
2993 if (IS_CHANNEL_ENABLED(*inst
, CHAN_W
)) {
2994 STORE(&d
[CHAN_Y
], 0, CHAN_W
);
2998 case TGSI_OPCODE_ARA
:
3002 case TGSI_OPCODE_ARR
:
3003 exec_vector_unary(mach
, inst
, micro_arr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
3006 case TGSI_OPCODE_BRA
:
3010 case TGSI_OPCODE_CAL
:
3011 /* skip the call if no execution channels are enabled */
3012 if (mach
->ExecMask
) {
3015 /* First, record the depths of the execution stacks.
3016 * This is important for deeply nested/looped return statements.
3017 * We have to unwind the stacks by the correct amount. For a
3018 * real code generator, we could determine the number of entries
3019 * to pop off each stack with simple static analysis and avoid
3020 * implementing this data structure at run time.
3022 mach
->CallStack
[mach
->CallStackTop
].CondStackTop
= mach
->CondStackTop
;
3023 mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
= mach
->LoopStackTop
;
3024 mach
->CallStack
[mach
->CallStackTop
].ContStackTop
= mach
->ContStackTop
;
3025 mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
= mach
->SwitchStackTop
;
3026 mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
= mach
->BreakStackTop
;
3027 /* note that PC was already incremented above */
3028 mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
= *pc
;
3030 mach
->CallStackTop
++;
3032 /* Second, push the Cond, Loop, Cont, Func stacks */
3033 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
3034 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3035 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3036 assert(mach
->SwitchStackTop
< TGSI_EXEC_MAX_SWITCH_NESTING
);
3037 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
3038 assert(mach
->FuncStackTop
< TGSI_EXEC_MAX_CALL_NESTING
);
3040 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
3041 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
3042 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
3043 mach
->SwitchStack
[mach
->SwitchStackTop
++] = mach
->Switch
;
3044 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
3045 mach
->FuncStack
[mach
->FuncStackTop
++] = mach
->FuncMask
;
3047 /* Finally, jump to the subroutine */
3048 *pc
= inst
->Label
.Label
;
3052 case TGSI_OPCODE_RET
:
3053 mach
->FuncMask
&= ~mach
->ExecMask
;
3054 UPDATE_EXEC_MASK(mach
);
3056 if (mach
->FuncMask
== 0x0) {
3057 /* really return now (otherwise, keep executing */
3059 if (mach
->CallStackTop
== 0) {
3060 /* returning from main() */
3065 assert(mach
->CallStackTop
> 0);
3066 mach
->CallStackTop
--;
3068 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
3069 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
3071 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
3072 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
3074 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
3075 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
3077 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
3078 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
3080 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
3081 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
3083 assert(mach
->FuncStackTop
> 0);
3084 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
3086 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
3088 UPDATE_EXEC_MASK(mach
);
3092 case TGSI_OPCODE_SSG
:
3093 exec_vector_unary(mach
, inst
, micro_sgn
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3096 case TGSI_OPCODE_CMP
:
3097 exec_vector_trinary(mach
, inst
, micro_cmp
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3100 case TGSI_OPCODE_SCS
:
3101 if( IS_CHANNEL_ENABLED( *inst
, CHAN_X
) || IS_CHANNEL_ENABLED( *inst
, CHAN_Y
) ) {
3102 FETCH( &r
[0], 0, CHAN_X
);
3103 if (IS_CHANNEL_ENABLED(*inst
, CHAN_X
)) {
3104 micro_cos(&r
[1], &r
[0]);
3105 STORE(&r
[1], 0, CHAN_X
);
3107 if (IS_CHANNEL_ENABLED(*inst
, CHAN_Y
)) {
3108 micro_sin(&r
[1], &r
[0]);
3109 STORE(&r
[1], 0, CHAN_Y
);
3112 if( IS_CHANNEL_ENABLED( *inst
, CHAN_Z
) ) {
3113 STORE( &mach
->Temps
[TEMP_0_I
].xyzw
[TEMP_0_C
], 0, CHAN_Z
);
3115 if( IS_CHANNEL_ENABLED( *inst
, CHAN_W
) ) {
3116 STORE( &mach
->Temps
[TEMP_1_I
].xyzw
[TEMP_1_C
], 0, CHAN_W
);
3120 case TGSI_OPCODE_NRM
:
3121 exec_nrm3(mach
, inst
);
3124 case TGSI_OPCODE_NRM4
:
3125 exec_nrm4(mach
, inst
);
3128 case TGSI_OPCODE_DIV
:
3132 case TGSI_OPCODE_DP2
:
3133 exec_dp2(mach
, inst
);
3136 case TGSI_OPCODE_IF
:
3138 assert(mach
->CondStackTop
< TGSI_EXEC_MAX_COND_NESTING
);
3139 mach
->CondStack
[mach
->CondStackTop
++] = mach
->CondMask
;
3140 FETCH( &r
[0], 0, CHAN_X
);
3141 /* update CondMask */
3143 mach
->CondMask
&= ~0x1;
3146 mach
->CondMask
&= ~0x2;
3149 mach
->CondMask
&= ~0x4;
3152 mach
->CondMask
&= ~0x8;
3154 UPDATE_EXEC_MASK(mach
);
3155 /* Todo: If CondMask==0, jump to ELSE */
3158 case TGSI_OPCODE_ELSE
:
3159 /* invert CondMask wrt previous mask */
3162 assert(mach
->CondStackTop
> 0);
3163 prevMask
= mach
->CondStack
[mach
->CondStackTop
- 1];
3164 mach
->CondMask
= ~mach
->CondMask
& prevMask
;
3165 UPDATE_EXEC_MASK(mach
);
3166 /* Todo: If CondMask==0, jump to ENDIF */
3170 case TGSI_OPCODE_ENDIF
:
3172 assert(mach
->CondStackTop
> 0);
3173 mach
->CondMask
= mach
->CondStack
[--mach
->CondStackTop
];
3174 UPDATE_EXEC_MASK(mach
);
3177 case TGSI_OPCODE_END
:
3178 /* halt execution */
3182 case TGSI_OPCODE_PUSHA
:
3186 case TGSI_OPCODE_POPA
:
3190 case TGSI_OPCODE_CEIL
:
3191 exec_vector_unary(mach
, inst
, micro_ceil
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3194 case TGSI_OPCODE_I2F
:
3195 exec_vector_unary(mach
, inst
, micro_i2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_INT
);
3198 case TGSI_OPCODE_NOT
:
3199 exec_vector_unary(mach
, inst
, micro_not
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3202 case TGSI_OPCODE_TRUNC
:
3203 exec_vector_unary(mach
, inst
, micro_trunc
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_FLOAT
);
3206 case TGSI_OPCODE_SHL
:
3207 exec_vector_binary(mach
, inst
, micro_shl
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3210 case TGSI_OPCODE_AND
:
3211 exec_vector_binary(mach
, inst
, micro_and
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3214 case TGSI_OPCODE_OR
:
3215 exec_vector_binary(mach
, inst
, micro_or
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3218 case TGSI_OPCODE_MOD
:
3222 case TGSI_OPCODE_XOR
:
3223 exec_vector_binary(mach
, inst
, micro_xor
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3226 case TGSI_OPCODE_SAD
:
3230 case TGSI_OPCODE_TXF
:
3234 case TGSI_OPCODE_TXQ
:
3238 case TGSI_OPCODE_EMIT
:
3242 case TGSI_OPCODE_ENDPRIM
:
3243 emit_primitive(mach
);
3246 case TGSI_OPCODE_BGNLOOP
:
3247 /* push LoopMask and ContMasks */
3248 assert(mach
->LoopStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3249 assert(mach
->ContStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3250 assert(mach
->LoopLabelStackTop
< TGSI_EXEC_MAX_LOOP_NESTING
);
3251 assert(mach
->BreakStackTop
< TGSI_EXEC_MAX_BREAK_STACK
);
3253 mach
->LoopStack
[mach
->LoopStackTop
++] = mach
->LoopMask
;
3254 mach
->ContStack
[mach
->ContStackTop
++] = mach
->ContMask
;
3255 mach
->LoopLabelStack
[mach
->LoopLabelStackTop
++] = *pc
- 1;
3256 mach
->BreakStack
[mach
->BreakStackTop
++] = mach
->BreakType
;
3257 mach
->BreakType
= TGSI_EXEC_BREAK_INSIDE_LOOP
;
3260 case TGSI_OPCODE_ENDLOOP
:
3261 /* Restore ContMask, but don't pop */
3262 assert(mach
->ContStackTop
> 0);
3263 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
- 1];
3264 UPDATE_EXEC_MASK(mach
);
3265 if (mach
->ExecMask
) {
3266 /* repeat loop: jump to instruction just past BGNLOOP */
3267 assert(mach
->LoopLabelStackTop
> 0);
3268 *pc
= mach
->LoopLabelStack
[mach
->LoopLabelStackTop
- 1] + 1;
3271 /* exit loop: pop LoopMask */
3272 assert(mach
->LoopStackTop
> 0);
3273 mach
->LoopMask
= mach
->LoopStack
[--mach
->LoopStackTop
];
3275 assert(mach
->ContStackTop
> 0);
3276 mach
->ContMask
= mach
->ContStack
[--mach
->ContStackTop
];
3277 assert(mach
->LoopLabelStackTop
> 0);
3278 --mach
->LoopLabelStackTop
;
3280 mach
->BreakType
= mach
->BreakStack
[--mach
->BreakStackTop
];
3282 UPDATE_EXEC_MASK(mach
);
3285 case TGSI_OPCODE_BRK
:
3289 case TGSI_OPCODE_CONT
:
3290 /* turn off cont channels for each enabled exec channel */
3291 mach
->ContMask
&= ~mach
->ExecMask
;
3292 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3293 UPDATE_EXEC_MASK(mach
);
3296 case TGSI_OPCODE_BGNSUB
:
3300 case TGSI_OPCODE_ENDSUB
:
3302 * XXX: This really should be a no-op. We should never reach this opcode.
3305 assert(mach
->CallStackTop
> 0);
3306 mach
->CallStackTop
--;
3308 mach
->CondStackTop
= mach
->CallStack
[mach
->CallStackTop
].CondStackTop
;
3309 mach
->CondMask
= mach
->CondStack
[mach
->CondStackTop
];
3311 mach
->LoopStackTop
= mach
->CallStack
[mach
->CallStackTop
].LoopStackTop
;
3312 mach
->LoopMask
= mach
->LoopStack
[mach
->LoopStackTop
];
3314 mach
->ContStackTop
= mach
->CallStack
[mach
->CallStackTop
].ContStackTop
;
3315 mach
->ContMask
= mach
->ContStack
[mach
->ContStackTop
];
3317 mach
->SwitchStackTop
= mach
->CallStack
[mach
->CallStackTop
].SwitchStackTop
;
3318 mach
->Switch
= mach
->SwitchStack
[mach
->SwitchStackTop
];
3320 mach
->BreakStackTop
= mach
->CallStack
[mach
->CallStackTop
].BreakStackTop
;
3321 mach
->BreakType
= mach
->BreakStack
[mach
->BreakStackTop
];
3323 assert(mach
->FuncStackTop
> 0);
3324 mach
->FuncMask
= mach
->FuncStack
[--mach
->FuncStackTop
];
3326 *pc
= mach
->CallStack
[mach
->CallStackTop
].ReturnAddr
;
3328 UPDATE_EXEC_MASK(mach
);
3331 case TGSI_OPCODE_NOP
:
3334 case TGSI_OPCODE_BREAKC
:
3335 FETCH(&r
[0], 0, CHAN_X
);
3336 /* update CondMask */
3337 if (r
[0].u
[0] && (mach
->ExecMask
& 0x1)) {
3338 mach
->LoopMask
&= ~0x1;
3340 if (r
[0].u
[1] && (mach
->ExecMask
& 0x2)) {
3341 mach
->LoopMask
&= ~0x2;
3343 if (r
[0].u
[2] && (mach
->ExecMask
& 0x4)) {
3344 mach
->LoopMask
&= ~0x4;
3346 if (r
[0].u
[3] && (mach
->ExecMask
& 0x8)) {
3347 mach
->LoopMask
&= ~0x8;
3349 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3350 UPDATE_EXEC_MASK(mach
);
3353 case TGSI_OPCODE_F2I
:
3354 exec_vector_unary(mach
, inst
, micro_f2i
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_FLOAT
);
3357 case TGSI_OPCODE_IDIV
:
3358 exec_vector_binary(mach
, inst
, micro_idiv
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3361 case TGSI_OPCODE_IMAX
:
3362 exec_vector_binary(mach
, inst
, micro_imax
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3365 case TGSI_OPCODE_IMIN
:
3366 exec_vector_binary(mach
, inst
, micro_imin
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3369 case TGSI_OPCODE_INEG
:
3370 exec_vector_unary(mach
, inst
, micro_ineg
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3373 case TGSI_OPCODE_ISGE
:
3374 exec_vector_binary(mach
, inst
, micro_isge
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3377 case TGSI_OPCODE_ISHR
:
3378 exec_vector_binary(mach
, inst
, micro_ishr
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3381 case TGSI_OPCODE_ISLT
:
3382 exec_vector_binary(mach
, inst
, micro_islt
, TGSI_EXEC_DATA_INT
, TGSI_EXEC_DATA_INT
);
3385 case TGSI_OPCODE_F2U
:
3386 exec_vector_unary(mach
, inst
, micro_f2u
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_FLOAT
);
3389 case TGSI_OPCODE_U2F
:
3390 exec_vector_unary(mach
, inst
, micro_u2f
, TGSI_EXEC_DATA_FLOAT
, TGSI_EXEC_DATA_UINT
);
3393 case TGSI_OPCODE_UADD
:
3394 exec_vector_binary(mach
, inst
, micro_uadd
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3397 case TGSI_OPCODE_UDIV
:
3398 exec_vector_binary(mach
, inst
, micro_udiv
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3401 case TGSI_OPCODE_UMAD
:
3402 exec_vector_trinary(mach
, inst
, micro_umad
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3405 case TGSI_OPCODE_UMAX
:
3406 exec_vector_binary(mach
, inst
, micro_umax
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3409 case TGSI_OPCODE_UMIN
:
3410 exec_vector_binary(mach
, inst
, micro_umin
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3413 case TGSI_OPCODE_UMOD
:
3414 exec_vector_binary(mach
, inst
, micro_umod
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3417 case TGSI_OPCODE_UMUL
:
3418 exec_vector_binary(mach
, inst
, micro_umul
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3421 case TGSI_OPCODE_USEQ
:
3422 exec_vector_binary(mach
, inst
, micro_useq
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3425 case TGSI_OPCODE_USGE
:
3426 exec_vector_binary(mach
, inst
, micro_usge
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3429 case TGSI_OPCODE_USHR
:
3430 exec_vector_binary(mach
, inst
, micro_ushr
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3433 case TGSI_OPCODE_USLT
:
3434 exec_vector_binary(mach
, inst
, micro_uslt
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3437 case TGSI_OPCODE_USNE
:
3438 exec_vector_binary(mach
, inst
, micro_usne
, TGSI_EXEC_DATA_UINT
, TGSI_EXEC_DATA_UINT
);
3441 case TGSI_OPCODE_SWITCH
:
3442 exec_switch(mach
, inst
);
3445 case TGSI_OPCODE_CASE
:
3446 exec_case(mach
, inst
);
3449 case TGSI_OPCODE_DEFAULT
:
3453 case TGSI_OPCODE_ENDSWITCH
:
3454 exec_endswitch(mach
);
3463 #define DEBUG_EXECUTION 0
3467 * Run TGSI interpreter.
3468 * \return bitmask of "alive" quad components
3471 tgsi_exec_machine_run( struct tgsi_exec_machine
*mach
)
3476 mach
->CondMask
= 0xf;
3477 mach
->LoopMask
= 0xf;
3478 mach
->ContMask
= 0xf;
3479 mach
->FuncMask
= 0xf;
3480 mach
->ExecMask
= 0xf;
3482 mach
->Switch
.mask
= 0xf;
3484 assert(mach
->CondStackTop
== 0);
3485 assert(mach
->LoopStackTop
== 0);
3486 assert(mach
->ContStackTop
== 0);
3487 assert(mach
->SwitchStackTop
== 0);
3488 assert(mach
->BreakStackTop
== 0);
3489 assert(mach
->CallStackTop
== 0);
3491 mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0] = 0;
3492 mach
->Temps
[TEMP_OUTPUT_I
].xyzw
[TEMP_OUTPUT_C
].u
[0] = 0;
3494 if( mach
->Processor
== TGSI_PROCESSOR_GEOMETRY
) {
3495 mach
->Temps
[TEMP_PRIMITIVE_I
].xyzw
[TEMP_PRIMITIVE_C
].u
[0] = 0;
3496 mach
->Primitives
[0] = 0;
3499 for (i
= 0; i
< QUAD_SIZE
; i
++) {
3500 mach
->Temps
[TEMP_CC_I
].xyzw
[TEMP_CC_C
].u
[i
] =
3501 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_X_SHIFT
) |
3502 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_Y_SHIFT
) |
3503 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_Z_SHIFT
) |
3504 (TGSI_EXEC_CC_EQ
<< TGSI_EXEC_CC_W_SHIFT
);
3507 /* execute declarations (interpolants) */
3508 for (i
= 0; i
< mach
->NumDeclarations
; i
++) {
3509 exec_declaration( mach
, mach
->Declarations
+i
);
3514 struct tgsi_exec_vector temps
[TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
];
3515 struct tgsi_exec_vector outputs
[PIPE_MAX_ATTRIBS
];
3518 memcpy(temps
, mach
->Temps
, sizeof(temps
));
3519 memcpy(outputs
, mach
->Outputs
, sizeof(outputs
));
3522 /* execute instructions, until pc is set to -1 */
3528 tgsi_dump_instruction(&mach
->Instructions
[pc
], inst
++);
3531 assert(pc
< (int) mach
->NumInstructions
);
3532 exec_instruction(mach
, mach
->Instructions
+ pc
, &pc
);
3535 for (i
= 0; i
< TGSI_EXEC_NUM_TEMPS
+ TGSI_EXEC_NUM_TEMP_EXTRAS
; i
++) {
3536 if (memcmp(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]))) {
3539 memcpy(&temps
[i
], &mach
->Temps
[i
], sizeof(temps
[i
]));
3540 debug_printf("TEMP[%2u] = ", i
);
3541 for (j
= 0; j
< 4; j
++) {
3545 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
3546 temps
[i
].xyzw
[0].f
[j
], temps
[i
].xyzw
[0].u
[j
],
3547 temps
[i
].xyzw
[1].f
[j
], temps
[i
].xyzw
[1].u
[j
],
3548 temps
[i
].xyzw
[2].f
[j
], temps
[i
].xyzw
[2].u
[j
],
3549 temps
[i
].xyzw
[3].f
[j
], temps
[i
].xyzw
[3].u
[j
]);
3553 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; i
++) {
3554 if (memcmp(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]))) {
3557 memcpy(&outputs
[i
], &mach
->Outputs
[i
], sizeof(outputs
[i
]));
3558 debug_printf("OUT[%2u] = ", i
);
3559 for (j
= 0; j
< 4; j
++) {
3563 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n",
3564 outputs
[i
].xyzw
[0].f
[j
], outputs
[i
].xyzw
[0].u
[j
],
3565 outputs
[i
].xyzw
[1].f
[j
], outputs
[i
].xyzw
[1].u
[j
],
3566 outputs
[i
].xyzw
[2].f
[j
], outputs
[i
].xyzw
[2].u
[j
],
3567 outputs
[i
].xyzw
[3].f
[j
], outputs
[i
].xyzw
[3].u
[j
]);
3576 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
3577 if (mach
->Processor
== TGSI_PROCESSOR_FRAGMENT
) {
3579 * Scale back depth component.
3581 for (i
= 0; i
< 4; i
++)
3582 mach
->Outputs
[0].xyzw
[2].f
[i
] *= ctx
->DrawBuffer
->_DepthMaxF
;
3586 assert(mach
->CondStackTop
== 0);
3587 assert(mach
->LoopStackTop
== 0);
3588 assert(mach
->ContStackTop
== 0);
3589 assert(mach
->SwitchStackTop
== 0);
3590 assert(mach
->BreakStackTop
== 0);
3591 assert(mach
->CallStackTop
== 0);
3593 return ~mach
->Temps
[TEMP_KILMASK_I
].xyzw
[TEMP_KILMASK_C
].u
[0];