projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
gallium: Add UMOD TGSI opcode.
[mesa.git] / src / gallium / auxiliary / tgsi / tgsi_exec.c
1 /**************************************************************************
2 *
3 * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 /**
29 * TGSI interpreter/executor.
30 *
31 * Flow control information:
32 *
33 * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
34 * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
35 * care since a condition may be true for some quad components but false
36 * for other components.
37 *
38 * We basically execute all statements (even if they're in the part of
39 * an IF/ELSE clause that's "not taken") and use a special mask to
40 * control writing to destination registers. This is the ExecMask.
41 * See store_dest().
42 *
43 * The ExecMask is computed from three other masks (CondMask, LoopMask and
44 * ContMask) which are controlled by the flow control instructions (namely:
45 * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
46 *
47 *
48 * Authors:
49 * Michal Krol
50 * Brian Paul
51 */
52
53 #include "pipe/p_compiler.h"
54 #include "pipe/p_state.h"
55 #include "pipe/p_shader_tokens.h"
56 #include "tgsi/tgsi_dump.h"
57 #include "tgsi/tgsi_parse.h"
58 #include "tgsi/tgsi_util.h"
59 #include "tgsi_exec.h"
60 #include "util/u_memory.h"
61 #include "util/u_math.h"
62
63 #define FAST_MATH 1
64
65 #define TILE_TOP_LEFT 0
66 #define TILE_TOP_RIGHT 1
67 #define TILE_BOTTOM_LEFT 2
68 #define TILE_BOTTOM_RIGHT 3
69
70 #define CHAN_X 0
71 #define CHAN_Y 1
72 #define CHAN_Z 2
73 #define CHAN_W 3
74
75 /*
76 * Shorthand locations of various utility registers (_I = Index, _C = Channel)
77 */
78 #define TEMP_0_I TGSI_EXEC_TEMP_00000000_I
79 #define TEMP_0_C TGSI_EXEC_TEMP_00000000_C
80 #define TEMP_7F_I TGSI_EXEC_TEMP_7FFFFFFF_I
81 #define TEMP_7F_C TGSI_EXEC_TEMP_7FFFFFFF_C
82 #define TEMP_80_I TGSI_EXEC_TEMP_80000000_I
83 #define TEMP_80_C TGSI_EXEC_TEMP_80000000_C
84 #define TEMP_FF_I TGSI_EXEC_TEMP_FFFFFFFF_I
85 #define TEMP_FF_C TGSI_EXEC_TEMP_FFFFFFFF_C
86 #define TEMP_1_I TGSI_EXEC_TEMP_ONE_I
87 #define TEMP_1_C TGSI_EXEC_TEMP_ONE_C
88 #define TEMP_2_I TGSI_EXEC_TEMP_TWO_I
89 #define TEMP_2_C TGSI_EXEC_TEMP_TWO_C
90 #define TEMP_128_I TGSI_EXEC_TEMP_128_I
91 #define TEMP_128_C TGSI_EXEC_TEMP_128_C
92 #define TEMP_M128_I TGSI_EXEC_TEMP_MINUS_128_I
93 #define TEMP_M128_C TGSI_EXEC_TEMP_MINUS_128_C
94 #define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I
95 #define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C
96 #define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I
97 #define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C
98 #define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I
99 #define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C
100 #define TEMP_CC_I TGSI_EXEC_TEMP_CC_I
101 #define TEMP_CC_C TGSI_EXEC_TEMP_CC_C
102 #define TEMP_3_I TGSI_EXEC_TEMP_THREE_I
103 #define TEMP_3_C TGSI_EXEC_TEMP_THREE_C
104 #define TEMP_HALF_I TGSI_EXEC_TEMP_HALF_I
105 #define TEMP_HALF_C TGSI_EXEC_TEMP_HALF_C
106 #define TEMP_R0 TGSI_EXEC_TEMP_R0
107 #define TEMP_P0 TGSI_EXEC_TEMP_P0
108
109 #define IS_CHANNEL_ENABLED(INST, CHAN)\
110 ((INST).Dst[0].Register.WriteMask & (1 << (CHAN)))
111
112 #define IS_CHANNEL_ENABLED2(INST, CHAN)\
113 ((INST).Dst[1].Register.WriteMask & (1 << (CHAN)))
114
115 #define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\
116 for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
117 if (IS_CHANNEL_ENABLED( INST, CHAN ))
118
119 #define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\
120 for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
121 if (IS_CHANNEL_ENABLED2( INST, CHAN ))
122
123
124 /** The execution mask depends on the conditional mask and the loop mask */
125 #define UPDATE_EXEC_MASK(MACH) \
126 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask
127
128
129 static const union tgsi_exec_channel ZeroVec =
130 { { 0.0, 0.0, 0.0, 0.0 } };
131
132
133 #ifdef DEBUG
134 static void
135 check_inf_or_nan(const union tgsi_exec_channel *chan)
136 {
137 assert(!util_is_inf_or_nan(chan->f[0]));
138 assert(!util_is_inf_or_nan(chan->f[1]));
139 assert(!util_is_inf_or_nan(chan->f[2]));
140 assert(!util_is_inf_or_nan(chan->f[3]));
141 }
142 #endif
143
144
145 #ifdef DEBUG
146 static void
147 print_chan(const char *msg, const union tgsi_exec_channel *chan)
148 {
149 debug_printf("%s = {%f, %f, %f, %f}\n",
150 msg, chan->f[0], chan->f[1], chan->f[2], chan->f[3]);
151 }
152 #endif
153
154
155 #ifdef DEBUG
156 static void
157 print_temp(const struct tgsi_exec_machine *mach, uint index)
158 {
159 const struct tgsi_exec_vector *tmp = &mach->Temps[index];
160 int i;
161 debug_printf("Temp[%u] =\n", index);
162 for (i = 0; i < 4; i++) {
163 debug_printf(" %c: { %f, %f, %f, %f }\n",
164 "XYZW"[i],
165 tmp->xyzw[i].f[0],
166 tmp->xyzw[i].f[1],
167 tmp->xyzw[i].f[2],
168 tmp->xyzw[i].f[3]);
169 }
170 }
171 #endif
172
173
174 /**
175 * Check if there's a potential src/dst register data dependency when
176 * using SOA execution.
177 * Example:
178 * MOV T, T.yxwz;
179 * This would expand into:
180 * MOV t0, t1;
181 * MOV t1, t0;
182 * MOV t2, t3;
183 * MOV t3, t2;
184 * The second instruction will have the wrong value for t0 if executed as-is.
185 */
186 boolean
187 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst)
188 {
189 uint i, chan;
190
191 uint writemask = inst->Dst[0].Register.WriteMask;
192 if (writemask == TGSI_WRITEMASK_X ||
193 writemask == TGSI_WRITEMASK_Y ||
194 writemask == TGSI_WRITEMASK_Z ||
195 writemask == TGSI_WRITEMASK_W ||
196 writemask == TGSI_WRITEMASK_NONE) {
197 /* no chance of data dependency */
198 return FALSE;
199 }
200
201 /* loop over src regs */
202 for (i = 0; i < inst->Instruction.NumSrcRegs; i++) {
203 if ((inst->Src[i].Register.File ==
204 inst->Dst[0].Register.File) &&
205 (inst->Src[i].Register.Index ==
206 inst->Dst[0].Register.Index)) {
207 /* loop over dest channels */
208 uint channelsWritten = 0x0;
209 FOR_EACH_ENABLED_CHANNEL(*inst, chan) {
210 /* check if we're reading a channel that's been written */
211 uint swizzle = tgsi_util_get_full_src_register_swizzle(&inst->Src[i], chan);
212 if (channelsWritten & (1 << swizzle)) {
213 return TRUE;
214 }
215
216 channelsWritten |= (1 << chan);
217 }
218 }
219 }
220 return FALSE;
221 }
222
223
224 /**
225 * Initialize machine state by expanding tokens to full instructions,
226 * allocating temporary storage, setting up constants, etc.
227 * After this, we can call tgsi_exec_machine_run() many times.
228 */
229 void
230 tgsi_exec_machine_bind_shader(
231 struct tgsi_exec_machine *mach,
232 const struct tgsi_token *tokens,
233 uint numSamplers,
234 struct tgsi_sampler **samplers)
235 {
236 uint k;
237 struct tgsi_parse_context parse;
238 struct tgsi_exec_labels *labels = &mach->Labels;
239 struct tgsi_full_instruction *instructions;
240 struct tgsi_full_declaration *declarations;
241 uint maxInstructions = 10, numInstructions = 0;
242 uint maxDeclarations = 10, numDeclarations = 0;
243 uint instno = 0;
244
245 #if 0
246 tgsi_dump(tokens, 0);
247 #endif
248
249 util_init_math();
250
251 mach->Tokens = tokens;
252 mach->Samplers = samplers;
253
254 k = tgsi_parse_init (&parse, mach->Tokens);
255 if (k != TGSI_PARSE_OK) {
256 debug_printf( "Problem parsing!\n" );
257 return;
258 }
259
260 mach->Processor = parse.FullHeader.Processor.Processor;
261 mach->ImmLimit = 0;
262 labels->count = 0;
263
264 declarations = (struct tgsi_full_declaration *)
265 MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) );
266
267 if (!declarations) {
268 return;
269 }
270
271 instructions = (struct tgsi_full_instruction *)
272 MALLOC( maxInstructions * sizeof(struct tgsi_full_instruction) );
273
274 if (!instructions) {
275 FREE( declarations );
276 return;
277 }
278
279 while( !tgsi_parse_end_of_tokens( &parse ) ) {
280 uint pointer = parse.Position;
281 uint i;
282
283 tgsi_parse_token( &parse );
284 switch( parse.FullToken.Token.Type ) {
285 case TGSI_TOKEN_TYPE_DECLARATION:
286 /* save expanded declaration */
287 if (numDeclarations == maxDeclarations) {
288 declarations = REALLOC(declarations,
289 maxDeclarations
290 * sizeof(struct tgsi_full_declaration),
291 (maxDeclarations + 10)
292 * sizeof(struct tgsi_full_declaration));
293 maxDeclarations += 10;
294 }
295 if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_OUTPUT) {
296 unsigned reg;
297 for (reg = parse.FullToken.FullDeclaration.Range.First;
298 reg <= parse.FullToken.FullDeclaration.Range.Last;
299 ++reg) {
300 ++mach->NumOutputs;
301 }
302 }
303 memcpy(declarations + numDeclarations,
304 &parse.FullToken.FullDeclaration,
305 sizeof(declarations[0]));
306 numDeclarations++;
307 break;
308
309 case TGSI_TOKEN_TYPE_IMMEDIATE:
310 {
311 uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
312 assert( size <= 4 );
313 assert( mach->ImmLimit + 1 <= TGSI_EXEC_NUM_IMMEDIATES );
314
315 for( i = 0; i < size; i++ ) {
316 mach->Imms[mach->ImmLimit][i] =
317 parse.FullToken.FullImmediate.u[i].Float;
318 }
319 mach->ImmLimit += 1;
320 }
321 break;
322
323 case TGSI_TOKEN_TYPE_INSTRUCTION:
324 assert( labels->count < MAX_LABELS );
325
326 labels->labels[labels->count][0] = instno;
327 labels->labels[labels->count][1] = pointer;
328 labels->count++;
329
330 /* save expanded instruction */
331 if (numInstructions == maxInstructions) {
332 instructions = REALLOC(instructions,
333 maxInstructions
334 * sizeof(struct tgsi_full_instruction),
335 (maxInstructions + 10)
336 * sizeof(struct tgsi_full_instruction));
337 maxInstructions += 10;
338 }
339
340 memcpy(instructions + numInstructions,
341 &parse.FullToken.FullInstruction,
342 sizeof(instructions[0]));
343
344 numInstructions++;
345 break;
346
347 case TGSI_TOKEN_TYPE_PROPERTY:
348 break;
349
350 default:
351 assert( 0 );
352 }
353 }
354 tgsi_parse_free (&parse);
355
356 if (mach->Declarations) {
357 FREE( mach->Declarations );
358 }
359 mach->Declarations = declarations;
360 mach->NumDeclarations = numDeclarations;
361
362 if (mach->Instructions) {
363 FREE( mach->Instructions );
364 }
365 mach->Instructions = instructions;
366 mach->NumInstructions = numInstructions;
367 }
368
369
370 struct tgsi_exec_machine *
371 tgsi_exec_machine_create( void )
372 {
373 struct tgsi_exec_machine *mach;
374 uint i;
375
376 mach = align_malloc( sizeof *mach, 16 );
377 if (!mach)
378 goto fail;
379
380 memset(mach, 0, sizeof(*mach));
381
382 mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR];
383 mach->MaxGeometryShaderOutputs = TGSI_MAX_TOTAL_VERTICES;
384 mach->Predicates = &mach->Temps[TGSI_EXEC_TEMP_P0];
385
386 /* Setup constants. */
387 for( i = 0; i < 4; i++ ) {
388 mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].u[i] = 0x00000000;
389 mach->Temps[TEMP_7F_I].xyzw[TEMP_7F_C].u[i] = 0x7FFFFFFF;
390 mach->Temps[TEMP_80_I].xyzw[TEMP_80_C].u[i] = 0x80000000;
391 mach->Temps[TEMP_FF_I].xyzw[TEMP_FF_C].u[i] = 0xFFFFFFFF;
392 mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].f[i] = 1.0f;
393 mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].f[i] = 2.0f;
394 mach->Temps[TEMP_128_I].xyzw[TEMP_128_C].f[i] = 128.0f;
395 mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C].f[i] = -128.0f;
396 mach->Temps[TEMP_3_I].xyzw[TEMP_3_C].f[i] = 3.0f;
397 mach->Temps[TEMP_HALF_I].xyzw[TEMP_HALF_C].f[i] = 0.5f;
398 }
399
400 #ifdef DEBUG
401 /* silence warnings */
402 (void) print_chan;
403 (void) print_temp;
404 #endif
405
406 return mach;
407
408 fail:
409 align_free(mach);
410 return NULL;
411 }
412
413
414 void
415 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach)
416 {
417 if (mach) {
418 FREE(mach->Instructions);
419 FREE(mach->Declarations);
420 }
421
422 align_free(mach);
423 }
424
425
426 static void
427 micro_abs(
428 union tgsi_exec_channel *dst,
429 const union tgsi_exec_channel *src )
430 {
431 dst->f[0] = fabsf( src->f[0] );
432 dst->f[1] = fabsf( src->f[1] );
433 dst->f[2] = fabsf( src->f[2] );
434 dst->f[3] = fabsf( src->f[3] );
435 }
436
437 static void
438 micro_add(
439 union tgsi_exec_channel *dst,
440 const union tgsi_exec_channel *src0,
441 const union tgsi_exec_channel *src1 )
442 {
443 dst->f[0] = src0->f[0] + src1->f[0];
444 dst->f[1] = src0->f[1] + src1->f[1];
445 dst->f[2] = src0->f[2] + src1->f[2];
446 dst->f[3] = src0->f[3] + src1->f[3];
447 }
448
449 static void
450 micro_ceil(
451 union tgsi_exec_channel *dst,
452 const union tgsi_exec_channel *src )
453 {
454 dst->f[0] = ceilf( src->f[0] );
455 dst->f[1] = ceilf( src->f[1] );
456 dst->f[2] = ceilf( src->f[2] );
457 dst->f[3] = ceilf( src->f[3] );
458 }
459
460 static void
461 micro_cos(
462 union tgsi_exec_channel *dst,
463 const union tgsi_exec_channel *src )
464 {
465 dst->f[0] = cosf( src->f[0] );
466 dst->f[1] = cosf( src->f[1] );
467 dst->f[2] = cosf( src->f[2] );
468 dst->f[3] = cosf( src->f[3] );
469 }
470
471 static void
472 micro_ddx(
473 union tgsi_exec_channel *dst,
474 const union tgsi_exec_channel *src )
475 {
476 dst->f[0] =
477 dst->f[1] =
478 dst->f[2] =
479 dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_BOTTOM_LEFT];
480 }
481
482 static void
483 micro_ddy(
484 union tgsi_exec_channel *dst,
485 const union tgsi_exec_channel *src )
486 {
487 dst->f[0] =
488 dst->f[1] =
489 dst->f[2] =
490 dst->f[3] = src->f[TILE_BOTTOM_LEFT] - src->f[TILE_TOP_LEFT];
491 }
492
493 static void
494 micro_div(
495 union tgsi_exec_channel *dst,
496 const union tgsi_exec_channel *src0,
497 const union tgsi_exec_channel *src1 )
498 {
499 if (src1->f[0] != 0) {
500 dst->f[0] = src0->f[0] / src1->f[0];
501 }
502 if (src1->f[1] != 0) {
503 dst->f[1] = src0->f[1] / src1->f[1];
504 }
505 if (src1->f[2] != 0) {
506 dst->f[2] = src0->f[2] / src1->f[2];
507 }
508 if (src1->f[3] != 0) {
509 dst->f[3] = src0->f[3] / src1->f[3];
510 }
511 }
512
513 static void
514 micro_eq(
515 union tgsi_exec_channel *dst,
516 const union tgsi_exec_channel *src0,
517 const union tgsi_exec_channel *src1,
518 const union tgsi_exec_channel *src2,
519 const union tgsi_exec_channel *src3 )
520 {
521 dst->f[0] = src0->f[0] == src1->f[0] ? src2->f[0] : src3->f[0];
522 dst->f[1] = src0->f[1] == src1->f[1] ? src2->f[1] : src3->f[1];
523 dst->f[2] = src0->f[2] == src1->f[2] ? src2->f[2] : src3->f[2];
524 dst->f[3] = src0->f[3] == src1->f[3] ? src2->f[3] : src3->f[3];
525 }
526
527 static void
528 micro_exp2(
529 union tgsi_exec_channel *dst,
530 const union tgsi_exec_channel *src)
531 {
532 #if FAST_MATH
533 dst->f[0] = util_fast_exp2( src->f[0] );
534 dst->f[1] = util_fast_exp2( src->f[1] );
535 dst->f[2] = util_fast_exp2( src->f[2] );
536 dst->f[3] = util_fast_exp2( src->f[3] );
537 #else
538
539 #if DEBUG
540 /* Inf is okay for this instruction, so clamp it to silence assertions. */
541 uint i;
542 union tgsi_exec_channel clamped;
543
544 for (i = 0; i < 4; i++) {
545 if (src->f[i] > 127.99999f) {
546 clamped.f[i] = 127.99999f;
547 } else if (src->f[i] < -126.99999f) {
548 clamped.f[i] = -126.99999f;
549 } else {
550 clamped.f[i] = src->f[i];
551 }
552 }
553 src = &clamped;
554 #endif
555
556 dst->f[0] = powf( 2.0f, src->f[0] );
557 dst->f[1] = powf( 2.0f, src->f[1] );
558 dst->f[2] = powf( 2.0f, src->f[2] );
559 dst->f[3] = powf( 2.0f, src->f[3] );
560 #endif
561 }
562
563 static void
564 micro_float_clamp(union tgsi_exec_channel *dst,
565 const union tgsi_exec_channel *src)
566 {
567 uint i;
568
569 for (i = 0; i < 4; i++) {
570 if (src->f[i] > 0.0f) {
571 if (src->f[i] > 1.884467e+019f)
572 dst->f[i] = 1.884467e+019f;
573 else if (src->f[i] < 5.42101e-020f)
574 dst->f[i] = 5.42101e-020f;
575 else
576 dst->f[i] = src->f[i];
577 }
578 else {
579 if (src->f[i] < -1.884467e+019f)
580 dst->f[i] = -1.884467e+019f;
581 else if (src->f[i] > -5.42101e-020f)
582 dst->f[i] = -5.42101e-020f;
583 else
584 dst->f[i] = src->f[i];
585 }
586 }
587 }
588
589 static void
590 micro_flr(
591 union tgsi_exec_channel *dst,
592 const union tgsi_exec_channel *src )
593 {
594 dst->f[0] = floorf( src->f[0] );
595 dst->f[1] = floorf( src->f[1] );
596 dst->f[2] = floorf( src->f[2] );
597 dst->f[3] = floorf( src->f[3] );
598 }
599
600 static void
601 micro_frc(
602 union tgsi_exec_channel *dst,
603 const union tgsi_exec_channel *src )
604 {
605 dst->f[0] = src->f[0] - floorf( src->f[0] );
606 dst->f[1] = src->f[1] - floorf( src->f[1] );
607 dst->f[2] = src->f[2] - floorf( src->f[2] );
608 dst->f[3] = src->f[3] - floorf( src->f[3] );
609 }
610
611 static void
612 micro_lg2(
613 union tgsi_exec_channel *dst,
614 const union tgsi_exec_channel *src )
615 {
616 #if FAST_MATH
617 dst->f[0] = util_fast_log2( src->f[0] );
618 dst->f[1] = util_fast_log2( src->f[1] );
619 dst->f[2] = util_fast_log2( src->f[2] );
620 dst->f[3] = util_fast_log2( src->f[3] );
621 #else
622 dst->f[0] = logf( src->f[0] ) * 1.442695f;
623 dst->f[1] = logf( src->f[1] ) * 1.442695f;
624 dst->f[2] = logf( src->f[2] ) * 1.442695f;
625 dst->f[3] = logf( src->f[3] ) * 1.442695f;
626 #endif
627 }
628
629 static void
630 micro_le(
631 union tgsi_exec_channel *dst,
632 const union tgsi_exec_channel *src0,
633 const union tgsi_exec_channel *src1,
634 const union tgsi_exec_channel *src2,
635 const union tgsi_exec_channel *src3 )
636 {
637 dst->f[0] = src0->f[0] <= src1->f[0] ? src2->f[0] : src3->f[0];
638 dst->f[1] = src0->f[1] <= src1->f[1] ? src2->f[1] : src3->f[1];
639 dst->f[2] = src0->f[2] <= src1->f[2] ? src2->f[2] : src3->f[2];
640 dst->f[3] = src0->f[3] <= src1->f[3] ? src2->f[3] : src3->f[3];
641 }
642
643 static void
644 micro_lt(
645 union tgsi_exec_channel *dst,
646 const union tgsi_exec_channel *src0,
647 const union tgsi_exec_channel *src1,
648 const union tgsi_exec_channel *src2,
649 const union tgsi_exec_channel *src3 )
650 {
651 dst->f[0] = src0->f[0] < src1->f[0] ? src2->f[0] : src3->f[0];
652 dst->f[1] = src0->f[1] < src1->f[1] ? src2->f[1] : src3->f[1];
653 dst->f[2] = src0->f[2] < src1->f[2] ? src2->f[2] : src3->f[2];
654 dst->f[3] = src0->f[3] < src1->f[3] ? src2->f[3] : src3->f[3];
655 }
656
657 static void
658 micro_max(
659 union tgsi_exec_channel *dst,
660 const union tgsi_exec_channel *src0,
661 const union tgsi_exec_channel *src1 )
662 {
663 dst->f[0] = src0->f[0] > src1->f[0] ? src0->f[0] : src1->f[0];
664 dst->f[1] = src0->f[1] > src1->f[1] ? src0->f[1] : src1->f[1];
665 dst->f[2] = src0->f[2] > src1->f[2] ? src0->f[2] : src1->f[2];
666 dst->f[3] = src0->f[3] > src1->f[3] ? src0->f[3] : src1->f[3];
667 }
668
669 static void
670 micro_min(
671 union tgsi_exec_channel *dst,
672 const union tgsi_exec_channel *src0,
673 const union tgsi_exec_channel *src1 )
674 {
675 dst->f[0] = src0->f[0] < src1->f[0] ? src0->f[0] : src1->f[0];
676 dst->f[1] = src0->f[1] < src1->f[1] ? src0->f[1] : src1->f[1];
677 dst->f[2] = src0->f[2] < src1->f[2] ? src0->f[2] : src1->f[2];
678 dst->f[3] = src0->f[3] < src1->f[3] ? src0->f[3] : src1->f[3];
679 }
680
681 #if 0
682 static void
683 micro_umod(
684 union tgsi_exec_channel *dst,
685 const union tgsi_exec_channel *src0,
686 const union tgsi_exec_channel *src1 )
687 {
688 dst->u[0] = src0->u[0] % src1->u[0];
689 dst->u[1] = src0->u[1] % src1->u[1];
690 dst->u[2] = src0->u[2] % src1->u[2];
691 dst->u[3] = src0->u[3] % src1->u[3];
692 }
693 #endif
694
695 static void
696 micro_mul(
697 union tgsi_exec_channel *dst,
698 const union tgsi_exec_channel *src0,
699 const union tgsi_exec_channel *src1 )
700 {
701 dst->f[0] = src0->f[0] * src1->f[0];
702 dst->f[1] = src0->f[1] * src1->f[1];
703 dst->f[2] = src0->f[2] * src1->f[2];
704 dst->f[3] = src0->f[3] * src1->f[3];
705 }
706
707 #if 0
708 static void
709 micro_imul64(
710 union tgsi_exec_channel *dst0,
711 union tgsi_exec_channel *dst1,
712 const union tgsi_exec_channel *src0,
713 const union tgsi_exec_channel *src1 )
714 {
715 dst1->i[0] = src0->i[0] * src1->i[0];
716 dst1->i[1] = src0->i[1] * src1->i[1];
717 dst1->i[2] = src0->i[2] * src1->i[2];
718 dst1->i[3] = src0->i[3] * src1->i[3];
719 dst0->i[0] = 0;
720 dst0->i[1] = 0;
721 dst0->i[2] = 0;
722 dst0->i[3] = 0;
723 }
724 #endif
725
726 #if 0
727 static void
728 micro_umul64(
729 union tgsi_exec_channel *dst0,
730 union tgsi_exec_channel *dst1,
731 const union tgsi_exec_channel *src0,
732 const union tgsi_exec_channel *src1 )
733 {
734 dst1->u[0] = src0->u[0] * src1->u[0];
735 dst1->u[1] = src0->u[1] * src1->u[1];
736 dst1->u[2] = src0->u[2] * src1->u[2];
737 dst1->u[3] = src0->u[3] * src1->u[3];
738 dst0->u[0] = 0;
739 dst0->u[1] = 0;
740 dst0->u[2] = 0;
741 dst0->u[3] = 0;
742 }
743 #endif
744
745
746 #if 0
747 static void
748 micro_movc(
749 union tgsi_exec_channel *dst,
750 const union tgsi_exec_channel *src0,
751 const union tgsi_exec_channel *src1,
752 const union tgsi_exec_channel *src2 )
753 {
754 dst->u[0] = src0->u[0] ? src1->u[0] : src2->u[0];
755 dst->u[1] = src0->u[1] ? src1->u[1] : src2->u[1];
756 dst->u[2] = src0->u[2] ? src1->u[2] : src2->u[2];
757 dst->u[3] = src0->u[3] ? src1->u[3] : src2->u[3];
758 }
759 #endif
760
761 static void
762 micro_neg(
763 union tgsi_exec_channel *dst,
764 const union tgsi_exec_channel *src )
765 {
766 dst->f[0] = -src->f[0];
767 dst->f[1] = -src->f[1];
768 dst->f[2] = -src->f[2];
769 dst->f[3] = -src->f[3];
770 }
771
772 static void
773 micro_pow(
774 union tgsi_exec_channel *dst,
775 const union tgsi_exec_channel *src0,
776 const union tgsi_exec_channel *src1 )
777 {
778 #if FAST_MATH
779 dst->f[0] = util_fast_pow( src0->f[0], src1->f[0] );
780 dst->f[1] = util_fast_pow( src0->f[1], src1->f[1] );
781 dst->f[2] = util_fast_pow( src0->f[2], src1->f[2] );
782 dst->f[3] = util_fast_pow( src0->f[3], src1->f[3] );
783 #else
784 dst->f[0] = powf( src0->f[0], src1->f[0] );
785 dst->f[1] = powf( src0->f[1], src1->f[1] );
786 dst->f[2] = powf( src0->f[2], src1->f[2] );
787 dst->f[3] = powf( src0->f[3], src1->f[3] );
788 #endif
789 }
790
791 static void
792 micro_rnd(
793 union tgsi_exec_channel *dst,
794 const union tgsi_exec_channel *src )
795 {
796 dst->f[0] = floorf( src->f[0] + 0.5f );
797 dst->f[1] = floorf( src->f[1] + 0.5f );
798 dst->f[2] = floorf( src->f[2] + 0.5f );
799 dst->f[3] = floorf( src->f[3] + 0.5f );
800 }
801
802 static void
803 micro_sgn(
804 union tgsi_exec_channel *dst,
805 const union tgsi_exec_channel *src )
806 {
807 dst->f[0] = src->f[0] < 0.0f ? -1.0f : src->f[0] > 0.0f ? 1.0f : 0.0f;
808 dst->f[1] = src->f[1] < 0.0f ? -1.0f : src->f[1] > 0.0f ? 1.0f : 0.0f;
809 dst->f[2] = src->f[2] < 0.0f ? -1.0f : src->f[2] > 0.0f ? 1.0f : 0.0f;
810 dst->f[3] = src->f[3] < 0.0f ? -1.0f : src->f[3] > 0.0f ? 1.0f : 0.0f;
811 }
812
813 static void
814 micro_trunc(
815 union tgsi_exec_channel *dst,
816 const union tgsi_exec_channel *src0 )
817 {
818 dst->f[0] = (float) (int) src0->f[0];
819 dst->f[1] = (float) (int) src0->f[1];
820 dst->f[2] = (float) (int) src0->f[2];
821 dst->f[3] = (float) (int) src0->f[3];
822 }
823
824 static void
825 micro_sin(
826 union tgsi_exec_channel *dst,
827 const union tgsi_exec_channel *src )
828 {
829 dst->f[0] = sinf( src->f[0] );
830 dst->f[1] = sinf( src->f[1] );
831 dst->f[2] = sinf( src->f[2] );
832 dst->f[3] = sinf( src->f[3] );
833 }
834
835 static void
836 micro_sqrt( union tgsi_exec_channel *dst,
837 const union tgsi_exec_channel *src )
838 {
839 dst->f[0] = sqrtf( src->f[0] );
840 dst->f[1] = sqrtf( src->f[1] );
841 dst->f[2] = sqrtf( src->f[2] );
842 dst->f[3] = sqrtf( src->f[3] );
843 }
844
845 static void
846 micro_sub(
847 union tgsi_exec_channel *dst,
848 const union tgsi_exec_channel *src0,
849 const union tgsi_exec_channel *src1 )
850 {
851 dst->f[0] = src0->f[0] - src1->f[0];
852 dst->f[1] = src0->f[1] - src1->f[1];
853 dst->f[2] = src0->f[2] - src1->f[2];
854 dst->f[3] = src0->f[3] - src1->f[3];
855 }
856
857 static void
858 fetch_src_file_channel(
859 const struct tgsi_exec_machine *mach,
860 const uint file,
861 const uint swizzle,
862 const union tgsi_exec_channel *index,
863 union tgsi_exec_channel *chan )
864 {
865 switch( swizzle ) {
866 case TGSI_SWIZZLE_X:
867 case TGSI_SWIZZLE_Y:
868 case TGSI_SWIZZLE_Z:
869 case TGSI_SWIZZLE_W:
870 switch( file ) {
871 case TGSI_FILE_CONSTANT:
872 assert(mach->Consts);
873 if (index->i[0] < 0)
874 chan->f[0] = 0.0f;
875 else
876 chan->f[0] = mach->Consts[index->i[0]][swizzle];
877 if (index->i[1] < 0)
878 chan->f[1] = 0.0f;
879 else
880 chan->f[1] = mach->Consts[index->i[1]][swizzle];
881 if (index->i[2] < 0)
882 chan->f[2] = 0.0f;
883 else
884 chan->f[2] = mach->Consts[index->i[2]][swizzle];
885 if (index->i[3] < 0)
886 chan->f[3] = 0.0f;
887 else
888 chan->f[3] = mach->Consts[index->i[3]][swizzle];
889 break;
890
891 case TGSI_FILE_INPUT:
892 case TGSI_FILE_SYSTEM_VALUE:
893 chan->u[0] = mach->Inputs[index->i[0]].xyzw[swizzle].u[0];
894 chan->u[1] = mach->Inputs[index->i[1]].xyzw[swizzle].u[1];
895 chan->u[2] = mach->Inputs[index->i[2]].xyzw[swizzle].u[2];
896 chan->u[3] = mach->Inputs[index->i[3]].xyzw[swizzle].u[3];
897 break;
898
899 case TGSI_FILE_TEMPORARY:
900 assert(index->i[0] < TGSI_EXEC_NUM_TEMPS);
901 chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0];
902 chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1];
903 chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2];
904 chan->u[3] = mach->Temps[index->i[3]].xyzw[swizzle].u[3];
905 break;
906
907 case TGSI_FILE_IMMEDIATE:
908 assert( index->i[0] < (int) mach->ImmLimit );
909 chan->f[0] = mach->Imms[index->i[0]][swizzle];
910 assert( index->i[1] < (int) mach->ImmLimit );
911 chan->f[1] = mach->Imms[index->i[1]][swizzle];
912 assert( index->i[2] < (int) mach->ImmLimit );
913 chan->f[2] = mach->Imms[index->i[2]][swizzle];
914 assert( index->i[3] < (int) mach->ImmLimit );
915 chan->f[3] = mach->Imms[index->i[3]][swizzle];
916 break;
917
918 case TGSI_FILE_ADDRESS:
919 chan->u[0] = mach->Addrs[index->i[0]].xyzw[swizzle].u[0];
920 chan->u[1] = mach->Addrs[index->i[1]].xyzw[swizzle].u[1];
921 chan->u[2] = mach->Addrs[index->i[2]].xyzw[swizzle].u[2];
922 chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3];
923 break;
924
925 case TGSI_FILE_PREDICATE:
926 assert(index->i[0] < TGSI_EXEC_NUM_PREDS);
927 assert(index->i[1] < TGSI_EXEC_NUM_PREDS);
928 assert(index->i[2] < TGSI_EXEC_NUM_PREDS);
929 assert(index->i[3] < TGSI_EXEC_NUM_PREDS);
930 chan->u[0] = mach->Predicates[0].xyzw[swizzle].u[0];
931 chan->u[1] = mach->Predicates[0].xyzw[swizzle].u[1];
932 chan->u[2] = mach->Predicates[0].xyzw[swizzle].u[2];
933 chan->u[3] = mach->Predicates[0].xyzw[swizzle].u[3];
934 break;
935
936 case TGSI_FILE_OUTPUT:
937 /* vertex/fragment output vars can be read too */
938 chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0];
939 chan->u[1] = mach->Outputs[index->i[1]].xyzw[swizzle].u[1];
940 chan->u[2] = mach->Outputs[index->i[2]].xyzw[swizzle].u[2];
941 chan->u[3] = mach->Outputs[index->i[3]].xyzw[swizzle].u[3];
942 break;
943
944 default:
945 assert( 0 );
946 }
947 break;
948
949 default:
950 assert( 0 );
951 }
952 }
953
954 static void
955 fetch_source(
956 const struct tgsi_exec_machine *mach,
957 union tgsi_exec_channel *chan,
958 const struct tgsi_full_src_register *reg,
959 const uint chan_index )
960 {
961 union tgsi_exec_channel index;
962 uint swizzle;
963
964 /* We start with a direct index into a register file.
965 *
966 * file[1],
967 * where:
968 * file = Register.File
969 * [1] = Register.Index
970 */
971 index.i[0] =
972 index.i[1] =
973 index.i[2] =
974 index.i[3] = reg->Register.Index;
975
976 /* There is an extra source register that indirectly subscripts
977 * a register file. The direct index now becomes an offset
978 * that is being added to the indirect register.
979 *
980 * file[ind[2].x+1],
981 * where:
982 * ind = Indirect.File
983 * [2] = Indirect.Index
984 * .x = Indirect.SwizzleX
985 */
986 if (reg->Register.Indirect) {
987 union tgsi_exec_channel index2;
988 union tgsi_exec_channel indir_index;
989 const uint execmask = mach->ExecMask;
990 uint i;
991
992 /* which address register (always zero now) */
993 index2.i[0] =
994 index2.i[1] =
995 index2.i[2] =
996 index2.i[3] = reg->Indirect.Index;
997
998 /* get current value of address register[swizzle] */
999 swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
1000 fetch_src_file_channel(
1001 mach,
1002 reg->Indirect.File,
1003 swizzle,
1004 &index2,
1005 &indir_index );
1006
1007 /* add value of address register to the offset */
1008 index.i[0] += (int) indir_index.f[0];
1009 index.i[1] += (int) indir_index.f[1];
1010 index.i[2] += (int) indir_index.f[2];
1011 index.i[3] += (int) indir_index.f[3];
1012
1013 /* for disabled execution channels, zero-out the index to
1014 * avoid using a potential garbage value.
1015 */
1016 for (i = 0; i < QUAD_SIZE; i++) {
1017 if ((execmask & (1 << i)) == 0)
1018 index.i[i] = 0;
1019 }
1020 }
1021
1022 /* There is an extra source register that is a second
1023 * subscript to a register file. Effectively it means that
1024 * the register file is actually a 2D array of registers.
1025 *
1026 * file[1][3] == file[1*sizeof(file[1])+3],
1027 * where:
1028 * [3] = Dimension.Index
1029 */
1030 if (reg->Register.Dimension) {
1031 /* The size of the first-order array depends on the register file type.
1032 * We need to multiply the index to the first array to get an effective,
1033 * "flat" index that points to the beginning of the second-order array.
1034 */
1035 switch (reg->Register.File) {
1036 case TGSI_FILE_INPUT:
1037 case TGSI_FILE_SYSTEM_VALUE:
1038 index.i[0] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
1039 index.i[1] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
1040 index.i[2] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
1041 index.i[3] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
1042 break;
1043 case TGSI_FILE_CONSTANT:
1044 index.i[0] *= TGSI_EXEC_MAX_CONST_BUFFER;
1045 index.i[1] *= TGSI_EXEC_MAX_CONST_BUFFER;
1046 index.i[2] *= TGSI_EXEC_MAX_CONST_BUFFER;
1047 index.i[3] *= TGSI_EXEC_MAX_CONST_BUFFER;
1048 break;
1049 default:
1050 assert( 0 );
1051 }
1052
1053 index.i[0] += reg->Dimension.Index;
1054 index.i[1] += reg->Dimension.Index;
1055 index.i[2] += reg->Dimension.Index;
1056 index.i[3] += reg->Dimension.Index;
1057
1058 /* Again, the second subscript index can be addressed indirectly
1059 * identically to the first one.
1060 * Nothing stops us from indirectly addressing the indirect register,
1061 * but there is no need for that, so we won't exercise it.
1062 *
1063 * file[1][ind[4].y+3],
1064 * where:
1065 * ind = DimIndirect.File
1066 * [4] = DimIndirect.Index
1067 * .y = DimIndirect.SwizzleX
1068 */
1069 if (reg->Dimension.Indirect) {
1070 union tgsi_exec_channel index2;
1071 union tgsi_exec_channel indir_index;
1072 const uint execmask = mach->ExecMask;
1073 uint i;
1074
1075 index2.i[0] =
1076 index2.i[1] =
1077 index2.i[2] =
1078 index2.i[3] = reg->DimIndirect.Index;
1079
1080 swizzle = tgsi_util_get_src_register_swizzle( &reg->DimIndirect, CHAN_X );
1081 fetch_src_file_channel(
1082 mach,
1083 reg->DimIndirect.File,
1084 swizzle,
1085 &index2,
1086 &indir_index );
1087
1088 index.i[0] += (int) indir_index.f[0];
1089 index.i[1] += (int) indir_index.f[1];
1090 index.i[2] += (int) indir_index.f[2];
1091 index.i[3] += (int) indir_index.f[3];
1092
1093 /* for disabled execution channels, zero-out the index to
1094 * avoid using a potential garbage value.
1095 */
1096 for (i = 0; i < QUAD_SIZE; i++) {
1097 if ((execmask & (1 << i)) == 0)
1098 index.i[i] = 0;
1099 }
1100 }
1101
1102 /* If by any chance there was a need for a 3D array of register
1103 * files, we would have to check whether Dimension is followed
1104 * by a dimension register and continue the saga.
1105 */
1106 }
1107
1108 swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index );
1109 fetch_src_file_channel(
1110 mach,
1111 reg->Register.File,
1112 swizzle,
1113 &index,
1114 chan );
1115
1116 switch (tgsi_util_get_full_src_register_sign_mode( reg, chan_index )) {
1117 case TGSI_UTIL_SIGN_CLEAR:
1118 micro_abs( chan, chan );
1119 break;
1120
1121 case TGSI_UTIL_SIGN_SET:
1122 micro_abs( chan, chan );
1123 micro_neg( chan, chan );
1124 break;
1125
1126 case TGSI_UTIL_SIGN_TOGGLE:
1127 micro_neg( chan, chan );
1128 break;
1129
1130 case TGSI_UTIL_SIGN_KEEP:
1131 break;
1132 }
1133 }
1134
1135 static void
1136 store_dest(
1137 struct tgsi_exec_machine *mach,
1138 const union tgsi_exec_channel *chan,
1139 const struct tgsi_full_dst_register *reg,
1140 const struct tgsi_full_instruction *inst,
1141 uint chan_index )
1142 {
1143 uint i;
1144 union tgsi_exec_channel null;
1145 union tgsi_exec_channel *dst;
1146 uint execmask = mach->ExecMask;
1147 int offset = 0; /* indirection offset */
1148 int index;
1149
1150 #ifdef DEBUG
1151 check_inf_or_nan(chan);
1152 #endif
1153
1154 /* There is an extra source register that indirectly subscripts
1155 * a register file. The direct index now becomes an offset
1156 * that is being added to the indirect register.
1157 *
1158 * file[ind[2].x+1],
1159 * where:
1160 * ind = Indirect.File
1161 * [2] = Indirect.Index
1162 * .x = Indirect.SwizzleX
1163 */
1164 if (reg->Register.Indirect) {
1165 union tgsi_exec_channel index;
1166 union tgsi_exec_channel indir_index;
1167 uint swizzle;
1168
1169 /* which address register (always zero for now) */
1170 index.i[0] =
1171 index.i[1] =
1172 index.i[2] =
1173 index.i[3] = reg->Indirect.Index;
1174
1175 /* get current value of address register[swizzle] */
1176 swizzle = tgsi_util_get_src_register_swizzle( &reg->Indirect, CHAN_X );
1177
1178 /* fetch values from the address/indirection register */
1179 fetch_src_file_channel(
1180 mach,
1181 reg->Indirect.File,
1182 swizzle,
1183 &index,
1184 &indir_index );
1185
1186 /* save indirection offset */
1187 offset = (int) indir_index.f[0];
1188 }
1189
1190 switch (reg->Register.File) {
1191 case TGSI_FILE_NULL:
1192 dst = &null;
1193 break;
1194
1195 case TGSI_FILE_OUTPUT:
1196 index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0]
1197 + reg->Register.Index;
1198 dst = &mach->Outputs[offset + index].xyzw[chan_index];
1199 #if 0
1200 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
1201 fprintf(stderr, "STORING OUT[%d] mask(%d), = (", offset + index, execmask);
1202 for (i = 0; i < QUAD_SIZE; i++)
1203 if (execmask & (1 << i))
1204 fprintf(stderr, "%f, ", chan->f[i]);
1205 fprintf(stderr, ")\n");
1206 }
1207 #endif
1208 break;
1209
1210 case TGSI_FILE_TEMPORARY:
1211 index = reg->Register.Index;
1212 assert( index < TGSI_EXEC_NUM_TEMPS );
1213 dst = &mach->Temps[offset + index].xyzw[chan_index];
1214 break;
1215
1216 case TGSI_FILE_ADDRESS:
1217 index = reg->Register.Index;
1218 dst = &mach->Addrs[index].xyzw[chan_index];
1219 break;
1220
1221 case TGSI_FILE_LOOP:
1222 assert(reg->Register.Index == 0);
1223 assert(mach->LoopCounterStackTop > 0);
1224 assert(chan_index == CHAN_X);
1225 dst = &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[chan_index];
1226 break;
1227
1228 case TGSI_FILE_PREDICATE:
1229 index = reg->Register.Index;
1230 assert(index < TGSI_EXEC_NUM_PREDS);
1231 dst = &mach->Predicates[index].xyzw[chan_index];
1232 break;
1233
1234 default:
1235 assert( 0 );
1236 return;
1237 }
1238
1239 if (inst->Instruction.Predicate) {
1240 uint swizzle;
1241 union tgsi_exec_channel *pred;
1242
1243 switch (chan_index) {
1244 case CHAN_X:
1245 swizzle = inst->Predicate.SwizzleX;
1246 break;
1247 case CHAN_Y:
1248 swizzle = inst->Predicate.SwizzleY;
1249 break;
1250 case CHAN_Z:
1251 swizzle = inst->Predicate.SwizzleZ;
1252 break;
1253 case CHAN_W:
1254 swizzle = inst->Predicate.SwizzleW;
1255 break;
1256 default:
1257 assert(0);
1258 return;
1259 }
1260
1261 assert(inst->Predicate.Index == 0);
1262
1263 pred = &mach->Predicates[inst->Predicate.Index].xyzw[swizzle];
1264
1265 if (inst->Predicate.Negate) {
1266 for (i = 0; i < QUAD_SIZE; i++) {
1267 if (pred->u[i]) {
1268 execmask &= ~(1 << i);
1269 }
1270 }
1271 } else {
1272 for (i = 0; i < QUAD_SIZE; i++) {
1273 if (!pred->u[i]) {
1274 execmask &= ~(1 << i);
1275 }
1276 }
1277 }
1278 }
1279
1280 switch (inst->Instruction.Saturate) {
1281 case TGSI_SAT_NONE:
1282 for (i = 0; i < QUAD_SIZE; i++)
1283 if (execmask & (1 << i))
1284 dst->i[i] = chan->i[i];
1285 break;
1286
1287 case TGSI_SAT_ZERO_ONE:
1288 for (i = 0; i < QUAD_SIZE; i++)
1289 if (execmask & (1 << i)) {
1290 if (chan->f[i] < 0.0f)
1291 dst->f[i] = 0.0f;
1292 else if (chan->f[i] > 1.0f)
1293 dst->f[i] = 1.0f;
1294 else
1295 dst->i[i] = chan->i[i];
1296 }
1297 break;
1298
1299 case TGSI_SAT_MINUS_PLUS_ONE:
1300 for (i = 0; i < QUAD_SIZE; i++)
1301 if (execmask & (1 << i)) {
1302 if (chan->f[i] < -1.0f)
1303 dst->f[i] = -1.0f;
1304 else if (chan->f[i] > 1.0f)
1305 dst->f[i] = 1.0f;
1306 else
1307 dst->i[i] = chan->i[i];
1308 }
1309 break;
1310
1311 default:
1312 assert( 0 );
1313 }
1314 }
1315
1316 #define FETCH(VAL,INDEX,CHAN)\
1317 fetch_source (mach, VAL, &inst->Src[INDEX], CHAN)
1318
1319 #define STORE(VAL,INDEX,CHAN)\
1320 store_dest (mach, VAL, &inst->Dst[INDEX], inst, CHAN )
1321
1322
1323 /**
1324 * Execute ARB-style KIL which is predicated by a src register.
1325 * Kill fragment if any of the four values is less than zero.
1326 */
1327 static void
1328 exec_kil(struct tgsi_exec_machine *mach,
1329 const struct tgsi_full_instruction *inst)
1330 {
1331 uint uniquemask;
1332 uint chan_index;
1333 uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1334 union tgsi_exec_channel r[1];
1335
1336 /* This mask stores component bits that were already tested. */
1337 uniquemask = 0;
1338
1339 for (chan_index = 0; chan_index < 4; chan_index++)
1340 {
1341 uint swizzle;
1342 uint i;
1343
1344 /* unswizzle channel */
1345 swizzle = tgsi_util_get_full_src_register_swizzle (
1346 &inst->Src[0],
1347 chan_index);
1348
1349 /* check if the component has not been already tested */
1350 if (uniquemask & (1 << swizzle))
1351 continue;
1352 uniquemask |= 1 << swizzle;
1353
1354 FETCH(&r[0], 0, chan_index);
1355 for (i = 0; i < 4; i++)
1356 if (r[0].f[i] < 0.0f)
1357 kilmask |= 1 << i;
1358 }
1359
1360 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
1361 }
1362
1363 /**
1364 * Execute NVIDIA-style KIL which is predicated by a condition code.
1365 * Kill fragment if the condition code is TRUE.
1366 */
1367 static void
1368 exec_kilp(struct tgsi_exec_machine *mach,
1369 const struct tgsi_full_instruction *inst)
1370 {
1371 uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
1372
1373 /* "unconditional" kil */
1374 kilmask = mach->ExecMask;
1375 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
1376 }
1377
1378 static void
1379 emit_vertex(struct tgsi_exec_machine *mach)
1380 {
1381 /* FIXME: check for exec mask correctly
1382 unsigned i;
1383 for (i = 0; i < QUAD_SIZE; ++i) {
1384 if ((mach->ExecMask & (1 << i)))
1385 */
1386 if (mach->ExecMask) {
1387 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += mach->NumOutputs;
1388 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++;
1389 }
1390 }
1391
1392 static void
1393 emit_primitive(struct tgsi_exec_machine *mach)
1394 {
1395 unsigned *prim_count = &mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0];
1396 /* FIXME: check for exec mask correctly
1397 unsigned i;
1398 for (i = 0; i < QUAD_SIZE; ++i) {
1399 if ((mach->ExecMask & (1 << i)))
1400 */
1401 if (mach->ExecMask) {
1402 ++(*prim_count);
1403 debug_assert((*prim_count * mach->NumOutputs) < mach->MaxGeometryShaderOutputs);
1404 mach->Primitives[*prim_count] = 0;
1405 }
1406 }
1407
1408 /*
1409 * Fetch a four texture samples using STR texture coordinates.
1410 */
1411 static void
1412 fetch_texel( struct tgsi_sampler *sampler,
1413 const union tgsi_exec_channel *s,
1414 const union tgsi_exec_channel *t,
1415 const union tgsi_exec_channel *p,
1416 float lodbias, /* XXX should be float[4] */
1417 union tgsi_exec_channel *r,
1418 union tgsi_exec_channel *g,
1419 union tgsi_exec_channel *b,
1420 union tgsi_exec_channel *a )
1421 {
1422 uint j;
1423 float rgba[NUM_CHANNELS][QUAD_SIZE];
1424
1425 sampler->get_samples(sampler, s->f, t->f, p->f, lodbias, rgba);
1426
1427 for (j = 0; j < 4; j++) {
1428 r->f[j] = rgba[0][j];
1429 g->f[j] = rgba[1][j];
1430 b->f[j] = rgba[2][j];
1431 a->f[j] = rgba[3][j];
1432 }
1433 }
1434
1435
1436 static void
1437 exec_tex(struct tgsi_exec_machine *mach,
1438 const struct tgsi_full_instruction *inst,
1439 boolean biasLod,
1440 boolean projected)
1441 {
1442 const uint unit = inst->Src[1].Register.Index;
1443 union tgsi_exec_channel r[4];
1444 uint chan_index;
1445 float lodBias;
1446
1447 /* debug_printf("Sampler %u unit %u\n", sampler, unit); */
1448
1449 switch (inst->Texture.Texture) {
1450 case TGSI_TEXTURE_1D:
1451 case TGSI_TEXTURE_SHADOW1D:
1452
1453 FETCH(&r[0], 0, CHAN_X);
1454
1455 if (projected) {
1456 FETCH(&r[1], 0, CHAN_W);
1457 micro_div( &r[0], &r[0], &r[1] );
1458 }
1459
1460 if (biasLod) {
1461 FETCH(&r[1], 0, CHAN_W);
1462 lodBias = r[2].f[0];
1463 }
1464 else
1465 lodBias = 0.0;
1466
1467 fetch_texel(mach->Samplers[unit],
1468 &r[0], &ZeroVec, &ZeroVec, lodBias, /* S, T, P, BIAS */
1469 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
1470 break;
1471
1472 case TGSI_TEXTURE_2D:
1473 case TGSI_TEXTURE_RECT:
1474 case TGSI_TEXTURE_SHADOW2D:
1475 case TGSI_TEXTURE_SHADOWRECT:
1476
1477 FETCH(&r[0], 0, CHAN_X);
1478 FETCH(&r[1], 0, CHAN_Y);
1479 FETCH(&r[2], 0, CHAN_Z);
1480
1481 if (projected) {
1482 FETCH(&r[3], 0, CHAN_W);
1483 micro_div( &r[0], &r[0], &r[3] );
1484 micro_div( &r[1], &r[1], &r[3] );
1485 micro_div( &r[2], &r[2], &r[3] );
1486 }
1487
1488 if (biasLod) {
1489 FETCH(&r[3], 0, CHAN_W);
1490 lodBias = r[3].f[0];
1491 }
1492 else
1493 lodBias = 0.0;
1494
1495 fetch_texel(mach->Samplers[unit],
1496 &r[0], &r[1], &r[2], lodBias, /* inputs */
1497 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1498 break;
1499
1500 case TGSI_TEXTURE_3D:
1501 case TGSI_TEXTURE_CUBE:
1502
1503 FETCH(&r[0], 0, CHAN_X);
1504 FETCH(&r[1], 0, CHAN_Y);
1505 FETCH(&r[2], 0, CHAN_Z);
1506
1507 if (projected) {
1508 FETCH(&r[3], 0, CHAN_W);
1509 micro_div( &r[0], &r[0], &r[3] );
1510 micro_div( &r[1], &r[1], &r[3] );
1511 micro_div( &r[2], &r[2], &r[3] );
1512 }
1513
1514 if (biasLod) {
1515 FETCH(&r[3], 0, CHAN_W);
1516 lodBias = r[3].f[0];
1517 }
1518 else
1519 lodBias = 0.0;
1520
1521 fetch_texel(mach->Samplers[unit],
1522 &r[0], &r[1], &r[2], lodBias,
1523 &r[0], &r[1], &r[2], &r[3]);
1524 break;
1525
1526 default:
1527 assert (0);
1528 }
1529
1530 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
1531 STORE( &r[chan_index], 0, chan_index );
1532 }
1533 }
1534
1535 static void
1536 exec_txd(struct tgsi_exec_machine *mach,
1537 const struct tgsi_full_instruction *inst)
1538 {
1539 const uint unit = inst->Src[3].Register.Index;
1540 union tgsi_exec_channel r[4];
1541 uint chan_index;
1542
1543 /*
1544 * XXX: This is fake TXD -- the derivatives are not taken into account, yet.
1545 */
1546
1547 switch (inst->Texture.Texture) {
1548 case TGSI_TEXTURE_1D:
1549 case TGSI_TEXTURE_SHADOW1D:
1550
1551 FETCH(&r[0], 0, CHAN_X);
1552
1553 fetch_texel(mach->Samplers[unit],
1554 &r[0], &ZeroVec, &ZeroVec, 0.0f, /* S, T, P, BIAS */
1555 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
1556 break;
1557
1558 case TGSI_TEXTURE_2D:
1559 case TGSI_TEXTURE_RECT:
1560 case TGSI_TEXTURE_SHADOW2D:
1561 case TGSI_TEXTURE_SHADOWRECT:
1562
1563 FETCH(&r[0], 0, CHAN_X);
1564 FETCH(&r[1], 0, CHAN_Y);
1565 FETCH(&r[2], 0, CHAN_Z);
1566
1567 fetch_texel(mach->Samplers[unit],
1568 &r[0], &r[1], &r[2], 0.0f, /* inputs */
1569 &r[0], &r[1], &r[2], &r[3]); /* outputs */
1570 break;
1571
1572 case TGSI_TEXTURE_3D:
1573 case TGSI_TEXTURE_CUBE:
1574
1575 FETCH(&r[0], 0, CHAN_X);
1576 FETCH(&r[1], 0, CHAN_Y);
1577 FETCH(&r[2], 0, CHAN_Z);
1578
1579 fetch_texel(mach->Samplers[unit],
1580 &r[0], &r[1], &r[2], 0.0f,
1581 &r[0], &r[1], &r[2], &r[3]);
1582 break;
1583
1584 default:
1585 assert(0);
1586 }
1587
1588 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
1589 STORE(&r[chan_index], 0, chan_index);
1590 }
1591 }
1592
1593
1594 /**
1595 * Evaluate a constant-valued coefficient at the position of the
1596 * current quad.
1597 */
1598 static void
1599 eval_constant_coef(
1600 struct tgsi_exec_machine *mach,
1601 unsigned attrib,
1602 unsigned chan )
1603 {
1604 unsigned i;
1605
1606 for( i = 0; i < QUAD_SIZE; i++ ) {
1607 mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan];
1608 }
1609 }
1610
1611 /**
1612 * Evaluate a linear-valued coefficient at the position of the
1613 * current quad.
1614 */
1615 static void
1616 eval_linear_coef(
1617 struct tgsi_exec_machine *mach,
1618 unsigned attrib,
1619 unsigned chan )
1620 {
1621 const float x = mach->QuadPos.xyzw[0].f[0];
1622 const float y = mach->QuadPos.xyzw[1].f[0];
1623 const float dadx = mach->InterpCoefs[attrib].dadx[chan];
1624 const float dady = mach->InterpCoefs[attrib].dady[chan];
1625 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
1626 mach->Inputs[attrib].xyzw[chan].f[0] = a0;
1627 mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx;
1628 mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady;
1629 mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady;
1630 }
1631
1632 /**
1633 * Evaluate a perspective-valued coefficient at the position of the
1634 * current quad.
1635 */
1636 static void
1637 eval_perspective_coef(
1638 struct tgsi_exec_machine *mach,
1639 unsigned attrib,
1640 unsigned chan )
1641 {
1642 const float x = mach->QuadPos.xyzw[0].f[0];
1643 const float y = mach->QuadPos.xyzw[1].f[0];
1644 const float dadx = mach->InterpCoefs[attrib].dadx[chan];
1645 const float dady = mach->InterpCoefs[attrib].dady[chan];
1646 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
1647 const float *w = mach->QuadPos.xyzw[3].f;
1648 /* divide by W here */
1649 mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0];
1650 mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1];
1651 mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2];
1652 mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3];
1653 }
1654
1655
1656 typedef void (* eval_coef_func)(
1657 struct tgsi_exec_machine *mach,
1658 unsigned attrib,
1659 unsigned chan );
1660
1661 static void
1662 exec_declaration(struct tgsi_exec_machine *mach,
1663 const struct tgsi_full_declaration *decl)
1664 {
1665 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
1666 if (decl->Declaration.File == TGSI_FILE_INPUT ||
1667 decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) {
1668 uint first, last, mask;
1669
1670 first = decl->Range.First;
1671 last = decl->Range.Last;
1672 mask = decl->Declaration.UsageMask;
1673
1674 if (decl->Semantic.Name == TGSI_SEMANTIC_POSITION) {
1675 assert(decl->Semantic.Index == 0);
1676 assert(first == last);
1677 assert(mask == TGSI_WRITEMASK_XYZW);
1678
1679 mach->Inputs[first] = mach->QuadPos;
1680 } else if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) {
1681 uint i;
1682
1683 assert(decl->Semantic.Index == 0);
1684 assert(first == last);
1685
1686 for (i = 0; i < QUAD_SIZE; i++) {
1687 mach->Inputs[first].xyzw[0].f[i] = mach->Face;
1688 }
1689 } else {
1690 eval_coef_func eval;
1691 uint i, j;
1692
1693 switch (decl->Declaration.Interpolate) {
1694 case TGSI_INTERPOLATE_CONSTANT:
1695 eval = eval_constant_coef;
1696 break;
1697
1698 case TGSI_INTERPOLATE_LINEAR:
1699 eval = eval_linear_coef;
1700 break;
1701
1702 case TGSI_INTERPOLATE_PERSPECTIVE:
1703 eval = eval_perspective_coef;
1704 break;
1705
1706 default:
1707 assert(0);
1708 return;
1709 }
1710
1711 for (j = 0; j < NUM_CHANNELS; j++) {
1712 if (mask & (1 << j)) {
1713 for (i = first; i <= last; i++) {
1714 eval(mach, i, j);
1715 }
1716 }
1717 }
1718 }
1719 }
1720 }
1721 }
1722
1723 typedef void (* micro_op)(union tgsi_exec_channel *dst,
1724 const union tgsi_exec_channel *src);
1725
1726 static void
1727 exec_vector_unary(struct tgsi_exec_machine *mach,
1728 const struct tgsi_full_instruction *inst,
1729 micro_op op)
1730 {
1731 unsigned int chan;
1732 struct tgsi_exec_vector dst;
1733
1734 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1735 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1736 union tgsi_exec_channel src;
1737
1738 fetch_source(mach, &src, &inst->Src[0], chan);
1739 op(&dst.xyzw[chan], &src);
1740 }
1741 }
1742 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1743 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1744 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan);
1745 }
1746 }
1747 }
1748
1749 static void
1750 exec_vector_binary(struct tgsi_exec_machine *mach,
1751 const struct tgsi_full_instruction *inst,
1752 micro_op op)
1753 {
1754 unsigned int chan;
1755 struct tgsi_exec_vector dst;
1756
1757 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1758 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1759 union tgsi_exec_channel src[2];
1760
1761 fetch_source(mach, &src[0], &inst->Src[0], chan);
1762 fetch_source(mach, &src[1], &inst->Src[1], chan);
1763 op(&dst.xyzw[chan], src);
1764 }
1765 }
1766 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1767 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1768 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan);
1769 }
1770 }
1771 }
1772
1773 static void
1774 exec_vector_trinary(struct tgsi_exec_machine *mach,
1775 const struct tgsi_full_instruction *inst,
1776 micro_op op)
1777 {
1778 unsigned int chan;
1779 struct tgsi_exec_vector dst;
1780
1781 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1782 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1783 union tgsi_exec_channel src[3];
1784
1785 fetch_source(mach, &src[0], &inst->Src[0], chan);
1786 fetch_source(mach, &src[1], &inst->Src[1], chan);
1787 fetch_source(mach, &src[2], &inst->Src[2], chan);
1788 op(&dst.xyzw[chan], src);
1789 }
1790 }
1791 for (chan = 0; chan < NUM_CHANNELS; chan++) {
1792 if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
1793 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan);
1794 }
1795 }
1796 }
1797
1798 static void
1799 exec_break(struct tgsi_exec_machine *mach)
1800 {
1801 if (mach->BreakType == TGSI_EXEC_BREAK_INSIDE_LOOP) {
1802 /* turn off loop channels for each enabled exec channel */
1803 mach->LoopMask &= ~mach->ExecMask;
1804 /* Todo: if mach->LoopMask == 0, jump to end of loop */
1805 UPDATE_EXEC_MASK(mach);
1806 } else {
1807 assert(mach->BreakType == TGSI_EXEC_BREAK_INSIDE_SWITCH);
1808
1809 mach->Switch.mask = 0x0;
1810
1811 UPDATE_EXEC_MASK(mach);
1812 }
1813 }
1814
1815 static void
1816 exec_switch(struct tgsi_exec_machine *mach,
1817 const struct tgsi_full_instruction *inst)
1818 {
1819 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING);
1820 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
1821
1822 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch;
1823 fetch_source(mach, &mach->Switch.selector, &inst->Src[0], CHAN_X);
1824 mach->Switch.mask = 0x0;
1825 mach->Switch.defaultMask = 0x0;
1826
1827 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
1828 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_SWITCH;
1829
1830 UPDATE_EXEC_MASK(mach);
1831 }
1832
1833 static void
1834 exec_case(struct tgsi_exec_machine *mach,
1835 const struct tgsi_full_instruction *inst)
1836 {
1837 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask;
1838 union tgsi_exec_channel src;
1839 uint mask = 0;
1840
1841 fetch_source(mach, &src, &inst->Src[0], CHAN_X);
1842
1843 if (mach->Switch.selector.u[0] == src.u[0]) {
1844 mask |= 0x1;
1845 }
1846 if (mach->Switch.selector.u[1] == src.u[1]) {
1847 mask |= 0x2;
1848 }
1849 if (mach->Switch.selector.u[2] == src.u[2]) {
1850 mask |= 0x4;
1851 }
1852 if (mach->Switch.selector.u[3] == src.u[3]) {
1853 mask |= 0x8;
1854 }
1855
1856 mach->Switch.defaultMask |= mask;
1857
1858 mach->Switch.mask |= mask & prevMask;
1859
1860 UPDATE_EXEC_MASK(mach);
1861 }
1862
1863 static void
1864 exec_default(struct tgsi_exec_machine *mach)
1865 {
1866 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask;
1867
1868 mach->Switch.mask |= ~mach->Switch.defaultMask & prevMask;
1869
1870 UPDATE_EXEC_MASK(mach);
1871 }
1872
1873 static void
1874 exec_endswitch(struct tgsi_exec_machine *mach)
1875 {
1876 mach->Switch = mach->SwitchStack[--mach->SwitchStackTop];
1877 mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
1878
1879 UPDATE_EXEC_MASK(mach);
1880 }
1881
1882 static void
1883 micro_i2f(union tgsi_exec_channel *dst,
1884 const union tgsi_exec_channel *src)
1885 {
1886 dst->f[0] = (float)src->i[0];
1887 dst->f[1] = (float)src->i[1];
1888 dst->f[2] = (float)src->i[2];
1889 dst->f[3] = (float)src->i[3];
1890 }
1891
1892 static void
1893 micro_not(union tgsi_exec_channel *dst,
1894 const union tgsi_exec_channel *src)
1895 {
1896 dst->u[0] = ~src->u[0];
1897 dst->u[1] = ~src->u[1];
1898 dst->u[2] = ~src->u[2];
1899 dst->u[3] = ~src->u[3];
1900 }
1901
1902 static void
1903 micro_shl(union tgsi_exec_channel *dst,
1904 const union tgsi_exec_channel *src)
1905 {
1906 dst->u[0] = src[0].u[0] << src[1].u[0];
1907 dst->u[1] = src[0].u[1] << src[1].u[1];
1908 dst->u[2] = src[0].u[2] << src[1].u[2];
1909 dst->u[3] = src[0].u[3] << src[1].u[3];
1910 }
1911
1912 static void
1913 micro_and(union tgsi_exec_channel *dst,
1914 const union tgsi_exec_channel *src)
1915 {
1916 dst->u[0] = src[0].u[0] & src[1].u[0];
1917 dst->u[1] = src[0].u[1] & src[1].u[1];
1918 dst->u[2] = src[0].u[2] & src[1].u[2];
1919 dst->u[3] = src[0].u[3] & src[1].u[3];
1920 }
1921
1922 static void
1923 micro_or(union tgsi_exec_channel *dst,
1924 const union tgsi_exec_channel *src)
1925 {
1926 dst->u[0] = src[0].u[0] | src[1].u[0];
1927 dst->u[1] = src[0].u[1] | src[1].u[1];
1928 dst->u[2] = src[0].u[2] | src[1].u[2];
1929 dst->u[3] = src[0].u[3] | src[1].u[3];
1930 }
1931
1932 static void
1933 micro_xor(union tgsi_exec_channel *dst,
1934 const union tgsi_exec_channel *src)
1935 {
1936 dst->u[0] = src[0].u[0] ^ src[1].u[0];
1937 dst->u[1] = src[0].u[1] ^ src[1].u[1];
1938 dst->u[2] = src[0].u[2] ^ src[1].u[2];
1939 dst->u[3] = src[0].u[3] ^ src[1].u[3];
1940 }
1941
1942 static void
1943 micro_f2i(union tgsi_exec_channel *dst,
1944 const union tgsi_exec_channel *src)
1945 {
1946 dst->i[0] = (int)src->f[0];
1947 dst->i[1] = (int)src->f[1];
1948 dst->i[2] = (int)src->f[2];
1949 dst->i[3] = (int)src->f[3];
1950 }
1951
1952 static void
1953 micro_idiv(union tgsi_exec_channel *dst,
1954 const union tgsi_exec_channel *src)
1955 {
1956 dst->i[0] = src[0].i[0] / src[1].i[0];
1957 dst->i[1] = src[0].i[1] / src[1].i[1];
1958 dst->i[2] = src[0].i[2] / src[1].i[2];
1959 dst->i[3] = src[0].i[3] / src[1].i[3];
1960 }
1961
1962 static void
1963 micro_imax(union tgsi_exec_channel *dst,
1964 const union tgsi_exec_channel *src)
1965 {
1966 dst->i[0] = src[0].i[0] > src[1].i[0] ? src[0].i[0] : src[1].i[0];
1967 dst->i[1] = src[0].i[1] > src[1].i[1] ? src[0].i[1] : src[1].i[1];
1968 dst->i[2] = src[0].i[2] > src[1].i[2] ? src[0].i[2] : src[1].i[2];
1969 dst->i[3] = src[0].i[3] > src[1].i[3] ? src[0].i[3] : src[1].i[3];
1970 }
1971
1972 static void
1973 micro_imin(union tgsi_exec_channel *dst,
1974 const union tgsi_exec_channel *src)
1975 {
1976 dst->i[0] = src[0].i[0] < src[1].i[0] ? src[0].i[0] : src[1].i[0];
1977 dst->i[1] = src[0].i[1] < src[1].i[1] ? src[0].i[1] : src[1].i[1];
1978 dst->i[2] = src[0].i[2] < src[1].i[2] ? src[0].i[2] : src[1].i[2];
1979 dst->i[3] = src[0].i[3] < src[1].i[3] ? src[0].i[3] : src[1].i[3];
1980 }
1981
1982 static void
1983 micro_ineg(union tgsi_exec_channel *dst,
1984 const union tgsi_exec_channel *src)
1985 {
1986 dst->i[0] = -src->i[0];
1987 dst->i[1] = -src->i[1];
1988 dst->i[2] = -src->i[2];
1989 dst->i[3] = -src->i[3];
1990 }
1991
1992 static void
1993 micro_isge(union tgsi_exec_channel *dst,
1994 const union tgsi_exec_channel *src)
1995 {
1996 dst->i[0] = src[0].i[0] >= src[1].i[0] ? -1 : 0;
1997 dst->i[1] = src[0].i[1] >= src[1].i[1] ? -1 : 0;
1998 dst->i[2] = src[0].i[2] >= src[1].i[2] ? -1 : 0;
1999 dst->i[3] = src[0].i[3] >= src[1].i[3] ? -1 : 0;
2000 }
2001
2002 static void
2003 micro_ishr(union tgsi_exec_channel *dst,
2004 const union tgsi_exec_channel *src)
2005 {
2006 dst->i[0] = src[0].i[0] >> src[1].i[0];
2007 dst->i[1] = src[0].i[1] >> src[1].i[1];
2008 dst->i[2] = src[0].i[2] >> src[1].i[2];
2009 dst->i[3] = src[0].i[3] >> src[1].i[3];
2010 }
2011
2012 static void
2013 micro_islt(union tgsi_exec_channel *dst,
2014 const union tgsi_exec_channel *src)
2015 {
2016 dst->i[0] = src[0].i[0] < src[1].i[0] ? -1 : 0;
2017 dst->i[1] = src[0].i[1] < src[1].i[1] ? -1 : 0;
2018 dst->i[2] = src[0].i[2] < src[1].i[2] ? -1 : 0;
2019 dst->i[3] = src[0].i[3] < src[1].i[3] ? -1 : 0;
2020 }
2021
2022 static void
2023 micro_f2u(union tgsi_exec_channel *dst,
2024 const union tgsi_exec_channel *src)
2025 {
2026 dst->u[0] = (uint)src->f[0];
2027 dst->u[1] = (uint)src->f[1];
2028 dst->u[2] = (uint)src->f[2];
2029 dst->u[3] = (uint)src->f[3];
2030 }
2031
2032 static void
2033 micro_u2f(union tgsi_exec_channel *dst,
2034 const union tgsi_exec_channel *src)
2035 {
2036 dst->f[0] = (float)src->u[0];
2037 dst->f[1] = (float)src->u[1];
2038 dst->f[2] = (float)src->u[2];
2039 dst->f[3] = (float)src->u[3];
2040 }
2041
2042 static void
2043 micro_uadd(union tgsi_exec_channel *dst,
2044 const union tgsi_exec_channel *src)
2045 {
2046 dst->u[0] = src[0].u[0] + src[1].u[0];
2047 dst->u[1] = src[0].u[1] + src[1].u[1];
2048 dst->u[2] = src[0].u[2] + src[1].u[2];
2049 dst->u[3] = src[0].u[3] + src[1].u[3];
2050 }
2051
2052 static void
2053 micro_udiv(union tgsi_exec_channel *dst,
2054 const union tgsi_exec_channel *src)
2055 {
2056 dst->u[0] = src[0].u[0] / src[1].u[0];
2057 dst->u[1] = src[0].u[1] / src[1].u[1];
2058 dst->u[2] = src[0].u[2] / src[1].u[2];
2059 dst->u[3] = src[0].u[3] / src[1].u[3];
2060 }
2061
2062 static void
2063 micro_umad(union tgsi_exec_channel *dst,
2064 const union tgsi_exec_channel *src)
2065 {
2066 dst->u[0] = src[0].u[0] * src[1].u[0] + src[2].u[0];
2067 dst->u[1] = src[0].u[1] * src[1].u[1] + src[2].u[1];
2068 dst->u[2] = src[0].u[2] * src[1].u[2] + src[2].u[2];
2069 dst->u[3] = src[0].u[3] * src[1].u[3] + src[2].u[3];
2070 }
2071
2072 static void
2073 micro_umax(union tgsi_exec_channel *dst,
2074 const union tgsi_exec_channel *src)
2075 {
2076 dst->u[0] = src[0].u[0] > src[1].u[0] ? src[0].u[0] : src[1].u[0];
2077 dst->u[1] = src[0].u[1] > src[1].u[1] ? src[0].u[1] : src[1].u[1];
2078 dst->u[2] = src[0].u[2] > src[1].u[2] ? src[0].u[2] : src[1].u[2];
2079 dst->u[3] = src[0].u[3] > src[1].u[3] ? src[0].u[3] : src[1].u[3];
2080 }
2081
2082 static void
2083 micro_umin(union tgsi_exec_channel *dst,
2084 const union tgsi_exec_channel *src)
2085 {
2086 dst->u[0] = src[0].u[0] < src[1].u[0] ? src[0].u[0] : src[1].u[0];
2087 dst->u[1] = src[0].u[1] < src[1].u[1] ? src[0].u[1] : src[1].u[1];
2088 dst->u[2] = src[0].u[2] < src[1].u[2] ? src[0].u[2] : src[1].u[2];
2089 dst->u[3] = src[0].u[3] < src[1].u[3] ? src[0].u[3] : src[1].u[3];
2090 }
2091
2092 static void
2093 micro_umod(union tgsi_exec_channel *dst,
2094 const union tgsi_exec_channel *src)
2095 {
2096 dst->u[0] = src[0].u[0] % src[1].u[0];
2097 dst->u[1] = src[0].u[1] % src[1].u[1];
2098 dst->u[2] = src[0].u[2] % src[1].u[2];
2099 dst->u[3] = src[0].u[3] % src[1].u[3];
2100 }
2101
2102 static void
2103 micro_umul(union tgsi_exec_channel *dst,
2104 const union tgsi_exec_channel *src)
2105 {
2106 dst->u[0] = src[0].u[0] * src[1].u[0];
2107 dst->u[1] = src[0].u[1] * src[1].u[1];
2108 dst->u[2] = src[0].u[2] * src[1].u[2];
2109 dst->u[3] = src[0].u[3] * src[1].u[3];
2110 }
2111
2112 static void
2113 micro_useq(union tgsi_exec_channel *dst,
2114 const union tgsi_exec_channel *src)
2115 {
2116 dst->u[0] = src[0].u[0] == src[1].u[0] ? ~0 : 0;
2117 dst->u[1] = src[0].u[1] == src[1].u[1] ? ~0 : 0;
2118 dst->u[2] = src[0].u[2] == src[1].u[2] ? ~0 : 0;
2119 dst->u[3] = src[0].u[3] == src[1].u[3] ? ~0 : 0;
2120 }
2121
2122 static void
2123 micro_usge(union tgsi_exec_channel *dst,
2124 const union tgsi_exec_channel *src)
2125 {
2126 dst->u[0] = src[0].u[0] >= src[1].u[0] ? ~0 : 0;
2127 dst->u[1] = src[0].u[1] >= src[1].u[1] ? ~0 : 0;
2128 dst->u[2] = src[0].u[2] >= src[1].u[2] ? ~0 : 0;
2129 dst->u[3] = src[0].u[3] >= src[1].u[3] ? ~0 : 0;
2130 }
2131
2132 static void
2133 micro_ushr(union tgsi_exec_channel *dst,
2134 const union tgsi_exec_channel *src)
2135 {
2136 dst->u[0] = src[0].u[0] >> src[1].u[0];
2137 dst->u[1] = src[0].u[1] >> src[1].u[1];
2138 dst->u[2] = src[0].u[2] >> src[1].u[2];
2139 dst->u[3] = src[0].u[3] >> src[1].u[3];
2140 }
2141
2142 static void
2143 micro_uslt(union tgsi_exec_channel *dst,
2144 const union tgsi_exec_channel *src)
2145 {
2146 dst->u[0] = src[0].u[0] < src[1].u[0] ? ~0 : 0;
2147 dst->u[1] = src[0].u[1] < src[1].u[1] ? ~0 : 0;
2148 dst->u[2] = src[0].u[2] < src[1].u[2] ? ~0 : 0;
2149 dst->u[3] = src[0].u[3] < src[1].u[3] ? ~0 : 0;
2150 }
2151
2152 static void
2153 micro_usne(union tgsi_exec_channel *dst,
2154 const union tgsi_exec_channel *src)
2155 {
2156 dst->u[0] = src[0].u[0] != src[1].u[0] ? ~0 : 0;
2157 dst->u[1] = src[0].u[1] != src[1].u[1] ? ~0 : 0;
2158 dst->u[2] = src[0].u[2] != src[1].u[2] ? ~0 : 0;
2159 dst->u[3] = src[0].u[3] != src[1].u[3] ? ~0 : 0;
2160 }
2161
2162 static void
2163 exec_instruction(
2164 struct tgsi_exec_machine *mach,
2165 const struct tgsi_full_instruction *inst,
2166 int *pc )
2167 {
2168 uint chan_index;
2169 union tgsi_exec_channel r[10];
2170 union tgsi_exec_channel d[8];
2171
2172 (*pc)++;
2173
2174 switch (inst->Instruction.Opcode) {
2175 case TGSI_OPCODE_ARL:
2176 case TGSI_OPCODE_FLR:
2177 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2178 FETCH( &r[0], 0, chan_index );
2179 micro_flr(&d[chan_index], &r[0]);
2180 }
2181 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2182 STORE(&d[chan_index], 0, chan_index);
2183 }
2184 break;
2185
2186 case TGSI_OPCODE_MOV:
2187 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2188 FETCH(&d[chan_index], 0, chan_index);
2189 }
2190 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2191 STORE(&d[chan_index], 0, chan_index);
2192 }
2193 break;
2194
2195 case TGSI_OPCODE_LIT:
2196 if (IS_CHANNEL_ENABLED( *inst, CHAN_Y ) || IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
2197 FETCH( &r[0], 0, CHAN_X );
2198 if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
2199 micro_max(&d[CHAN_Y], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]);
2200 }
2201
2202 if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
2203 FETCH( &r[1], 0, CHAN_Y );
2204 micro_max( &r[1], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
2205
2206 FETCH( &r[2], 0, CHAN_W );
2207 micro_min( &r[2], &r[2], &mach->Temps[TEMP_128_I].xyzw[TEMP_128_C] );
2208 micro_max( &r[2], &r[2], &mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C] );
2209 micro_pow( &r[1], &r[1], &r[2] );
2210 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]);
2211 }
2212
2213 if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
2214 STORE(&d[CHAN_Y], 0, CHAN_Y);
2215 }
2216 if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
2217 STORE(&d[CHAN_Z], 0, CHAN_Z);
2218 }
2219 }
2220 if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
2221 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X );
2222 }
2223 if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
2224 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
2225 }
2226 break;
2227
2228 case TGSI_OPCODE_RCP:
2229 /* TGSI_OPCODE_RECIP */
2230 FETCH( &r[0], 0, CHAN_X );
2231 micro_div( &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &r[0] );
2232 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2233 STORE( &r[0], 0, chan_index );
2234 }
2235 break;
2236
2237 case TGSI_OPCODE_RSQ:
2238 /* TGSI_OPCODE_RECIPSQRT */
2239 FETCH( &r[0], 0, CHAN_X );
2240 micro_abs( &r[0], &r[0] );
2241 micro_sqrt( &r[0], &r[0] );
2242 micro_div( &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &r[0] );
2243 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2244 STORE( &r[0], 0, chan_index );
2245 }
2246 break;
2247
2248 case TGSI_OPCODE_EXP:
2249 FETCH( &r[0], 0, CHAN_X );
2250 micro_flr( &r[1], &r[0] ); /* r1 = floor(r0) */
2251 if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
2252 micro_exp2( &r[2], &r[1] ); /* r2 = 2 ^ r1 */
2253 STORE( &r[2], 0, CHAN_X ); /* store r2 */
2254 }
2255 if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
2256 micro_sub( &r[2], &r[0], &r[1] ); /* r2 = r0 - r1 */
2257 STORE( &r[2], 0, CHAN_Y ); /* store r2 */
2258 }
2259 if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
2260 micro_exp2( &r[2], &r[0] ); /* r2 = 2 ^ r0 */
2261 STORE( &r[2], 0, CHAN_Z ); /* store r2 */
2262 }
2263 if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
2264 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
2265 }
2266 break;
2267
2268 case TGSI_OPCODE_LOG:
2269 FETCH( &r[0], 0, CHAN_X );
2270 micro_abs( &r[2], &r[0] ); /* r2 = abs(r0) */
2271 micro_lg2( &r[1], &r[2] ); /* r1 = lg2(r2) */
2272 micro_flr( &r[0], &r[1] ); /* r0 = floor(r1) */
2273 if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
2274 STORE( &r[0], 0, CHAN_X );
2275 }
2276 if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
2277 micro_exp2( &r[0], &r[0] ); /* r0 = 2 ^ r0 */
2278 micro_div( &r[0], &r[2], &r[0] ); /* r0 = r2 / r0 */
2279 STORE( &r[0], 0, CHAN_Y );
2280 }
2281 if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
2282 STORE( &r[1], 0, CHAN_Z );
2283 }
2284 if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
2285 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
2286 }
2287 break;
2288
2289 case TGSI_OPCODE_MUL:
2290 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2291 FETCH(&r[0], 0, chan_index);
2292 FETCH(&r[1], 1, chan_index);
2293 micro_mul(&d[chan_index], &r[0], &r[1]);
2294 }
2295 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2296 STORE(&d[chan_index], 0, chan_index);
2297 }
2298 break;
2299
2300 case TGSI_OPCODE_ADD:
2301 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2302 FETCH( &r[0], 0, chan_index );
2303 FETCH( &r[1], 1, chan_index );
2304 micro_add(&d[chan_index], &r[0], &r[1]);
2305 }
2306 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2307 STORE(&d[chan_index], 0, chan_index);
2308 }
2309 break;
2310
2311 case TGSI_OPCODE_DP3:
2312 /* TGSI_OPCODE_DOT3 */
2313 FETCH( &r[0], 0, CHAN_X );
2314 FETCH( &r[1], 1, CHAN_X );
2315 micro_mul( &r[0], &r[0], &r[1] );
2316
2317 FETCH( &r[1], 0, CHAN_Y );
2318 FETCH( &r[2], 1, CHAN_Y );
2319 micro_mul( &r[1], &r[1], &r[2] );
2320 micro_add( &r[0], &r[0], &r[1] );
2321
2322 FETCH( &r[1], 0, CHAN_Z );
2323 FETCH( &r[2], 1, CHAN_Z );
2324 micro_mul( &r[1], &r[1], &r[2] );
2325 micro_add( &r[0], &r[0], &r[1] );
2326
2327 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2328 STORE( &r[0], 0, chan_index );
2329 }
2330 break;
2331
2332 case TGSI_OPCODE_DP4:
2333 /* TGSI_OPCODE_DOT4 */
2334 FETCH(&r[0], 0, CHAN_X);
2335 FETCH(&r[1], 1, CHAN_X);
2336
2337 micro_mul( &r[0], &r[0], &r[1] );
2338
2339 FETCH(&r[1], 0, CHAN_Y);
2340 FETCH(&r[2], 1, CHAN_Y);
2341
2342 micro_mul( &r[1], &r[1], &r[2] );
2343 micro_add( &r[0], &r[0], &r[1] );
2344
2345 FETCH(&r[1], 0, CHAN_Z);
2346 FETCH(&r[2], 1, CHAN_Z);
2347
2348 micro_mul( &r[1], &r[1], &r[2] );
2349 micro_add( &r[0], &r[0], &r[1] );
2350
2351 FETCH(&r[1], 0, CHAN_W);
2352 FETCH(&r[2], 1, CHAN_W);
2353
2354 micro_mul( &r[1], &r[1], &r[2] );
2355 micro_add( &r[0], &r[0], &r[1] );
2356
2357 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2358 STORE( &r[0], 0, chan_index );
2359 }
2360 break;
2361
2362 case TGSI_OPCODE_DST:
2363 if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
2364 FETCH( &r[0], 0, CHAN_Y );
2365 FETCH( &r[1], 1, CHAN_Y);
2366 micro_mul(&d[CHAN_Y], &r[0], &r[1]);
2367 }
2368 if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
2369 FETCH(&d[CHAN_Z], 0, CHAN_Z);
2370 }
2371 if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
2372 FETCH(&d[CHAN_W], 1, CHAN_W);
2373 }
2374
2375 if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
2376 STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X);
2377 }
2378 if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
2379 STORE(&d[CHAN_Y], 0, CHAN_Y);
2380 }
2381 if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
2382 STORE(&d[CHAN_Z], 0, CHAN_Z);
2383 }
2384 if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
2385 STORE(&d[CHAN_W], 0, CHAN_W);
2386 }
2387 break;
2388
2389 case TGSI_OPCODE_MIN:
2390 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2391 FETCH(&r[0], 0, chan_index);
2392 FETCH(&r[1], 1, chan_index);
2393
2394 /* XXX use micro_min()?? */
2395 micro_lt(&d[chan_index], &r[0], &r[1], &r[0], &r[1]);
2396 }
2397 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2398 STORE(&d[chan_index], 0, chan_index);
2399 }
2400 break;
2401
2402 case TGSI_OPCODE_MAX:
2403 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2404 FETCH(&r[0], 0, chan_index);
2405 FETCH(&r[1], 1, chan_index);
2406
2407 /* XXX use micro_max()?? */
2408 micro_lt(&d[chan_index], &r[0], &r[1], &r[1], &r[0] );
2409 }
2410 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2411 STORE(&d[chan_index], 0, chan_index);
2412 }
2413 break;
2414
2415 case TGSI_OPCODE_SLT:
2416 /* TGSI_OPCODE_SETLT */
2417 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2418 FETCH( &r[0], 0, chan_index );
2419 FETCH( &r[1], 1, chan_index );
2420 micro_lt(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]);
2421 }
2422 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2423 STORE(&d[chan_index], 0, chan_index);
2424 }
2425 break;
2426
2427 case TGSI_OPCODE_SGE:
2428 /* TGSI_OPCODE_SETGE */
2429 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2430 FETCH( &r[0], 0, chan_index );
2431 FETCH( &r[1], 1, chan_index );
2432 micro_le(&d[chan_index], &r[1], &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]);
2433 }
2434 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2435 STORE(&d[chan_index], 0, chan_index);
2436 }
2437 break;
2438
2439 case TGSI_OPCODE_MAD:
2440 /* TGSI_OPCODE_MADD */
2441 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2442 FETCH( &r[0], 0, chan_index );
2443 FETCH( &r[1], 1, chan_index );
2444 micro_mul( &r[0], &r[0], &r[1] );
2445 FETCH( &r[1], 2, chan_index );
2446 micro_add(&d[chan_index], &r[0], &r[1]);
2447 }
2448 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2449 STORE(&d[chan_index], 0, chan_index);
2450 }
2451 break;
2452
2453 case TGSI_OPCODE_SUB:
2454 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2455 FETCH(&r[0], 0, chan_index);
2456 FETCH(&r[1], 1, chan_index);
2457 micro_sub(&d[chan_index], &r[0], &r[1]);
2458 }
2459 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2460 STORE(&d[chan_index], 0, chan_index);
2461 }
2462 break;
2463
2464 case TGSI_OPCODE_LRP:
2465 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2466 FETCH(&r[0], 0, chan_index);
2467 FETCH(&r[1], 1, chan_index);
2468 FETCH(&r[2], 2, chan_index);
2469 micro_sub( &r[1], &r[1], &r[2] );
2470 micro_mul( &r[0], &r[0], &r[1] );
2471 micro_add(&d[chan_index], &r[0], &r[2]);
2472 }
2473 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2474 STORE(&d[chan_index], 0, chan_index);
2475 }
2476 break;
2477
2478 case TGSI_OPCODE_CND:
2479 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2480 FETCH(&r[0], 0, chan_index);
2481 FETCH(&r[1], 1, chan_index);
2482 FETCH(&r[2], 2, chan_index);
2483 micro_lt(&d[chan_index], &mach->Temps[TEMP_HALF_I].xyzw[TEMP_HALF_C], &r[2], &r[0], &r[1]);
2484 }
2485 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2486 STORE(&d[chan_index], 0, chan_index);
2487 }
2488 break;
2489
2490 case TGSI_OPCODE_DP2A:
2491 FETCH( &r[0], 0, CHAN_X );
2492 FETCH( &r[1], 1, CHAN_X );
2493 micro_mul( &r[0], &r[0], &r[1] );
2494
2495 FETCH( &r[1], 0, CHAN_Y );
2496 FETCH( &r[2], 1, CHAN_Y );
2497 micro_mul( &r[1], &r[1], &r[2] );
2498 micro_add( &r[0], &r[0], &r[1] );
2499
2500 FETCH( &r[2], 2, CHAN_X );
2501 micro_add( &r[0], &r[0], &r[2] );
2502
2503 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2504 STORE( &r[0], 0, chan_index );
2505 }
2506 break;
2507
2508 case TGSI_OPCODE_FRC:
2509 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2510 FETCH( &r[0], 0, chan_index );
2511 micro_frc(&d[chan_index], &r[0]);
2512 }
2513 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2514 STORE(&d[chan_index], 0, chan_index);
2515 }
2516 break;
2517
2518 case TGSI_OPCODE_CLAMP:
2519 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2520 FETCH(&r[0], 0, chan_index);
2521 FETCH(&r[1], 1, chan_index);
2522 micro_max(&r[0], &r[0], &r[1]);
2523 FETCH(&r[1], 2, chan_index);
2524 micro_min(&d[chan_index], &r[0], &r[1]);
2525 }
2526 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2527 STORE(&d[chan_index], 0, chan_index);
2528 }
2529 break;
2530
2531 case TGSI_OPCODE_ROUND:
2532 case TGSI_OPCODE_ARR:
2533 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2534 FETCH( &r[0], 0, chan_index );
2535 micro_rnd(&d[chan_index], &r[0]);
2536 }
2537 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2538 STORE(&d[chan_index], 0, chan_index);
2539 }
2540 break;
2541
2542 case TGSI_OPCODE_EX2:
2543 FETCH(&r[0], 0, CHAN_X);
2544
2545 micro_exp2( &r[0], &r[0] );
2546
2547 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2548 STORE( &r[0], 0, chan_index );
2549 }
2550 break;
2551
2552 case TGSI_OPCODE_LG2:
2553 FETCH( &r[0], 0, CHAN_X );
2554 micro_lg2( &r[0], &r[0] );
2555 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2556 STORE( &r[0], 0, chan_index );
2557 }
2558 break;
2559
2560 case TGSI_OPCODE_POW:
2561 FETCH(&r[0], 0, CHAN_X);
2562 FETCH(&r[1], 1, CHAN_X);
2563
2564 micro_pow( &r[0], &r[0], &r[1] );
2565
2566 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2567 STORE( &r[0], 0, chan_index );
2568 }
2569 break;
2570
2571 case TGSI_OPCODE_XPD:
2572 FETCH(&r[0], 0, CHAN_Y);
2573 FETCH(&r[1], 1, CHAN_Z);
2574
2575 micro_mul( &r[2], &r[0], &r[1] );
2576
2577 FETCH(&r[3], 0, CHAN_Z);
2578 FETCH(&r[4], 1, CHAN_Y);
2579
2580 micro_mul( &r[5], &r[3], &r[4] );
2581 micro_sub(&d[CHAN_X], &r[2], &r[5]);
2582
2583 FETCH(&r[2], 1, CHAN_X);
2584
2585 micro_mul( &r[3], &r[3], &r[2] );
2586
2587 FETCH(&r[5], 0, CHAN_X);
2588
2589 micro_mul( &r[1], &r[1], &r[5] );
2590 micro_sub(&d[CHAN_Y], &r[3], &r[1]);
2591
2592 micro_mul( &r[5], &r[5], &r[4] );
2593 micro_mul( &r[0], &r[0], &r[2] );
2594 micro_sub(&d[CHAN_Z], &r[5], &r[0]);
2595
2596 if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
2597 STORE(&d[CHAN_X], 0, CHAN_X);
2598 }
2599 if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
2600 STORE(&d[CHAN_Y], 0, CHAN_Y);
2601 }
2602 if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
2603 STORE(&d[CHAN_Z], 0, CHAN_Z);
2604 }
2605 if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
2606 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
2607 }
2608 break;
2609
2610 case TGSI_OPCODE_ABS:
2611 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2612 FETCH(&r[0], 0, chan_index);
2613 micro_abs(&d[chan_index], &r[0]);
2614 }
2615 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2616 STORE(&d[chan_index], 0, chan_index);
2617 }
2618 break;
2619
2620 case TGSI_OPCODE_RCC:
2621 FETCH(&r[0], 0, CHAN_X);
2622 micro_div(&r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &r[0]);
2623 micro_float_clamp(&r[0], &r[0]);
2624 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2625 STORE(&r[0], 0, chan_index);
2626 }
2627 break;
2628
2629 case TGSI_OPCODE_DPH:
2630 FETCH(&r[0], 0, CHAN_X);
2631 FETCH(&r[1], 1, CHAN_X);
2632
2633 micro_mul( &r[0], &r[0], &r[1] );
2634
2635 FETCH(&r[1], 0, CHAN_Y);
2636 FETCH(&r[2], 1, CHAN_Y);
2637
2638 micro_mul( &r[1], &r[1], &r[2] );
2639 micro_add( &r[0], &r[0], &r[1] );
2640
2641 FETCH(&r[1], 0, CHAN_Z);
2642 FETCH(&r[2], 1, CHAN_Z);
2643
2644 micro_mul( &r[1], &r[1], &r[2] );
2645 micro_add( &r[0], &r[0], &r[1] );
2646
2647 FETCH(&r[1], 1, CHAN_W);
2648
2649 micro_add( &r[0], &r[0], &r[1] );
2650
2651 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2652 STORE( &r[0], 0, chan_index );
2653 }
2654 break;
2655
2656 case TGSI_OPCODE_COS:
2657 FETCH(&r[0], 0, CHAN_X);
2658
2659 micro_cos( &r[0], &r[0] );
2660
2661 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2662 STORE( &r[0], 0, chan_index );
2663 }
2664 break;
2665
2666 case TGSI_OPCODE_DDX:
2667 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2668 FETCH( &r[0], 0, chan_index );
2669 micro_ddx(&d[chan_index], &r[0]);
2670 }
2671 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2672 STORE(&d[chan_index], 0, chan_index);
2673 }
2674 break;
2675
2676 case TGSI_OPCODE_DDY:
2677 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2678 FETCH( &r[0], 0, chan_index );
2679 micro_ddy(&d[chan_index], &r[0]);
2680 }
2681 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2682 STORE(&d[chan_index], 0, chan_index);
2683 }
2684 break;
2685
2686 case TGSI_OPCODE_KILP:
2687 exec_kilp (mach, inst);
2688 break;
2689
2690 case TGSI_OPCODE_KIL:
2691 exec_kil (mach, inst);
2692 break;
2693
2694 case TGSI_OPCODE_PK2H:
2695 assert (0);
2696 break;
2697
2698 case TGSI_OPCODE_PK2US:
2699 assert (0);
2700 break;
2701
2702 case TGSI_OPCODE_PK4B:
2703 assert (0);
2704 break;
2705
2706 case TGSI_OPCODE_PK4UB:
2707 assert (0);
2708 break;
2709
2710 case TGSI_OPCODE_RFL:
2711 if (IS_CHANNEL_ENABLED(*inst, CHAN_X) ||
2712 IS_CHANNEL_ENABLED(*inst, CHAN_Y) ||
2713 IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
2714 /* r0 = dp3(src0, src0) */
2715 FETCH(&r[2], 0, CHAN_X);
2716 micro_mul(&r[0], &r[2], &r[2]);
2717 FETCH(&r[4], 0, CHAN_Y);
2718 micro_mul(&r[8], &r[4], &r[4]);
2719 micro_add(&r[0], &r[0], &r[8]);
2720 FETCH(&r[6], 0, CHAN_Z);
2721 micro_mul(&r[8], &r[6], &r[6]);
2722 micro_add(&r[0], &r[0], &r[8]);
2723
2724 /* r1 = dp3(src0, src1) */
2725 FETCH(&r[3], 1, CHAN_X);
2726 micro_mul(&r[1], &r[2], &r[3]);
2727 FETCH(&r[5], 1, CHAN_Y);
2728 micro_mul(&r[8], &r[4], &r[5]);
2729 micro_add(&r[1], &r[1], &r[8]);
2730 FETCH(&r[7], 1, CHAN_Z);
2731 micro_mul(&r[8], &r[6], &r[7]);
2732 micro_add(&r[1], &r[1], &r[8]);
2733
2734 /* r1 = 2 * r1 / r0 */
2735 micro_add(&r[1], &r[1], &r[1]);
2736 micro_div(&r[1], &r[1], &r[0]);
2737
2738 if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
2739 micro_mul(&r[2], &r[2], &r[1]);
2740 micro_sub(&r[2], &r[2], &r[3]);
2741 STORE(&r[2], 0, CHAN_X);
2742 }
2743 if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
2744 micro_mul(&r[4], &r[4], &r[1]);
2745 micro_sub(&r[4], &r[4], &r[5]);
2746 STORE(&r[4], 0, CHAN_Y);
2747 }
2748 if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
2749 micro_mul(&r[6], &r[6], &r[1]);
2750 micro_sub(&r[6], &r[6], &r[7]);
2751 STORE(&r[6], 0, CHAN_Z);
2752 }
2753 }
2754 if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
2755 STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W);
2756 }
2757 break;
2758
2759 case TGSI_OPCODE_SEQ:
2760 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2761 FETCH( &r[0], 0, chan_index );
2762 FETCH( &r[1], 1, chan_index );
2763 micro_eq(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]);
2764 }
2765 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2766 STORE(&d[chan_index], 0, chan_index);
2767 }
2768 break;
2769
2770 case TGSI_OPCODE_SFL:
2771 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2772 STORE(&mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], 0, chan_index);
2773 }
2774 break;
2775
2776 case TGSI_OPCODE_SGT:
2777 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2778 FETCH( &r[0], 0, chan_index );
2779 FETCH( &r[1], 1, chan_index );
2780 micro_le(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]);
2781 }
2782 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2783 STORE(&d[chan_index], 0, chan_index);
2784 }
2785 break;
2786
2787 case TGSI_OPCODE_SIN:
2788 FETCH( &r[0], 0, CHAN_X );
2789 micro_sin( &r[0], &r[0] );
2790 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2791 STORE( &r[0], 0, chan_index );
2792 }
2793 break;
2794
2795 case TGSI_OPCODE_SLE:
2796 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2797 FETCH( &r[0], 0, chan_index );
2798 FETCH( &r[1], 1, chan_index );
2799 micro_le(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C]);
2800 }
2801 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2802 STORE(&d[chan_index], 0, chan_index);
2803 }
2804 break;
2805
2806 case TGSI_OPCODE_SNE:
2807 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
2808 FETCH( &r[0], 0, chan_index );
2809 FETCH( &r[1], 1, chan_index );
2810 micro_eq(&d[chan_index], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]);
2811 }
2812 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2813 STORE(&d[chan_index], 0, chan_index);
2814 }
2815 break;
2816
2817 case TGSI_OPCODE_STR:
2818 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
2819 STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, chan_index);
2820 }
2821 break;
2822
2823 case TGSI_OPCODE_TEX:
2824 /* simple texture lookup */
2825 /* src[0] = texcoord */
2826 /* src[1] = sampler unit */
2827 exec_tex(mach, inst, FALSE, FALSE);
2828 break;
2829
2830 case TGSI_OPCODE_TXB:
2831 /* Texture lookup with lod bias */
2832 /* src[0] = texcoord (src[0].w = LOD bias) */
2833 /* src[1] = sampler unit */
2834 exec_tex(mach, inst, TRUE, FALSE);
2835 break;
2836
2837 case TGSI_OPCODE_TXD:
2838 /* Texture lookup with explict partial derivatives */
2839 /* src[0] = texcoord */
2840 /* src[1] = d[strq]/dx */
2841 /* src[2] = d[strq]/dy */
2842 /* src[3] = sampler unit */
2843 exec_txd(mach, inst);
2844 break;
2845
2846 case TGSI_OPCODE_TXL:
2847 /* Texture lookup with explit LOD */
2848 /* src[0] = texcoord (src[0].w = LOD) */
2849 /* src[1] = sampler unit */
2850 exec_tex(mach, inst, TRUE, FALSE);
2851 break;
2852
2853 case TGSI_OPCODE_TXP:
2854 /* Texture lookup with projection */
2855 /* src[0] = texcoord (src[0].w = projection) */
2856 /* src[1] = sampler unit */
2857 exec_tex(mach, inst, FALSE, TRUE);
2858 break;
2859
2860 case TGSI_OPCODE_UP2H:
2861 assert (0);
2862 break;
2863
2864 case TGSI_OPCODE_UP2US:
2865 assert (0);
2866 break;
2867
2868 case TGSI_OPCODE_UP4B:
2869 assert (0);
2870 break;
2871
2872 case TGSI_OPCODE_UP4UB:
2873 assert (0);
2874 break;
2875
2876 case TGSI_OPCODE_X2D:
2877 FETCH(&r[0], 1, CHAN_X);
2878 FETCH(&r[1], 1, CHAN_Y);
2879 if (IS_CHANNEL_ENABLED(*inst, CHAN_X) ||
2880 IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
2881 FETCH(&r[2], 2, CHAN_X);
2882 micro_mul(&r[2], &r[2], &r[0]);
2883 FETCH(&r[3], 2, CHAN_Y);
2884 micro_mul(&r[3], &r[3], &r[1]);
2885 micro_add(&r[2], &r[2], &r[3]);
2886 FETCH(&r[3], 0, CHAN_X);
2887 micro_add(&d[CHAN_X], &r[2], &r[3]);
2888
2889 }
2890 if (IS_CHANNEL_ENABLED(*inst, CHAN_Y) ||
2891 IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
2892 FETCH(&r[2], 2, CHAN_Z);
2893 micro_mul(&r[2], &r[2], &r[0]);
2894 FETCH(&r[3], 2, CHAN_W);
2895 micro_mul(&r[3], &r[3], &r[1]);
2896 micro_add(&r[2], &r[2], &r[3]);
2897 FETCH(&r[3], 0, CHAN_Y);
2898 micro_add(&d[CHAN_Y], &r[2], &r[3]);
2899
2900 }
2901 if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
2902 STORE(&d[CHAN_X], 0, CHAN_X);
2903 }
2904 if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
2905 STORE(&d[CHAN_Y], 0, CHAN_Y);
2906 }
2907 if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
2908 STORE(&d[CHAN_X], 0, CHAN_Z);
2909 }
2910 if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
2911 STORE(&d[CHAN_Y], 0, CHAN_W);
2912 }
2913 break;
2914
2915 case TGSI_OPCODE_ARA:
2916 assert (0);
2917 break;
2918
2919 case TGSI_OPCODE_BRA:
2920 assert (0);
2921 break;
2922
2923 case TGSI_OPCODE_CAL:
2924 /* skip the call if no execution channels are enabled */
2925 if (mach->ExecMask) {
2926 /* do the call */
2927
2928 /* First, record the depths of the execution stacks.
2929 * This is important for deeply nested/looped return statements.
2930 * We have to unwind the stacks by the correct amount. For a
2931 * real code generator, we could determine the number of entries
2932 * to pop off each stack with simple static analysis and avoid
2933 * implementing this data structure at run time.
2934 */
2935 mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop;
2936 mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop;
2937 mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop;
2938 mach->CallStack[mach->CallStackTop].SwitchStackTop = mach->SwitchStackTop;
2939 mach->CallStack[mach->CallStackTop].BreakStackTop = mach->BreakStackTop;
2940 /* note that PC was already incremented above */
2941 mach->CallStack[mach->CallStackTop].ReturnAddr = *pc;
2942
2943 mach->CallStackTop++;
2944
2945 /* Second, push the Cond, Loop, Cont, Func stacks */
2946 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
2947 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
2948 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
2949 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING);
2950 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
2951 assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
2952
2953 mach->CondStack[mach->CondStackTop++] = mach->CondMask;
2954 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
2955 mach->ContStack[mach->ContStackTop++] = mach->ContMask;
2956 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch;
2957 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
2958 mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
2959
2960 /* Finally, jump to the subroutine */
2961 *pc = inst->Label.Label;
2962 }
2963 break;
2964
2965 case TGSI_OPCODE_RET:
2966 mach->FuncMask &= ~mach->ExecMask;
2967 UPDATE_EXEC_MASK(mach);
2968
2969 if (mach->FuncMask == 0x0) {
2970 /* really return now (otherwise, keep executing */
2971
2972 if (mach->CallStackTop == 0) {
2973 /* returning from main() */
2974 *pc = -1;
2975 return;
2976 }
2977
2978 assert(mach->CallStackTop > 0);
2979 mach->CallStackTop--;
2980
2981 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
2982 mach->CondMask = mach->CondStack[mach->CondStackTop];
2983
2984 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
2985 mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
2986
2987 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
2988 mach->ContMask = mach->ContStack[mach->ContStackTop];
2989
2990 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop;
2991 mach->Switch = mach->SwitchStack[mach->SwitchStackTop];
2992
2993 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop;
2994 mach->BreakType = mach->BreakStack[mach->BreakStackTop];
2995
2996 assert(mach->FuncStackTop > 0);
2997 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
2998
2999 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
3000
3001 UPDATE_EXEC_MASK(mach);
3002 }
3003 break;
3004
3005 case TGSI_OPCODE_SSG:
3006 /* TGSI_OPCODE_SGN */
3007 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
3008 FETCH( &r[0], 0, chan_index );
3009 micro_sgn(&d[chan_index], &r[0]);
3010 }
3011 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
3012 STORE(&d[chan_index], 0, chan_index);
3013 }
3014 break;
3015
3016 case TGSI_OPCODE_CMP:
3017 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
3018 FETCH(&r[0], 0, chan_index);
3019 FETCH(&r[1], 1, chan_index);
3020 FETCH(&r[2], 2, chan_index);
3021 micro_lt(&d[chan_index], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[1], &r[2]);
3022 }
3023 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
3024 STORE(&d[chan_index], 0, chan_index);
3025 }
3026 break;
3027
3028 case TGSI_OPCODE_SCS:
3029 if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) || IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
3030 FETCH( &r[0], 0, CHAN_X );
3031 if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
3032 micro_cos(&r[1], &r[0]);
3033 STORE(&r[1], 0, CHAN_X);
3034 }
3035 if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
3036 micro_sin(&r[1], &r[0]);
3037 STORE(&r[1], 0, CHAN_Y);
3038 }
3039 }
3040 if( IS_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
3041 STORE( &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], 0, CHAN_Z );
3042 }
3043 if( IS_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
3044 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
3045 }
3046 break;
3047
3048 case TGSI_OPCODE_NRM:
3049 /* 3-component vector normalize */
3050 if(IS_CHANNEL_ENABLED(*inst, CHAN_X) ||
3051 IS_CHANNEL_ENABLED(*inst, CHAN_Y) ||
3052 IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
3053 /* r3 = sqrt(dp3(src0, src0)) */
3054 FETCH(&r[0], 0, CHAN_X);
3055 micro_mul(&r[3], &r[0], &r[0]);
3056 FETCH(&r[1], 0, CHAN_Y);
3057 micro_mul(&r[4], &r[1], &r[1]);
3058 micro_add(&r[3], &r[3], &r[4]);
3059 FETCH(&r[2], 0, CHAN_Z);
3060 micro_mul(&r[4], &r[2], &r[2]);
3061 micro_add(&r[3], &r[3], &r[4]);
3062 micro_sqrt(&r[3], &r[3]);
3063
3064 if (IS_CHANNEL_ENABLED(*inst, CHAN_X)) {
3065 micro_div(&r[0], &r[0], &r[3]);
3066 STORE(&r[0], 0, CHAN_X);
3067 }
3068 if (IS_CHANNEL_ENABLED(*inst, CHAN_Y)) {
3069 micro_div(&r[1], &r[1], &r[3]);
3070 STORE(&r[1], 0, CHAN_Y);
3071 }
3072 if (IS_CHANNEL_ENABLED(*inst, CHAN_Z)) {
3073 micro_div(&r[2], &r[2], &r[3]);
3074 STORE(&r[2], 0, CHAN_Z);
3075 }
3076 }
3077 if (IS_CHANNEL_ENABLED(*inst, CHAN_W)) {
3078 STORE(&mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W);
3079 }
3080 break;
3081
3082 case TGSI_OPCODE_NRM4:
3083 /* 4-component vector normalize */
3084 {
3085 union tgsi_exec_channel tmp, dot;
3086
3087 /* tmp = dp4(src0, src0): */
3088 FETCH( &r[0], 0, CHAN_X );
3089 micro_mul( &tmp, &r[0], &r[0] );
3090
3091 FETCH( &r[1], 0, CHAN_Y );
3092 micro_mul( &dot, &r[1], &r[1] );
3093 micro_add( &tmp, &tmp, &dot );
3094
3095 FETCH( &r[2], 0, CHAN_Z );
3096 micro_mul( &dot, &r[2], &r[2] );
3097 micro_add( &tmp, &tmp, &dot );
3098
3099 FETCH( &r[3], 0, CHAN_W );
3100 micro_mul( &dot, &r[3], &r[3] );
3101 micro_add( &tmp, &tmp, &dot );
3102
3103 /* tmp = 1 / sqrt(tmp) */
3104 micro_sqrt( &tmp, &tmp );
3105 micro_div( &tmp, &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &tmp );
3106
3107 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
3108 /* chan = chan * tmp */
3109 micro_mul( &r[chan_index], &tmp, &r[chan_index] );
3110 STORE( &r[chan_index], 0, chan_index );
3111 }
3112 }
3113 break;
3114
3115 case TGSI_OPCODE_DIV:
3116 assert( 0 );
3117 break;
3118
3119 case TGSI_OPCODE_DP2:
3120 FETCH( &r[0], 0, CHAN_X );
3121 FETCH( &r[1], 1, CHAN_X );
3122 micro_mul( &r[0], &r[0], &r[1] );
3123
3124 FETCH( &r[1], 0, CHAN_Y );
3125 FETCH( &r[2], 1, CHAN_Y );
3126 micro_mul( &r[1], &r[1], &r[2] );
3127 micro_add( &r[0], &r[0], &r[1] );
3128
3129 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
3130 STORE( &r[0], 0, chan_index );
3131 }
3132 break;
3133
3134 case TGSI_OPCODE_IF:
3135 /* push CondMask */
3136 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
3137 mach->CondStack[mach->CondStackTop++] = mach->CondMask;
3138 FETCH( &r[0], 0, CHAN_X );
3139 /* update CondMask */
3140 if( ! r[0].u[0] ) {
3141 mach->CondMask &= ~0x1;
3142 }
3143 if( ! r[0].u[1] ) {
3144 mach->CondMask &= ~0x2;
3145 }
3146 if( ! r[0].u[2] ) {
3147 mach->CondMask &= ~0x4;
3148 }
3149 if( ! r[0].u[3] ) {
3150 mach->CondMask &= ~0x8;
3151 }
3152 UPDATE_EXEC_MASK(mach);
3153 /* Todo: If CondMask==0, jump to ELSE */
3154 break;
3155
3156 case TGSI_OPCODE_ELSE:
3157 /* invert CondMask wrt previous mask */
3158 {
3159 uint prevMask;
3160 assert(mach->CondStackTop > 0);
3161 prevMask = mach->CondStack[mach->CondStackTop - 1];
3162 mach->CondMask = ~mach->CondMask & prevMask;
3163 UPDATE_EXEC_MASK(mach);
3164 /* Todo: If CondMask==0, jump to ENDIF */
3165 }
3166 break;
3167
3168 case TGSI_OPCODE_ENDIF:
3169 /* pop CondMask */
3170 assert(mach->CondStackTop > 0);
3171 mach->CondMask = mach->CondStack[--mach->CondStackTop];
3172 UPDATE_EXEC_MASK(mach);
3173 break;
3174
3175 case TGSI_OPCODE_END:
3176 /* halt execution */
3177 *pc = -1;
3178 break;
3179
3180 case TGSI_OPCODE_REP:
3181 assert (0);
3182 break;
3183
3184 case TGSI_OPCODE_ENDREP:
3185 assert (0);
3186 break;
3187
3188 case TGSI_OPCODE_PUSHA:
3189 assert (0);
3190 break;
3191
3192 case TGSI_OPCODE_POPA:
3193 assert (0);
3194 break;
3195
3196 case TGSI_OPCODE_CEIL:
3197 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
3198 FETCH( &r[0], 0, chan_index );
3199 micro_ceil(&d[chan_index], &r[0]);
3200 }
3201 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
3202 STORE(&d[chan_index], 0, chan_index);
3203 }
3204 break;
3205
3206 case TGSI_OPCODE_I2F:
3207 exec_vector_unary(mach, inst, micro_i2f);
3208 break;
3209
3210 case TGSI_OPCODE_NOT:
3211 exec_vector_unary(mach, inst, micro_not);
3212 break;
3213
3214 case TGSI_OPCODE_TRUNC:
3215 FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
3216 FETCH( &r[0], 0, chan_index );
3217 micro_trunc(&d[chan_index], &r[0]);
3218 }
3219 FOR_EACH_ENABLED_CHANNEL(*inst, chan_index) {
3220 STORE(&d[chan_index], 0, chan_index);
3221 }
3222 break;
3223
3224 case TGSI_OPCODE_SHL:
3225 exec_vector_binary(mach, inst, micro_shl);
3226 break;
3227
3228 case TGSI_OPCODE_AND:
3229 exec_vector_binary(mach, inst, micro_and);
3230 break;
3231
3232 case TGSI_OPCODE_OR:
3233 exec_vector_binary(mach, inst, micro_or);
3234 break;
3235
3236 case TGSI_OPCODE_MOD:
3237 assert (0);
3238 break;
3239
3240 case TGSI_OPCODE_XOR:
3241 exec_vector_binary(mach, inst, micro_xor);
3242 break;
3243
3244 case TGSI_OPCODE_SAD:
3245 assert (0);
3246 break;
3247
3248 case TGSI_OPCODE_TXF:
3249 assert (0);
3250 break;
3251
3252 case TGSI_OPCODE_TXQ:
3253 assert (0);
3254 break;
3255
3256 case TGSI_OPCODE_EMIT:
3257 emit_vertex(mach);
3258 break;
3259
3260 case TGSI_OPCODE_ENDPRIM:
3261 emit_primitive(mach);
3262 break;
3263
3264 case TGSI_OPCODE_BGNFOR:
3265 assert(mach->LoopCounterStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3266 for (chan_index = 0; chan_index < 3; chan_index++) {
3267 FETCH( &mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[chan_index], 0, chan_index );
3268 }
3269 ++mach->LoopCounterStackTop;
3270 STORE(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X], 0, CHAN_X);
3271 /* update LoopMask */
3272 if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[0] <= 0.0f) {
3273 mach->LoopMask &= ~0x1;
3274 }
3275 if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[1] <= 0.0f) {
3276 mach->LoopMask &= ~0x2;
3277 }
3278 if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[2] <= 0.0f) {
3279 mach->LoopMask &= ~0x4;
3280 }
3281 if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[3] <= 0.0f) {
3282 mach->LoopMask &= ~0x8;
3283 }
3284 /* TODO: if mach->LoopMask == 0, jump to end of loop */
3285 UPDATE_EXEC_MASK(mach);
3286 /* fall-through (for now) */
3287 case TGSI_OPCODE_BGNLOOP:
3288 /* push LoopMask and ContMasks */
3289 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3290 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3291 assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
3292 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK);
3293
3294 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
3295 mach->ContStack[mach->ContStackTop++] = mach->ContMask;
3296 mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1;
3297 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType;
3298 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP;
3299 break;
3300
3301 case TGSI_OPCODE_ENDFOR:
3302 assert(mach->LoopCounterStackTop > 0);
3303 micro_sub(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y],
3304 &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y],
3305 &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C]);
3306 /* update LoopMask */
3307 if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[0] <= 0.0f) {
3308 mach->LoopMask &= ~0x1;
3309 }
3310 if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[1] <= 0.0f) {
3311 mach->LoopMask &= ~0x2;
3312 }
3313 if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[2] <= 0.0f) {
3314 mach->LoopMask &= ~0x4;
3315 }
3316 if (mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Y].f[3] <= 0.0f) {
3317 mach->LoopMask &= ~0x8;
3318 }
3319 micro_add(&mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X],
3320 &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_X],
3321 &mach->LoopCounterStack[mach->LoopCounterStackTop - 1].xyzw[CHAN_Z]);
3322 assert(mach->LoopLabelStackTop > 0);
3323 inst = mach->Instructions + mach->LoopLabelStack[mach->LoopLabelStackTop - 1];
3324 STORE(&mach->LoopCounterStack[mach->LoopCounterStackTop].xyzw[CHAN_X], 0, CHAN_X);
3325 /* Restore ContMask, but don't pop */
3326 assert(mach->ContStackTop > 0);
3327 mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
3328 UPDATE_EXEC_MASK(mach);
3329 if (mach->ExecMask) {
3330 /* repeat loop: jump to instruction just past BGNLOOP */
3331 assert(mach->LoopLabelStackTop > 0);
3332 *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1;
3333 }
3334 else {
3335 /* exit loop: pop LoopMask */
3336 assert(mach->LoopStackTop > 0);
3337 mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
3338 /* pop ContMask */
3339 assert(mach->ContStackTop > 0);
3340 mach->ContMask = mach->ContStack[--mach->ContStackTop];
3341 assert(mach->LoopLabelStackTop > 0);
3342 --mach->LoopLabelStackTop;
3343 assert(mach->LoopCounterStackTop > 0);
3344 --mach->LoopCounterStackTop;
3345
3346 mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
3347 }
3348 UPDATE_EXEC_MASK(mach);
3349 break;
3350
3351 case TGSI_OPCODE_ENDLOOP:
3352 /* Restore ContMask, but don't pop */
3353 assert(mach->ContStackTop > 0);
3354 mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
3355 UPDATE_EXEC_MASK(mach);
3356 if (mach->ExecMask) {
3357 /* repeat loop: jump to instruction just past BGNLOOP */
3358 assert(mach->LoopLabelStackTop > 0);
3359 *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1;
3360 }
3361 else {
3362 /* exit loop: pop LoopMask */
3363 assert(mach->LoopStackTop > 0);
3364 mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
3365 /* pop ContMask */
3366 assert(mach->ContStackTop > 0);
3367 mach->ContMask = mach->ContStack[--mach->ContStackTop];
3368 assert(mach->LoopLabelStackTop > 0);
3369 --mach->LoopLabelStackTop;
3370
3371 mach->BreakType = mach->BreakStack[--mach->BreakStackTop];
3372 }
3373 UPDATE_EXEC_MASK(mach);
3374 break;
3375
3376 case TGSI_OPCODE_BRK:
3377 exec_break(mach);
3378 break;
3379
3380 case TGSI_OPCODE_CONT:
3381 /* turn off cont channels for each enabled exec channel */
3382 mach->ContMask &= ~mach->ExecMask;
3383 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3384 UPDATE_EXEC_MASK(mach);
3385 break;
3386
3387 case TGSI_OPCODE_BGNSUB:
3388 /* no-op */
3389 break;
3390
3391 case TGSI_OPCODE_ENDSUB:
3392 /*
3393 * XXX: This really should be a no-op. We should never reach this opcode.
3394 */
3395
3396 assert(mach->CallStackTop > 0);
3397 mach->CallStackTop--;
3398
3399 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop;
3400 mach->CondMask = mach->CondStack[mach->CondStackTop];
3401
3402 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop;
3403 mach->LoopMask = mach->LoopStack[mach->LoopStackTop];
3404
3405 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop;
3406 mach->ContMask = mach->ContStack[mach->ContStackTop];
3407
3408 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop;
3409 mach->Switch = mach->SwitchStack[mach->SwitchStackTop];
3410
3411 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop;
3412 mach->BreakType = mach->BreakStack[mach->BreakStackTop];
3413
3414 assert(mach->FuncStackTop > 0);
3415 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
3416
3417 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr;
3418
3419 UPDATE_EXEC_MASK(mach);
3420 break;
3421
3422 case TGSI_OPCODE_NOP:
3423 break;
3424
3425 case TGSI_OPCODE_BREAKC:
3426 FETCH(&r[0], 0, CHAN_X);
3427 /* update CondMask */
3428 if (r[0].u[0] && (mach->ExecMask & 0x1)) {
3429 mach->LoopMask &= ~0x1;
3430 }
3431 if (r[0].u[1] && (mach->ExecMask & 0x2)) {
3432 mach->LoopMask &= ~0x2;
3433 }
3434 if (r[0].u[2] && (mach->ExecMask & 0x4)) {
3435 mach->LoopMask &= ~0x4;
3436 }
3437 if (r[0].u[3] && (mach->ExecMask & 0x8)) {
3438 mach->LoopMask &= ~0x8;
3439 }
3440 /* Todo: if mach->LoopMask == 0, jump to end of loop */
3441 UPDATE_EXEC_MASK(mach);
3442 break;
3443
3444 case TGSI_OPCODE_F2I:
3445 exec_vector_unary(mach, inst, micro_f2i);
3446 break;
3447
3448 case TGSI_OPCODE_IDIV:
3449 exec_vector_binary(mach, inst, micro_idiv);
3450 break;
3451
3452 case TGSI_OPCODE_IMAX:
3453 exec_vector_binary(mach, inst, micro_imax);
3454 break;
3455
3456 case TGSI_OPCODE_IMIN:
3457 exec_vector_binary(mach, inst, micro_imin);
3458 break;
3459
3460 case TGSI_OPCODE_INEG:
3461 exec_vector_unary(mach, inst, micro_ineg);
3462 break;
3463
3464 case TGSI_OPCODE_ISGE:
3465 exec_vector_binary(mach, inst, micro_isge);
3466 break;
3467
3468 case TGSI_OPCODE_ISHR:
3469 exec_vector_binary(mach, inst, micro_ishr);
3470 break;
3471
3472 case TGSI_OPCODE_ISLT:
3473 exec_vector_binary(mach, inst, micro_islt);
3474 break;
3475
3476 case TGSI_OPCODE_F2U:
3477 exec_vector_unary(mach, inst, micro_f2u);
3478 break;
3479
3480 case TGSI_OPCODE_U2F:
3481 exec_vector_unary(mach, inst, micro_u2f);
3482 break;
3483
3484 case TGSI_OPCODE_UADD:
3485 exec_vector_binary(mach, inst, micro_uadd);
3486 break;
3487
3488 case TGSI_OPCODE_UDIV:
3489 exec_vector_binary(mach, inst, micro_udiv);
3490 break;
3491
3492 case TGSI_OPCODE_UMAD:
3493 exec_vector_trinary(mach, inst, micro_umad);
3494 break;
3495
3496 case TGSI_OPCODE_UMAX:
3497 exec_vector_binary(mach, inst, micro_umax);
3498 break;
3499
3500 case TGSI_OPCODE_UMIN:
3501 exec_vector_binary(mach, inst, micro_umin);
3502 break;
3503
3504 case TGSI_OPCODE_UMOD:
3505 exec_vector_binary(mach, inst, micro_umod);
3506 break;
3507
3508 case TGSI_OPCODE_UMUL:
3509 exec_vector_binary(mach, inst, micro_umul);
3510 break;
3511
3512 case TGSI_OPCODE_USEQ:
3513 exec_vector_binary(mach, inst, micro_useq);
3514 break;
3515
3516 case TGSI_OPCODE_USGE:
3517 exec_vector_binary(mach, inst, micro_usge);
3518 break;
3519
3520 case TGSI_OPCODE_USHR:
3521 exec_vector_binary(mach, inst, micro_ushr);
3522 break;
3523
3524 case TGSI_OPCODE_USLT:
3525 exec_vector_binary(mach, inst, micro_uslt);
3526 break;
3527
3528 case TGSI_OPCODE_USNE:
3529 exec_vector_binary(mach, inst, micro_usne);
3530 break;
3531
3532 case TGSI_OPCODE_SWITCH:
3533 exec_switch(mach, inst);
3534 break;
3535
3536 case TGSI_OPCODE_CASE:
3537 exec_case(mach, inst);
3538 break;
3539
3540 case TGSI_OPCODE_DEFAULT:
3541 exec_default(mach);
3542 break;
3543
3544 case TGSI_OPCODE_ENDSWITCH:
3545 exec_endswitch(mach);
3546 break;
3547
3548 default:
3549 assert( 0 );
3550 }
3551 }
3552
3553 #define DEBUG_EXECUTION 0
3554
3555
3556 /**
3557 * Run TGSI interpreter.
3558 * \return bitmask of "alive" quad components
3559 */
3560 uint
3561 tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
3562 {
3563 uint i;
3564 int pc = 0;
3565
3566 mach->CondMask = 0xf;
3567 mach->LoopMask = 0xf;
3568 mach->ContMask = 0xf;
3569 mach->FuncMask = 0xf;
3570 mach->ExecMask = 0xf;
3571
3572 mach->Switch.mask = 0xf;
3573
3574 assert(mach->CondStackTop == 0);
3575 assert(mach->LoopStackTop == 0);
3576 assert(mach->ContStackTop == 0);
3577 assert(mach->SwitchStackTop == 0);
3578 assert(mach->BreakStackTop == 0);
3579 assert(mach->CallStackTop == 0);
3580
3581 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0;
3582 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0;
3583
3584 if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) {
3585 mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0;
3586 mach->Primitives[0] = 0;
3587 }
3588
3589 for (i = 0; i < QUAD_SIZE; i++) {
3590 mach->Temps[TEMP_CC_I].xyzw[TEMP_CC_C].u[i] =
3591 (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_X_SHIFT) |
3592 (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_Y_SHIFT) |
3593 (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_Z_SHIFT) |
3594 (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_W_SHIFT);
3595 }
3596
3597 /* execute declarations (interpolants) */
3598 for (i = 0; i < mach->NumDeclarations; i++) {
3599 exec_declaration( mach, mach->Declarations+i );
3600 }
3601
3602 {
3603 #if DEBUG_EXECUTION
3604 struct tgsi_exec_vector temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS];
3605 struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS];
3606 uint inst = 1;
3607
3608 memcpy(temps, mach->Temps, sizeof(temps));
3609 memcpy(outputs, mach->Outputs, sizeof(outputs));
3610 #endif
3611
3612 /* execute instructions, until pc is set to -1 */
3613 while (pc != -1) {
3614
3615 #if DEBUG_EXECUTION
3616 uint i;
3617
3618 tgsi_dump_instruction(&mach->Instructions[pc], inst++);
3619 #endif
3620
3621 assert(pc < (int) mach->NumInstructions);
3622 exec_instruction(mach, mach->Instructions + pc, &pc);
3623
3624 #if DEBUG_EXECUTION
3625 for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) {
3626 if (memcmp(&temps[i], &mach->Temps[i], sizeof(temps[i]))) {
3627 uint j;
3628
3629 memcpy(&temps[i], &mach->Temps[i], sizeof(temps[i]));
3630 debug_printf("TEMP[%2u] = ", i);
3631 for (j = 0; j < 4; j++) {
3632 if (j > 0) {
3633 debug_printf(" ");
3634 }
3635 debug_printf("(%6f, %6f, %6f, %6f)\n",
3636 temps[i].xyzw[0].f[j],
3637 temps[i].xyzw[1].f[j],
3638 temps[i].xyzw[2].f[j],
3639 temps[i].xyzw[3].f[j]);
3640 }
3641 }
3642 }
3643 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {
3644 if (memcmp(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]))) {
3645 uint j;
3646
3647 memcpy(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]));
3648 debug_printf("OUT[%2u] = ", i);
3649 for (j = 0; j < 4; j++) {
3650 if (j > 0) {
3651 debug_printf(" ");
3652 }
3653 debug_printf("{%6f, %6f, %6f, %6f}\n",
3654 outputs[i].xyzw[0].f[j],
3655 outputs[i].xyzw[1].f[j],
3656 outputs[i].xyzw[2].f[j],
3657 outputs[i].xyzw[3].f[j]);
3658 }
3659 }
3660 }
3661 #endif
3662 }
3663 }
3664
3665 #if 0
3666 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
3667 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
3668 /*
3669 * Scale back depth component.
3670 */
3671 for (i = 0; i < 4; i++)
3672 mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF;
3673 }
3674 #endif
3675
3676 assert(mach->CondStackTop == 0);
3677 assert(mach->LoopStackTop == 0);
3678 assert(mach->ContStackTop == 0);
3679 assert(mach->SwitchStackTop == 0);
3680 assert(mach->BreakStackTop == 0);
3681 assert(mach->CallStackTop == 0);
3682
3683 return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
3684 }