projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
mesa: rework ParameterList to allow packing
[mesa.git] / src / mesa / program / prog_execute.c
1 /*
2 * Mesa 3-D graphics library
3 *
4 * Copyright (C) 1999-2008 Brian Paul 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 "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25 /**
26 * \file prog_execute.c
27 * Software interpreter for vertex/fragment programs.
28 * \author Brian Paul
29 */
30
31 /*
32 * NOTE: we do everything in single-precision floating point; we don't
33 * currently observe the single/half/fixed-precision qualifiers.
34 *
35 */
36
37
38 #include "c99_math.h"
39 #include "main/glheader.h"
40 #include "main/macros.h"
41 #include "prog_execute.h"
42 #include "prog_instruction.h"
43 #include "prog_parameter.h"
44 #include "prog_print.h"
45 #include "prog_noise.h"
46
47
48 /* debug predicate */
49 #define DEBUG_PROG 0
50
51
52 /**
53 * Set x to positive or negative infinity.
54 */
55 #define SET_POS_INFINITY(x) \
56 do { \
57 fi_type fi; \
58 fi.i = 0x7F800000; \
59 x = fi.f; \
60 } while (0)
61 #define SET_NEG_INFINITY(x) \
62 do { \
63 fi_type fi; \
64 fi.i = 0xFF800000; \
65 x = fi.f; \
66 } while (0)
67
68 #define SET_FLOAT_BITS(x, bits) ((fi_type *) (void *) &(x))->i = bits
69
70
71 static const GLfloat ZeroVec[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
72
73
74 /**
75 * Return a pointer to the 4-element float vector specified by the given
76 * source register.
77 */
78 static inline const GLfloat *
79 get_src_register_pointer(const struct prog_src_register *source,
80 const struct gl_program_machine *machine)
81 {
82 const struct gl_program *prog = machine->CurProgram;
83 GLint reg = source->Index;
84
85 if (source->RelAddr) {
86 /* add address register value to src index/offset */
87 reg += machine->AddressReg[0][0];
88 if (reg < 0) {
89 return ZeroVec;
90 }
91 }
92
93 switch (source->File) {
94 case PROGRAM_TEMPORARY:
95 if (reg >= MAX_PROGRAM_TEMPS)
96 return ZeroVec;
97 return machine->Temporaries[reg];
98
99 case PROGRAM_INPUT:
100 if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
101 if (reg >= VERT_ATTRIB_MAX)
102 return ZeroVec;
103 return machine->VertAttribs[reg];
104 }
105 else {
106 if (reg >= VARYING_SLOT_MAX)
107 return ZeroVec;
108 return machine->Attribs[reg][machine->CurElement];
109 }
110
111 case PROGRAM_OUTPUT:
112 if (reg >= MAX_PROGRAM_OUTPUTS)
113 return ZeroVec;
114 return machine->Outputs[reg];
115
116 case PROGRAM_STATE_VAR:
117 /* Fallthrough */
118 case PROGRAM_CONSTANT:
119 /* Fallthrough */
120 case PROGRAM_UNIFORM: {
121 if (reg >= (GLint) prog->Parameters->NumParameters)
122 return ZeroVec;
123
124 unsigned pvo = prog->Parameters->ParameterValueOffset[reg];
125 return (GLfloat *) prog->Parameters->ParameterValues + pvo;
126 }
127 case PROGRAM_SYSTEM_VALUE:
128 assert(reg < (GLint) ARRAY_SIZE(machine->SystemValues));
129 return machine->SystemValues[reg];
130
131 default:
132 _mesa_problem(NULL,
133 "Invalid src register file %d in get_src_register_pointer()",
134 source->File);
135 return ZeroVec;
136 }
137 }
138
139
140 /**
141 * Return a pointer to the 4-element float vector specified by the given
142 * destination register.
143 */
144 static inline GLfloat *
145 get_dst_register_pointer(const struct prog_dst_register *dest,
146 struct gl_program_machine *machine)
147 {
148 static GLfloat dummyReg[4];
149 GLint reg = dest->Index;
150
151 if (dest->RelAddr) {
152 /* add address register value to src index/offset */
153 reg += machine->AddressReg[0][0];
154 if (reg < 0) {
155 return dummyReg;
156 }
157 }
158
159 switch (dest->File) {
160 case PROGRAM_TEMPORARY:
161 if (reg >= MAX_PROGRAM_TEMPS)
162 return dummyReg;
163 return machine->Temporaries[reg];
164
165 case PROGRAM_OUTPUT:
166 if (reg >= MAX_PROGRAM_OUTPUTS)
167 return dummyReg;
168 return machine->Outputs[reg];
169
170 default:
171 _mesa_problem(NULL,
172 "Invalid dest register file %d in get_dst_register_pointer()",
173 dest->File);
174 return dummyReg;
175 }
176 }
177
178
179
180 /**
181 * Fetch a 4-element float vector from the given source register.
182 * Apply swizzling and negating as needed.
183 */
184 static void
185 fetch_vector4(const struct prog_src_register *source,
186 const struct gl_program_machine *machine, GLfloat result[4])
187 {
188 const GLfloat *src = get_src_register_pointer(source, machine);
189
190 if (source->Swizzle == SWIZZLE_NOOP) {
191 /* no swizzling */
192 COPY_4V(result, src);
193 }
194 else {
195 assert(GET_SWZ(source->Swizzle, 0) <= 3);
196 assert(GET_SWZ(source->Swizzle, 1) <= 3);
197 assert(GET_SWZ(source->Swizzle, 2) <= 3);
198 assert(GET_SWZ(source->Swizzle, 3) <= 3);
199 result[0] = src[GET_SWZ(source->Swizzle, 0)];
200 result[1] = src[GET_SWZ(source->Swizzle, 1)];
201 result[2] = src[GET_SWZ(source->Swizzle, 2)];
202 result[3] = src[GET_SWZ(source->Swizzle, 3)];
203 }
204
205 if (source->Negate) {
206 assert(source->Negate == NEGATE_XYZW);
207 result[0] = -result[0];
208 result[1] = -result[1];
209 result[2] = -result[2];
210 result[3] = -result[3];
211 }
212
213 #ifdef NAN_CHECK
214 assert(!IS_INF_OR_NAN(result[0]));
215 assert(!IS_INF_OR_NAN(result[0]));
216 assert(!IS_INF_OR_NAN(result[0]));
217 assert(!IS_INF_OR_NAN(result[0]));
218 #endif
219 }
220
221
222 /**
223 * Fetch the derivative with respect to X or Y for the given register.
224 * XXX this currently only works for fragment program input attribs.
225 */
226 static void
227 fetch_vector4_deriv(const struct prog_src_register *source,
228 const struct gl_program_machine *machine,
229 char xOrY, GLfloat result[4])
230 {
231 if (source->File == PROGRAM_INPUT &&
232 source->Index < (GLint) machine->NumDeriv) {
233 const GLint col = machine->CurElement;
234 const GLfloat w = machine->Attribs[VARYING_SLOT_POS][col][3];
235 const GLfloat invQ = 1.0f / w;
236 GLfloat deriv[4];
237
238 if (xOrY == 'X') {
239 deriv[0] = machine->DerivX[source->Index][0] * invQ;
240 deriv[1] = machine->DerivX[source->Index][1] * invQ;
241 deriv[2] = machine->DerivX[source->Index][2] * invQ;
242 deriv[3] = machine->DerivX[source->Index][3] * invQ;
243 }
244 else {
245 deriv[0] = machine->DerivY[source->Index][0] * invQ;
246 deriv[1] = machine->DerivY[source->Index][1] * invQ;
247 deriv[2] = machine->DerivY[source->Index][2] * invQ;
248 deriv[3] = machine->DerivY[source->Index][3] * invQ;
249 }
250
251 result[0] = deriv[GET_SWZ(source->Swizzle, 0)];
252 result[1] = deriv[GET_SWZ(source->Swizzle, 1)];
253 result[2] = deriv[GET_SWZ(source->Swizzle, 2)];
254 result[3] = deriv[GET_SWZ(source->Swizzle, 3)];
255
256 if (source->Negate) {
257 assert(source->Negate == NEGATE_XYZW);
258 result[0] = -result[0];
259 result[1] = -result[1];
260 result[2] = -result[2];
261 result[3] = -result[3];
262 }
263 }
264 else {
265 ASSIGN_4V(result, 0.0, 0.0, 0.0, 0.0);
266 }
267 }
268
269
270 /**
271 * As above, but only return result[0] element.
272 */
273 static void
274 fetch_vector1(const struct prog_src_register *source,
275 const struct gl_program_machine *machine, GLfloat result[4])
276 {
277 const GLfloat *src = get_src_register_pointer(source, machine);
278
279 result[0] = src[GET_SWZ(source->Swizzle, 0)];
280
281 if (source->Negate) {
282 result[0] = -result[0];
283 }
284 }
285
286
287 /**
288 * Fetch texel from texture. Use partial derivatives when possible.
289 */
290 static inline void
291 fetch_texel(struct gl_context *ctx,
292 const struct gl_program_machine *machine,
293 const struct prog_instruction *inst,
294 const GLfloat texcoord[4], GLfloat lodBias,
295 GLfloat color[4])
296 {
297 const GLuint unit = machine->Samplers[inst->TexSrcUnit];
298
299 /* Note: we only have the right derivatives for fragment input attribs.
300 */
301 if (machine->NumDeriv > 0 &&
302 inst->SrcReg[0].File == PROGRAM_INPUT &&
303 inst->SrcReg[0].Index == VARYING_SLOT_TEX0 + inst->TexSrcUnit) {
304 /* simple texture fetch for which we should have derivatives */
305 GLuint attr = inst->SrcReg[0].Index;
306 machine->FetchTexelDeriv(ctx, texcoord,
307 machine->DerivX[attr],
308 machine->DerivY[attr],
309 lodBias, unit, color);
310 }
311 else {
312 machine->FetchTexelLod(ctx, texcoord, lodBias, unit, color);
313 }
314 }
315
316
317 /**
318 * Store 4 floats into a register. Observe the instructions saturate and
319 * set-condition-code flags.
320 */
321 static void
322 store_vector4(const struct prog_instruction *inst,
323 struct gl_program_machine *machine, const GLfloat value[4])
324 {
325 const struct prog_dst_register *dstReg = &(inst->DstReg);
326 const GLboolean clamp = inst->Saturate;
327 GLuint writeMask = dstReg->WriteMask;
328 GLfloat clampedValue[4];
329 GLfloat *dst = get_dst_register_pointer(dstReg, machine);
330
331 #if 0
332 if (value[0] > 1.0e10 ||
333 IS_INF_OR_NAN(value[0]) ||
334 IS_INF_OR_NAN(value[1]) ||
335 IS_INF_OR_NAN(value[2]) || IS_INF_OR_NAN(value[3]))
336 printf("store %g %g %g %g\n", value[0], value[1], value[2], value[3]);
337 #endif
338
339 if (clamp) {
340 clampedValue[0] = CLAMP(value[0], 0.0F, 1.0F);
341 clampedValue[1] = CLAMP(value[1], 0.0F, 1.0F);
342 clampedValue[2] = CLAMP(value[2], 0.0F, 1.0F);
343 clampedValue[3] = CLAMP(value[3], 0.0F, 1.0F);
344 value = clampedValue;
345 }
346
347 #ifdef NAN_CHECK
348 assert(!IS_INF_OR_NAN(value[0]));
349 assert(!IS_INF_OR_NAN(value[0]));
350 assert(!IS_INF_OR_NAN(value[0]));
351 assert(!IS_INF_OR_NAN(value[0]));
352 #endif
353
354 if (writeMask & WRITEMASK_X)
355 dst[0] = value[0];
356 if (writeMask & WRITEMASK_Y)
357 dst[1] = value[1];
358 if (writeMask & WRITEMASK_Z)
359 dst[2] = value[2];
360 if (writeMask & WRITEMASK_W)
361 dst[3] = value[3];
362 }
363
364
365 /**
366 * Execute the given vertex/fragment program.
367 *
368 * \param ctx rendering context
369 * \param program the program to execute
370 * \param machine machine state (must be initialized)
371 * \return GL_TRUE if program completed or GL_FALSE if program executed KIL.
372 */
373 GLboolean
374 _mesa_execute_program(struct gl_context * ctx,
375 const struct gl_program *program,
376 struct gl_program_machine *machine)
377 {
378 const GLuint numInst = program->arb.NumInstructions;
379 const GLuint maxExec = 65536;
380 GLuint pc, numExec = 0;
381
382 machine->CurProgram = program;
383
384 if (DEBUG_PROG) {
385 printf("execute program %u --------------------\n", program->Id);
386 }
387
388 if (program->Target == GL_VERTEX_PROGRAM_ARB) {
389 machine->EnvParams = ctx->VertexProgram.Parameters;
390 }
391 else {
392 machine->EnvParams = ctx->FragmentProgram.Parameters;
393 }
394
395 for (pc = 0; pc < numInst; pc++) {
396 const struct prog_instruction *inst = program->arb.Instructions + pc;
397
398 if (DEBUG_PROG) {
399 _mesa_print_instruction(inst);
400 }
401
402 switch (inst->Opcode) {
403 case OPCODE_ABS:
404 {
405 GLfloat a[4], result[4];
406 fetch_vector4(&inst->SrcReg[0], machine, a);
407 result[0] = fabsf(a[0]);
408 result[1] = fabsf(a[1]);
409 result[2] = fabsf(a[2]);
410 result[3] = fabsf(a[3]);
411 store_vector4(inst, machine, result);
412 }
413 break;
414 case OPCODE_ADD:
415 {
416 GLfloat a[4], b[4], result[4];
417 fetch_vector4(&inst->SrcReg[0], machine, a);
418 fetch_vector4(&inst->SrcReg[1], machine, b);
419 result[0] = a[0] + b[0];
420 result[1] = a[1] + b[1];
421 result[2] = a[2] + b[2];
422 result[3] = a[3] + b[3];
423 store_vector4(inst, machine, result);
424 if (DEBUG_PROG) {
425 printf("ADD (%g %g %g %g) = (%g %g %g %g) + (%g %g %g %g)\n",
426 result[0], result[1], result[2], result[3],
427 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
428 }
429 }
430 break;
431 case OPCODE_ARL:
432 {
433 GLfloat t[4];
434 fetch_vector4(&inst->SrcReg[0], machine, t);
435 machine->AddressReg[0][0] = IFLOOR(t[0]);
436 if (DEBUG_PROG) {
437 printf("ARL %d\n", machine->AddressReg[0][0]);
438 }
439 }
440 break;
441 case OPCODE_BGNLOOP:
442 /* no-op */
443 assert(program->arb.Instructions[inst->BranchTarget].Opcode
444 == OPCODE_ENDLOOP);
445 break;
446 case OPCODE_ENDLOOP:
447 /* subtract 1 here since pc is incremented by for(pc) loop */
448 assert(program->arb.Instructions[inst->BranchTarget].Opcode
449 == OPCODE_BGNLOOP);
450 pc = inst->BranchTarget - 1; /* go to matching BNGLOOP */
451 break;
452 case OPCODE_BGNSUB: /* begin subroutine */
453 break;
454 case OPCODE_ENDSUB: /* end subroutine */
455 break;
456 case OPCODE_BRK: /* break out of loop (conditional) */
457 assert(program->arb.Instructions[inst->BranchTarget].Opcode
458 == OPCODE_ENDLOOP);
459 /* break out of loop */
460 /* pc++ at end of for-loop will put us after the ENDLOOP inst */
461 pc = inst->BranchTarget;
462 break;
463 case OPCODE_CONT: /* continue loop (conditional) */
464 assert(program->arb.Instructions[inst->BranchTarget].Opcode
465 == OPCODE_ENDLOOP);
466 /* continue at ENDLOOP */
467 /* Subtract 1 here since we'll do pc++ at end of for-loop */
468 pc = inst->BranchTarget - 1;
469 break;
470 case OPCODE_CAL: /* Call subroutine (conditional) */
471 /* call the subroutine */
472 if (machine->StackDepth >= MAX_PROGRAM_CALL_DEPTH) {
473 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
474 }
475 machine->CallStack[machine->StackDepth++] = pc + 1; /* next inst */
476 /* Subtract 1 here since we'll do pc++ at end of for-loop */
477 pc = inst->BranchTarget - 1;
478 break;
479 case OPCODE_CMP:
480 {
481 GLfloat a[4], b[4], c[4], result[4];
482 fetch_vector4(&inst->SrcReg[0], machine, a);
483 fetch_vector4(&inst->SrcReg[1], machine, b);
484 fetch_vector4(&inst->SrcReg[2], machine, c);
485 result[0] = a[0] < 0.0F ? b[0] : c[0];
486 result[1] = a[1] < 0.0F ? b[1] : c[1];
487 result[2] = a[2] < 0.0F ? b[2] : c[2];
488 result[3] = a[3] < 0.0F ? b[3] : c[3];
489 store_vector4(inst, machine, result);
490 if (DEBUG_PROG) {
491 printf("CMP (%g %g %g %g) = (%g %g %g %g) < 0 ? (%g %g %g %g) : (%g %g %g %g)\n",
492 result[0], result[1], result[2], result[3],
493 a[0], a[1], a[2], a[3],
494 b[0], b[1], b[2], b[3],
495 c[0], c[1], c[2], c[3]);
496 }
497 }
498 break;
499 case OPCODE_COS:
500 {
501 GLfloat a[4], result[4];
502 fetch_vector1(&inst->SrcReg[0], machine, a);
503 result[0] = result[1] = result[2] = result[3]
504 = cosf(a[0]);
505 store_vector4(inst, machine, result);
506 }
507 break;
508 case OPCODE_DDX: /* Partial derivative with respect to X */
509 {
510 GLfloat result[4];
511 fetch_vector4_deriv(&inst->SrcReg[0], machine, 'X', result);
512 store_vector4(inst, machine, result);
513 }
514 break;
515 case OPCODE_DDY: /* Partial derivative with respect to Y */
516 {
517 GLfloat result[4];
518 fetch_vector4_deriv(&inst->SrcReg[0], machine, 'Y', result);
519 store_vector4(inst, machine, result);
520 }
521 break;
522 case OPCODE_DP2:
523 {
524 GLfloat a[4], b[4], result[4];
525 fetch_vector4(&inst->SrcReg[0], machine, a);
526 fetch_vector4(&inst->SrcReg[1], machine, b);
527 result[0] = result[1] = result[2] = result[3] = DOT2(a, b);
528 store_vector4(inst, machine, result);
529 if (DEBUG_PROG) {
530 printf("DP2 %g = (%g %g) . (%g %g)\n",
531 result[0], a[0], a[1], b[0], b[1]);
532 }
533 }
534 break;
535 case OPCODE_DP3:
536 {
537 GLfloat a[4], b[4], result[4];
538 fetch_vector4(&inst->SrcReg[0], machine, a);
539 fetch_vector4(&inst->SrcReg[1], machine, b);
540 result[0] = result[1] = result[2] = result[3] = DOT3(a, b);
541 store_vector4(inst, machine, result);
542 if (DEBUG_PROG) {
543 printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
544 result[0], a[0], a[1], a[2], b[0], b[1], b[2]);
545 }
546 }
547 break;
548 case OPCODE_DP4:
549 {
550 GLfloat a[4], b[4], result[4];
551 fetch_vector4(&inst->SrcReg[0], machine, a);
552 fetch_vector4(&inst->SrcReg[1], machine, b);
553 result[0] = result[1] = result[2] = result[3] = DOT4(a, b);
554 store_vector4(inst, machine, result);
555 if (DEBUG_PROG) {
556 printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
557 result[0], a[0], a[1], a[2], a[3],
558 b[0], b[1], b[2], b[3]);
559 }
560 }
561 break;
562 case OPCODE_DPH:
563 {
564 GLfloat a[4], b[4], result[4];
565 fetch_vector4(&inst->SrcReg[0], machine, a);
566 fetch_vector4(&inst->SrcReg[1], machine, b);
567 result[0] = result[1] = result[2] = result[3] = DOT3(a, b) + b[3];
568 store_vector4(inst, machine, result);
569 }
570 break;
571 case OPCODE_DST: /* Distance vector */
572 {
573 GLfloat a[4], b[4], result[4];
574 fetch_vector4(&inst->SrcReg[0], machine, a);
575 fetch_vector4(&inst->SrcReg[1], machine, b);
576 result[0] = 1.0F;
577 result[1] = a[1] * b[1];
578 result[2] = a[2];
579 result[3] = b[3];
580 store_vector4(inst, machine, result);
581 }
582 break;
583 case OPCODE_EXP:
584 {
585 GLfloat t[4], q[4], floor_t0;
586 fetch_vector1(&inst->SrcReg[0], machine, t);
587 floor_t0 = floorf(t[0]);
588 if (floor_t0 > FLT_MAX_EXP) {
589 SET_POS_INFINITY(q[0]);
590 SET_POS_INFINITY(q[2]);
591 }
592 else if (floor_t0 < FLT_MIN_EXP) {
593 q[0] = 0.0F;
594 q[2] = 0.0F;
595 }
596 else {
597 q[0] = ldexpf(1.0, (int) floor_t0);
598 /* Note: GL_NV_vertex_program expects
599 * result.z = result.x * APPX(result.y)
600 * We do what the ARB extension says.
601 */
602 q[2] = exp2f(t[0]);
603 }
604 q[1] = t[0] - floor_t0;
605 q[3] = 1.0F;
606 store_vector4( inst, machine, q );
607 }
608 break;
609 case OPCODE_EX2: /* Exponential base 2 */
610 {
611 GLfloat a[4], result[4], val;
612 fetch_vector1(&inst->SrcReg[0], machine, a);
613 val = exp2f(a[0]);
614 /*
615 if (IS_INF_OR_NAN(val))
616 val = 1.0e10;
617 */
618 result[0] = result[1] = result[2] = result[3] = val;
619 store_vector4(inst, machine, result);
620 }
621 break;
622 case OPCODE_FLR:
623 {
624 GLfloat a[4], result[4];
625 fetch_vector4(&inst->SrcReg[0], machine, a);
626 result[0] = floorf(a[0]);
627 result[1] = floorf(a[1]);
628 result[2] = floorf(a[2]);
629 result[3] = floorf(a[3]);
630 store_vector4(inst, machine, result);
631 }
632 break;
633 case OPCODE_FRC:
634 {
635 GLfloat a[4], result[4];
636 fetch_vector4(&inst->SrcReg[0], machine, a);
637 result[0] = a[0] - floorf(a[0]);
638 result[1] = a[1] - floorf(a[1]);
639 result[2] = a[2] - floorf(a[2]);
640 result[3] = a[3] - floorf(a[3]);
641 store_vector4(inst, machine, result);
642 }
643 break;
644 case OPCODE_IF:
645 {
646 GLboolean cond;
647 assert(program->arb.Instructions[inst->BranchTarget].Opcode
648 == OPCODE_ELSE ||
649 program->arb.Instructions[inst->BranchTarget].Opcode
650 == OPCODE_ENDIF);
651 /* eval condition */
652 GLfloat a[4];
653 fetch_vector1(&inst->SrcReg[0], machine, a);
654 cond = (a[0] != 0.0F);
655 if (DEBUG_PROG) {
656 printf("IF: %d\n", cond);
657 }
658 /* do if/else */
659 if (cond) {
660 /* do if-clause (just continue execution) */
661 }
662 else {
663 /* go to the instruction after ELSE or ENDIF */
664 assert(inst->BranchTarget >= 0);
665 pc = inst->BranchTarget;
666 }
667 }
668 break;
669 case OPCODE_ELSE:
670 /* goto ENDIF */
671 assert(program->arb.Instructions[inst->BranchTarget].Opcode
672 == OPCODE_ENDIF);
673 assert(inst->BranchTarget >= 0);
674 pc = inst->BranchTarget;
675 break;
676 case OPCODE_ENDIF:
677 /* nothing */
678 break;
679 case OPCODE_KIL: /* ARB_f_p only */
680 {
681 GLfloat a[4];
682 fetch_vector4(&inst->SrcReg[0], machine, a);
683 if (DEBUG_PROG) {
684 printf("KIL if (%g %g %g %g) <= 0.0\n",
685 a[0], a[1], a[2], a[3]);
686 }
687
688 if (a[0] < 0.0F || a[1] < 0.0F || a[2] < 0.0F || a[3] < 0.0F) {
689 return GL_FALSE;
690 }
691 }
692 break;
693 case OPCODE_LG2: /* log base 2 */
694 {
695 GLfloat a[4], result[4], val;
696 fetch_vector1(&inst->SrcReg[0], machine, a);
697 /* The fast LOG2 macro doesn't meet the precision requirements.
698 */
699 if (a[0] == 0.0F) {
700 val = -FLT_MAX;
701 }
702 else {
703 val = logf(a[0]) * 1.442695F;
704 }
705 result[0] = result[1] = result[2] = result[3] = val;
706 store_vector4(inst, machine, result);
707 }
708 break;
709 case OPCODE_LIT:
710 {
711 const GLfloat epsilon = 1.0F / 256.0F; /* from NV VP spec */
712 GLfloat a[4], result[4];
713 fetch_vector4(&inst->SrcReg[0], machine, a);
714 a[0] = MAX2(a[0], 0.0F);
715 a[1] = MAX2(a[1], 0.0F);
716 /* XXX ARB version clamps a[3], NV version doesn't */
717 a[3] = CLAMP(a[3], -(128.0F - epsilon), (128.0F - epsilon));
718 result[0] = 1.0F;
719 result[1] = a[0];
720 /* XXX we could probably just use pow() here */
721 if (a[0] > 0.0F) {
722 if (a[1] == 0.0F && a[3] == 0.0F)
723 result[2] = 1.0F;
724 else
725 result[2] = powf(a[1], a[3]);
726 }
727 else {
728 result[2] = 0.0F;
729 }
730 result[3] = 1.0F;
731 store_vector4(inst, machine, result);
732 if (DEBUG_PROG) {
733 printf("LIT (%g %g %g %g) : (%g %g %g %g)\n",
734 result[0], result[1], result[2], result[3],
735 a[0], a[1], a[2], a[3]);
736 }
737 }
738 break;
739 case OPCODE_LOG:
740 {
741 GLfloat t[4], q[4], abs_t0;
742 fetch_vector1(&inst->SrcReg[0], machine, t);
743 abs_t0 = fabsf(t[0]);
744 if (abs_t0 != 0.0F) {
745 if (IS_INF_OR_NAN(abs_t0))
746 {
747 SET_POS_INFINITY(q[0]);
748 q[1] = 1.0F;
749 SET_POS_INFINITY(q[2]);
750 }
751 else {
752 int exponent;
753 GLfloat mantissa = frexpf(t[0], &exponent);
754 q[0] = (GLfloat) (exponent - 1);
755 q[1] = 2.0F * mantissa; /* map [.5, 1) -> [1, 2) */
756
757 /* The fast LOG2 macro doesn't meet the precision
758 * requirements.
759 */
760 q[2] = logf(t[0]) * 1.442695F;
761 }
762 }
763 else {
764 SET_NEG_INFINITY(q[0]);
765 q[1] = 1.0F;
766 SET_NEG_INFINITY(q[2]);
767 }
768 q[3] = 1.0;
769 store_vector4(inst, machine, q);
770 }
771 break;
772 case OPCODE_LRP:
773 {
774 GLfloat a[4], b[4], c[4], result[4];
775 fetch_vector4(&inst->SrcReg[0], machine, a);
776 fetch_vector4(&inst->SrcReg[1], machine, b);
777 fetch_vector4(&inst->SrcReg[2], machine, c);
778 result[0] = a[0] * b[0] + (1.0F - a[0]) * c[0];
779 result[1] = a[1] * b[1] + (1.0F - a[1]) * c[1];
780 result[2] = a[2] * b[2] + (1.0F - a[2]) * c[2];
781 result[3] = a[3] * b[3] + (1.0F - a[3]) * c[3];
782 store_vector4(inst, machine, result);
783 if (DEBUG_PROG) {
784 printf("LRP (%g %g %g %g) = (%g %g %g %g), "
785 "(%g %g %g %g), (%g %g %g %g)\n",
786 result[0], result[1], result[2], result[3],
787 a[0], a[1], a[2], a[3],
788 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
789 }
790 }
791 break;
792 case OPCODE_MAD:
793 {
794 GLfloat a[4], b[4], c[4], result[4];
795 fetch_vector4(&inst->SrcReg[0], machine, a);
796 fetch_vector4(&inst->SrcReg[1], machine, b);
797 fetch_vector4(&inst->SrcReg[2], machine, c);
798 result[0] = a[0] * b[0] + c[0];
799 result[1] = a[1] * b[1] + c[1];
800 result[2] = a[2] * b[2] + c[2];
801 result[3] = a[3] * b[3] + c[3];
802 store_vector4(inst, machine, result);
803 if (DEBUG_PROG) {
804 printf("MAD (%g %g %g %g) = (%g %g %g %g) * "
805 "(%g %g %g %g) + (%g %g %g %g)\n",
806 result[0], result[1], result[2], result[3],
807 a[0], a[1], a[2], a[3],
808 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
809 }
810 }
811 break;
812 case OPCODE_MAX:
813 {
814 GLfloat a[4], b[4], result[4];
815 fetch_vector4(&inst->SrcReg[0], machine, a);
816 fetch_vector4(&inst->SrcReg[1], machine, b);
817 result[0] = MAX2(a[0], b[0]);
818 result[1] = MAX2(a[1], b[1]);
819 result[2] = MAX2(a[2], b[2]);
820 result[3] = MAX2(a[3], b[3]);
821 store_vector4(inst, machine, result);
822 if (DEBUG_PROG) {
823 printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
824 result[0], result[1], result[2], result[3],
825 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
826 }
827 }
828 break;
829 case OPCODE_MIN:
830 {
831 GLfloat a[4], b[4], result[4];
832 fetch_vector4(&inst->SrcReg[0], machine, a);
833 fetch_vector4(&inst->SrcReg[1], machine, b);
834 result[0] = MIN2(a[0], b[0]);
835 result[1] = MIN2(a[1], b[1]);
836 result[2] = MIN2(a[2], b[2]);
837 result[3] = MIN2(a[3], b[3]);
838 store_vector4(inst, machine, result);
839 }
840 break;
841 case OPCODE_MOV:
842 {
843 GLfloat result[4];
844 fetch_vector4(&inst->SrcReg[0], machine, result);
845 store_vector4(inst, machine, result);
846 if (DEBUG_PROG) {
847 printf("MOV (%g %g %g %g)\n",
848 result[0], result[1], result[2], result[3]);
849 }
850 }
851 break;
852 case OPCODE_MUL:
853 {
854 GLfloat a[4], b[4], result[4];
855 fetch_vector4(&inst->SrcReg[0], machine, a);
856 fetch_vector4(&inst->SrcReg[1], machine, b);
857 result[0] = a[0] * b[0];
858 result[1] = a[1] * b[1];
859 result[2] = a[2] * b[2];
860 result[3] = a[3] * b[3];
861 store_vector4(inst, machine, result);
862 if (DEBUG_PROG) {
863 printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
864 result[0], result[1], result[2], result[3],
865 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
866 }
867 }
868 break;
869 case OPCODE_NOISE1:
870 {
871 GLfloat a[4], result[4];
872 fetch_vector1(&inst->SrcReg[0], machine, a);
873 result[0] =
874 result[1] =
875 result[2] =
876 result[3] = _mesa_noise1(a[0]);
877 store_vector4(inst, machine, result);
878 }
879 break;
880 case OPCODE_NOISE2:
881 {
882 GLfloat a[4], result[4];
883 fetch_vector4(&inst->SrcReg[0], machine, a);
884 result[0] =
885 result[1] =
886 result[2] = result[3] = _mesa_noise2(a[0], a[1]);
887 store_vector4(inst, machine, result);
888 }
889 break;
890 case OPCODE_NOISE3:
891 {
892 GLfloat a[4], result[4];
893 fetch_vector4(&inst->SrcReg[0], machine, a);
894 result[0] =
895 result[1] =
896 result[2] =
897 result[3] = _mesa_noise3(a[0], a[1], a[2]);
898 store_vector4(inst, machine, result);
899 }
900 break;
901 case OPCODE_NOISE4:
902 {
903 GLfloat a[4], result[4];
904 fetch_vector4(&inst->SrcReg[0], machine, a);
905 result[0] =
906 result[1] =
907 result[2] =
908 result[3] = _mesa_noise4(a[0], a[1], a[2], a[3]);
909 store_vector4(inst, machine, result);
910 }
911 break;
912 case OPCODE_NOP:
913 break;
914 case OPCODE_POW:
915 {
916 GLfloat a[4], b[4], result[4];
917 fetch_vector1(&inst->SrcReg[0], machine, a);
918 fetch_vector1(&inst->SrcReg[1], machine, b);
919 result[0] = result[1] = result[2] = result[3]
920 = powf(a[0], b[0]);
921 store_vector4(inst, machine, result);
922 }
923 break;
924
925 case OPCODE_RCP:
926 {
927 GLfloat a[4], result[4];
928 fetch_vector1(&inst->SrcReg[0], machine, a);
929 if (DEBUG_PROG) {
930 if (a[0] == 0)
931 printf("RCP(0)\n");
932 else if (IS_INF_OR_NAN(a[0]))
933 printf("RCP(inf)\n");
934 }
935 result[0] = result[1] = result[2] = result[3] = 1.0F / a[0];
936 store_vector4(inst, machine, result);
937 }
938 break;
939 case OPCODE_RET: /* return from subroutine (conditional) */
940 if (machine->StackDepth == 0) {
941 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
942 }
943 /* subtract one because of pc++ in the for loop */
944 pc = machine->CallStack[--machine->StackDepth] - 1;
945 break;
946 case OPCODE_RSQ: /* 1 / sqrt() */
947 {
948 GLfloat a[4], result[4];
949 fetch_vector1(&inst->SrcReg[0], machine, a);
950 a[0] = fabsf(a[0]);
951 result[0] = result[1] = result[2] = result[3] = 1.0f / sqrtf(a[0]);
952 store_vector4(inst, machine, result);
953 if (DEBUG_PROG) {
954 printf("RSQ %g = 1/sqrt(|%g|)\n", result[0], a[0]);
955 }
956 }
957 break;
958 case OPCODE_SCS: /* sine and cos */
959 {
960 GLfloat a[4], result[4];
961 fetch_vector1(&inst->SrcReg[0], machine, a);
962 result[0] = cosf(a[0]);
963 result[1] = sinf(a[0]);
964 result[2] = 0.0F; /* undefined! */
965 result[3] = 0.0F; /* undefined! */
966 store_vector4(inst, machine, result);
967 }
968 break;
969 case OPCODE_SGE: /* set on greater or equal */
970 {
971 GLfloat a[4], b[4], result[4];
972 fetch_vector4(&inst->SrcReg[0], machine, a);
973 fetch_vector4(&inst->SrcReg[1], machine, b);
974 result[0] = (a[0] >= b[0]) ? 1.0F : 0.0F;
975 result[1] = (a[1] >= b[1]) ? 1.0F : 0.0F;
976 result[2] = (a[2] >= b[2]) ? 1.0F : 0.0F;
977 result[3] = (a[3] >= b[3]) ? 1.0F : 0.0F;
978 store_vector4(inst, machine, result);
979 if (DEBUG_PROG) {
980 printf("SGE (%g %g %g %g) = (%g %g %g %g) >= (%g %g %g %g)\n",
981 result[0], result[1], result[2], result[3],
982 a[0], a[1], a[2], a[3],
983 b[0], b[1], b[2], b[3]);
984 }
985 }
986 break;
987 case OPCODE_SIN:
988 {
989 GLfloat a[4], result[4];
990 fetch_vector1(&inst->SrcReg[0], machine, a);
991 result[0] = result[1] = result[2] = result[3]
992 = sinf(a[0]);
993 store_vector4(inst, machine, result);
994 }
995 break;
996 case OPCODE_SLT: /* set on less */
997 {
998 GLfloat a[4], b[4], result[4];
999 fetch_vector4(&inst->SrcReg[0], machine, a);
1000 fetch_vector4(&inst->SrcReg[1], machine, b);
1001 result[0] = (a[0] < b[0]) ? 1.0F : 0.0F;
1002 result[1] = (a[1] < b[1]) ? 1.0F : 0.0F;
1003 result[2] = (a[2] < b[2]) ? 1.0F : 0.0F;
1004 result[3] = (a[3] < b[3]) ? 1.0F : 0.0F;
1005 store_vector4(inst, machine, result);
1006 if (DEBUG_PROG) {
1007 printf("SLT (%g %g %g %g) = (%g %g %g %g) < (%g %g %g %g)\n",
1008 result[0], result[1], result[2], result[3],
1009 a[0], a[1], a[2], a[3],
1010 b[0], b[1], b[2], b[3]);
1011 }
1012 }
1013 break;
1014 case OPCODE_SSG: /* set sign (-1, 0 or +1) */
1015 {
1016 GLfloat a[4], result[4];
1017 fetch_vector4(&inst->SrcReg[0], machine, a);
1018 result[0] = (GLfloat) ((a[0] > 0.0F) - (a[0] < 0.0F));
1019 result[1] = (GLfloat) ((a[1] > 0.0F) - (a[1] < 0.0F));
1020 result[2] = (GLfloat) ((a[2] > 0.0F) - (a[2] < 0.0F));
1021 result[3] = (GLfloat) ((a[3] > 0.0F) - (a[3] < 0.0F));
1022 store_vector4(inst, machine, result);
1023 }
1024 break;
1025 case OPCODE_SUB:
1026 {
1027 GLfloat a[4], b[4], result[4];
1028 fetch_vector4(&inst->SrcReg[0], machine, a);
1029 fetch_vector4(&inst->SrcReg[1], machine, b);
1030 result[0] = a[0] - b[0];
1031 result[1] = a[1] - b[1];
1032 result[2] = a[2] - b[2];
1033 result[3] = a[3] - b[3];
1034 store_vector4(inst, machine, result);
1035 if (DEBUG_PROG) {
1036 printf("SUB (%g %g %g %g) = (%g %g %g %g) - (%g %g %g %g)\n",
1037 result[0], result[1], result[2], result[3],
1038 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1039 }
1040 }
1041 break;
1042 case OPCODE_SWZ: /* extended swizzle */
1043 {
1044 const struct prog_src_register *source = &inst->SrcReg[0];
1045 const GLfloat *src = get_src_register_pointer(source, machine);
1046 GLfloat result[4];
1047 GLuint i;
1048 for (i = 0; i < 4; i++) {
1049 const GLuint swz = GET_SWZ(source->Swizzle, i);
1050 if (swz == SWIZZLE_ZERO)
1051 result[i] = 0.0;
1052 else if (swz == SWIZZLE_ONE)
1053 result[i] = 1.0;
1054 else {
1055 assert(swz <= 3);
1056 result[i] = src[swz];
1057 }
1058 if (source->Negate & (1 << i))
1059 result[i] = -result[i];
1060 }
1061 store_vector4(inst, machine, result);
1062 }
1063 break;
1064 case OPCODE_TEX: /* Both ARB and NV frag prog */
1065 /* Simple texel lookup */
1066 {
1067 GLfloat texcoord[4], color[4];
1068 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1069
1070 /* For TEX, texcoord.Q should not be used and its value should not
1071 * matter (at most, we pass coord.xyz to texture3D() in GLSL).
1072 * Set Q=1 so that FetchTexelDeriv() doesn't get a garbage value
1073 * which is effectively what happens when the texcoord swizzle
1074 * is .xyzz
1075 */
1076 texcoord[3] = 1.0f;
1077
1078 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1079
1080 if (DEBUG_PROG) {
1081 printf("TEX (%g, %g, %g, %g) = texture[%d][%g, %g, %g, %g]\n",
1082 color[0], color[1], color[2], color[3],
1083 inst->TexSrcUnit,
1084 texcoord[0], texcoord[1], texcoord[2], texcoord[3]);
1085 }
1086 store_vector4(inst, machine, color);
1087 }
1088 break;
1089 case OPCODE_TXB: /* GL_ARB_fragment_program only */
1090 /* Texel lookup with LOD bias */
1091 {
1092 GLfloat texcoord[4], color[4], lodBias;
1093
1094 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1095
1096 /* texcoord[3] is the bias to add to lambda */
1097 lodBias = texcoord[3];
1098
1099 fetch_texel(ctx, machine, inst, texcoord, lodBias, color);
1100
1101 if (DEBUG_PROG) {
1102 printf("TXB (%g, %g, %g, %g) = texture[%d][%g %g %g %g]"
1103 " bias %g\n",
1104 color[0], color[1], color[2], color[3],
1105 inst->TexSrcUnit,
1106 texcoord[0],
1107 texcoord[1],
1108 texcoord[2],
1109 texcoord[3],
1110 lodBias);
1111 }
1112
1113 store_vector4(inst, machine, color);
1114 }
1115 break;
1116 case OPCODE_TXD:
1117 /* Texture lookup w/ partial derivatives for LOD */
1118 {
1119 GLfloat texcoord[4], dtdx[4], dtdy[4], color[4];
1120 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1121 fetch_vector4(&inst->SrcReg[1], machine, dtdx);
1122 fetch_vector4(&inst->SrcReg[2], machine, dtdy);
1123 machine->FetchTexelDeriv(ctx, texcoord, dtdx, dtdy,
1124 0.0, /* lodBias */
1125 inst->TexSrcUnit, color);
1126 store_vector4(inst, machine, color);
1127 }
1128 break;
1129 case OPCODE_TXL:
1130 /* Texel lookup with explicit LOD */
1131 {
1132 GLfloat texcoord[4], color[4], lod;
1133
1134 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1135
1136 /* texcoord[3] is the LOD */
1137 lod = texcoord[3];
1138
1139 machine->FetchTexelLod(ctx, texcoord, lod,
1140 machine->Samplers[inst->TexSrcUnit], color);
1141
1142 store_vector4(inst, machine, color);
1143 }
1144 break;
1145 case OPCODE_TXP: /* GL_ARB_fragment_program only */
1146 /* Texture lookup w/ projective divide */
1147 {
1148 GLfloat texcoord[4], color[4];
1149
1150 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1151 /* Not so sure about this test - if texcoord[3] is
1152 * zero, we'd probably be fine except for an assert in
1153 * IROUND_POS() which gets triggered by the inf values created.
1154 */
1155 if (texcoord[3] != 0.0F) {
1156 texcoord[0] /= texcoord[3];
1157 texcoord[1] /= texcoord[3];
1158 texcoord[2] /= texcoord[3];
1159 }
1160
1161 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1162
1163 store_vector4(inst, machine, color);
1164 }
1165 break;
1166 case OPCODE_TRUNC: /* truncate toward zero */
1167 {
1168 GLfloat a[4], result[4];
1169 fetch_vector4(&inst->SrcReg[0], machine, a);
1170 result[0] = (GLfloat) (GLint) a[0];
1171 result[1] = (GLfloat) (GLint) a[1];
1172 result[2] = (GLfloat) (GLint) a[2];
1173 result[3] = (GLfloat) (GLint) a[3];
1174 store_vector4(inst, machine, result);
1175 }
1176 break;
1177 case OPCODE_XPD: /* cross product */
1178 {
1179 GLfloat a[4], b[4], result[4];
1180 fetch_vector4(&inst->SrcReg[0], machine, a);
1181 fetch_vector4(&inst->SrcReg[1], machine, b);
1182 result[0] = a[1] * b[2] - a[2] * b[1];
1183 result[1] = a[2] * b[0] - a[0] * b[2];
1184 result[2] = a[0] * b[1] - a[1] * b[0];
1185 result[3] = 1.0;
1186 store_vector4(inst, machine, result);
1187 if (DEBUG_PROG) {
1188 printf("XPD (%g %g %g %g) = (%g %g %g) X (%g %g %g)\n",
1189 result[0], result[1], result[2], result[3],
1190 a[0], a[1], a[2], b[0], b[1], b[2]);
1191 }
1192 }
1193 break;
1194 case OPCODE_END:
1195 return GL_TRUE;
1196 default:
1197 _mesa_problem(ctx, "Bad opcode %d in _mesa_execute_program",
1198 inst->Opcode);
1199 return GL_TRUE; /* return value doesn't matter */
1200 }
1201
1202 numExec++;
1203 if (numExec > maxExec) {
1204 static GLboolean reported = GL_FALSE;
1205 if (!reported) {
1206 _mesa_problem(ctx, "Infinite loop detected in fragment program");
1207 reported = GL_TRUE;
1208 }
1209 return GL_TRUE;
1210 }
1211
1212 } /* for pc */
1213
1214 return GL_TRUE;
1215 }