projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
mesa: remove INV_SQRTF() macro
[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/colormac.h"
41 #include "main/macros.h"
42 #include "prog_execute.h"
43 #include "prog_instruction.h"
44 #include "prog_parameter.h"
45 #include "prog_print.h"
46 #include "prog_noise.h"
47
48
49 /* debug predicate */
50 #define DEBUG_PROG 0
51
52
53 /**
54 * Set x to positive or negative infinity.
55 */
56 #define SET_POS_INFINITY(x) \
57 do { \
58 fi_type fi; \
59 fi.i = 0x7F800000; \
60 x = fi.f; \
61 } while (0)
62 #define SET_NEG_INFINITY(x) \
63 do { \
64 fi_type fi; \
65 fi.i = 0xFF800000; \
66 x = fi.f; \
67 } while (0)
68
69 #define SET_FLOAT_BITS(x, bits) ((fi_type *) (void *) &(x))->i = bits
70
71
72 static const GLfloat ZeroVec[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
73
74
75 /**
76 * Return a pointer to the 4-element float vector specified by the given
77 * source register.
78 */
79 static inline const GLfloat *
80 get_src_register_pointer(const struct prog_src_register *source,
81 const struct gl_program_machine *machine)
82 {
83 const struct gl_program *prog = machine->CurProgram;
84 GLint reg = source->Index;
85
86 if (source->RelAddr) {
87 /* add address register value to src index/offset */
88 reg += machine->AddressReg[0][0];
89 if (reg < 0) {
90 return ZeroVec;
91 }
92 }
93
94 switch (source->File) {
95 case PROGRAM_TEMPORARY:
96 if (reg >= MAX_PROGRAM_TEMPS)
97 return ZeroVec;
98 return machine->Temporaries[reg];
99
100 case PROGRAM_INPUT:
101 if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
102 if (reg >= VERT_ATTRIB_MAX)
103 return ZeroVec;
104 return machine->VertAttribs[reg];
105 }
106 else {
107 if (reg >= VARYING_SLOT_MAX)
108 return ZeroVec;
109 return machine->Attribs[reg][machine->CurElement];
110 }
111
112 case PROGRAM_OUTPUT:
113 if (reg >= MAX_PROGRAM_OUTPUTS)
114 return ZeroVec;
115 return machine->Outputs[reg];
116
117 case PROGRAM_STATE_VAR:
118 /* Fallthrough */
119 case PROGRAM_CONSTANT:
120 /* Fallthrough */
121 case PROGRAM_UNIFORM:
122 if (reg >= (GLint) prog->Parameters->NumParameters)
123 return ZeroVec;
124 return (GLfloat *) prog->Parameters->ParameterValues[reg];
125
126 case PROGRAM_SYSTEM_VALUE:
127 assert(reg < (GLint) Elements(machine->SystemValues));
128 return machine->SystemValues[reg];
129
130 default:
131 _mesa_problem(NULL,
132 "Invalid src register file %d in get_src_register_pointer()",
133 source->File);
134 return ZeroVec;
135 }
136 }
137
138
139 /**
140 * Return a pointer to the 4-element float vector specified by the given
141 * destination register.
142 */
143 static inline GLfloat *
144 get_dst_register_pointer(const struct prog_dst_register *dest,
145 struct gl_program_machine *machine)
146 {
147 static GLfloat dummyReg[4];
148 GLint reg = dest->Index;
149
150 if (dest->RelAddr) {
151 /* add address register value to src index/offset */
152 reg += machine->AddressReg[0][0];
153 if (reg < 0) {
154 return dummyReg;
155 }
156 }
157
158 switch (dest->File) {
159 case PROGRAM_TEMPORARY:
160 if (reg >= MAX_PROGRAM_TEMPS)
161 return dummyReg;
162 return machine->Temporaries[reg];
163
164 case PROGRAM_OUTPUT:
165 if (reg >= MAX_PROGRAM_OUTPUTS)
166 return dummyReg;
167 return machine->Outputs[reg];
168
169 default:
170 _mesa_problem(NULL,
171 "Invalid dest register file %d in get_dst_register_pointer()",
172 dest->File);
173 return dummyReg;
174 }
175 }
176
177
178
179 /**
180 * Fetch a 4-element float vector from the given source register.
181 * Apply swizzling and negating as needed.
182 */
183 static void
184 fetch_vector4(const struct prog_src_register *source,
185 const struct gl_program_machine *machine, GLfloat result[4])
186 {
187 const GLfloat *src = get_src_register_pointer(source, machine);
188
189 if (source->Swizzle == SWIZZLE_NOOP) {
190 /* no swizzling */
191 COPY_4V(result, src);
192 }
193 else {
194 assert(GET_SWZ(source->Swizzle, 0) <= 3);
195 assert(GET_SWZ(source->Swizzle, 1) <= 3);
196 assert(GET_SWZ(source->Swizzle, 2) <= 3);
197 assert(GET_SWZ(source->Swizzle, 3) <= 3);
198 result[0] = src[GET_SWZ(source->Swizzle, 0)];
199 result[1] = src[GET_SWZ(source->Swizzle, 1)];
200 result[2] = src[GET_SWZ(source->Swizzle, 2)];
201 result[3] = src[GET_SWZ(source->Swizzle, 3)];
202 }
203
204 if (source->Abs) {
205 result[0] = fabsf(result[0]);
206 result[1] = fabsf(result[1]);
207 result[2] = fabsf(result[2]);
208 result[3] = fabsf(result[3]);
209 }
210 if (source->Negate) {
211 assert(source->Negate == NEGATE_XYZW);
212 result[0] = -result[0];
213 result[1] = -result[1];
214 result[2] = -result[2];
215 result[3] = -result[3];
216 }
217
218 #ifdef NAN_CHECK
219 assert(!IS_INF_OR_NAN(result[0]));
220 assert(!IS_INF_OR_NAN(result[0]));
221 assert(!IS_INF_OR_NAN(result[0]));
222 assert(!IS_INF_OR_NAN(result[0]));
223 #endif
224 }
225
226
227 /**
228 * Fetch the derivative with respect to X or Y for the given register.
229 * XXX this currently only works for fragment program input attribs.
230 */
231 static void
232 fetch_vector4_deriv(struct gl_context * ctx,
233 const struct prog_src_register *source,
234 const struct gl_program_machine *machine,
235 char xOrY, GLfloat result[4])
236 {
237 if (source->File == PROGRAM_INPUT &&
238 source->Index < (GLint) machine->NumDeriv) {
239 const GLint col = machine->CurElement;
240 const GLfloat w = machine->Attribs[VARYING_SLOT_POS][col][3];
241 const GLfloat invQ = 1.0f / w;
242 GLfloat deriv[4];
243
244 if (xOrY == 'X') {
245 deriv[0] = machine->DerivX[source->Index][0] * invQ;
246 deriv[1] = machine->DerivX[source->Index][1] * invQ;
247 deriv[2] = machine->DerivX[source->Index][2] * invQ;
248 deriv[3] = machine->DerivX[source->Index][3] * invQ;
249 }
250 else {
251 deriv[0] = machine->DerivY[source->Index][0] * invQ;
252 deriv[1] = machine->DerivY[source->Index][1] * invQ;
253 deriv[2] = machine->DerivY[source->Index][2] * invQ;
254 deriv[3] = machine->DerivY[source->Index][3] * invQ;
255 }
256
257 result[0] = deriv[GET_SWZ(source->Swizzle, 0)];
258 result[1] = deriv[GET_SWZ(source->Swizzle, 1)];
259 result[2] = deriv[GET_SWZ(source->Swizzle, 2)];
260 result[3] = deriv[GET_SWZ(source->Swizzle, 3)];
261
262 if (source->Abs) {
263 result[0] = fabsf(result[0]);
264 result[1] = fabsf(result[1]);
265 result[2] = fabsf(result[2]);
266 result[3] = fabsf(result[3]);
267 }
268 if (source->Negate) {
269 assert(source->Negate == NEGATE_XYZW);
270 result[0] = -result[0];
271 result[1] = -result[1];
272 result[2] = -result[2];
273 result[3] = -result[3];
274 }
275 }
276 else {
277 ASSIGN_4V(result, 0.0, 0.0, 0.0, 0.0);
278 }
279 }
280
281
282 /**
283 * As above, but only return result[0] element.
284 */
285 static void
286 fetch_vector1(const struct prog_src_register *source,
287 const struct gl_program_machine *machine, GLfloat result[4])
288 {
289 const GLfloat *src = get_src_register_pointer(source, machine);
290
291 result[0] = src[GET_SWZ(source->Swizzle, 0)];
292
293 if (source->Abs) {
294 result[0] = fabsf(result[0]);
295 }
296 if (source->Negate) {
297 result[0] = -result[0];
298 }
299 }
300
301
302 /**
303 * Fetch texel from texture. Use partial derivatives when possible.
304 */
305 static inline void
306 fetch_texel(struct gl_context *ctx,
307 const struct gl_program_machine *machine,
308 const struct prog_instruction *inst,
309 const GLfloat texcoord[4], GLfloat lodBias,
310 GLfloat color[4])
311 {
312 const GLuint unit = machine->Samplers[inst->TexSrcUnit];
313
314 /* Note: we only have the right derivatives for fragment input attribs.
315 */
316 if (machine->NumDeriv > 0 &&
317 inst->SrcReg[0].File == PROGRAM_INPUT &&
318 inst->SrcReg[0].Index == VARYING_SLOT_TEX0 + inst->TexSrcUnit) {
319 /* simple texture fetch for which we should have derivatives */
320 GLuint attr = inst->SrcReg[0].Index;
321 machine->FetchTexelDeriv(ctx, texcoord,
322 machine->DerivX[attr],
323 machine->DerivY[attr],
324 lodBias, unit, color);
325 }
326 else {
327 machine->FetchTexelLod(ctx, texcoord, lodBias, unit, color);
328 }
329 }
330
331
332 /**
333 * Test value against zero and return GT, LT, EQ or UN if NaN.
334 */
335 static inline GLuint
336 generate_cc(float value)
337 {
338 if (value != value)
339 return COND_UN; /* NaN */
340 if (value > 0.0F)
341 return COND_GT;
342 if (value < 0.0F)
343 return COND_LT;
344 return COND_EQ;
345 }
346
347
348 /**
349 * Test if the ccMaskRule is satisfied by the given condition code.
350 * Used to mask destination writes according to the current condition code.
351 */
352 static inline GLboolean
353 test_cc(GLuint condCode, GLuint ccMaskRule)
354 {
355 switch (ccMaskRule) {
356 case COND_EQ: return (condCode == COND_EQ);
357 case COND_NE: return (condCode != COND_EQ);
358 case COND_LT: return (condCode == COND_LT);
359 case COND_GE: return (condCode == COND_GT || condCode == COND_EQ);
360 case COND_LE: return (condCode == COND_LT || condCode == COND_EQ);
361 case COND_GT: return (condCode == COND_GT);
362 case COND_TR: return GL_TRUE;
363 case COND_FL: return GL_FALSE;
364 default: return GL_TRUE;
365 }
366 }
367
368
369 /**
370 * Evaluate the 4 condition codes against a predicate and return GL_TRUE
371 * or GL_FALSE to indicate result.
372 */
373 static inline GLboolean
374 eval_condition(const struct gl_program_machine *machine,
375 const struct prog_instruction *inst)
376 {
377 const GLuint swizzle = inst->DstReg.CondSwizzle;
378 const GLuint condMask = inst->DstReg.CondMask;
379 if (test_cc(machine->CondCodes[GET_SWZ(swizzle, 0)], condMask) ||
380 test_cc(machine->CondCodes[GET_SWZ(swizzle, 1)], condMask) ||
381 test_cc(machine->CondCodes[GET_SWZ(swizzle, 2)], condMask) ||
382 test_cc(machine->CondCodes[GET_SWZ(swizzle, 3)], condMask)) {
383 return GL_TRUE;
384 }
385 else {
386 return GL_FALSE;
387 }
388 }
389
390
391
392 /**
393 * Store 4 floats into a register. Observe the instructions saturate and
394 * set-condition-code flags.
395 */
396 static void
397 store_vector4(const struct prog_instruction *inst,
398 struct gl_program_machine *machine, const GLfloat value[4])
399 {
400 const struct prog_dst_register *dstReg = &(inst->DstReg);
401 const GLboolean clamp = inst->SaturateMode == SATURATE_ZERO_ONE;
402 GLuint writeMask = dstReg->WriteMask;
403 GLfloat clampedValue[4];
404 GLfloat *dst = get_dst_register_pointer(dstReg, machine);
405
406 #if 0
407 if (value[0] > 1.0e10 ||
408 IS_INF_OR_NAN(value[0]) ||
409 IS_INF_OR_NAN(value[1]) ||
410 IS_INF_OR_NAN(value[2]) || IS_INF_OR_NAN(value[3]))
411 printf("store %g %g %g %g\n", value[0], value[1], value[2], value[3]);
412 #endif
413
414 if (clamp) {
415 clampedValue[0] = CLAMP(value[0], 0.0F, 1.0F);
416 clampedValue[1] = CLAMP(value[1], 0.0F, 1.0F);
417 clampedValue[2] = CLAMP(value[2], 0.0F, 1.0F);
418 clampedValue[3] = CLAMP(value[3], 0.0F, 1.0F);
419 value = clampedValue;
420 }
421
422 if (dstReg->CondMask != COND_TR) {
423 /* condition codes may turn off some writes */
424 if (writeMask & WRITEMASK_X) {
425 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 0)],
426 dstReg->CondMask))
427 writeMask &= ~WRITEMASK_X;
428 }
429 if (writeMask & WRITEMASK_Y) {
430 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 1)],
431 dstReg->CondMask))
432 writeMask &= ~WRITEMASK_Y;
433 }
434 if (writeMask & WRITEMASK_Z) {
435 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 2)],
436 dstReg->CondMask))
437 writeMask &= ~WRITEMASK_Z;
438 }
439 if (writeMask & WRITEMASK_W) {
440 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 3)],
441 dstReg->CondMask))
442 writeMask &= ~WRITEMASK_W;
443 }
444 }
445
446 #ifdef NAN_CHECK
447 assert(!IS_INF_OR_NAN(value[0]));
448 assert(!IS_INF_OR_NAN(value[0]));
449 assert(!IS_INF_OR_NAN(value[0]));
450 assert(!IS_INF_OR_NAN(value[0]));
451 #endif
452
453 if (writeMask & WRITEMASK_X)
454 dst[0] = value[0];
455 if (writeMask & WRITEMASK_Y)
456 dst[1] = value[1];
457 if (writeMask & WRITEMASK_Z)
458 dst[2] = value[2];
459 if (writeMask & WRITEMASK_W)
460 dst[3] = value[3];
461
462 if (inst->CondUpdate) {
463 if (writeMask & WRITEMASK_X)
464 machine->CondCodes[0] = generate_cc(value[0]);
465 if (writeMask & WRITEMASK_Y)
466 machine->CondCodes[1] = generate_cc(value[1]);
467 if (writeMask & WRITEMASK_Z)
468 machine->CondCodes[2] = generate_cc(value[2]);
469 if (writeMask & WRITEMASK_W)
470 machine->CondCodes[3] = generate_cc(value[3]);
471 #if DEBUG_PROG
472 printf("CondCodes=(%s,%s,%s,%s) for:\n",
473 _mesa_condcode_string(machine->CondCodes[0]),
474 _mesa_condcode_string(machine->CondCodes[1]),
475 _mesa_condcode_string(machine->CondCodes[2]),
476 _mesa_condcode_string(machine->CondCodes[3]));
477 #endif
478 }
479 }
480
481
482 /**
483 * Execute the given vertex/fragment program.
484 *
485 * \param ctx rendering context
486 * \param program the program to execute
487 * \param machine machine state (must be initialized)
488 * \return GL_TRUE if program completed or GL_FALSE if program executed KIL.
489 */
490 GLboolean
491 _mesa_execute_program(struct gl_context * ctx,
492 const struct gl_program *program,
493 struct gl_program_machine *machine)
494 {
495 const GLuint numInst = program->NumInstructions;
496 const GLuint maxExec = 65536;
497 GLuint pc, numExec = 0;
498
499 machine->CurProgram = program;
500
501 if (DEBUG_PROG) {
502 printf("execute program %u --------------------\n", program->Id);
503 }
504
505 if (program->Target == GL_VERTEX_PROGRAM_ARB) {
506 machine->EnvParams = ctx->VertexProgram.Parameters;
507 }
508 else {
509 machine->EnvParams = ctx->FragmentProgram.Parameters;
510 }
511
512 for (pc = 0; pc < numInst; pc++) {
513 const struct prog_instruction *inst = program->Instructions + pc;
514
515 if (DEBUG_PROG) {
516 _mesa_print_instruction(inst);
517 }
518
519 switch (inst->Opcode) {
520 case OPCODE_ABS:
521 {
522 GLfloat a[4], result[4];
523 fetch_vector4(&inst->SrcReg[0], machine, a);
524 result[0] = fabsf(a[0]);
525 result[1] = fabsf(a[1]);
526 result[2] = fabsf(a[2]);
527 result[3] = fabsf(a[3]);
528 store_vector4(inst, machine, result);
529 }
530 break;
531 case OPCODE_ADD:
532 {
533 GLfloat a[4], b[4], result[4];
534 fetch_vector4(&inst->SrcReg[0], machine, a);
535 fetch_vector4(&inst->SrcReg[1], machine, b);
536 result[0] = a[0] + b[0];
537 result[1] = a[1] + b[1];
538 result[2] = a[2] + b[2];
539 result[3] = a[3] + b[3];
540 store_vector4(inst, machine, result);
541 if (DEBUG_PROG) {
542 printf("ADD (%g %g %g %g) = (%g %g %g %g) + (%g %g %g %g)\n",
543 result[0], result[1], result[2], result[3],
544 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
545 }
546 }
547 break;
548 case OPCODE_ARL:
549 {
550 GLfloat t[4];
551 fetch_vector4(&inst->SrcReg[0], machine, t);
552 machine->AddressReg[0][0] = IFLOOR(t[0]);
553 if (DEBUG_PROG) {
554 printf("ARL %d\n", machine->AddressReg[0][0]);
555 }
556 }
557 break;
558 case OPCODE_BGNLOOP:
559 /* no-op */
560 assert(program->Instructions[inst->BranchTarget].Opcode
561 == OPCODE_ENDLOOP);
562 break;
563 case OPCODE_ENDLOOP:
564 /* subtract 1 here since pc is incremented by for(pc) loop */
565 assert(program->Instructions[inst->BranchTarget].Opcode
566 == OPCODE_BGNLOOP);
567 pc = inst->BranchTarget - 1; /* go to matching BNGLOOP */
568 break;
569 case OPCODE_BGNSUB: /* begin subroutine */
570 break;
571 case OPCODE_ENDSUB: /* end subroutine */
572 break;
573 case OPCODE_BRK: /* break out of loop (conditional) */
574 assert(program->Instructions[inst->BranchTarget].Opcode
575 == OPCODE_ENDLOOP);
576 if (eval_condition(machine, inst)) {
577 /* break out of loop */
578 /* pc++ at end of for-loop will put us after the ENDLOOP inst */
579 pc = inst->BranchTarget;
580 }
581 break;
582 case OPCODE_CONT: /* continue loop (conditional) */
583 assert(program->Instructions[inst->BranchTarget].Opcode
584 == OPCODE_ENDLOOP);
585 if (eval_condition(machine, inst)) {
586 /* continue at ENDLOOP */
587 /* Subtract 1 here since we'll do pc++ at end of for-loop */
588 pc = inst->BranchTarget - 1;
589 }
590 break;
591 case OPCODE_CAL: /* Call subroutine (conditional) */
592 if (eval_condition(machine, inst)) {
593 /* call the subroutine */
594 if (machine->StackDepth >= MAX_PROGRAM_CALL_DEPTH) {
595 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
596 }
597 machine->CallStack[machine->StackDepth++] = pc + 1; /* next inst */
598 /* Subtract 1 here since we'll do pc++ at end of for-loop */
599 pc = inst->BranchTarget - 1;
600 }
601 break;
602 case OPCODE_CMP:
603 {
604 GLfloat a[4], b[4], c[4], result[4];
605 fetch_vector4(&inst->SrcReg[0], machine, a);
606 fetch_vector4(&inst->SrcReg[1], machine, b);
607 fetch_vector4(&inst->SrcReg[2], machine, c);
608 result[0] = a[0] < 0.0F ? b[0] : c[0];
609 result[1] = a[1] < 0.0F ? b[1] : c[1];
610 result[2] = a[2] < 0.0F ? b[2] : c[2];
611 result[3] = a[3] < 0.0F ? b[3] : c[3];
612 store_vector4(inst, machine, result);
613 if (DEBUG_PROG) {
614 printf("CMP (%g %g %g %g) = (%g %g %g %g) < 0 ? (%g %g %g %g) : (%g %g %g %g)\n",
615 result[0], result[1], result[2], result[3],
616 a[0], a[1], a[2], a[3],
617 b[0], b[1], b[2], b[3],
618 c[0], c[1], c[2], c[3]);
619 }
620 }
621 break;
622 case OPCODE_COS:
623 {
624 GLfloat a[4], result[4];
625 fetch_vector1(&inst->SrcReg[0], machine, a);
626 result[0] = result[1] = result[2] = result[3]
627 = (GLfloat) cos(a[0]);
628 store_vector4(inst, machine, result);
629 }
630 break;
631 case OPCODE_DDX: /* Partial derivative with respect to X */
632 {
633 GLfloat result[4];
634 fetch_vector4_deriv(ctx, &inst->SrcReg[0], machine,
635 'X', result);
636 store_vector4(inst, machine, result);
637 }
638 break;
639 case OPCODE_DDY: /* Partial derivative with respect to Y */
640 {
641 GLfloat result[4];
642 fetch_vector4_deriv(ctx, &inst->SrcReg[0], machine,
643 'Y', result);
644 store_vector4(inst, machine, result);
645 }
646 break;
647 case OPCODE_DP2:
648 {
649 GLfloat a[4], b[4], result[4];
650 fetch_vector4(&inst->SrcReg[0], machine, a);
651 fetch_vector4(&inst->SrcReg[1], machine, b);
652 result[0] = result[1] = result[2] = result[3] = DOT2(a, b);
653 store_vector4(inst, machine, result);
654 if (DEBUG_PROG) {
655 printf("DP2 %g = (%g %g) . (%g %g)\n",
656 result[0], a[0], a[1], b[0], b[1]);
657 }
658 }
659 break;
660 case OPCODE_DP3:
661 {
662 GLfloat a[4], b[4], result[4];
663 fetch_vector4(&inst->SrcReg[0], machine, a);
664 fetch_vector4(&inst->SrcReg[1], machine, b);
665 result[0] = result[1] = result[2] = result[3] = DOT3(a, b);
666 store_vector4(inst, machine, result);
667 if (DEBUG_PROG) {
668 printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
669 result[0], a[0], a[1], a[2], b[0], b[1], b[2]);
670 }
671 }
672 break;
673 case OPCODE_DP4:
674 {
675 GLfloat a[4], b[4], result[4];
676 fetch_vector4(&inst->SrcReg[0], machine, a);
677 fetch_vector4(&inst->SrcReg[1], machine, b);
678 result[0] = result[1] = result[2] = result[3] = DOT4(a, b);
679 store_vector4(inst, machine, result);
680 if (DEBUG_PROG) {
681 printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
682 result[0], a[0], a[1], a[2], a[3],
683 b[0], b[1], b[2], b[3]);
684 }
685 }
686 break;
687 case OPCODE_DPH:
688 {
689 GLfloat a[4], b[4], result[4];
690 fetch_vector4(&inst->SrcReg[0], machine, a);
691 fetch_vector4(&inst->SrcReg[1], machine, b);
692 result[0] = result[1] = result[2] = result[3] = DOT3(a, b) + b[3];
693 store_vector4(inst, machine, result);
694 }
695 break;
696 case OPCODE_DST: /* Distance vector */
697 {
698 GLfloat a[4], b[4], result[4];
699 fetch_vector4(&inst->SrcReg[0], machine, a);
700 fetch_vector4(&inst->SrcReg[1], machine, b);
701 result[0] = 1.0F;
702 result[1] = a[1] * b[1];
703 result[2] = a[2];
704 result[3] = b[3];
705 store_vector4(inst, machine, result);
706 }
707 break;
708 case OPCODE_EXP:
709 {
710 GLfloat t[4], q[4], floor_t0;
711 fetch_vector1(&inst->SrcReg[0], machine, t);
712 floor_t0 = floorf(t[0]);
713 if (floor_t0 > FLT_MAX_EXP) {
714 SET_POS_INFINITY(q[0]);
715 SET_POS_INFINITY(q[2]);
716 }
717 else if (floor_t0 < FLT_MIN_EXP) {
718 q[0] = 0.0F;
719 q[2] = 0.0F;
720 }
721 else {
722 q[0] = ldexpf(1.0, (int) floor_t0);
723 /* Note: GL_NV_vertex_program expects
724 * result.z = result.x * APPX(result.y)
725 * We do what the ARB extension says.
726 */
727 q[2] = (GLfloat) pow(2.0, t[0]);
728 }
729 q[1] = t[0] - floor_t0;
730 q[3] = 1.0F;
731 store_vector4( inst, machine, q );
732 }
733 break;
734 case OPCODE_EX2: /* Exponential base 2 */
735 {
736 GLfloat a[4], result[4], val;
737 fetch_vector1(&inst->SrcReg[0], machine, a);
738 val = (GLfloat) pow(2.0, a[0]);
739 /*
740 if (IS_INF_OR_NAN(val))
741 val = 1.0e10;
742 */
743 result[0] = result[1] = result[2] = result[3] = val;
744 store_vector4(inst, machine, result);
745 }
746 break;
747 case OPCODE_FLR:
748 {
749 GLfloat a[4], result[4];
750 fetch_vector4(&inst->SrcReg[0], machine, a);
751 result[0] = floorf(a[0]);
752 result[1] = floorf(a[1]);
753 result[2] = floorf(a[2]);
754 result[3] = floorf(a[3]);
755 store_vector4(inst, machine, result);
756 }
757 break;
758 case OPCODE_FRC:
759 {
760 GLfloat a[4], result[4];
761 fetch_vector4(&inst->SrcReg[0], machine, a);
762 result[0] = a[0] - floorf(a[0]);
763 result[1] = a[1] - floorf(a[1]);
764 result[2] = a[2] - floorf(a[2]);
765 result[3] = a[3] - floorf(a[3]);
766 store_vector4(inst, machine, result);
767 }
768 break;
769 case OPCODE_IF:
770 {
771 GLboolean cond;
772 assert(program->Instructions[inst->BranchTarget].Opcode
773 == OPCODE_ELSE ||
774 program->Instructions[inst->BranchTarget].Opcode
775 == OPCODE_ENDIF);
776 /* eval condition */
777 if (inst->SrcReg[0].File != PROGRAM_UNDEFINED) {
778 GLfloat a[4];
779 fetch_vector1(&inst->SrcReg[0], machine, a);
780 cond = (a[0] != 0.0);
781 }
782 else {
783 cond = eval_condition(machine, inst);
784 }
785 if (DEBUG_PROG) {
786 printf("IF: %d\n", cond);
787 }
788 /* do if/else */
789 if (cond) {
790 /* do if-clause (just continue execution) */
791 }
792 else {
793 /* go to the instruction after ELSE or ENDIF */
794 assert(inst->BranchTarget >= 0);
795 pc = inst->BranchTarget;
796 }
797 }
798 break;
799 case OPCODE_ELSE:
800 /* goto ENDIF */
801 assert(program->Instructions[inst->BranchTarget].Opcode
802 == OPCODE_ENDIF);
803 assert(inst->BranchTarget >= 0);
804 pc = inst->BranchTarget;
805 break;
806 case OPCODE_ENDIF:
807 /* nothing */
808 break;
809 case OPCODE_KIL_NV: /* NV_f_p only (conditional) */
810 if (eval_condition(machine, inst)) {
811 return GL_FALSE;
812 }
813 break;
814 case OPCODE_KIL: /* ARB_f_p only */
815 {
816 GLfloat a[4];
817 fetch_vector4(&inst->SrcReg[0], machine, a);
818 if (DEBUG_PROG) {
819 printf("KIL if (%g %g %g %g) <= 0.0\n",
820 a[0], a[1], a[2], a[3]);
821 }
822
823 if (a[0] < 0.0F || a[1] < 0.0F || a[2] < 0.0F || a[3] < 0.0F) {
824 return GL_FALSE;
825 }
826 }
827 break;
828 case OPCODE_LG2: /* log base 2 */
829 {
830 GLfloat a[4], result[4], val;
831 fetch_vector1(&inst->SrcReg[0], machine, a);
832 /* The fast LOG2 macro doesn't meet the precision requirements.
833 */
834 if (a[0] == 0.0F) {
835 val = -FLT_MAX;
836 }
837 else {
838 val = (float)(log(a[0]) * 1.442695F);
839 }
840 result[0] = result[1] = result[2] = result[3] = val;
841 store_vector4(inst, machine, result);
842 }
843 break;
844 case OPCODE_LIT:
845 {
846 const GLfloat epsilon = 1.0F / 256.0F; /* from NV VP spec */
847 GLfloat a[4], result[4];
848 fetch_vector4(&inst->SrcReg[0], machine, a);
849 a[0] = MAX2(a[0], 0.0F);
850 a[1] = MAX2(a[1], 0.0F);
851 /* XXX ARB version clamps a[3], NV version doesn't */
852 a[3] = CLAMP(a[3], -(128.0F - epsilon), (128.0F - epsilon));
853 result[0] = 1.0F;
854 result[1] = a[0];
855 /* XXX we could probably just use pow() here */
856 if (a[0] > 0.0F) {
857 if (a[1] == 0.0 && a[3] == 0.0)
858 result[2] = 1.0F;
859 else
860 result[2] = (GLfloat) pow(a[1], a[3]);
861 }
862 else {
863 result[2] = 0.0F;
864 }
865 result[3] = 1.0F;
866 store_vector4(inst, machine, result);
867 if (DEBUG_PROG) {
868 printf("LIT (%g %g %g %g) : (%g %g %g %g)\n",
869 result[0], result[1], result[2], result[3],
870 a[0], a[1], a[2], a[3]);
871 }
872 }
873 break;
874 case OPCODE_LOG:
875 {
876 GLfloat t[4], q[4], abs_t0;
877 fetch_vector1(&inst->SrcReg[0], machine, t);
878 abs_t0 = fabsf(t[0]);
879 if (abs_t0 != 0.0F) {
880 if (IS_INF_OR_NAN(abs_t0))
881 {
882 SET_POS_INFINITY(q[0]);
883 q[1] = 1.0F;
884 SET_POS_INFINITY(q[2]);
885 }
886 else {
887 int exponent;
888 GLfloat mantissa = frexpf(t[0], &exponent);
889 q[0] = (GLfloat) (exponent - 1);
890 q[1] = (GLfloat) (2.0 * mantissa); /* map [.5, 1) -> [1, 2) */
891
892 /* The fast LOG2 macro doesn't meet the precision
893 * requirements.
894 */
895 q[2] = (float)(log(t[0]) * 1.442695F);
896 }
897 }
898 else {
899 SET_NEG_INFINITY(q[0]);
900 q[1] = 1.0F;
901 SET_NEG_INFINITY(q[2]);
902 }
903 q[3] = 1.0;
904 store_vector4(inst, machine, q);
905 }
906 break;
907 case OPCODE_LRP:
908 {
909 GLfloat a[4], b[4], c[4], result[4];
910 fetch_vector4(&inst->SrcReg[0], machine, a);
911 fetch_vector4(&inst->SrcReg[1], machine, b);
912 fetch_vector4(&inst->SrcReg[2], machine, c);
913 result[0] = a[0] * b[0] + (1.0F - a[0]) * c[0];
914 result[1] = a[1] * b[1] + (1.0F - a[1]) * c[1];
915 result[2] = a[2] * b[2] + (1.0F - a[2]) * c[2];
916 result[3] = a[3] * b[3] + (1.0F - a[3]) * c[3];
917 store_vector4(inst, machine, result);
918 if (DEBUG_PROG) {
919 printf("LRP (%g %g %g %g) = (%g %g %g %g), "
920 "(%g %g %g %g), (%g %g %g %g)\n",
921 result[0], result[1], result[2], result[3],
922 a[0], a[1], a[2], a[3],
923 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
924 }
925 }
926 break;
927 case OPCODE_MAD:
928 {
929 GLfloat a[4], b[4], c[4], result[4];
930 fetch_vector4(&inst->SrcReg[0], machine, a);
931 fetch_vector4(&inst->SrcReg[1], machine, b);
932 fetch_vector4(&inst->SrcReg[2], machine, c);
933 result[0] = a[0] * b[0] + c[0];
934 result[1] = a[1] * b[1] + c[1];
935 result[2] = a[2] * b[2] + c[2];
936 result[3] = a[3] * b[3] + c[3];
937 store_vector4(inst, machine, result);
938 if (DEBUG_PROG) {
939 printf("MAD (%g %g %g %g) = (%g %g %g %g) * "
940 "(%g %g %g %g) + (%g %g %g %g)\n",
941 result[0], result[1], result[2], result[3],
942 a[0], a[1], a[2], a[3],
943 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
944 }
945 }
946 break;
947 case OPCODE_MAX:
948 {
949 GLfloat a[4], b[4], result[4];
950 fetch_vector4(&inst->SrcReg[0], machine, a);
951 fetch_vector4(&inst->SrcReg[1], machine, b);
952 result[0] = MAX2(a[0], b[0]);
953 result[1] = MAX2(a[1], b[1]);
954 result[2] = MAX2(a[2], b[2]);
955 result[3] = MAX2(a[3], b[3]);
956 store_vector4(inst, machine, result);
957 if (DEBUG_PROG) {
958 printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
959 result[0], result[1], result[2], result[3],
960 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
961 }
962 }
963 break;
964 case OPCODE_MIN:
965 {
966 GLfloat a[4], b[4], result[4];
967 fetch_vector4(&inst->SrcReg[0], machine, a);
968 fetch_vector4(&inst->SrcReg[1], machine, b);
969 result[0] = MIN2(a[0], b[0]);
970 result[1] = MIN2(a[1], b[1]);
971 result[2] = MIN2(a[2], b[2]);
972 result[3] = MIN2(a[3], b[3]);
973 store_vector4(inst, machine, result);
974 }
975 break;
976 case OPCODE_MOV:
977 {
978 GLfloat result[4];
979 fetch_vector4(&inst->SrcReg[0], machine, result);
980 store_vector4(inst, machine, result);
981 if (DEBUG_PROG) {
982 printf("MOV (%g %g %g %g)\n",
983 result[0], result[1], result[2], result[3]);
984 }
985 }
986 break;
987 case OPCODE_MUL:
988 {
989 GLfloat a[4], b[4], result[4];
990 fetch_vector4(&inst->SrcReg[0], machine, a);
991 fetch_vector4(&inst->SrcReg[1], machine, b);
992 result[0] = a[0] * b[0];
993 result[1] = a[1] * b[1];
994 result[2] = a[2] * b[2];
995 result[3] = a[3] * b[3];
996 store_vector4(inst, machine, result);
997 if (DEBUG_PROG) {
998 printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
999 result[0], result[1], result[2], result[3],
1000 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1001 }
1002 }
1003 break;
1004 case OPCODE_NOISE1:
1005 {
1006 GLfloat a[4], result[4];
1007 fetch_vector1(&inst->SrcReg[0], machine, a);
1008 result[0] =
1009 result[1] =
1010 result[2] =
1011 result[3] = _mesa_noise1(a[0]);
1012 store_vector4(inst, machine, result);
1013 }
1014 break;
1015 case OPCODE_NOISE2:
1016 {
1017 GLfloat a[4], result[4];
1018 fetch_vector4(&inst->SrcReg[0], machine, a);
1019 result[0] =
1020 result[1] =
1021 result[2] = result[3] = _mesa_noise2(a[0], a[1]);
1022 store_vector4(inst, machine, result);
1023 }
1024 break;
1025 case OPCODE_NOISE3:
1026 {
1027 GLfloat a[4], result[4];
1028 fetch_vector4(&inst->SrcReg[0], machine, a);
1029 result[0] =
1030 result[1] =
1031 result[2] =
1032 result[3] = _mesa_noise3(a[0], a[1], a[2]);
1033 store_vector4(inst, machine, result);
1034 }
1035 break;
1036 case OPCODE_NOISE4:
1037 {
1038 GLfloat a[4], result[4];
1039 fetch_vector4(&inst->SrcReg[0], machine, a);
1040 result[0] =
1041 result[1] =
1042 result[2] =
1043 result[3] = _mesa_noise4(a[0], a[1], a[2], a[3]);
1044 store_vector4(inst, machine, result);
1045 }
1046 break;
1047 case OPCODE_NOP:
1048 break;
1049 case OPCODE_POW:
1050 {
1051 GLfloat a[4], b[4], result[4];
1052 fetch_vector1(&inst->SrcReg[0], machine, a);
1053 fetch_vector1(&inst->SrcReg[1], machine, b);
1054 result[0] = result[1] = result[2] = result[3]
1055 = (GLfloat) pow(a[0], b[0]);
1056 store_vector4(inst, machine, result);
1057 }
1058 break;
1059
1060 case OPCODE_RCP:
1061 {
1062 GLfloat a[4], result[4];
1063 fetch_vector1(&inst->SrcReg[0], machine, a);
1064 if (DEBUG_PROG) {
1065 if (a[0] == 0)
1066 printf("RCP(0)\n");
1067 else if (IS_INF_OR_NAN(a[0]))
1068 printf("RCP(inf)\n");
1069 }
1070 result[0] = result[1] = result[2] = result[3] = 1.0F / a[0];
1071 store_vector4(inst, machine, result);
1072 }
1073 break;
1074 case OPCODE_RET: /* return from subroutine (conditional) */
1075 if (eval_condition(machine, inst)) {
1076 if (machine->StackDepth == 0) {
1077 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
1078 }
1079 /* subtract one because of pc++ in the for loop */
1080 pc = machine->CallStack[--machine->StackDepth] - 1;
1081 }
1082 break;
1083 case OPCODE_RSQ: /* 1 / sqrt() */
1084 {
1085 GLfloat a[4], result[4];
1086 fetch_vector1(&inst->SrcReg[0], machine, a);
1087 a[0] = fabsf(a[0]);
1088 result[0] = result[1] = result[2] = result[3] = 1.0f / sqrtf(a[0]);
1089 store_vector4(inst, machine, result);
1090 if (DEBUG_PROG) {
1091 printf("RSQ %g = 1/sqrt(|%g|)\n", result[0], a[0]);
1092 }
1093 }
1094 break;
1095 case OPCODE_SCS: /* sine and cos */
1096 {
1097 GLfloat a[4], result[4];
1098 fetch_vector1(&inst->SrcReg[0], machine, a);
1099 result[0] = (GLfloat) cos(a[0]);
1100 result[1] = (GLfloat) sin(a[0]);
1101 result[2] = 0.0; /* undefined! */
1102 result[3] = 0.0; /* undefined! */
1103 store_vector4(inst, machine, result);
1104 }
1105 break;
1106 case OPCODE_SEQ: /* set on equal */
1107 {
1108 GLfloat a[4], b[4], result[4];
1109 fetch_vector4(&inst->SrcReg[0], machine, a);
1110 fetch_vector4(&inst->SrcReg[1], machine, b);
1111 result[0] = (a[0] == b[0]) ? 1.0F : 0.0F;
1112 result[1] = (a[1] == b[1]) ? 1.0F : 0.0F;
1113 result[2] = (a[2] == b[2]) ? 1.0F : 0.0F;
1114 result[3] = (a[3] == b[3]) ? 1.0F : 0.0F;
1115 store_vector4(inst, machine, result);
1116 if (DEBUG_PROG) {
1117 printf("SEQ (%g %g %g %g) = (%g %g %g %g) == (%g %g %g %g)\n",
1118 result[0], result[1], result[2], result[3],
1119 a[0], a[1], a[2], a[3],
1120 b[0], b[1], b[2], b[3]);
1121 }
1122 }
1123 break;
1124 case OPCODE_SGE: /* set on greater or equal */
1125 {
1126 GLfloat a[4], b[4], result[4];
1127 fetch_vector4(&inst->SrcReg[0], machine, a);
1128 fetch_vector4(&inst->SrcReg[1], machine, b);
1129 result[0] = (a[0] >= b[0]) ? 1.0F : 0.0F;
1130 result[1] = (a[1] >= b[1]) ? 1.0F : 0.0F;
1131 result[2] = (a[2] >= b[2]) ? 1.0F : 0.0F;
1132 result[3] = (a[3] >= b[3]) ? 1.0F : 0.0F;
1133 store_vector4(inst, machine, result);
1134 if (DEBUG_PROG) {
1135 printf("SGE (%g %g %g %g) = (%g %g %g %g) >= (%g %g %g %g)\n",
1136 result[0], result[1], result[2], result[3],
1137 a[0], a[1], a[2], a[3],
1138 b[0], b[1], b[2], b[3]);
1139 }
1140 }
1141 break;
1142 case OPCODE_SGT: /* set on greater */
1143 {
1144 GLfloat a[4], b[4], result[4];
1145 fetch_vector4(&inst->SrcReg[0], machine, a);
1146 fetch_vector4(&inst->SrcReg[1], machine, b);
1147 result[0] = (a[0] > b[0]) ? 1.0F : 0.0F;
1148 result[1] = (a[1] > b[1]) ? 1.0F : 0.0F;
1149 result[2] = (a[2] > b[2]) ? 1.0F : 0.0F;
1150 result[3] = (a[3] > b[3]) ? 1.0F : 0.0F;
1151 store_vector4(inst, machine, result);
1152 if (DEBUG_PROG) {
1153 printf("SGT (%g %g %g %g) = (%g %g %g %g) > (%g %g %g %g)\n",
1154 result[0], result[1], result[2], result[3],
1155 a[0], a[1], a[2], a[3],
1156 b[0], b[1], b[2], b[3]);
1157 }
1158 }
1159 break;
1160 case OPCODE_SIN:
1161 {
1162 GLfloat a[4], result[4];
1163 fetch_vector1(&inst->SrcReg[0], machine, a);
1164 result[0] = result[1] = result[2] = result[3]
1165 = (GLfloat) sin(a[0]);
1166 store_vector4(inst, machine, result);
1167 }
1168 break;
1169 case OPCODE_SLE: /* set on less or equal */
1170 {
1171 GLfloat a[4], b[4], result[4];
1172 fetch_vector4(&inst->SrcReg[0], machine, a);
1173 fetch_vector4(&inst->SrcReg[1], machine, b);
1174 result[0] = (a[0] <= b[0]) ? 1.0F : 0.0F;
1175 result[1] = (a[1] <= b[1]) ? 1.0F : 0.0F;
1176 result[2] = (a[2] <= b[2]) ? 1.0F : 0.0F;
1177 result[3] = (a[3] <= b[3]) ? 1.0F : 0.0F;
1178 store_vector4(inst, machine, result);
1179 if (DEBUG_PROG) {
1180 printf("SLE (%g %g %g %g) = (%g %g %g %g) <= (%g %g %g %g)\n",
1181 result[0], result[1], result[2], result[3],
1182 a[0], a[1], a[2], a[3],
1183 b[0], b[1], b[2], b[3]);
1184 }
1185 }
1186 break;
1187 case OPCODE_SLT: /* set on less */
1188 {
1189 GLfloat a[4], b[4], result[4];
1190 fetch_vector4(&inst->SrcReg[0], machine, a);
1191 fetch_vector4(&inst->SrcReg[1], machine, b);
1192 result[0] = (a[0] < b[0]) ? 1.0F : 0.0F;
1193 result[1] = (a[1] < b[1]) ? 1.0F : 0.0F;
1194 result[2] = (a[2] < b[2]) ? 1.0F : 0.0F;
1195 result[3] = (a[3] < b[3]) ? 1.0F : 0.0F;
1196 store_vector4(inst, machine, result);
1197 if (DEBUG_PROG) {
1198 printf("SLT (%g %g %g %g) = (%g %g %g %g) < (%g %g %g %g)\n",
1199 result[0], result[1], result[2], result[3],
1200 a[0], a[1], a[2], a[3],
1201 b[0], b[1], b[2], b[3]);
1202 }
1203 }
1204 break;
1205 case OPCODE_SNE: /* set on not equal */
1206 {
1207 GLfloat a[4], b[4], result[4];
1208 fetch_vector4(&inst->SrcReg[0], machine, a);
1209 fetch_vector4(&inst->SrcReg[1], machine, b);
1210 result[0] = (a[0] != b[0]) ? 1.0F : 0.0F;
1211 result[1] = (a[1] != b[1]) ? 1.0F : 0.0F;
1212 result[2] = (a[2] != b[2]) ? 1.0F : 0.0F;
1213 result[3] = (a[3] != b[3]) ? 1.0F : 0.0F;
1214 store_vector4(inst, machine, result);
1215 if (DEBUG_PROG) {
1216 printf("SNE (%g %g %g %g) = (%g %g %g %g) != (%g %g %g %g)\n",
1217 result[0], result[1], result[2], result[3],
1218 a[0], a[1], a[2], a[3],
1219 b[0], b[1], b[2], b[3]);
1220 }
1221 }
1222 break;
1223 case OPCODE_SSG: /* set sign (-1, 0 or +1) */
1224 {
1225 GLfloat a[4], result[4];
1226 fetch_vector4(&inst->SrcReg[0], machine, a);
1227 result[0] = (GLfloat) ((a[0] > 0.0F) - (a[0] < 0.0F));
1228 result[1] = (GLfloat) ((a[1] > 0.0F) - (a[1] < 0.0F));
1229 result[2] = (GLfloat) ((a[2] > 0.0F) - (a[2] < 0.0F));
1230 result[3] = (GLfloat) ((a[3] > 0.0F) - (a[3] < 0.0F));
1231 store_vector4(inst, machine, result);
1232 }
1233 break;
1234 case OPCODE_SUB:
1235 {
1236 GLfloat a[4], b[4], result[4];
1237 fetch_vector4(&inst->SrcReg[0], machine, a);
1238 fetch_vector4(&inst->SrcReg[1], machine, b);
1239 result[0] = a[0] - b[0];
1240 result[1] = a[1] - b[1];
1241 result[2] = a[2] - b[2];
1242 result[3] = a[3] - b[3];
1243 store_vector4(inst, machine, result);
1244 if (DEBUG_PROG) {
1245 printf("SUB (%g %g %g %g) = (%g %g %g %g) - (%g %g %g %g)\n",
1246 result[0], result[1], result[2], result[3],
1247 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1248 }
1249 }
1250 break;
1251 case OPCODE_SWZ: /* extended swizzle */
1252 {
1253 const struct prog_src_register *source = &inst->SrcReg[0];
1254 const GLfloat *src = get_src_register_pointer(source, machine);
1255 GLfloat result[4];
1256 GLuint i;
1257 for (i = 0; i < 4; i++) {
1258 const GLuint swz = GET_SWZ(source->Swizzle, i);
1259 if (swz == SWIZZLE_ZERO)
1260 result[i] = 0.0;
1261 else if (swz == SWIZZLE_ONE)
1262 result[i] = 1.0;
1263 else {
1264 assert(swz <= 3);
1265 result[i] = src[swz];
1266 }
1267 if (source->Negate & (1 << i))
1268 result[i] = -result[i];
1269 }
1270 store_vector4(inst, machine, result);
1271 }
1272 break;
1273 case OPCODE_TEX: /* Both ARB and NV frag prog */
1274 /* Simple texel lookup */
1275 {
1276 GLfloat texcoord[4], color[4];
1277 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1278
1279 /* For TEX, texcoord.Q should not be used and its value should not
1280 * matter (at most, we pass coord.xyz to texture3D() in GLSL).
1281 * Set Q=1 so that FetchTexelDeriv() doesn't get a garbage value
1282 * which is effectively what happens when the texcoord swizzle
1283 * is .xyzz
1284 */
1285 texcoord[3] = 1.0f;
1286
1287 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1288
1289 if (DEBUG_PROG) {
1290 printf("TEX (%g, %g, %g, %g) = texture[%d][%g, %g, %g, %g]\n",
1291 color[0], color[1], color[2], color[3],
1292 inst->TexSrcUnit,
1293 texcoord[0], texcoord[1], texcoord[2], texcoord[3]);
1294 }
1295 store_vector4(inst, machine, color);
1296 }
1297 break;
1298 case OPCODE_TXB: /* GL_ARB_fragment_program only */
1299 /* Texel lookup with LOD bias */
1300 {
1301 GLfloat texcoord[4], color[4], lodBias;
1302
1303 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1304
1305 /* texcoord[3] is the bias to add to lambda */
1306 lodBias = texcoord[3];
1307
1308 fetch_texel(ctx, machine, inst, texcoord, lodBias, color);
1309
1310 if (DEBUG_PROG) {
1311 printf("TXB (%g, %g, %g, %g) = texture[%d][%g %g %g %g]"
1312 " bias %g\n",
1313 color[0], color[1], color[2], color[3],
1314 inst->TexSrcUnit,
1315 texcoord[0],
1316 texcoord[1],
1317 texcoord[2],
1318 texcoord[3],
1319 lodBias);
1320 }
1321
1322 store_vector4(inst, machine, color);
1323 }
1324 break;
1325 case OPCODE_TXD: /* GL_NV_fragment_program only */
1326 /* Texture lookup w/ partial derivatives for LOD */
1327 {
1328 GLfloat texcoord[4], dtdx[4], dtdy[4], color[4];
1329 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1330 fetch_vector4(&inst->SrcReg[1], machine, dtdx);
1331 fetch_vector4(&inst->SrcReg[2], machine, dtdy);
1332 machine->FetchTexelDeriv(ctx, texcoord, dtdx, dtdy,
1333 0.0, /* lodBias */
1334 inst->TexSrcUnit, color);
1335 store_vector4(inst, machine, color);
1336 }
1337 break;
1338 case OPCODE_TXL:
1339 /* Texel lookup with explicit LOD */
1340 {
1341 GLfloat texcoord[4], color[4], lod;
1342
1343 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1344
1345 /* texcoord[3] is the LOD */
1346 lod = texcoord[3];
1347
1348 machine->FetchTexelLod(ctx, texcoord, lod,
1349 machine->Samplers[inst->TexSrcUnit], color);
1350
1351 store_vector4(inst, machine, color);
1352 }
1353 break;
1354 case OPCODE_TXP: /* GL_ARB_fragment_program only */
1355 /* Texture lookup w/ projective divide */
1356 {
1357 GLfloat texcoord[4], color[4];
1358
1359 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1360 /* Not so sure about this test - if texcoord[3] is
1361 * zero, we'd probably be fine except for an assert in
1362 * IROUND_POS() which gets triggered by the inf values created.
1363 */
1364 if (texcoord[3] != 0.0) {
1365 texcoord[0] /= texcoord[3];
1366 texcoord[1] /= texcoord[3];
1367 texcoord[2] /= texcoord[3];
1368 }
1369
1370 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1371
1372 store_vector4(inst, machine, color);
1373 }
1374 break;
1375 case OPCODE_TXP_NV: /* GL_NV_fragment_program only */
1376 /* Texture lookup w/ projective divide, as above, but do not
1377 * do the divide by w if sampling from a cube map.
1378 */
1379 {
1380 GLfloat texcoord[4], color[4];
1381
1382 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1383 if (inst->TexSrcTarget != TEXTURE_CUBE_INDEX &&
1384 texcoord[3] != 0.0) {
1385 texcoord[0] /= texcoord[3];
1386 texcoord[1] /= texcoord[3];
1387 texcoord[2] /= texcoord[3];
1388 }
1389
1390 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1391
1392 store_vector4(inst, machine, color);
1393 }
1394 break;
1395 case OPCODE_TRUNC: /* truncate toward zero */
1396 {
1397 GLfloat a[4], result[4];
1398 fetch_vector4(&inst->SrcReg[0], machine, a);
1399 result[0] = (GLfloat) (GLint) a[0];
1400 result[1] = (GLfloat) (GLint) a[1];
1401 result[2] = (GLfloat) (GLint) a[2];
1402 result[3] = (GLfloat) (GLint) a[3];
1403 store_vector4(inst, machine, result);
1404 }
1405 break;
1406 case OPCODE_XPD: /* cross product */
1407 {
1408 GLfloat a[4], b[4], result[4];
1409 fetch_vector4(&inst->SrcReg[0], machine, a);
1410 fetch_vector4(&inst->SrcReg[1], machine, b);
1411 result[0] = a[1] * b[2] - a[2] * b[1];
1412 result[1] = a[2] * b[0] - a[0] * b[2];
1413 result[2] = a[0] * b[1] - a[1] * b[0];
1414 result[3] = 1.0;
1415 store_vector4(inst, machine, result);
1416 if (DEBUG_PROG) {
1417 printf("XPD (%g %g %g %g) = (%g %g %g) X (%g %g %g)\n",
1418 result[0], result[1], result[2], result[3],
1419 a[0], a[1], a[2], b[0], b[1], b[2]);
1420 }
1421 }
1422 break;
1423 case OPCODE_END:
1424 return GL_TRUE;
1425 default:
1426 _mesa_problem(ctx, "Bad opcode %d in _mesa_execute_program",
1427 inst->Opcode);
1428 return GL_TRUE; /* return value doesn't matter */
1429 }
1430
1431 numExec++;
1432 if (numExec > maxExec) {
1433 static GLboolean reported = GL_FALSE;
1434 if (!reported) {
1435 _mesa_problem(ctx, "Infinite loop detected in fragment program");
1436 reported = GL_TRUE;
1437 }
1438 return GL_TRUE;
1439 }
1440
1441 } /* for pc */
1442
1443 return GL_TRUE;
1444 }