projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
mesa: fix shader prog_execute strict aliasing violations
[mesa.git] / src / mesa / shader / prog_execute.c
1 /*
2 * Mesa 3-D graphics library
3 * Version: 7.3
4 *
5 * Copyright (C) 1999-2008 Brian Paul All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR 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 "main/glheader.h"
39 #include "main/colormac.h"
40 #include "main/context.h"
41 #include "program.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 #if defined(USE_IEEE) || defined(_WIN32)
57 #define SET_POS_INFINITY(x) \
58 do { \
59 fi_type fi; \
60 fi.i = 0x7F800000; \
61 x = fi.f; \
62 } while (0)
63 #define SET_NEG_INFINITY(x) \
64 do { \
65 fi_type fi; \
66 fi.i = 0xFF800000; \
67 x = fi.f; \
68 } while (0)
69 #elif defined(VMS)
70 #define SET_POS_INFINITY(x) x = __MAXFLOAT
71 #define SET_NEG_INFINITY(x) x = -__MAXFLOAT
72 #else
73 #define SET_POS_INFINITY(x) x = (GLfloat) HUGE_VAL
74 #define SET_NEG_INFINITY(x) x = (GLfloat) -HUGE_VAL
75 #endif
76
77 #define SET_FLOAT_BITS(x, bits) ((fi_type *) (void *) &(x))->i = bits
78
79
80 static const GLfloat ZeroVec[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
81
82
83
84 /**
85 * Return a pointer to the 4-element float vector specified by the given
86 * source register.
87 */
88 static INLINE const GLfloat *
89 get_src_register_pointer(const struct prog_src_register *source,
90 const struct gl_program_machine *machine)
91 {
92 const struct gl_program *prog = machine->CurProgram;
93 GLint reg = source->Index;
94
95 if (source->RelAddr) {
96 /* add address register value to src index/offset */
97 reg += machine->AddressReg[0][0];
98 if (reg < 0) {
99 return ZeroVec;
100 }
101 }
102
103 switch (source->File) {
104 case PROGRAM_TEMPORARY:
105 if (reg >= MAX_PROGRAM_TEMPS)
106 return ZeroVec;
107 return machine->Temporaries[reg];
108
109 case PROGRAM_INPUT:
110 if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
111 if (reg >= VERT_ATTRIB_MAX)
112 return ZeroVec;
113 return machine->VertAttribs[reg];
114 }
115 else {
116 if (reg >= FRAG_ATTRIB_MAX)
117 return ZeroVec;
118 return machine->Attribs[reg][machine->CurElement];
119 }
120
121 case PROGRAM_OUTPUT:
122 if (reg >= MAX_PROGRAM_OUTPUTS)
123 return ZeroVec;
124 return machine->Outputs[reg];
125
126 case PROGRAM_LOCAL_PARAM:
127 if (reg >= MAX_PROGRAM_LOCAL_PARAMS)
128 return ZeroVec;
129 return machine->CurProgram->LocalParams[reg];
130
131 case PROGRAM_ENV_PARAM:
132 if (reg >= MAX_PROGRAM_ENV_PARAMS)
133 return ZeroVec;
134 return machine->EnvParams[reg];
135
136 case PROGRAM_STATE_VAR:
137 /* Fallthrough */
138 case PROGRAM_CONSTANT:
139 /* Fallthrough */
140 case PROGRAM_UNIFORM:
141 /* Fallthrough */
142 case PROGRAM_NAMED_PARAM:
143 if (reg >= (GLint) prog->Parameters->NumParameters)
144 return ZeroVec;
145 return prog->Parameters->ParameterValues[reg];
146
147 default:
148 _mesa_problem(NULL,
149 "Invalid src register file %d in get_src_register_pointer()",
150 source->File);
151 return NULL;
152 }
153 }
154
155
156 /**
157 * Return a pointer to the 4-element float vector specified by the given
158 * destination register.
159 */
160 static INLINE GLfloat *
161 get_dst_register_pointer(const struct prog_dst_register *dest,
162 struct gl_program_machine *machine)
163 {
164 static GLfloat dummyReg[4];
165 GLint reg = dest->Index;
166
167 if (dest->RelAddr) {
168 /* add address register value to src index/offset */
169 reg += machine->AddressReg[0][0];
170 if (reg < 0) {
171 return dummyReg;
172 }
173 }
174
175 switch (dest->File) {
176 case PROGRAM_TEMPORARY:
177 if (reg >= MAX_PROGRAM_TEMPS)
178 return dummyReg;
179 return machine->Temporaries[reg];
180
181 case PROGRAM_OUTPUT:
182 if (reg >= MAX_PROGRAM_OUTPUTS)
183 return dummyReg;
184 return machine->Outputs[reg];
185
186 case PROGRAM_WRITE_ONLY:
187 return dummyReg;
188
189 default:
190 _mesa_problem(NULL,
191 "Invalid dest register file %d in get_dst_register_pointer()",
192 dest->File);
193 return NULL;
194 }
195 }
196
197
198
199 /**
200 * Fetch a 4-element float vector from the given source register.
201 * Apply swizzling and negating as needed.
202 */
203 static void
204 fetch_vector4(const struct prog_src_register *source,
205 const struct gl_program_machine *machine, GLfloat result[4])
206 {
207 const GLfloat *src = get_src_register_pointer(source, machine);
208 ASSERT(src);
209
210 if (source->Swizzle == SWIZZLE_NOOP) {
211 /* no swizzling */
212 COPY_4V(result, src);
213 }
214 else {
215 ASSERT(GET_SWZ(source->Swizzle, 0) <= 3);
216 ASSERT(GET_SWZ(source->Swizzle, 1) <= 3);
217 ASSERT(GET_SWZ(source->Swizzle, 2) <= 3);
218 ASSERT(GET_SWZ(source->Swizzle, 3) <= 3);
219 result[0] = src[GET_SWZ(source->Swizzle, 0)];
220 result[1] = src[GET_SWZ(source->Swizzle, 1)];
221 result[2] = src[GET_SWZ(source->Swizzle, 2)];
222 result[3] = src[GET_SWZ(source->Swizzle, 3)];
223 }
224
225 if (source->Abs) {
226 result[0] = FABSF(result[0]);
227 result[1] = FABSF(result[1]);
228 result[2] = FABSF(result[2]);
229 result[3] = FABSF(result[3]);
230 }
231 if (source->Negate) {
232 ASSERT(source->Negate == NEGATE_XYZW);
233 result[0] = -result[0];
234 result[1] = -result[1];
235 result[2] = -result[2];
236 result[3] = -result[3];
237 }
238
239 #ifdef NAN_CHECK
240 assert(!IS_INF_OR_NAN(result[0]));
241 assert(!IS_INF_OR_NAN(result[0]));
242 assert(!IS_INF_OR_NAN(result[0]));
243 assert(!IS_INF_OR_NAN(result[0]));
244 #endif
245 }
246
247
248 /**
249 * Fetch a 4-element uint vector from the given source register.
250 * Apply swizzling but not negation/abs.
251 */
252 static void
253 fetch_vector4ui(const struct prog_src_register *source,
254 const struct gl_program_machine *machine, GLuint result[4])
255 {
256 const GLuint *src = (GLuint *) get_src_register_pointer(source, machine);
257 ASSERT(src);
258
259 if (source->Swizzle == SWIZZLE_NOOP) {
260 /* no swizzling */
261 COPY_4V(result, src);
262 }
263 else {
264 ASSERT(GET_SWZ(source->Swizzle, 0) <= 3);
265 ASSERT(GET_SWZ(source->Swizzle, 1) <= 3);
266 ASSERT(GET_SWZ(source->Swizzle, 2) <= 3);
267 ASSERT(GET_SWZ(source->Swizzle, 3) <= 3);
268 result[0] = src[GET_SWZ(source->Swizzle, 0)];
269 result[1] = src[GET_SWZ(source->Swizzle, 1)];
270 result[2] = src[GET_SWZ(source->Swizzle, 2)];
271 result[3] = src[GET_SWZ(source->Swizzle, 3)];
272 }
273
274 /* Note: no Negate or Abs here */
275 }
276
277
278
279 /**
280 * Fetch the derivative with respect to X or Y for the given register.
281 * XXX this currently only works for fragment program input attribs.
282 */
283 static void
284 fetch_vector4_deriv(GLcontext * ctx,
285 const struct prog_src_register *source,
286 const struct gl_program_machine *machine,
287 char xOrY, GLfloat result[4])
288 {
289 if (source->File == PROGRAM_INPUT &&
290 source->Index < (GLint) machine->NumDeriv) {
291 const GLint col = machine->CurElement;
292 const GLfloat w = machine->Attribs[FRAG_ATTRIB_WPOS][col][3];
293 const GLfloat invQ = 1.0f / w;
294 GLfloat deriv[4];
295
296 if (xOrY == 'X') {
297 deriv[0] = machine->DerivX[source->Index][0] * invQ;
298 deriv[1] = machine->DerivX[source->Index][1] * invQ;
299 deriv[2] = machine->DerivX[source->Index][2] * invQ;
300 deriv[3] = machine->DerivX[source->Index][3] * invQ;
301 }
302 else {
303 deriv[0] = machine->DerivY[source->Index][0] * invQ;
304 deriv[1] = machine->DerivY[source->Index][1] * invQ;
305 deriv[2] = machine->DerivY[source->Index][2] * invQ;
306 deriv[3] = machine->DerivY[source->Index][3] * invQ;
307 }
308
309 result[0] = deriv[GET_SWZ(source->Swizzle, 0)];
310 result[1] = deriv[GET_SWZ(source->Swizzle, 1)];
311 result[2] = deriv[GET_SWZ(source->Swizzle, 2)];
312 result[3] = deriv[GET_SWZ(source->Swizzle, 3)];
313
314 if (source->Abs) {
315 result[0] = FABSF(result[0]);
316 result[1] = FABSF(result[1]);
317 result[2] = FABSF(result[2]);
318 result[3] = FABSF(result[3]);
319 }
320 if (source->Negate) {
321 ASSERT(source->Negate == NEGATE_XYZW);
322 result[0] = -result[0];
323 result[1] = -result[1];
324 result[2] = -result[2];
325 result[3] = -result[3];
326 }
327 }
328 else {
329 ASSIGN_4V(result, 0.0, 0.0, 0.0, 0.0);
330 }
331 }
332
333
334 /**
335 * As above, but only return result[0] element.
336 */
337 static void
338 fetch_vector1(const struct prog_src_register *source,
339 const struct gl_program_machine *machine, GLfloat result[4])
340 {
341 const GLfloat *src = get_src_register_pointer(source, machine);
342 ASSERT(src);
343
344 result[0] = src[GET_SWZ(source->Swizzle, 0)];
345
346 if (source->Abs) {
347 result[0] = FABSF(result[0]);
348 }
349 if (source->Negate) {
350 result[0] = -result[0];
351 }
352 }
353
354
355 /**
356 * Fetch texel from texture. Use partial derivatives when possible.
357 */
358 static INLINE void
359 fetch_texel(GLcontext *ctx,
360 const struct gl_program_machine *machine,
361 const struct prog_instruction *inst,
362 const GLfloat texcoord[4], GLfloat lodBias,
363 GLfloat color[4])
364 {
365 const GLuint unit = machine->Samplers[inst->TexSrcUnit];
366
367 /* Note: we only have the right derivatives for fragment input attribs.
368 */
369 if (machine->NumDeriv > 0 &&
370 inst->SrcReg[0].File == PROGRAM_INPUT &&
371 inst->SrcReg[0].Index == FRAG_ATTRIB_TEX0 + inst->TexSrcUnit) {
372 /* simple texture fetch for which we should have derivatives */
373 GLuint attr = inst->SrcReg[0].Index;
374 machine->FetchTexelDeriv(ctx, texcoord,
375 machine->DerivX[attr],
376 machine->DerivY[attr],
377 lodBias, unit, color);
378 }
379 else {
380 machine->FetchTexelLod(ctx, texcoord, lodBias, unit, color);
381 }
382 }
383
384
385 /**
386 * Test value against zero and return GT, LT, EQ or UN if NaN.
387 */
388 static INLINE GLuint
389 generate_cc(float value)
390 {
391 if (value != value)
392 return COND_UN; /* NaN */
393 if (value > 0.0F)
394 return COND_GT;
395 if (value < 0.0F)
396 return COND_LT;
397 return COND_EQ;
398 }
399
400
401 /**
402 * Test if the ccMaskRule is satisfied by the given condition code.
403 * Used to mask destination writes according to the current condition code.
404 */
405 static INLINE GLboolean
406 test_cc(GLuint condCode, GLuint ccMaskRule)
407 {
408 switch (ccMaskRule) {
409 case COND_EQ: return (condCode == COND_EQ);
410 case COND_NE: return (condCode != COND_EQ);
411 case COND_LT: return (condCode == COND_LT);
412 case COND_GE: return (condCode == COND_GT || condCode == COND_EQ);
413 case COND_LE: return (condCode == COND_LT || condCode == COND_EQ);
414 case COND_GT: return (condCode == COND_GT);
415 case COND_TR: return GL_TRUE;
416 case COND_FL: return GL_FALSE;
417 default: return GL_TRUE;
418 }
419 }
420
421
422 /**
423 * Evaluate the 4 condition codes against a predicate and return GL_TRUE
424 * or GL_FALSE to indicate result.
425 */
426 static INLINE GLboolean
427 eval_condition(const struct gl_program_machine *machine,
428 const struct prog_instruction *inst)
429 {
430 const GLuint swizzle = inst->DstReg.CondSwizzle;
431 const GLuint condMask = inst->DstReg.CondMask;
432 if (test_cc(machine->CondCodes[GET_SWZ(swizzle, 0)], condMask) ||
433 test_cc(machine->CondCodes[GET_SWZ(swizzle, 1)], condMask) ||
434 test_cc(machine->CondCodes[GET_SWZ(swizzle, 2)], condMask) ||
435 test_cc(machine->CondCodes[GET_SWZ(swizzle, 3)], condMask)) {
436 return GL_TRUE;
437 }
438 else {
439 return GL_FALSE;
440 }
441 }
442
443
444
445 /**
446 * Store 4 floats into a register. Observe the instructions saturate and
447 * set-condition-code flags.
448 */
449 static void
450 store_vector4(const struct prog_instruction *inst,
451 struct gl_program_machine *machine, const GLfloat value[4])
452 {
453 const struct prog_dst_register *dstReg = &(inst->DstReg);
454 const GLboolean clamp = inst->SaturateMode == SATURATE_ZERO_ONE;
455 GLuint writeMask = dstReg->WriteMask;
456 GLfloat clampedValue[4];
457 GLfloat *dst = get_dst_register_pointer(dstReg, machine);
458
459 #if 0
460 if (value[0] > 1.0e10 ||
461 IS_INF_OR_NAN(value[0]) ||
462 IS_INF_OR_NAN(value[1]) ||
463 IS_INF_OR_NAN(value[2]) || IS_INF_OR_NAN(value[3]))
464 printf("store %g %g %g %g\n", value[0], value[1], value[2], value[3]);
465 #endif
466
467 if (clamp) {
468 clampedValue[0] = CLAMP(value[0], 0.0F, 1.0F);
469 clampedValue[1] = CLAMP(value[1], 0.0F, 1.0F);
470 clampedValue[2] = CLAMP(value[2], 0.0F, 1.0F);
471 clampedValue[3] = CLAMP(value[3], 0.0F, 1.0F);
472 value = clampedValue;
473 }
474
475 if (dstReg->CondMask != COND_TR) {
476 /* condition codes may turn off some writes */
477 if (writeMask & WRITEMASK_X) {
478 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 0)],
479 dstReg->CondMask))
480 writeMask &= ~WRITEMASK_X;
481 }
482 if (writeMask & WRITEMASK_Y) {
483 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 1)],
484 dstReg->CondMask))
485 writeMask &= ~WRITEMASK_Y;
486 }
487 if (writeMask & WRITEMASK_Z) {
488 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 2)],
489 dstReg->CondMask))
490 writeMask &= ~WRITEMASK_Z;
491 }
492 if (writeMask & WRITEMASK_W) {
493 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 3)],
494 dstReg->CondMask))
495 writeMask &= ~WRITEMASK_W;
496 }
497 }
498
499 #ifdef NAN_CHECK
500 assert(!IS_INF_OR_NAN(value[0]));
501 assert(!IS_INF_OR_NAN(value[0]));
502 assert(!IS_INF_OR_NAN(value[0]));
503 assert(!IS_INF_OR_NAN(value[0]));
504 #endif
505
506 if (writeMask & WRITEMASK_X)
507 dst[0] = value[0];
508 if (writeMask & WRITEMASK_Y)
509 dst[1] = value[1];
510 if (writeMask & WRITEMASK_Z)
511 dst[2] = value[2];
512 if (writeMask & WRITEMASK_W)
513 dst[3] = value[3];
514
515 if (inst->CondUpdate) {
516 if (writeMask & WRITEMASK_X)
517 machine->CondCodes[0] = generate_cc(value[0]);
518 if (writeMask & WRITEMASK_Y)
519 machine->CondCodes[1] = generate_cc(value[1]);
520 if (writeMask & WRITEMASK_Z)
521 machine->CondCodes[2] = generate_cc(value[2]);
522 if (writeMask & WRITEMASK_W)
523 machine->CondCodes[3] = generate_cc(value[3]);
524 #if DEBUG_PROG
525 printf("CondCodes=(%s,%s,%s,%s) for:\n",
526 _mesa_condcode_string(machine->CondCodes[0]),
527 _mesa_condcode_string(machine->CondCodes[1]),
528 _mesa_condcode_string(machine->CondCodes[2]),
529 _mesa_condcode_string(machine->CondCodes[3]));
530 #endif
531 }
532 }
533
534
535 /**
536 * Store 4 uints into a register. Observe the set-condition-code flags.
537 */
538 static void
539 store_vector4ui(const struct prog_instruction *inst,
540 struct gl_program_machine *machine, const GLuint value[4])
541 {
542 const struct prog_dst_register *dstReg = &(inst->DstReg);
543 GLuint writeMask = dstReg->WriteMask;
544 GLuint *dst = (GLuint *) get_dst_register_pointer(dstReg, machine);
545
546 if (dstReg->CondMask != COND_TR) {
547 /* condition codes may turn off some writes */
548 if (writeMask & WRITEMASK_X) {
549 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 0)],
550 dstReg->CondMask))
551 writeMask &= ~WRITEMASK_X;
552 }
553 if (writeMask & WRITEMASK_Y) {
554 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 1)],
555 dstReg->CondMask))
556 writeMask &= ~WRITEMASK_Y;
557 }
558 if (writeMask & WRITEMASK_Z) {
559 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 2)],
560 dstReg->CondMask))
561 writeMask &= ~WRITEMASK_Z;
562 }
563 if (writeMask & WRITEMASK_W) {
564 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 3)],
565 dstReg->CondMask))
566 writeMask &= ~WRITEMASK_W;
567 }
568 }
569
570 if (writeMask & WRITEMASK_X)
571 dst[0] = value[0];
572 if (writeMask & WRITEMASK_Y)
573 dst[1] = value[1];
574 if (writeMask & WRITEMASK_Z)
575 dst[2] = value[2];
576 if (writeMask & WRITEMASK_W)
577 dst[3] = value[3];
578
579 if (inst->CondUpdate) {
580 if (writeMask & WRITEMASK_X)
581 machine->CondCodes[0] = generate_cc(value[0]);
582 if (writeMask & WRITEMASK_Y)
583 machine->CondCodes[1] = generate_cc(value[1]);
584 if (writeMask & WRITEMASK_Z)
585 machine->CondCodes[2] = generate_cc(value[2]);
586 if (writeMask & WRITEMASK_W)
587 machine->CondCodes[3] = generate_cc(value[3]);
588 #if DEBUG_PROG
589 printf("CondCodes=(%s,%s,%s,%s) for:\n",
590 _mesa_condcode_string(machine->CondCodes[0]),
591 _mesa_condcode_string(machine->CondCodes[1]),
592 _mesa_condcode_string(machine->CondCodes[2]),
593 _mesa_condcode_string(machine->CondCodes[3]));
594 #endif
595 }
596 }
597
598
599
600 /**
601 * Execute the given vertex/fragment program.
602 *
603 * \param ctx rendering context
604 * \param program the program to execute
605 * \param machine machine state (must be initialized)
606 * \return GL_TRUE if program completed or GL_FALSE if program executed KIL.
607 */
608 GLboolean
609 _mesa_execute_program(GLcontext * ctx,
610 const struct gl_program *program,
611 struct gl_program_machine *machine)
612 {
613 const GLuint numInst = program->NumInstructions;
614 const GLuint maxExec = 10000;
615 GLuint pc, numExec = 0;
616
617 machine->CurProgram = program;
618
619 if (DEBUG_PROG) {
620 printf("execute program %u --------------------\n", program->Id);
621 }
622
623 if (program->Target == GL_VERTEX_PROGRAM_ARB) {
624 machine->EnvParams = ctx->VertexProgram.Parameters;
625 }
626 else {
627 machine->EnvParams = ctx->FragmentProgram.Parameters;
628 }
629
630 for (pc = 0; pc < numInst; pc++) {
631 const struct prog_instruction *inst = program->Instructions + pc;
632
633 if (DEBUG_PROG) {
634 _mesa_print_instruction(inst);
635 }
636
637 switch (inst->Opcode) {
638 case OPCODE_ABS:
639 {
640 GLfloat a[4], result[4];
641 fetch_vector4(&inst->SrcReg[0], machine, a);
642 result[0] = FABSF(a[0]);
643 result[1] = FABSF(a[1]);
644 result[2] = FABSF(a[2]);
645 result[3] = FABSF(a[3]);
646 store_vector4(inst, machine, result);
647 }
648 break;
649 case OPCODE_ADD:
650 {
651 GLfloat a[4], b[4], result[4];
652 fetch_vector4(&inst->SrcReg[0], machine, a);
653 fetch_vector4(&inst->SrcReg[1], machine, b);
654 result[0] = a[0] + b[0];
655 result[1] = a[1] + b[1];
656 result[2] = a[2] + b[2];
657 result[3] = a[3] + b[3];
658 store_vector4(inst, machine, result);
659 if (DEBUG_PROG) {
660 printf("ADD (%g %g %g %g) = (%g %g %g %g) + (%g %g %g %g)\n",
661 result[0], result[1], result[2], result[3],
662 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
663 }
664 }
665 break;
666 case OPCODE_AND: /* bitwise AND */
667 {
668 GLuint a[4], b[4], result[4];
669 fetch_vector4ui(&inst->SrcReg[0], machine, a);
670 fetch_vector4ui(&inst->SrcReg[1], machine, b);
671 result[0] = a[0] & b[0];
672 result[1] = a[1] & b[1];
673 result[2] = a[2] & b[2];
674 result[3] = a[3] & b[3];
675 store_vector4ui(inst, machine, result);
676 }
677 break;
678 case OPCODE_ARL:
679 {
680 GLfloat t[4];
681 fetch_vector4(&inst->SrcReg[0], machine, t);
682 machine->AddressReg[0][0] = IFLOOR(t[0]);
683 }
684 break;
685 case OPCODE_BGNLOOP:
686 /* no-op */
687 break;
688 case OPCODE_ENDLOOP:
689 /* subtract 1 here since pc is incremented by for(pc) loop */
690 pc = inst->BranchTarget - 1; /* go to matching BNGLOOP */
691 break;
692 case OPCODE_BGNSUB: /* begin subroutine */
693 break;
694 case OPCODE_ENDSUB: /* end subroutine */
695 break;
696 case OPCODE_BRA: /* branch (conditional) */
697 /* fall-through */
698 case OPCODE_BRK: /* break out of loop (conditional) */
699 /* fall-through */
700 case OPCODE_CONT: /* continue loop (conditional) */
701 if (eval_condition(machine, inst)) {
702 /* take branch */
703 /* Subtract 1 here since we'll do pc++ at end of for-loop */
704 pc = inst->BranchTarget - 1;
705 }
706 break;
707 case OPCODE_CAL: /* Call subroutine (conditional) */
708 if (eval_condition(machine, inst)) {
709 /* call the subroutine */
710 if (machine->StackDepth >= MAX_PROGRAM_CALL_DEPTH) {
711 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
712 }
713 machine->CallStack[machine->StackDepth++] = pc + 1; /* next inst */
714 /* Subtract 1 here since we'll do pc++ at end of for-loop */
715 pc = inst->BranchTarget - 1;
716 }
717 break;
718 case OPCODE_CMP:
719 {
720 GLfloat a[4], b[4], c[4], result[4];
721 fetch_vector4(&inst->SrcReg[0], machine, a);
722 fetch_vector4(&inst->SrcReg[1], machine, b);
723 fetch_vector4(&inst->SrcReg[2], machine, c);
724 result[0] = a[0] < 0.0F ? b[0] : c[0];
725 result[1] = a[1] < 0.0F ? b[1] : c[1];
726 result[2] = a[2] < 0.0F ? b[2] : c[2];
727 result[3] = a[3] < 0.0F ? b[3] : c[3];
728 store_vector4(inst, machine, result);
729 }
730 break;
731 case OPCODE_COS:
732 {
733 GLfloat a[4], result[4];
734 fetch_vector1(&inst->SrcReg[0], machine, a);
735 result[0] = result[1] = result[2] = result[3]
736 = (GLfloat) _mesa_cos(a[0]);
737 store_vector4(inst, machine, result);
738 }
739 break;
740 case OPCODE_DDX: /* Partial derivative with respect to X */
741 {
742 GLfloat result[4];
743 fetch_vector4_deriv(ctx, &inst->SrcReg[0], machine,
744 'X', result);
745 store_vector4(inst, machine, result);
746 }
747 break;
748 case OPCODE_DDY: /* Partial derivative with respect to Y */
749 {
750 GLfloat result[4];
751 fetch_vector4_deriv(ctx, &inst->SrcReg[0], machine,
752 'Y', result);
753 store_vector4(inst, machine, result);
754 }
755 break;
756 case OPCODE_DP2:
757 {
758 GLfloat a[4], b[4], result[4];
759 fetch_vector4(&inst->SrcReg[0], machine, a);
760 fetch_vector4(&inst->SrcReg[1], machine, b);
761 result[0] = result[1] = result[2] = result[3] = DOT2(a, b);
762 store_vector4(inst, machine, result);
763 if (DEBUG_PROG) {
764 printf("DP2 %g = (%g %g) . (%g %g)\n",
765 result[0], a[0], a[1], b[0], b[1]);
766 }
767 }
768 break;
769 case OPCODE_DP2A:
770 {
771 GLfloat a[4], b[4], c, result[4];
772 fetch_vector4(&inst->SrcReg[0], machine, a);
773 fetch_vector4(&inst->SrcReg[1], machine, b);
774 fetch_vector1(&inst->SrcReg[1], machine, &c);
775 result[0] = result[1] = result[2] = result[3] = DOT2(a, b) + c;
776 store_vector4(inst, machine, result);
777 if (DEBUG_PROG) {
778 printf("DP2A %g = (%g %g) . (%g %g) + %g\n",
779 result[0], a[0], a[1], b[0], b[1], c);
780 }
781 }
782 break;
783 case OPCODE_DP3:
784 {
785 GLfloat a[4], b[4], result[4];
786 fetch_vector4(&inst->SrcReg[0], machine, a);
787 fetch_vector4(&inst->SrcReg[1], machine, b);
788 result[0] = result[1] = result[2] = result[3] = DOT3(a, b);
789 store_vector4(inst, machine, result);
790 if (DEBUG_PROG) {
791 printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
792 result[0], a[0], a[1], a[2], b[0], b[1], b[2]);
793 }
794 }
795 break;
796 case OPCODE_DP4:
797 {
798 GLfloat a[4], b[4], result[4];
799 fetch_vector4(&inst->SrcReg[0], machine, a);
800 fetch_vector4(&inst->SrcReg[1], machine, b);
801 result[0] = result[1] = result[2] = result[3] = DOT4(a, b);
802 store_vector4(inst, machine, result);
803 if (DEBUG_PROG) {
804 printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
805 result[0], a[0], a[1], a[2], a[3],
806 b[0], b[1], b[2], b[3]);
807 }
808 }
809 break;
810 case OPCODE_DPH:
811 {
812 GLfloat a[4], b[4], result[4];
813 fetch_vector4(&inst->SrcReg[0], machine, a);
814 fetch_vector4(&inst->SrcReg[1], machine, b);
815 result[0] = result[1] = result[2] = result[3] = DOT3(a, b) + b[3];
816 store_vector4(inst, machine, result);
817 }
818 break;
819 case OPCODE_DST: /* Distance vector */
820 {
821 GLfloat a[4], b[4], result[4];
822 fetch_vector4(&inst->SrcReg[0], machine, a);
823 fetch_vector4(&inst->SrcReg[1], machine, b);
824 result[0] = 1.0F;
825 result[1] = a[1] * b[1];
826 result[2] = a[2];
827 result[3] = b[3];
828 store_vector4(inst, machine, result);
829 }
830 break;
831 case OPCODE_EXP:
832 {
833 GLfloat t[4], q[4], floor_t0;
834 fetch_vector1(&inst->SrcReg[0], machine, t);
835 floor_t0 = FLOORF(t[0]);
836 if (floor_t0 > FLT_MAX_EXP) {
837 SET_POS_INFINITY(q[0]);
838 SET_POS_INFINITY(q[2]);
839 }
840 else if (floor_t0 < FLT_MIN_EXP) {
841 q[0] = 0.0F;
842 q[2] = 0.0F;
843 }
844 else {
845 q[0] = LDEXPF(1.0, (int) floor_t0);
846 /* Note: GL_NV_vertex_program expects
847 * result.z = result.x * APPX(result.y)
848 * We do what the ARB extension says.
849 */
850 q[2] = (GLfloat) _mesa_pow(2.0, t[0]);
851 }
852 q[1] = t[0] - floor_t0;
853 q[3] = 1.0F;
854 store_vector4( inst, machine, q );
855 }
856 break;
857 case OPCODE_EX2: /* Exponential base 2 */
858 {
859 GLfloat a[4], result[4], val;
860 fetch_vector1(&inst->SrcReg[0], machine, a);
861 val = (GLfloat) _mesa_pow(2.0, a[0]);
862 /*
863 if (IS_INF_OR_NAN(val))
864 val = 1.0e10;
865 */
866 result[0] = result[1] = result[2] = result[3] = val;
867 store_vector4(inst, machine, result);
868 }
869 break;
870 case OPCODE_FLR:
871 {
872 GLfloat a[4], result[4];
873 fetch_vector4(&inst->SrcReg[0], machine, a);
874 result[0] = FLOORF(a[0]);
875 result[1] = FLOORF(a[1]);
876 result[2] = FLOORF(a[2]);
877 result[3] = FLOORF(a[3]);
878 store_vector4(inst, machine, result);
879 }
880 break;
881 case OPCODE_FRC:
882 {
883 GLfloat a[4], result[4];
884 fetch_vector4(&inst->SrcReg[0], machine, a);
885 result[0] = a[0] - FLOORF(a[0]);
886 result[1] = a[1] - FLOORF(a[1]);
887 result[2] = a[2] - FLOORF(a[2]);
888 result[3] = a[3] - FLOORF(a[3]);
889 store_vector4(inst, machine, result);
890 }
891 break;
892 case OPCODE_IF:
893 {
894 GLboolean cond;
895 /* eval condition */
896 if (inst->SrcReg[0].File != PROGRAM_UNDEFINED) {
897 GLfloat a[4];
898 fetch_vector1(&inst->SrcReg[0], machine, a);
899 cond = (a[0] != 0.0);
900 }
901 else {
902 cond = eval_condition(machine, inst);
903 }
904 if (DEBUG_PROG) {
905 printf("IF: %d\n", cond);
906 }
907 /* do if/else */
908 if (cond) {
909 /* do if-clause (just continue execution) */
910 }
911 else {
912 /* go to the instruction after ELSE or ENDIF */
913 assert(inst->BranchTarget >= 0);
914 pc = inst->BranchTarget - 1;
915 }
916 }
917 break;
918 case OPCODE_ELSE:
919 /* goto ENDIF */
920 assert(inst->BranchTarget >= 0);
921 pc = inst->BranchTarget - 1;
922 break;
923 case OPCODE_ENDIF:
924 /* nothing */
925 break;
926 case OPCODE_KIL_NV: /* NV_f_p only (conditional) */
927 if (eval_condition(machine, inst)) {
928 return GL_FALSE;
929 }
930 break;
931 case OPCODE_KIL: /* ARB_f_p only */
932 {
933 GLfloat a[4];
934 fetch_vector4(&inst->SrcReg[0], machine, a);
935 if (DEBUG_PROG) {
936 printf("KIL if (%g %g %g %g) <= 0.0\n",
937 a[0], a[1], a[2], a[3]);
938 }
939
940 if (a[0] < 0.0F || a[1] < 0.0F || a[2] < 0.0F || a[3] < 0.0F) {
941 return GL_FALSE;
942 }
943 }
944 break;
945 case OPCODE_LG2: /* log base 2 */
946 {
947 GLfloat a[4], result[4], val;
948 fetch_vector1(&inst->SrcReg[0], machine, a);
949 /* The fast LOG2 macro doesn't meet the precision requirements.
950 */
951 if (a[0] == 0.0F) {
952 val = -FLT_MAX;
953 }
954 else {
955 val = log(a[0]) * 1.442695F;
956 }
957 result[0] = result[1] = result[2] = result[3] = val;
958 store_vector4(inst, machine, result);
959 }
960 break;
961 case OPCODE_LIT:
962 {
963 const GLfloat epsilon = 1.0F / 256.0F; /* from NV VP spec */
964 GLfloat a[4], result[4];
965 fetch_vector4(&inst->SrcReg[0], machine, a);
966 a[0] = MAX2(a[0], 0.0F);
967 a[1] = MAX2(a[1], 0.0F);
968 /* XXX ARB version clamps a[3], NV version doesn't */
969 a[3] = CLAMP(a[3], -(128.0F - epsilon), (128.0F - epsilon));
970 result[0] = 1.0F;
971 result[1] = a[0];
972 /* XXX we could probably just use pow() here */
973 if (a[0] > 0.0F) {
974 if (a[1] == 0.0 && a[3] == 0.0)
975 result[2] = 1.0;
976 else
977 result[2] = (GLfloat) _mesa_pow(a[1], a[3]);
978 }
979 else {
980 result[2] = 0.0;
981 }
982 result[3] = 1.0F;
983 store_vector4(inst, machine, result);
984 if (DEBUG_PROG) {
985 printf("LIT (%g %g %g %g) : (%g %g %g %g)\n",
986 result[0], result[1], result[2], result[3],
987 a[0], a[1], a[2], a[3]);
988 }
989 }
990 break;
991 case OPCODE_LOG:
992 {
993 GLfloat t[4], q[4], abs_t0;
994 fetch_vector1(&inst->SrcReg[0], machine, t);
995 abs_t0 = FABSF(t[0]);
996 if (abs_t0 != 0.0F) {
997 /* Since we really can't handle infinite values on VMS
998 * like other OSes we'll use __MAXFLOAT to represent
999 * infinity. This may need some tweaking.
1000 */
1001 #ifdef VMS
1002 if (abs_t0 == __MAXFLOAT)
1003 #else
1004 if (IS_INF_OR_NAN(abs_t0))
1005 #endif
1006 {
1007 SET_POS_INFINITY(q[0]);
1008 q[1] = 1.0F;
1009 SET_POS_INFINITY(q[2]);
1010 }
1011 else {
1012 int exponent;
1013 GLfloat mantissa = FREXPF(t[0], &exponent);
1014 q[0] = (GLfloat) (exponent - 1);
1015 q[1] = (GLfloat) (2.0 * mantissa); /* map [.5, 1) -> [1, 2) */
1016
1017 /* The fast LOG2 macro doesn't meet the precision
1018 * requirements.
1019 */
1020 q[2] = (log(t[0]) * 1.442695F);
1021 }
1022 }
1023 else {
1024 SET_NEG_INFINITY(q[0]);
1025 q[1] = 1.0F;
1026 SET_NEG_INFINITY(q[2]);
1027 }
1028 q[3] = 1.0;
1029 store_vector4(inst, machine, q);
1030 }
1031 break;
1032 case OPCODE_LRP:
1033 {
1034 GLfloat a[4], b[4], c[4], result[4];
1035 fetch_vector4(&inst->SrcReg[0], machine, a);
1036 fetch_vector4(&inst->SrcReg[1], machine, b);
1037 fetch_vector4(&inst->SrcReg[2], machine, c);
1038 result[0] = a[0] * b[0] + (1.0F - a[0]) * c[0];
1039 result[1] = a[1] * b[1] + (1.0F - a[1]) * c[1];
1040 result[2] = a[2] * b[2] + (1.0F - a[2]) * c[2];
1041 result[3] = a[3] * b[3] + (1.0F - a[3]) * c[3];
1042 store_vector4(inst, machine, result);
1043 if (DEBUG_PROG) {
1044 printf("LRP (%g %g %g %g) = (%g %g %g %g), "
1045 "(%g %g %g %g), (%g %g %g %g)\n",
1046 result[0], result[1], result[2], result[3],
1047 a[0], a[1], a[2], a[3],
1048 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
1049 }
1050 }
1051 break;
1052 case OPCODE_MAD:
1053 {
1054 GLfloat a[4], b[4], c[4], result[4];
1055 fetch_vector4(&inst->SrcReg[0], machine, a);
1056 fetch_vector4(&inst->SrcReg[1], machine, b);
1057 fetch_vector4(&inst->SrcReg[2], machine, c);
1058 result[0] = a[0] * b[0] + c[0];
1059 result[1] = a[1] * b[1] + c[1];
1060 result[2] = a[2] * b[2] + c[2];
1061 result[3] = a[3] * b[3] + c[3];
1062 store_vector4(inst, machine, result);
1063 if (DEBUG_PROG) {
1064 printf("MAD (%g %g %g %g) = (%g %g %g %g) * "
1065 "(%g %g %g %g) + (%g %g %g %g)\n",
1066 result[0], result[1], result[2], result[3],
1067 a[0], a[1], a[2], a[3],
1068 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
1069 }
1070 }
1071 break;
1072 case OPCODE_MAX:
1073 {
1074 GLfloat a[4], b[4], result[4];
1075 fetch_vector4(&inst->SrcReg[0], machine, a);
1076 fetch_vector4(&inst->SrcReg[1], machine, b);
1077 result[0] = MAX2(a[0], b[0]);
1078 result[1] = MAX2(a[1], b[1]);
1079 result[2] = MAX2(a[2], b[2]);
1080 result[3] = MAX2(a[3], b[3]);
1081 store_vector4(inst, machine, result);
1082 if (DEBUG_PROG) {
1083 printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
1084 result[0], result[1], result[2], result[3],
1085 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1086 }
1087 }
1088 break;
1089 case OPCODE_MIN:
1090 {
1091 GLfloat a[4], b[4], result[4];
1092 fetch_vector4(&inst->SrcReg[0], machine, a);
1093 fetch_vector4(&inst->SrcReg[1], machine, b);
1094 result[0] = MIN2(a[0], b[0]);
1095 result[1] = MIN2(a[1], b[1]);
1096 result[2] = MIN2(a[2], b[2]);
1097 result[3] = MIN2(a[3], b[3]);
1098 store_vector4(inst, machine, result);
1099 }
1100 break;
1101 case OPCODE_MOV:
1102 {
1103 GLfloat result[4];
1104 fetch_vector4(&inst->SrcReg[0], machine, result);
1105 store_vector4(inst, machine, result);
1106 if (DEBUG_PROG) {
1107 printf("MOV (%g %g %g %g)\n",
1108 result[0], result[1], result[2], result[3]);
1109 }
1110 }
1111 break;
1112 case OPCODE_MUL:
1113 {
1114 GLfloat a[4], b[4], result[4];
1115 fetch_vector4(&inst->SrcReg[0], machine, a);
1116 fetch_vector4(&inst->SrcReg[1], machine, b);
1117 result[0] = a[0] * b[0];
1118 result[1] = a[1] * b[1];
1119 result[2] = a[2] * b[2];
1120 result[3] = a[3] * b[3];
1121 store_vector4(inst, machine, result);
1122 if (DEBUG_PROG) {
1123 printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
1124 result[0], result[1], result[2], result[3],
1125 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1126 }
1127 }
1128 break;
1129 case OPCODE_NOISE1:
1130 {
1131 GLfloat a[4], result[4];
1132 fetch_vector1(&inst->SrcReg[0], machine, a);
1133 result[0] =
1134 result[1] =
1135 result[2] =
1136 result[3] = _mesa_noise1(a[0]);
1137 store_vector4(inst, machine, result);
1138 }
1139 break;
1140 case OPCODE_NOISE2:
1141 {
1142 GLfloat a[4], result[4];
1143 fetch_vector4(&inst->SrcReg[0], machine, a);
1144 result[0] =
1145 result[1] =
1146 result[2] = result[3] = _mesa_noise2(a[0], a[1]);
1147 store_vector4(inst, machine, result);
1148 }
1149 break;
1150 case OPCODE_NOISE3:
1151 {
1152 GLfloat a[4], result[4];
1153 fetch_vector4(&inst->SrcReg[0], machine, a);
1154 result[0] =
1155 result[1] =
1156 result[2] =
1157 result[3] = _mesa_noise3(a[0], a[1], a[2]);
1158 store_vector4(inst, machine, result);
1159 }
1160 break;
1161 case OPCODE_NOISE4:
1162 {
1163 GLfloat a[4], result[4];
1164 fetch_vector4(&inst->SrcReg[0], machine, a);
1165 result[0] =
1166 result[1] =
1167 result[2] =
1168 result[3] = _mesa_noise4(a[0], a[1], a[2], a[3]);
1169 store_vector4(inst, machine, result);
1170 }
1171 break;
1172 case OPCODE_NOP:
1173 break;
1174 case OPCODE_NOT: /* bitwise NOT */
1175 {
1176 GLuint a[4], result[4];
1177 fetch_vector4ui(&inst->SrcReg[0], machine, a);
1178 result[0] = ~a[0];
1179 result[1] = ~a[1];
1180 result[2] = ~a[2];
1181 result[3] = ~a[3];
1182 store_vector4ui(inst, machine, result);
1183 }
1184 break;
1185 case OPCODE_NRM3: /* 3-component normalization */
1186 {
1187 GLfloat a[4], result[4];
1188 GLfloat tmp;
1189 fetch_vector4(&inst->SrcReg[0], machine, a);
1190 tmp = a[0] * a[0] + a[1] * a[1] + a[2] * a[2];
1191 if (tmp != 0.0F)
1192 tmp = INV_SQRTF(tmp);
1193 result[0] = tmp * a[0];
1194 result[1] = tmp * a[1];
1195 result[2] = tmp * a[2];
1196 result[3] = 0.0; /* undefined, but prevent valgrind warnings */
1197 store_vector4(inst, machine, result);
1198 }
1199 break;
1200 case OPCODE_NRM4: /* 4-component normalization */
1201 {
1202 GLfloat a[4], result[4];
1203 GLfloat tmp;
1204 fetch_vector4(&inst->SrcReg[0], machine, a);
1205 tmp = a[0] * a[0] + a[1] * a[1] + a[2] * a[2] + a[3] * a[3];
1206 if (tmp != 0.0F)
1207 tmp = INV_SQRTF(tmp);
1208 result[0] = tmp * a[0];
1209 result[1] = tmp * a[1];
1210 result[2] = tmp * a[2];
1211 result[3] = tmp * a[3];
1212 store_vector4(inst, machine, result);
1213 }
1214 break;
1215 case OPCODE_OR: /* bitwise OR */
1216 {
1217 GLuint a[4], b[4], result[4];
1218 fetch_vector4ui(&inst->SrcReg[0], machine, a);
1219 fetch_vector4ui(&inst->SrcReg[1], machine, b);
1220 result[0] = a[0] | b[0];
1221 result[1] = a[1] | b[1];
1222 result[2] = a[2] | b[2];
1223 result[3] = a[3] | b[3];
1224 store_vector4ui(inst, machine, result);
1225 }
1226 break;
1227 case OPCODE_PK2H: /* pack two 16-bit floats in one 32-bit float */
1228 {
1229 GLfloat a[4];
1230 GLuint result[4];
1231 GLhalfNV hx, hy;
1232 fetch_vector4(&inst->SrcReg[0], machine, a);
1233 hx = _mesa_float_to_half(a[0]);
1234 hy = _mesa_float_to_half(a[1]);
1235 result[0] =
1236 result[1] =
1237 result[2] =
1238 result[3] = hx | (hy << 16);
1239 store_vector4ui(inst, machine, result);
1240 }
1241 break;
1242 case OPCODE_PK2US: /* pack two GLushorts into one 32-bit float */
1243 {
1244 GLfloat a[4];
1245 GLuint result[4], usx, usy;
1246 fetch_vector4(&inst->SrcReg[0], machine, a);
1247 a[0] = CLAMP(a[0], 0.0F, 1.0F);
1248 a[1] = CLAMP(a[1], 0.0F, 1.0F);
1249 usx = IROUND(a[0] * 65535.0F);
1250 usy = IROUND(a[1] * 65535.0F);
1251 result[0] =
1252 result[1] =
1253 result[2] =
1254 result[3] = usx | (usy << 16);
1255 store_vector4ui(inst, machine, result);
1256 }
1257 break;
1258 case OPCODE_PK4B: /* pack four GLbytes into one 32-bit float */
1259 {
1260 GLfloat a[4];
1261 GLuint result[4], ubx, uby, ubz, ubw;
1262 fetch_vector4(&inst->SrcReg[0], machine, a);
1263 a[0] = CLAMP(a[0], -128.0F / 127.0F, 1.0F);
1264 a[1] = CLAMP(a[1], -128.0F / 127.0F, 1.0F);
1265 a[2] = CLAMP(a[2], -128.0F / 127.0F, 1.0F);
1266 a[3] = CLAMP(a[3], -128.0F / 127.0F, 1.0F);
1267 ubx = IROUND(127.0F * a[0] + 128.0F);
1268 uby = IROUND(127.0F * a[1] + 128.0F);
1269 ubz = IROUND(127.0F * a[2] + 128.0F);
1270 ubw = IROUND(127.0F * a[3] + 128.0F);
1271 result[0] =
1272 result[1] =
1273 result[2] =
1274 result[3] = ubx | (uby << 8) | (ubz << 16) | (ubw << 24);
1275 store_vector4ui(inst, machine, result);
1276 }
1277 break;
1278 case OPCODE_PK4UB: /* pack four GLubytes into one 32-bit float */
1279 {
1280 GLfloat a[4];
1281 GLuint result[4], ubx, uby, ubz, ubw;
1282 fetch_vector4(&inst->SrcReg[0], machine, a);
1283 a[0] = CLAMP(a[0], 0.0F, 1.0F);
1284 a[1] = CLAMP(a[1], 0.0F, 1.0F);
1285 a[2] = CLAMP(a[2], 0.0F, 1.0F);
1286 a[3] = CLAMP(a[3], 0.0F, 1.0F);
1287 ubx = IROUND(255.0F * a[0]);
1288 uby = IROUND(255.0F * a[1]);
1289 ubz = IROUND(255.0F * a[2]);
1290 ubw = IROUND(255.0F * a[3]);
1291 result[0] =
1292 result[1] =
1293 result[2] =
1294 result[3] = ubx | (uby << 8) | (ubz << 16) | (ubw << 24);
1295 store_vector4ui(inst, machine, result);
1296 }
1297 break;
1298 case OPCODE_POW:
1299 {
1300 GLfloat a[4], b[4], result[4];
1301 fetch_vector1(&inst->SrcReg[0], machine, a);
1302 fetch_vector1(&inst->SrcReg[1], machine, b);
1303 result[0] = result[1] = result[2] = result[3]
1304 = (GLfloat) _mesa_pow(a[0], b[0]);
1305 store_vector4(inst, machine, result);
1306 }
1307 break;
1308 case OPCODE_RCP:
1309 {
1310 GLfloat a[4], result[4];
1311 fetch_vector1(&inst->SrcReg[0], machine, a);
1312 if (DEBUG_PROG) {
1313 if (a[0] == 0)
1314 printf("RCP(0)\n");
1315 else if (IS_INF_OR_NAN(a[0]))
1316 printf("RCP(inf)\n");
1317 }
1318 result[0] = result[1] = result[2] = result[3] = 1.0F / a[0];
1319 store_vector4(inst, machine, result);
1320 }
1321 break;
1322 case OPCODE_RET: /* return from subroutine (conditional) */
1323 if (eval_condition(machine, inst)) {
1324 if (machine->StackDepth == 0) {
1325 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
1326 }
1327 /* subtract one because of pc++ in the for loop */
1328 pc = machine->CallStack[--machine->StackDepth] - 1;
1329 }
1330 break;
1331 case OPCODE_RFL: /* reflection vector */
1332 {
1333 GLfloat axis[4], dir[4], result[4], tmpX, tmpW;
1334 fetch_vector4(&inst->SrcReg[0], machine, axis);
1335 fetch_vector4(&inst->SrcReg[1], machine, dir);
1336 tmpW = DOT3(axis, axis);
1337 tmpX = (2.0F * DOT3(axis, dir)) / tmpW;
1338 result[0] = tmpX * axis[0] - dir[0];
1339 result[1] = tmpX * axis[1] - dir[1];
1340 result[2] = tmpX * axis[2] - dir[2];
1341 /* result[3] is never written! XXX enforce in parser! */
1342 store_vector4(inst, machine, result);
1343 }
1344 break;
1345 case OPCODE_RSQ: /* 1 / sqrt() */
1346 {
1347 GLfloat a[4], result[4];
1348 fetch_vector1(&inst->SrcReg[0], machine, a);
1349 a[0] = FABSF(a[0]);
1350 result[0] = result[1] = result[2] = result[3] = INV_SQRTF(a[0]);
1351 store_vector4(inst, machine, result);
1352 if (DEBUG_PROG) {
1353 printf("RSQ %g = 1/sqrt(|%g|)\n", result[0], a[0]);
1354 }
1355 }
1356 break;
1357 case OPCODE_SCS: /* sine and cos */
1358 {
1359 GLfloat a[4], result[4];
1360 fetch_vector1(&inst->SrcReg[0], machine, a);
1361 result[0] = (GLfloat) _mesa_cos(a[0]);
1362 result[1] = (GLfloat) _mesa_sin(a[0]);
1363 result[2] = 0.0; /* undefined! */
1364 result[3] = 0.0; /* undefined! */
1365 store_vector4(inst, machine, result);
1366 }
1367 break;
1368 case OPCODE_SEQ: /* set on equal */
1369 {
1370 GLfloat a[4], b[4], result[4];
1371 fetch_vector4(&inst->SrcReg[0], machine, a);
1372 fetch_vector4(&inst->SrcReg[1], machine, b);
1373 result[0] = (a[0] == b[0]) ? 1.0F : 0.0F;
1374 result[1] = (a[1] == b[1]) ? 1.0F : 0.0F;
1375 result[2] = (a[2] == b[2]) ? 1.0F : 0.0F;
1376 result[3] = (a[3] == b[3]) ? 1.0F : 0.0F;
1377 store_vector4(inst, machine, result);
1378 if (DEBUG_PROG) {
1379 printf("SEQ (%g %g %g %g) = (%g %g %g %g) == (%g %g %g %g)\n",
1380 result[0], result[1], result[2], result[3],
1381 a[0], a[1], a[2], a[3],
1382 b[0], b[1], b[2], b[3]);
1383 }
1384 }
1385 break;
1386 case OPCODE_SFL: /* set false, operands ignored */
1387 {
1388 static const GLfloat result[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
1389 store_vector4(inst, machine, result);
1390 }
1391 break;
1392 case OPCODE_SGE: /* set on greater or equal */
1393 {
1394 GLfloat a[4], b[4], result[4];
1395 fetch_vector4(&inst->SrcReg[0], machine, a);
1396 fetch_vector4(&inst->SrcReg[1], machine, b);
1397 result[0] = (a[0] >= b[0]) ? 1.0F : 0.0F;
1398 result[1] = (a[1] >= b[1]) ? 1.0F : 0.0F;
1399 result[2] = (a[2] >= b[2]) ? 1.0F : 0.0F;
1400 result[3] = (a[3] >= b[3]) ? 1.0F : 0.0F;
1401 store_vector4(inst, machine, result);
1402 if (DEBUG_PROG) {
1403 printf("SGE (%g %g %g %g) = (%g %g %g %g) >= (%g %g %g %g)\n",
1404 result[0], result[1], result[2], result[3],
1405 a[0], a[1], a[2], a[3],
1406 b[0], b[1], b[2], b[3]);
1407 }
1408 }
1409 break;
1410 case OPCODE_SGT: /* set on greater */
1411 {
1412 GLfloat a[4], b[4], result[4];
1413 fetch_vector4(&inst->SrcReg[0], machine, a);
1414 fetch_vector4(&inst->SrcReg[1], machine, b);
1415 result[0] = (a[0] > b[0]) ? 1.0F : 0.0F;
1416 result[1] = (a[1] > b[1]) ? 1.0F : 0.0F;
1417 result[2] = (a[2] > b[2]) ? 1.0F : 0.0F;
1418 result[3] = (a[3] > b[3]) ? 1.0F : 0.0F;
1419 store_vector4(inst, machine, result);
1420 if (DEBUG_PROG) {
1421 printf("SGT (%g %g %g %g) = (%g %g %g %g) > (%g %g %g %g)\n",
1422 result[0], result[1], result[2], result[3],
1423 a[0], a[1], a[2], a[3],
1424 b[0], b[1], b[2], b[3]);
1425 }
1426 }
1427 break;
1428 case OPCODE_SIN:
1429 {
1430 GLfloat a[4], result[4];
1431 fetch_vector1(&inst->SrcReg[0], machine, a);
1432 result[0] = result[1] = result[2] = result[3]
1433 = (GLfloat) _mesa_sin(a[0]);
1434 store_vector4(inst, machine, result);
1435 }
1436 break;
1437 case OPCODE_SLE: /* set on less or equal */
1438 {
1439 GLfloat a[4], b[4], result[4];
1440 fetch_vector4(&inst->SrcReg[0], machine, a);
1441 fetch_vector4(&inst->SrcReg[1], machine, b);
1442 result[0] = (a[0] <= b[0]) ? 1.0F : 0.0F;
1443 result[1] = (a[1] <= b[1]) ? 1.0F : 0.0F;
1444 result[2] = (a[2] <= b[2]) ? 1.0F : 0.0F;
1445 result[3] = (a[3] <= b[3]) ? 1.0F : 0.0F;
1446 store_vector4(inst, machine, result);
1447 if (DEBUG_PROG) {
1448 printf("SLE (%g %g %g %g) = (%g %g %g %g) <= (%g %g %g %g)\n",
1449 result[0], result[1], result[2], result[3],
1450 a[0], a[1], a[2], a[3],
1451 b[0], b[1], b[2], b[3]);
1452 }
1453 }
1454 break;
1455 case OPCODE_SLT: /* set on less */
1456 {
1457 GLfloat a[4], b[4], result[4];
1458 fetch_vector4(&inst->SrcReg[0], machine, a);
1459 fetch_vector4(&inst->SrcReg[1], machine, b);
1460 result[0] = (a[0] < b[0]) ? 1.0F : 0.0F;
1461 result[1] = (a[1] < b[1]) ? 1.0F : 0.0F;
1462 result[2] = (a[2] < b[2]) ? 1.0F : 0.0F;
1463 result[3] = (a[3] < b[3]) ? 1.0F : 0.0F;
1464 store_vector4(inst, machine, result);
1465 if (DEBUG_PROG) {
1466 printf("SLT (%g %g %g %g) = (%g %g %g %g) < (%g %g %g %g)\n",
1467 result[0], result[1], result[2], result[3],
1468 a[0], a[1], a[2], a[3],
1469 b[0], b[1], b[2], b[3]);
1470 }
1471 }
1472 break;
1473 case OPCODE_SNE: /* set on not equal */
1474 {
1475 GLfloat a[4], b[4], result[4];
1476 fetch_vector4(&inst->SrcReg[0], machine, a);
1477 fetch_vector4(&inst->SrcReg[1], machine, b);
1478 result[0] = (a[0] != b[0]) ? 1.0F : 0.0F;
1479 result[1] = (a[1] != b[1]) ? 1.0F : 0.0F;
1480 result[2] = (a[2] != b[2]) ? 1.0F : 0.0F;
1481 result[3] = (a[3] != b[3]) ? 1.0F : 0.0F;
1482 store_vector4(inst, machine, result);
1483 if (DEBUG_PROG) {
1484 printf("SNE (%g %g %g %g) = (%g %g %g %g) != (%g %g %g %g)\n",
1485 result[0], result[1], result[2], result[3],
1486 a[0], a[1], a[2], a[3],
1487 b[0], b[1], b[2], b[3]);
1488 }
1489 }
1490 break;
1491 case OPCODE_SSG: /* set sign (-1, 0 or +1) */
1492 {
1493 GLfloat a[4], result[4];
1494 fetch_vector4(&inst->SrcReg[0], machine, a);
1495 result[0] = (GLfloat) ((a[0] > 0.0F) - (a[0] < 0.0F));
1496 result[1] = (GLfloat) ((a[1] > 0.0F) - (a[1] < 0.0F));
1497 result[2] = (GLfloat) ((a[2] > 0.0F) - (a[2] < 0.0F));
1498 result[3] = (GLfloat) ((a[3] > 0.0F) - (a[3] < 0.0F));
1499 store_vector4(inst, machine, result);
1500 }
1501 break;
1502 case OPCODE_STR: /* set true, operands ignored */
1503 {
1504 static const GLfloat result[4] = { 1.0F, 1.0F, 1.0F, 1.0F };
1505 store_vector4(inst, machine, result);
1506 }
1507 break;
1508 case OPCODE_SUB:
1509 {
1510 GLfloat a[4], b[4], result[4];
1511 fetch_vector4(&inst->SrcReg[0], machine, a);
1512 fetch_vector4(&inst->SrcReg[1], machine, b);
1513 result[0] = a[0] - b[0];
1514 result[1] = a[1] - b[1];
1515 result[2] = a[2] - b[2];
1516 result[3] = a[3] - b[3];
1517 store_vector4(inst, machine, result);
1518 if (DEBUG_PROG) {
1519 printf("SUB (%g %g %g %g) = (%g %g %g %g) - (%g %g %g %g)\n",
1520 result[0], result[1], result[2], result[3],
1521 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1522 }
1523 }
1524 break;
1525 case OPCODE_SWZ: /* extended swizzle */
1526 {
1527 const struct prog_src_register *source = &inst->SrcReg[0];
1528 const GLfloat *src = get_src_register_pointer(source, machine);
1529 GLfloat result[4];
1530 GLuint i;
1531 for (i = 0; i < 4; i++) {
1532 const GLuint swz = GET_SWZ(source->Swizzle, i);
1533 if (swz == SWIZZLE_ZERO)
1534 result[i] = 0.0;
1535 else if (swz == SWIZZLE_ONE)
1536 result[i] = 1.0;
1537 else {
1538 ASSERT(swz >= 0);
1539 ASSERT(swz <= 3);
1540 result[i] = src[swz];
1541 }
1542 if (source->Negate & (1 << i))
1543 result[i] = -result[i];
1544 }
1545 store_vector4(inst, machine, result);
1546 }
1547 break;
1548 case OPCODE_TEX: /* Both ARB and NV frag prog */
1549 /* Simple texel lookup */
1550 {
1551 GLfloat texcoord[4], color[4];
1552 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1553
1554 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1555
1556 if (DEBUG_PROG) {
1557 printf("TEX (%g, %g, %g, %g) = texture[%d][%g, %g, %g, %g]\n",
1558 color[0], color[1], color[2], color[3],
1559 inst->TexSrcUnit,
1560 texcoord[0], texcoord[1], texcoord[2], texcoord[3]);
1561 }
1562 store_vector4(inst, machine, color);
1563 }
1564 break;
1565 case OPCODE_TXB: /* GL_ARB_fragment_program only */
1566 /* Texel lookup with LOD bias */
1567 {
1568 GLfloat texcoord[4], color[4], lodBias;
1569
1570 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1571
1572 /* texcoord[3] is the bias to add to lambda */
1573 lodBias = texcoord[3];
1574
1575 fetch_texel(ctx, machine, inst, texcoord, lodBias, color);
1576
1577 store_vector4(inst, machine, color);
1578 }
1579 break;
1580 case OPCODE_TXD: /* GL_NV_fragment_program only */
1581 /* Texture lookup w/ partial derivatives for LOD */
1582 {
1583 GLfloat texcoord[4], dtdx[4], dtdy[4], color[4];
1584 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1585 fetch_vector4(&inst->SrcReg[1], machine, dtdx);
1586 fetch_vector4(&inst->SrcReg[2], machine, dtdy);
1587 machine->FetchTexelDeriv(ctx, texcoord, dtdx, dtdy,
1588 0.0, /* lodBias */
1589 inst->TexSrcUnit, color);
1590 store_vector4(inst, machine, color);
1591 }
1592 break;
1593 case OPCODE_TXP: /* GL_ARB_fragment_program only */
1594 /* Texture lookup w/ projective divide */
1595 {
1596 GLfloat texcoord[4], color[4];
1597
1598 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1599 /* Not so sure about this test - if texcoord[3] is
1600 * zero, we'd probably be fine except for an ASSERT in
1601 * IROUND_POS() which gets triggered by the inf values created.
1602 */
1603 if (texcoord[3] != 0.0) {
1604 texcoord[0] /= texcoord[3];
1605 texcoord[1] /= texcoord[3];
1606 texcoord[2] /= texcoord[3];
1607 }
1608
1609 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1610
1611 store_vector4(inst, machine, color);
1612 }
1613 break;
1614 case OPCODE_TXP_NV: /* GL_NV_fragment_program only */
1615 /* Texture lookup w/ projective divide, as above, but do not
1616 * do the divide by w if sampling from a cube map.
1617 */
1618 {
1619 GLfloat texcoord[4], color[4];
1620
1621 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1622 if (inst->TexSrcTarget != TEXTURE_CUBE_INDEX &&
1623 texcoord[3] != 0.0) {
1624 texcoord[0] /= texcoord[3];
1625 texcoord[1] /= texcoord[3];
1626 texcoord[2] /= texcoord[3];
1627 }
1628
1629 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1630
1631 store_vector4(inst, machine, color);
1632 }
1633 break;
1634 case OPCODE_TRUNC: /* truncate toward zero */
1635 {
1636 GLfloat a[4], result[4];
1637 fetch_vector4(&inst->SrcReg[0], machine, a);
1638 result[0] = (GLfloat) (GLint) a[0];
1639 result[1] = (GLfloat) (GLint) a[1];
1640 result[2] = (GLfloat) (GLint) a[2];
1641 result[3] = (GLfloat) (GLint) a[3];
1642 store_vector4(inst, machine, result);
1643 }
1644 break;
1645 case OPCODE_UP2H: /* unpack two 16-bit floats */
1646 {
1647 GLfloat a[4], result[4];
1648 fi_type fi;
1649 GLhalfNV hx, hy;
1650 fetch_vector1(&inst->SrcReg[0], machine, a);
1651 fi.f = a[0];
1652 hx = fi.i & 0xffff;
1653 hy = fi.i >> 16;
1654 result[0] = result[2] = _mesa_half_to_float(hx);
1655 result[1] = result[3] = _mesa_half_to_float(hy);
1656 store_vector4(inst, machine, result);
1657 }
1658 break;
1659 case OPCODE_UP2US: /* unpack two GLushorts */
1660 {
1661 GLfloat a[4], result[4];
1662 fi_type fi;
1663 GLushort usx, usy;
1664 fetch_vector1(&inst->SrcReg[0], machine, a);
1665 fi.f = a[0];
1666 usx = fi.i & 0xffff;
1667 usy = fi.i >> 16;
1668 result[0] = result[2] = usx * (1.0f / 65535.0f);
1669 result[1] = result[3] = usy * (1.0f / 65535.0f);
1670 store_vector4(inst, machine, result);
1671 }
1672 break;
1673 case OPCODE_UP4B: /* unpack four GLbytes */
1674 {
1675 GLfloat a[4], result[4];
1676 fi_type fi;
1677 fetch_vector1(&inst->SrcReg[0], machine, a);
1678 fi.f = a[0];
1679 result[0] = (((fi.i >> 0) & 0xff) - 128) / 127.0F;
1680 result[1] = (((fi.i >> 8) & 0xff) - 128) / 127.0F;
1681 result[2] = (((fi.i >> 16) & 0xff) - 128) / 127.0F;
1682 result[3] = (((fi.i >> 24) & 0xff) - 128) / 127.0F;
1683 store_vector4(inst, machine, result);
1684 }
1685 break;
1686 case OPCODE_UP4UB: /* unpack four GLubytes */
1687 {
1688 GLfloat a[4], result[4];
1689 fi_type fi;
1690 fetch_vector1(&inst->SrcReg[0], machine, a);
1691 fi.f = a[0];
1692 result[0] = ((fi.i >> 0) & 0xff) / 255.0F;
1693 result[1] = ((fi.i >> 8) & 0xff) / 255.0F;
1694 result[2] = ((fi.i >> 16) & 0xff) / 255.0F;
1695 result[3] = ((fi.i >> 24) & 0xff) / 255.0F;
1696 store_vector4(inst, machine, result);
1697 }
1698 break;
1699 case OPCODE_XOR: /* bitwise XOR */
1700 {
1701 GLuint a[4], b[4], result[4];
1702 fetch_vector4ui(&inst->SrcReg[0], machine, a);
1703 fetch_vector4ui(&inst->SrcReg[1], machine, b);
1704 result[0] = a[0] ^ b[0];
1705 result[1] = a[1] ^ b[1];
1706 result[2] = a[2] ^ b[2];
1707 result[3] = a[3] ^ b[3];
1708 store_vector4ui(inst, machine, result);
1709 }
1710 break;
1711 case OPCODE_XPD: /* cross product */
1712 {
1713 GLfloat a[4], b[4], result[4];
1714 fetch_vector4(&inst->SrcReg[0], machine, a);
1715 fetch_vector4(&inst->SrcReg[1], machine, b);
1716 result[0] = a[1] * b[2] - a[2] * b[1];
1717 result[1] = a[2] * b[0] - a[0] * b[2];
1718 result[2] = a[0] * b[1] - a[1] * b[0];
1719 result[3] = 1.0;
1720 store_vector4(inst, machine, result);
1721 if (DEBUG_PROG) {
1722 printf("XPD (%g %g %g %g) = (%g %g %g) X (%g %g %g)\n",
1723 result[0], result[1], result[2], result[3],
1724 a[0], a[1], a[2], b[0], b[1], b[2]);
1725 }
1726 }
1727 break;
1728 case OPCODE_X2D: /* 2-D matrix transform */
1729 {
1730 GLfloat a[4], b[4], c[4], result[4];
1731 fetch_vector4(&inst->SrcReg[0], machine, a);
1732 fetch_vector4(&inst->SrcReg[1], machine, b);
1733 fetch_vector4(&inst->SrcReg[2], machine, c);
1734 result[0] = a[0] + b[0] * c[0] + b[1] * c[1];
1735 result[1] = a[1] + b[0] * c[2] + b[1] * c[3];
1736 result[2] = a[2] + b[0] * c[0] + b[1] * c[1];
1737 result[3] = a[3] + b[0] * c[2] + b[1] * c[3];
1738 store_vector4(inst, machine, result);
1739 }
1740 break;
1741 case OPCODE_PRINT:
1742 {
1743 if (inst->SrcReg[0].File != -1) {
1744 GLfloat a[4];
1745 fetch_vector4(&inst->SrcReg[0], machine, a);
1746 _mesa_printf("%s%g, %g, %g, %g\n", (const char *) inst->Data,
1747 a[0], a[1], a[2], a[3]);
1748 }
1749 else {
1750 _mesa_printf("%s\n", (const char *) inst->Data);
1751 }
1752 }
1753 break;
1754 case OPCODE_END:
1755 return GL_TRUE;
1756 default:
1757 _mesa_problem(ctx, "Bad opcode %d in _mesa_execute_program",
1758 inst->Opcode);
1759 return GL_TRUE; /* return value doesn't matter */
1760 }
1761
1762 numExec++;
1763 if (numExec > maxExec) {
1764 _mesa_problem(ctx, "Infinite loop detected in fragment program");
1765 return GL_TRUE;
1766 }
1767
1768 } /* for pc */
1769
1770 return GL_TRUE;
1771 }