projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
mesa: Change "BRIAN PAUL" to "THE AUTHORS" in license text.
[mesa.git] / src / mesa / program / 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 * THE AUTHORS 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/macros.h"
41 #include "prog_execute.h"
42 #include "prog_instruction.h"
43 #include "prog_parameter.h"
44 #include "prog_print.h"
45 #include "prog_noise.h"
46
47
48 /* debug predicate */
49 #define DEBUG_PROG 0
50
51
52 /**
53 * Set x to positive or negative infinity.
54 */
55 #if defined(USE_IEEE) || defined(_WIN32)
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 #else
69 #define SET_POS_INFINITY(x) x = (GLfloat) HUGE_VAL
70 #define SET_NEG_INFINITY(x) x = (GLfloat) -HUGE_VAL
71 #endif
72
73 #define SET_FLOAT_BITS(x, bits) ((fi_type *) (void *) &(x))->i = bits
74
75
76 static const GLfloat ZeroVec[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
77
78
79 /**
80 * Return a pointer to the 4-element float vector specified by the given
81 * source register.
82 */
83 static inline const GLfloat *
84 get_src_register_pointer(const struct prog_src_register *source,
85 const struct gl_program_machine *machine)
86 {
87 const struct gl_program *prog = machine->CurProgram;
88 GLint reg = source->Index;
89
90 if (source->RelAddr) {
91 /* add address register value to src index/offset */
92 reg += machine->AddressReg[0][0];
93 if (reg < 0) {
94 return ZeroVec;
95 }
96 }
97
98 switch (source->File) {
99 case PROGRAM_TEMPORARY:
100 if (reg >= MAX_PROGRAM_TEMPS)
101 return ZeroVec;
102 return machine->Temporaries[reg];
103
104 case PROGRAM_INPUT:
105 if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
106 if (reg >= VERT_ATTRIB_MAX)
107 return ZeroVec;
108 return machine->VertAttribs[reg];
109 }
110 else {
111 if (reg >= VARYING_SLOT_MAX)
112 return ZeroVec;
113 return machine->Attribs[reg][machine->CurElement];
114 }
115
116 case PROGRAM_OUTPUT:
117 if (reg >= MAX_PROGRAM_OUTPUTS)
118 return ZeroVec;
119 return machine->Outputs[reg];
120
121 case PROGRAM_LOCAL_PARAM:
122 if (reg >= MAX_PROGRAM_LOCAL_PARAMS)
123 return ZeroVec;
124 return machine->CurProgram->LocalParams[reg];
125
126 case PROGRAM_ENV_PARAM:
127 if (reg >= MAX_PROGRAM_ENV_PARAMS)
128 return ZeroVec;
129 return machine->EnvParams[reg];
130
131 case PROGRAM_STATE_VAR:
132 /* Fallthrough */
133 case PROGRAM_CONSTANT:
134 /* Fallthrough */
135 case PROGRAM_UNIFORM:
136 if (reg >= (GLint) prog->Parameters->NumParameters)
137 return ZeroVec;
138 return (GLfloat *) prog->Parameters->ParameterValues[reg];
139
140 case PROGRAM_SYSTEM_VALUE:
141 assert(reg < Elements(machine->SystemValues));
142 return machine->SystemValues[reg];
143
144 default:
145 _mesa_problem(NULL,
146 "Invalid src register file %d in get_src_register_pointer()",
147 source->File);
148 return NULL;
149 }
150 }
151
152
153 /**
154 * Return a pointer to the 4-element float vector specified by the given
155 * destination register.
156 */
157 static inline GLfloat *
158 get_dst_register_pointer(const struct prog_dst_register *dest,
159 struct gl_program_machine *machine)
160 {
161 static GLfloat dummyReg[4];
162 GLint reg = dest->Index;
163
164 if (dest->RelAddr) {
165 /* add address register value to src index/offset */
166 reg += machine->AddressReg[0][0];
167 if (reg < 0) {
168 return dummyReg;
169 }
170 }
171
172 switch (dest->File) {
173 case PROGRAM_TEMPORARY:
174 if (reg >= MAX_PROGRAM_TEMPS)
175 return dummyReg;
176 return machine->Temporaries[reg];
177
178 case PROGRAM_OUTPUT:
179 if (reg >= MAX_PROGRAM_OUTPUTS)
180 return dummyReg;
181 return machine->Outputs[reg];
182
183 default:
184 _mesa_problem(NULL,
185 "Invalid dest register file %d in get_dst_register_pointer()",
186 dest->File);
187 return NULL;
188 }
189 }
190
191
192
193 /**
194 * Fetch a 4-element float vector from the given source register.
195 * Apply swizzling and negating as needed.
196 */
197 static void
198 fetch_vector4(const struct prog_src_register *source,
199 const struct gl_program_machine *machine, GLfloat result[4])
200 {
201 const GLfloat *src = get_src_register_pointer(source, machine);
202 ASSERT(src);
203
204 if (source->Swizzle == SWIZZLE_NOOP) {
205 /* no swizzling */
206 COPY_4V(result, src);
207 }
208 else {
209 ASSERT(GET_SWZ(source->Swizzle, 0) <= 3);
210 ASSERT(GET_SWZ(source->Swizzle, 1) <= 3);
211 ASSERT(GET_SWZ(source->Swizzle, 2) <= 3);
212 ASSERT(GET_SWZ(source->Swizzle, 3) <= 3);
213 result[0] = src[GET_SWZ(source->Swizzle, 0)];
214 result[1] = src[GET_SWZ(source->Swizzle, 1)];
215 result[2] = src[GET_SWZ(source->Swizzle, 2)];
216 result[3] = src[GET_SWZ(source->Swizzle, 3)];
217 }
218
219 if (source->Abs) {
220 result[0] = FABSF(result[0]);
221 result[1] = FABSF(result[1]);
222 result[2] = FABSF(result[2]);
223 result[3] = FABSF(result[3]);
224 }
225 if (source->Negate) {
226 ASSERT(source->Negate == NEGATE_XYZW);
227 result[0] = -result[0];
228 result[1] = -result[1];
229 result[2] = -result[2];
230 result[3] = -result[3];
231 }
232
233 #ifdef NAN_CHECK
234 assert(!IS_INF_OR_NAN(result[0]));
235 assert(!IS_INF_OR_NAN(result[0]));
236 assert(!IS_INF_OR_NAN(result[0]));
237 assert(!IS_INF_OR_NAN(result[0]));
238 #endif
239 }
240
241
242 /**
243 * Fetch the derivative with respect to X or Y for the given register.
244 * XXX this currently only works for fragment program input attribs.
245 */
246 static void
247 fetch_vector4_deriv(struct gl_context * ctx,
248 const struct prog_src_register *source,
249 const struct gl_program_machine *machine,
250 char xOrY, GLfloat result[4])
251 {
252 if (source->File == PROGRAM_INPUT &&
253 source->Index < (GLint) machine->NumDeriv) {
254 const GLint col = machine->CurElement;
255 const GLfloat w = machine->Attribs[VARYING_SLOT_POS][col][3];
256 const GLfloat invQ = 1.0f / w;
257 GLfloat deriv[4];
258
259 if (xOrY == 'X') {
260 deriv[0] = machine->DerivX[source->Index][0] * invQ;
261 deriv[1] = machine->DerivX[source->Index][1] * invQ;
262 deriv[2] = machine->DerivX[source->Index][2] * invQ;
263 deriv[3] = machine->DerivX[source->Index][3] * invQ;
264 }
265 else {
266 deriv[0] = machine->DerivY[source->Index][0] * invQ;
267 deriv[1] = machine->DerivY[source->Index][1] * invQ;
268 deriv[2] = machine->DerivY[source->Index][2] * invQ;
269 deriv[3] = machine->DerivY[source->Index][3] * invQ;
270 }
271
272 result[0] = deriv[GET_SWZ(source->Swizzle, 0)];
273 result[1] = deriv[GET_SWZ(source->Swizzle, 1)];
274 result[2] = deriv[GET_SWZ(source->Swizzle, 2)];
275 result[3] = deriv[GET_SWZ(source->Swizzle, 3)];
276
277 if (source->Abs) {
278 result[0] = FABSF(result[0]);
279 result[1] = FABSF(result[1]);
280 result[2] = FABSF(result[2]);
281 result[3] = FABSF(result[3]);
282 }
283 if (source->Negate) {
284 ASSERT(source->Negate == NEGATE_XYZW);
285 result[0] = -result[0];
286 result[1] = -result[1];
287 result[2] = -result[2];
288 result[3] = -result[3];
289 }
290 }
291 else {
292 ASSIGN_4V(result, 0.0, 0.0, 0.0, 0.0);
293 }
294 }
295
296
297 /**
298 * As above, but only return result[0] element.
299 */
300 static void
301 fetch_vector1(const struct prog_src_register *source,
302 const struct gl_program_machine *machine, GLfloat result[4])
303 {
304 const GLfloat *src = get_src_register_pointer(source, machine);
305 ASSERT(src);
306
307 result[0] = src[GET_SWZ(source->Swizzle, 0)];
308
309 if (source->Abs) {
310 result[0] = FABSF(result[0]);
311 }
312 if (source->Negate) {
313 result[0] = -result[0];
314 }
315 }
316
317
318 static GLuint
319 fetch_vector1ui(const struct prog_src_register *source,
320 const struct gl_program_machine *machine)
321 {
322 const GLuint *src = (GLuint *) get_src_register_pointer(source, machine);
323 return src[GET_SWZ(source->Swizzle, 0)];
324 }
325
326
327 /**
328 * Fetch texel from texture. Use partial derivatives when possible.
329 */
330 static inline void
331 fetch_texel(struct gl_context *ctx,
332 const struct gl_program_machine *machine,
333 const struct prog_instruction *inst,
334 const GLfloat texcoord[4], GLfloat lodBias,
335 GLfloat color[4])
336 {
337 const GLuint unit = machine->Samplers[inst->TexSrcUnit];
338
339 /* Note: we only have the right derivatives for fragment input attribs.
340 */
341 if (machine->NumDeriv > 0 &&
342 inst->SrcReg[0].File == PROGRAM_INPUT &&
343 inst->SrcReg[0].Index == VARYING_SLOT_TEX0 + inst->TexSrcUnit) {
344 /* simple texture fetch for which we should have derivatives */
345 GLuint attr = inst->SrcReg[0].Index;
346 machine->FetchTexelDeriv(ctx, texcoord,
347 machine->DerivX[attr],
348 machine->DerivY[attr],
349 lodBias, unit, color);
350 }
351 else {
352 machine->FetchTexelLod(ctx, texcoord, lodBias, unit, color);
353 }
354 }
355
356
357 /**
358 * Test value against zero and return GT, LT, EQ or UN if NaN.
359 */
360 static inline GLuint
361 generate_cc(float value)
362 {
363 if (value != value)
364 return COND_UN; /* NaN */
365 if (value > 0.0F)
366 return COND_GT;
367 if (value < 0.0F)
368 return COND_LT;
369 return COND_EQ;
370 }
371
372
373 /**
374 * Test if the ccMaskRule is satisfied by the given condition code.
375 * Used to mask destination writes according to the current condition code.
376 */
377 static inline GLboolean
378 test_cc(GLuint condCode, GLuint ccMaskRule)
379 {
380 switch (ccMaskRule) {
381 case COND_EQ: return (condCode == COND_EQ);
382 case COND_NE: return (condCode != COND_EQ);
383 case COND_LT: return (condCode == COND_LT);
384 case COND_GE: return (condCode == COND_GT || condCode == COND_EQ);
385 case COND_LE: return (condCode == COND_LT || condCode == COND_EQ);
386 case COND_GT: return (condCode == COND_GT);
387 case COND_TR: return GL_TRUE;
388 case COND_FL: return GL_FALSE;
389 default: return GL_TRUE;
390 }
391 }
392
393
394 /**
395 * Evaluate the 4 condition codes against a predicate and return GL_TRUE
396 * or GL_FALSE to indicate result.
397 */
398 static inline GLboolean
399 eval_condition(const struct gl_program_machine *machine,
400 const struct prog_instruction *inst)
401 {
402 const GLuint swizzle = inst->DstReg.CondSwizzle;
403 const GLuint condMask = inst->DstReg.CondMask;
404 if (test_cc(machine->CondCodes[GET_SWZ(swizzle, 0)], condMask) ||
405 test_cc(machine->CondCodes[GET_SWZ(swizzle, 1)], condMask) ||
406 test_cc(machine->CondCodes[GET_SWZ(swizzle, 2)], condMask) ||
407 test_cc(machine->CondCodes[GET_SWZ(swizzle, 3)], condMask)) {
408 return GL_TRUE;
409 }
410 else {
411 return GL_FALSE;
412 }
413 }
414
415
416
417 /**
418 * Store 4 floats into a register. Observe the instructions saturate and
419 * set-condition-code flags.
420 */
421 static void
422 store_vector4(const struct prog_instruction *inst,
423 struct gl_program_machine *machine, const GLfloat value[4])
424 {
425 const struct prog_dst_register *dstReg = &(inst->DstReg);
426 const GLboolean clamp = inst->SaturateMode == SATURATE_ZERO_ONE;
427 GLuint writeMask = dstReg->WriteMask;
428 GLfloat clampedValue[4];
429 GLfloat *dst = get_dst_register_pointer(dstReg, machine);
430
431 #if 0
432 if (value[0] > 1.0e10 ||
433 IS_INF_OR_NAN(value[0]) ||
434 IS_INF_OR_NAN(value[1]) ||
435 IS_INF_OR_NAN(value[2]) || IS_INF_OR_NAN(value[3]))
436 printf("store %g %g %g %g\n", value[0], value[1], value[2], value[3]);
437 #endif
438
439 if (clamp) {
440 clampedValue[0] = CLAMP(value[0], 0.0F, 1.0F);
441 clampedValue[1] = CLAMP(value[1], 0.0F, 1.0F);
442 clampedValue[2] = CLAMP(value[2], 0.0F, 1.0F);
443 clampedValue[3] = CLAMP(value[3], 0.0F, 1.0F);
444 value = clampedValue;
445 }
446
447 if (dstReg->CondMask != COND_TR) {
448 /* condition codes may turn off some writes */
449 if (writeMask & WRITEMASK_X) {
450 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 0)],
451 dstReg->CondMask))
452 writeMask &= ~WRITEMASK_X;
453 }
454 if (writeMask & WRITEMASK_Y) {
455 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 1)],
456 dstReg->CondMask))
457 writeMask &= ~WRITEMASK_Y;
458 }
459 if (writeMask & WRITEMASK_Z) {
460 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 2)],
461 dstReg->CondMask))
462 writeMask &= ~WRITEMASK_Z;
463 }
464 if (writeMask & WRITEMASK_W) {
465 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 3)],
466 dstReg->CondMask))
467 writeMask &= ~WRITEMASK_W;
468 }
469 }
470
471 #ifdef NAN_CHECK
472 assert(!IS_INF_OR_NAN(value[0]));
473 assert(!IS_INF_OR_NAN(value[0]));
474 assert(!IS_INF_OR_NAN(value[0]));
475 assert(!IS_INF_OR_NAN(value[0]));
476 #endif
477
478 if (writeMask & WRITEMASK_X)
479 dst[0] = value[0];
480 if (writeMask & WRITEMASK_Y)
481 dst[1] = value[1];
482 if (writeMask & WRITEMASK_Z)
483 dst[2] = value[2];
484 if (writeMask & WRITEMASK_W)
485 dst[3] = value[3];
486
487 if (inst->CondUpdate) {
488 if (writeMask & WRITEMASK_X)
489 machine->CondCodes[0] = generate_cc(value[0]);
490 if (writeMask & WRITEMASK_Y)
491 machine->CondCodes[1] = generate_cc(value[1]);
492 if (writeMask & WRITEMASK_Z)
493 machine->CondCodes[2] = generate_cc(value[2]);
494 if (writeMask & WRITEMASK_W)
495 machine->CondCodes[3] = generate_cc(value[3]);
496 #if DEBUG_PROG
497 printf("CondCodes=(%s,%s,%s,%s) for:\n",
498 _mesa_condcode_string(machine->CondCodes[0]),
499 _mesa_condcode_string(machine->CondCodes[1]),
500 _mesa_condcode_string(machine->CondCodes[2]),
501 _mesa_condcode_string(machine->CondCodes[3]));
502 #endif
503 }
504 }
505
506
507 /**
508 * Store 4 uints into a register. Observe the set-condition-code flags.
509 */
510 static void
511 store_vector4ui(const struct prog_instruction *inst,
512 struct gl_program_machine *machine, const GLuint value[4])
513 {
514 const struct prog_dst_register *dstReg = &(inst->DstReg);
515 GLuint writeMask = dstReg->WriteMask;
516 GLuint *dst = (GLuint *) get_dst_register_pointer(dstReg, machine);
517
518 if (dstReg->CondMask != COND_TR) {
519 /* condition codes may turn off some writes */
520 if (writeMask & WRITEMASK_X) {
521 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 0)],
522 dstReg->CondMask))
523 writeMask &= ~WRITEMASK_X;
524 }
525 if (writeMask & WRITEMASK_Y) {
526 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 1)],
527 dstReg->CondMask))
528 writeMask &= ~WRITEMASK_Y;
529 }
530 if (writeMask & WRITEMASK_Z) {
531 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 2)],
532 dstReg->CondMask))
533 writeMask &= ~WRITEMASK_Z;
534 }
535 if (writeMask & WRITEMASK_W) {
536 if (!test_cc(machine->CondCodes[GET_SWZ(dstReg->CondSwizzle, 3)],
537 dstReg->CondMask))
538 writeMask &= ~WRITEMASK_W;
539 }
540 }
541
542 if (writeMask & WRITEMASK_X)
543 dst[0] = value[0];
544 if (writeMask & WRITEMASK_Y)
545 dst[1] = value[1];
546 if (writeMask & WRITEMASK_Z)
547 dst[2] = value[2];
548 if (writeMask & WRITEMASK_W)
549 dst[3] = value[3];
550
551 if (inst->CondUpdate) {
552 if (writeMask & WRITEMASK_X)
553 machine->CondCodes[0] = generate_cc((float)value[0]);
554 if (writeMask & WRITEMASK_Y)
555 machine->CondCodes[1] = generate_cc((float)value[1]);
556 if (writeMask & WRITEMASK_Z)
557 machine->CondCodes[2] = generate_cc((float)value[2]);
558 if (writeMask & WRITEMASK_W)
559 machine->CondCodes[3] = generate_cc((float)value[3]);
560 #if DEBUG_PROG
561 printf("CondCodes=(%s,%s,%s,%s) for:\n",
562 _mesa_condcode_string(machine->CondCodes[0]),
563 _mesa_condcode_string(machine->CondCodes[1]),
564 _mesa_condcode_string(machine->CondCodes[2]),
565 _mesa_condcode_string(machine->CondCodes[3]));
566 #endif
567 }
568 }
569
570
571
572 /**
573 * Execute the given vertex/fragment program.
574 *
575 * \param ctx rendering context
576 * \param program the program to execute
577 * \param machine machine state (must be initialized)
578 * \return GL_TRUE if program completed or GL_FALSE if program executed KIL.
579 */
580 GLboolean
581 _mesa_execute_program(struct gl_context * ctx,
582 const struct gl_program *program,
583 struct gl_program_machine *machine)
584 {
585 const GLuint numInst = program->NumInstructions;
586 const GLuint maxExec = 65536;
587 GLuint pc, numExec = 0;
588
589 machine->CurProgram = program;
590
591 if (DEBUG_PROG) {
592 printf("execute program %u --------------------\n", program->Id);
593 }
594
595 if (program->Target == GL_VERTEX_PROGRAM_ARB) {
596 machine->EnvParams = ctx->VertexProgram.Parameters;
597 }
598 else {
599 machine->EnvParams = ctx->FragmentProgram.Parameters;
600 }
601
602 for (pc = 0; pc < numInst; pc++) {
603 const struct prog_instruction *inst = program->Instructions + pc;
604
605 if (DEBUG_PROG) {
606 _mesa_print_instruction(inst);
607 }
608
609 switch (inst->Opcode) {
610 case OPCODE_ABS:
611 {
612 GLfloat a[4], result[4];
613 fetch_vector4(&inst->SrcReg[0], machine, a);
614 result[0] = FABSF(a[0]);
615 result[1] = FABSF(a[1]);
616 result[2] = FABSF(a[2]);
617 result[3] = FABSF(a[3]);
618 store_vector4(inst, machine, result);
619 }
620 break;
621 case OPCODE_ADD:
622 {
623 GLfloat a[4], b[4], result[4];
624 fetch_vector4(&inst->SrcReg[0], machine, a);
625 fetch_vector4(&inst->SrcReg[1], machine, b);
626 result[0] = a[0] + b[0];
627 result[1] = a[1] + b[1];
628 result[2] = a[2] + b[2];
629 result[3] = a[3] + b[3];
630 store_vector4(inst, machine, result);
631 if (DEBUG_PROG) {
632 printf("ADD (%g %g %g %g) = (%g %g %g %g) + (%g %g %g %g)\n",
633 result[0], result[1], result[2], result[3],
634 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
635 }
636 }
637 break;
638 case OPCODE_ARL:
639 {
640 GLfloat t[4];
641 fetch_vector4(&inst->SrcReg[0], machine, t);
642 machine->AddressReg[0][0] = IFLOOR(t[0]);
643 if (DEBUG_PROG) {
644 printf("ARL %d\n", machine->AddressReg[0][0]);
645 }
646 }
647 break;
648 case OPCODE_BGNLOOP:
649 /* no-op */
650 ASSERT(program->Instructions[inst->BranchTarget].Opcode
651 == OPCODE_ENDLOOP);
652 break;
653 case OPCODE_ENDLOOP:
654 /* subtract 1 here since pc is incremented by for(pc) loop */
655 ASSERT(program->Instructions[inst->BranchTarget].Opcode
656 == OPCODE_BGNLOOP);
657 pc = inst->BranchTarget - 1; /* go to matching BNGLOOP */
658 break;
659 case OPCODE_BGNSUB: /* begin subroutine */
660 break;
661 case OPCODE_ENDSUB: /* end subroutine */
662 break;
663 case OPCODE_BRK: /* break out of loop (conditional) */
664 ASSERT(program->Instructions[inst->BranchTarget].Opcode
665 == OPCODE_ENDLOOP);
666 if (eval_condition(machine, inst)) {
667 /* break out of loop */
668 /* pc++ at end of for-loop will put us after the ENDLOOP inst */
669 pc = inst->BranchTarget;
670 }
671 break;
672 case OPCODE_CONT: /* continue loop (conditional) */
673 ASSERT(program->Instructions[inst->BranchTarget].Opcode
674 == OPCODE_ENDLOOP);
675 if (eval_condition(machine, inst)) {
676 /* continue at ENDLOOP */
677 /* Subtract 1 here since we'll do pc++ at end of for-loop */
678 pc = inst->BranchTarget - 1;
679 }
680 break;
681 case OPCODE_CAL: /* Call subroutine (conditional) */
682 if (eval_condition(machine, inst)) {
683 /* call the subroutine */
684 if (machine->StackDepth >= MAX_PROGRAM_CALL_DEPTH) {
685 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
686 }
687 machine->CallStack[machine->StackDepth++] = pc + 1; /* next inst */
688 /* Subtract 1 here since we'll do pc++ at end of for-loop */
689 pc = inst->BranchTarget - 1;
690 }
691 break;
692 case OPCODE_CMP:
693 {
694 GLfloat a[4], b[4], c[4], result[4];
695 fetch_vector4(&inst->SrcReg[0], machine, a);
696 fetch_vector4(&inst->SrcReg[1], machine, b);
697 fetch_vector4(&inst->SrcReg[2], machine, c);
698 result[0] = a[0] < 0.0F ? b[0] : c[0];
699 result[1] = a[1] < 0.0F ? b[1] : c[1];
700 result[2] = a[2] < 0.0F ? b[2] : c[2];
701 result[3] = a[3] < 0.0F ? b[3] : c[3];
702 store_vector4(inst, machine, result);
703 if (DEBUG_PROG) {
704 printf("CMP (%g %g %g %g) = (%g %g %g %g) < 0 ? (%g %g %g %g) : (%g %g %g %g)\n",
705 result[0], result[1], result[2], result[3],
706 a[0], a[1], a[2], a[3],
707 b[0], b[1], b[2], b[3],
708 c[0], c[1], c[2], c[3]);
709 }
710 }
711 break;
712 case OPCODE_COS:
713 {
714 GLfloat a[4], result[4];
715 fetch_vector1(&inst->SrcReg[0], machine, a);
716 result[0] = result[1] = result[2] = result[3]
717 = (GLfloat) cos(a[0]);
718 store_vector4(inst, machine, result);
719 }
720 break;
721 case OPCODE_DDX: /* Partial derivative with respect to X */
722 {
723 GLfloat result[4];
724 fetch_vector4_deriv(ctx, &inst->SrcReg[0], machine,
725 'X', result);
726 store_vector4(inst, machine, result);
727 }
728 break;
729 case OPCODE_DDY: /* Partial derivative with respect to Y */
730 {
731 GLfloat result[4];
732 fetch_vector4_deriv(ctx, &inst->SrcReg[0], machine,
733 'Y', result);
734 store_vector4(inst, machine, result);
735 }
736 break;
737 case OPCODE_DP2:
738 {
739 GLfloat a[4], b[4], result[4];
740 fetch_vector4(&inst->SrcReg[0], machine, a);
741 fetch_vector4(&inst->SrcReg[1], machine, b);
742 result[0] = result[1] = result[2] = result[3] = DOT2(a, b);
743 store_vector4(inst, machine, result);
744 if (DEBUG_PROG) {
745 printf("DP2 %g = (%g %g) . (%g %g)\n",
746 result[0], a[0], a[1], b[0], b[1]);
747 }
748 }
749 break;
750 case OPCODE_DP3:
751 {
752 GLfloat a[4], b[4], result[4];
753 fetch_vector4(&inst->SrcReg[0], machine, a);
754 fetch_vector4(&inst->SrcReg[1], machine, b);
755 result[0] = result[1] = result[2] = result[3] = DOT3(a, b);
756 store_vector4(inst, machine, result);
757 if (DEBUG_PROG) {
758 printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
759 result[0], a[0], a[1], a[2], b[0], b[1], b[2]);
760 }
761 }
762 break;
763 case OPCODE_DP4:
764 {
765 GLfloat a[4], b[4], result[4];
766 fetch_vector4(&inst->SrcReg[0], machine, a);
767 fetch_vector4(&inst->SrcReg[1], machine, b);
768 result[0] = result[1] = result[2] = result[3] = DOT4(a, b);
769 store_vector4(inst, machine, result);
770 if (DEBUG_PROG) {
771 printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
772 result[0], a[0], a[1], a[2], a[3],
773 b[0], b[1], b[2], b[3]);
774 }
775 }
776 break;
777 case OPCODE_DPH:
778 {
779 GLfloat a[4], b[4], result[4];
780 fetch_vector4(&inst->SrcReg[0], machine, a);
781 fetch_vector4(&inst->SrcReg[1], machine, b);
782 result[0] = result[1] = result[2] = result[3] = DOT3(a, b) + b[3];
783 store_vector4(inst, machine, result);
784 }
785 break;
786 case OPCODE_DST: /* Distance vector */
787 {
788 GLfloat a[4], b[4], result[4];
789 fetch_vector4(&inst->SrcReg[0], machine, a);
790 fetch_vector4(&inst->SrcReg[1], machine, b);
791 result[0] = 1.0F;
792 result[1] = a[1] * b[1];
793 result[2] = a[2];
794 result[3] = b[3];
795 store_vector4(inst, machine, result);
796 }
797 break;
798 case OPCODE_EXP:
799 {
800 GLfloat t[4], q[4], floor_t0;
801 fetch_vector1(&inst->SrcReg[0], machine, t);
802 floor_t0 = FLOORF(t[0]);
803 if (floor_t0 > FLT_MAX_EXP) {
804 SET_POS_INFINITY(q[0]);
805 SET_POS_INFINITY(q[2]);
806 }
807 else if (floor_t0 < FLT_MIN_EXP) {
808 q[0] = 0.0F;
809 q[2] = 0.0F;
810 }
811 else {
812 q[0] = LDEXPF(1.0, (int) floor_t0);
813 /* Note: GL_NV_vertex_program expects
814 * result.z = result.x * APPX(result.y)
815 * We do what the ARB extension says.
816 */
817 q[2] = (GLfloat) pow(2.0, t[0]);
818 }
819 q[1] = t[0] - floor_t0;
820 q[3] = 1.0F;
821 store_vector4( inst, machine, q );
822 }
823 break;
824 case OPCODE_EX2: /* Exponential base 2 */
825 {
826 GLfloat a[4], result[4], val;
827 fetch_vector1(&inst->SrcReg[0], machine, a);
828 val = (GLfloat) pow(2.0, a[0]);
829 /*
830 if (IS_INF_OR_NAN(val))
831 val = 1.0e10;
832 */
833 result[0] = result[1] = result[2] = result[3] = val;
834 store_vector4(inst, machine, result);
835 }
836 break;
837 case OPCODE_FLR:
838 {
839 GLfloat a[4], result[4];
840 fetch_vector4(&inst->SrcReg[0], machine, a);
841 result[0] = FLOORF(a[0]);
842 result[1] = FLOORF(a[1]);
843 result[2] = FLOORF(a[2]);
844 result[3] = FLOORF(a[3]);
845 store_vector4(inst, machine, result);
846 }
847 break;
848 case OPCODE_FRC:
849 {
850 GLfloat a[4], result[4];
851 fetch_vector4(&inst->SrcReg[0], machine, a);
852 result[0] = a[0] - FLOORF(a[0]);
853 result[1] = a[1] - FLOORF(a[1]);
854 result[2] = a[2] - FLOORF(a[2]);
855 result[3] = a[3] - FLOORF(a[3]);
856 store_vector4(inst, machine, result);
857 }
858 break;
859 case OPCODE_IF:
860 {
861 GLboolean cond;
862 ASSERT(program->Instructions[inst->BranchTarget].Opcode
863 == OPCODE_ELSE ||
864 program->Instructions[inst->BranchTarget].Opcode
865 == OPCODE_ENDIF);
866 /* eval condition */
867 if (inst->SrcReg[0].File != PROGRAM_UNDEFINED) {
868 GLfloat a[4];
869 fetch_vector1(&inst->SrcReg[0], machine, a);
870 cond = (a[0] != 0.0);
871 }
872 else {
873 cond = eval_condition(machine, inst);
874 }
875 if (DEBUG_PROG) {
876 printf("IF: %d\n", cond);
877 }
878 /* do if/else */
879 if (cond) {
880 /* do if-clause (just continue execution) */
881 }
882 else {
883 /* go to the instruction after ELSE or ENDIF */
884 assert(inst->BranchTarget >= 0);
885 pc = inst->BranchTarget;
886 }
887 }
888 break;
889 case OPCODE_ELSE:
890 /* goto ENDIF */
891 ASSERT(program->Instructions[inst->BranchTarget].Opcode
892 == OPCODE_ENDIF);
893 assert(inst->BranchTarget >= 0);
894 pc = inst->BranchTarget;
895 break;
896 case OPCODE_ENDIF:
897 /* nothing */
898 break;
899 case OPCODE_KIL_NV: /* NV_f_p only (conditional) */
900 if (eval_condition(machine, inst)) {
901 return GL_FALSE;
902 }
903 break;
904 case OPCODE_KIL: /* ARB_f_p only */
905 {
906 GLfloat a[4];
907 fetch_vector4(&inst->SrcReg[0], machine, a);
908 if (DEBUG_PROG) {
909 printf("KIL if (%g %g %g %g) <= 0.0\n",
910 a[0], a[1], a[2], a[3]);
911 }
912
913 if (a[0] < 0.0F || a[1] < 0.0F || a[2] < 0.0F || a[3] < 0.0F) {
914 return GL_FALSE;
915 }
916 }
917 break;
918 case OPCODE_LG2: /* log base 2 */
919 {
920 GLfloat a[4], result[4], val;
921 fetch_vector1(&inst->SrcReg[0], machine, a);
922 /* The fast LOG2 macro doesn't meet the precision requirements.
923 */
924 if (a[0] == 0.0F) {
925 val = -FLT_MAX;
926 }
927 else {
928 val = (float)(log(a[0]) * 1.442695F);
929 }
930 result[0] = result[1] = result[2] = result[3] = val;
931 store_vector4(inst, machine, result);
932 }
933 break;
934 case OPCODE_LIT:
935 {
936 const GLfloat epsilon = 1.0F / 256.0F; /* from NV VP spec */
937 GLfloat a[4], result[4];
938 fetch_vector4(&inst->SrcReg[0], machine, a);
939 a[0] = MAX2(a[0], 0.0F);
940 a[1] = MAX2(a[1], 0.0F);
941 /* XXX ARB version clamps a[3], NV version doesn't */
942 a[3] = CLAMP(a[3], -(128.0F - epsilon), (128.0F - epsilon));
943 result[0] = 1.0F;
944 result[1] = a[0];
945 /* XXX we could probably just use pow() here */
946 if (a[0] > 0.0F) {
947 if (a[1] == 0.0 && a[3] == 0.0)
948 result[2] = 1.0F;
949 else
950 result[2] = (GLfloat) pow(a[1], a[3]);
951 }
952 else {
953 result[2] = 0.0F;
954 }
955 result[3] = 1.0F;
956 store_vector4(inst, machine, result);
957 if (DEBUG_PROG) {
958 printf("LIT (%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]);
961 }
962 }
963 break;
964 case OPCODE_LOG:
965 {
966 GLfloat t[4], q[4], abs_t0;
967 fetch_vector1(&inst->SrcReg[0], machine, t);
968 abs_t0 = FABSF(t[0]);
969 if (abs_t0 != 0.0F) {
970 if (IS_INF_OR_NAN(abs_t0))
971 {
972 SET_POS_INFINITY(q[0]);
973 q[1] = 1.0F;
974 SET_POS_INFINITY(q[2]);
975 }
976 else {
977 int exponent;
978 GLfloat mantissa = FREXPF(t[0], &exponent);
979 q[0] = (GLfloat) (exponent - 1);
980 q[1] = (GLfloat) (2.0 * mantissa); /* map [.5, 1) -> [1, 2) */
981
982 /* The fast LOG2 macro doesn't meet the precision
983 * requirements.
984 */
985 q[2] = (float)(log(t[0]) * 1.442695F);
986 }
987 }
988 else {
989 SET_NEG_INFINITY(q[0]);
990 q[1] = 1.0F;
991 SET_NEG_INFINITY(q[2]);
992 }
993 q[3] = 1.0;
994 store_vector4(inst, machine, q);
995 }
996 break;
997 case OPCODE_LRP:
998 {
999 GLfloat a[4], b[4], c[4], result[4];
1000 fetch_vector4(&inst->SrcReg[0], machine, a);
1001 fetch_vector4(&inst->SrcReg[1], machine, b);
1002 fetch_vector4(&inst->SrcReg[2], machine, c);
1003 result[0] = a[0] * b[0] + (1.0F - a[0]) * c[0];
1004 result[1] = a[1] * b[1] + (1.0F - a[1]) * c[1];
1005 result[2] = a[2] * b[2] + (1.0F - a[2]) * c[2];
1006 result[3] = a[3] * b[3] + (1.0F - a[3]) * c[3];
1007 store_vector4(inst, machine, result);
1008 if (DEBUG_PROG) {
1009 printf("LRP (%g %g %g %g) = (%g %g %g %g), "
1010 "(%g %g %g %g), (%g %g %g %g)\n",
1011 result[0], result[1], result[2], result[3],
1012 a[0], a[1], a[2], a[3],
1013 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
1014 }
1015 }
1016 break;
1017 case OPCODE_MAD:
1018 {
1019 GLfloat a[4], b[4], c[4], result[4];
1020 fetch_vector4(&inst->SrcReg[0], machine, a);
1021 fetch_vector4(&inst->SrcReg[1], machine, b);
1022 fetch_vector4(&inst->SrcReg[2], machine, c);
1023 result[0] = a[0] * b[0] + c[0];
1024 result[1] = a[1] * b[1] + c[1];
1025 result[2] = a[2] * b[2] + c[2];
1026 result[3] = a[3] * b[3] + c[3];
1027 store_vector4(inst, machine, result);
1028 if (DEBUG_PROG) {
1029 printf("MAD (%g %g %g %g) = (%g %g %g %g) * "
1030 "(%g %g %g %g) + (%g %g %g %g)\n",
1031 result[0], result[1], result[2], result[3],
1032 a[0], a[1], a[2], a[3],
1033 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
1034 }
1035 }
1036 break;
1037 case OPCODE_MAX:
1038 {
1039 GLfloat a[4], b[4], result[4];
1040 fetch_vector4(&inst->SrcReg[0], machine, a);
1041 fetch_vector4(&inst->SrcReg[1], machine, b);
1042 result[0] = MAX2(a[0], b[0]);
1043 result[1] = MAX2(a[1], b[1]);
1044 result[2] = MAX2(a[2], b[2]);
1045 result[3] = MAX2(a[3], b[3]);
1046 store_vector4(inst, machine, result);
1047 if (DEBUG_PROG) {
1048 printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
1049 result[0], result[1], result[2], result[3],
1050 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1051 }
1052 }
1053 break;
1054 case OPCODE_MIN:
1055 {
1056 GLfloat a[4], b[4], result[4];
1057 fetch_vector4(&inst->SrcReg[0], machine, a);
1058 fetch_vector4(&inst->SrcReg[1], machine, b);
1059 result[0] = MIN2(a[0], b[0]);
1060 result[1] = MIN2(a[1], b[1]);
1061 result[2] = MIN2(a[2], b[2]);
1062 result[3] = MIN2(a[3], b[3]);
1063 store_vector4(inst, machine, result);
1064 }
1065 break;
1066 case OPCODE_MOV:
1067 {
1068 GLfloat result[4];
1069 fetch_vector4(&inst->SrcReg[0], machine, result);
1070 store_vector4(inst, machine, result);
1071 if (DEBUG_PROG) {
1072 printf("MOV (%g %g %g %g)\n",
1073 result[0], result[1], result[2], result[3]);
1074 }
1075 }
1076 break;
1077 case OPCODE_MUL:
1078 {
1079 GLfloat a[4], b[4], result[4];
1080 fetch_vector4(&inst->SrcReg[0], machine, a);
1081 fetch_vector4(&inst->SrcReg[1], machine, b);
1082 result[0] = a[0] * b[0];
1083 result[1] = a[1] * b[1];
1084 result[2] = a[2] * b[2];
1085 result[3] = a[3] * b[3];
1086 store_vector4(inst, machine, result);
1087 if (DEBUG_PROG) {
1088 printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
1089 result[0], result[1], result[2], result[3],
1090 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1091 }
1092 }
1093 break;
1094 case OPCODE_NOISE1:
1095 {
1096 GLfloat a[4], result[4];
1097 fetch_vector1(&inst->SrcReg[0], machine, a);
1098 result[0] =
1099 result[1] =
1100 result[2] =
1101 result[3] = _mesa_noise1(a[0]);
1102 store_vector4(inst, machine, result);
1103 }
1104 break;
1105 case OPCODE_NOISE2:
1106 {
1107 GLfloat a[4], result[4];
1108 fetch_vector4(&inst->SrcReg[0], machine, a);
1109 result[0] =
1110 result[1] =
1111 result[2] = result[3] = _mesa_noise2(a[0], a[1]);
1112 store_vector4(inst, machine, result);
1113 }
1114 break;
1115 case OPCODE_NOISE3:
1116 {
1117 GLfloat a[4], result[4];
1118 fetch_vector4(&inst->SrcReg[0], machine, a);
1119 result[0] =
1120 result[1] =
1121 result[2] =
1122 result[3] = _mesa_noise3(a[0], a[1], a[2]);
1123 store_vector4(inst, machine, result);
1124 }
1125 break;
1126 case OPCODE_NOISE4:
1127 {
1128 GLfloat a[4], result[4];
1129 fetch_vector4(&inst->SrcReg[0], machine, a);
1130 result[0] =
1131 result[1] =
1132 result[2] =
1133 result[3] = _mesa_noise4(a[0], a[1], a[2], a[3]);
1134 store_vector4(inst, machine, result);
1135 }
1136 break;
1137 case OPCODE_NOP:
1138 break;
1139 case OPCODE_PK2H: /* pack two 16-bit floats in one 32-bit float */
1140 {
1141 GLfloat a[4];
1142 GLuint result[4];
1143 GLhalfNV hx, hy;
1144 fetch_vector4(&inst->SrcReg[0], machine, a);
1145 hx = _mesa_float_to_half(a[0]);
1146 hy = _mesa_float_to_half(a[1]);
1147 result[0] =
1148 result[1] =
1149 result[2] =
1150 result[3] = hx | (hy << 16);
1151 store_vector4ui(inst, machine, result);
1152 }
1153 break;
1154 case OPCODE_PK2US: /* pack two GLushorts into one 32-bit float */
1155 {
1156 GLfloat a[4];
1157 GLuint result[4], usx, usy;
1158 fetch_vector4(&inst->SrcReg[0], machine, a);
1159 a[0] = CLAMP(a[0], 0.0F, 1.0F);
1160 a[1] = CLAMP(a[1], 0.0F, 1.0F);
1161 usx = F_TO_I(a[0] * 65535.0F);
1162 usy = F_TO_I(a[1] * 65535.0F);
1163 result[0] =
1164 result[1] =
1165 result[2] =
1166 result[3] = usx | (usy << 16);
1167 store_vector4ui(inst, machine, result);
1168 }
1169 break;
1170 case OPCODE_PK4B: /* pack four GLbytes into one 32-bit float */
1171 {
1172 GLfloat a[4];
1173 GLuint result[4], ubx, uby, ubz, ubw;
1174 fetch_vector4(&inst->SrcReg[0], machine, a);
1175 a[0] = CLAMP(a[0], -128.0F / 127.0F, 1.0F);
1176 a[1] = CLAMP(a[1], -128.0F / 127.0F, 1.0F);
1177 a[2] = CLAMP(a[2], -128.0F / 127.0F, 1.0F);
1178 a[3] = CLAMP(a[3], -128.0F / 127.0F, 1.0F);
1179 ubx = F_TO_I(127.0F * a[0] + 128.0F);
1180 uby = F_TO_I(127.0F * a[1] + 128.0F);
1181 ubz = F_TO_I(127.0F * a[2] + 128.0F);
1182 ubw = F_TO_I(127.0F * a[3] + 128.0F);
1183 result[0] =
1184 result[1] =
1185 result[2] =
1186 result[3] = ubx | (uby << 8) | (ubz << 16) | (ubw << 24);
1187 store_vector4ui(inst, machine, result);
1188 }
1189 break;
1190 case OPCODE_PK4UB: /* pack four GLubytes into one 32-bit float */
1191 {
1192 GLfloat a[4];
1193 GLuint result[4], ubx, uby, ubz, ubw;
1194 fetch_vector4(&inst->SrcReg[0], machine, a);
1195 a[0] = CLAMP(a[0], 0.0F, 1.0F);
1196 a[1] = CLAMP(a[1], 0.0F, 1.0F);
1197 a[2] = CLAMP(a[2], 0.0F, 1.0F);
1198 a[3] = CLAMP(a[3], 0.0F, 1.0F);
1199 ubx = F_TO_I(255.0F * a[0]);
1200 uby = F_TO_I(255.0F * a[1]);
1201 ubz = F_TO_I(255.0F * a[2]);
1202 ubw = F_TO_I(255.0F * a[3]);
1203 result[0] =
1204 result[1] =
1205 result[2] =
1206 result[3] = ubx | (uby << 8) | (ubz << 16) | (ubw << 24);
1207 store_vector4ui(inst, machine, result);
1208 }
1209 break;
1210 case OPCODE_POW:
1211 {
1212 GLfloat a[4], b[4], result[4];
1213 fetch_vector1(&inst->SrcReg[0], machine, a);
1214 fetch_vector1(&inst->SrcReg[1], machine, b);
1215 result[0] = result[1] = result[2] = result[3]
1216 = (GLfloat) pow(a[0], b[0]);
1217 store_vector4(inst, machine, result);
1218 }
1219 break;
1220
1221 case OPCODE_RCP:
1222 {
1223 GLfloat a[4], result[4];
1224 fetch_vector1(&inst->SrcReg[0], machine, a);
1225 if (DEBUG_PROG) {
1226 if (a[0] == 0)
1227 printf("RCP(0)\n");
1228 else if (IS_INF_OR_NAN(a[0]))
1229 printf("RCP(inf)\n");
1230 }
1231 result[0] = result[1] = result[2] = result[3] = 1.0F / a[0];
1232 store_vector4(inst, machine, result);
1233 }
1234 break;
1235 case OPCODE_RET: /* return from subroutine (conditional) */
1236 if (eval_condition(machine, inst)) {
1237 if (machine->StackDepth == 0) {
1238 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
1239 }
1240 /* subtract one because of pc++ in the for loop */
1241 pc = machine->CallStack[--machine->StackDepth] - 1;
1242 }
1243 break;
1244 case OPCODE_RFL: /* reflection vector */
1245 {
1246 GLfloat axis[4], dir[4], result[4], tmpX, tmpW;
1247 fetch_vector4(&inst->SrcReg[0], machine, axis);
1248 fetch_vector4(&inst->SrcReg[1], machine, dir);
1249 tmpW = DOT3(axis, axis);
1250 tmpX = (2.0F * DOT3(axis, dir)) / tmpW;
1251 result[0] = tmpX * axis[0] - dir[0];
1252 result[1] = tmpX * axis[1] - dir[1];
1253 result[2] = tmpX * axis[2] - dir[2];
1254 /* result[3] is never written! XXX enforce in parser! */
1255 store_vector4(inst, machine, result);
1256 }
1257 break;
1258 case OPCODE_RSQ: /* 1 / sqrt() */
1259 {
1260 GLfloat a[4], result[4];
1261 fetch_vector1(&inst->SrcReg[0], machine, a);
1262 a[0] = FABSF(a[0]);
1263 result[0] = result[1] = result[2] = result[3] = INV_SQRTF(a[0]);
1264 store_vector4(inst, machine, result);
1265 if (DEBUG_PROG) {
1266 printf("RSQ %g = 1/sqrt(|%g|)\n", result[0], a[0]);
1267 }
1268 }
1269 break;
1270 case OPCODE_SCS: /* sine and cos */
1271 {
1272 GLfloat a[4], result[4];
1273 fetch_vector1(&inst->SrcReg[0], machine, a);
1274 result[0] = (GLfloat) cos(a[0]);
1275 result[1] = (GLfloat) sin(a[0]);
1276 result[2] = 0.0; /* undefined! */
1277 result[3] = 0.0; /* undefined! */
1278 store_vector4(inst, machine, result);
1279 }
1280 break;
1281 case OPCODE_SEQ: /* set on equal */
1282 {
1283 GLfloat a[4], b[4], result[4];
1284 fetch_vector4(&inst->SrcReg[0], machine, a);
1285 fetch_vector4(&inst->SrcReg[1], machine, b);
1286 result[0] = (a[0] == b[0]) ? 1.0F : 0.0F;
1287 result[1] = (a[1] == b[1]) ? 1.0F : 0.0F;
1288 result[2] = (a[2] == b[2]) ? 1.0F : 0.0F;
1289 result[3] = (a[3] == b[3]) ? 1.0F : 0.0F;
1290 store_vector4(inst, machine, result);
1291 if (DEBUG_PROG) {
1292 printf("SEQ (%g %g %g %g) = (%g %g %g %g) == (%g %g %g %g)\n",
1293 result[0], result[1], result[2], result[3],
1294 a[0], a[1], a[2], a[3],
1295 b[0], b[1], b[2], b[3]);
1296 }
1297 }
1298 break;
1299 case OPCODE_SFL: /* set false, operands ignored */
1300 {
1301 static const GLfloat result[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
1302 store_vector4(inst, machine, result);
1303 }
1304 break;
1305 case OPCODE_SGE: /* set on greater or equal */
1306 {
1307 GLfloat a[4], b[4], result[4];
1308 fetch_vector4(&inst->SrcReg[0], machine, a);
1309 fetch_vector4(&inst->SrcReg[1], machine, b);
1310 result[0] = (a[0] >= b[0]) ? 1.0F : 0.0F;
1311 result[1] = (a[1] >= b[1]) ? 1.0F : 0.0F;
1312 result[2] = (a[2] >= b[2]) ? 1.0F : 0.0F;
1313 result[3] = (a[3] >= b[3]) ? 1.0F : 0.0F;
1314 store_vector4(inst, machine, result);
1315 if (DEBUG_PROG) {
1316 printf("SGE (%g %g %g %g) = (%g %g %g %g) >= (%g %g %g %g)\n",
1317 result[0], result[1], result[2], result[3],
1318 a[0], a[1], a[2], a[3],
1319 b[0], b[1], b[2], b[3]);
1320 }
1321 }
1322 break;
1323 case OPCODE_SGT: /* set on greater */
1324 {
1325 GLfloat a[4], b[4], result[4];
1326 fetch_vector4(&inst->SrcReg[0], machine, a);
1327 fetch_vector4(&inst->SrcReg[1], machine, b);
1328 result[0] = (a[0] > b[0]) ? 1.0F : 0.0F;
1329 result[1] = (a[1] > b[1]) ? 1.0F : 0.0F;
1330 result[2] = (a[2] > b[2]) ? 1.0F : 0.0F;
1331 result[3] = (a[3] > b[3]) ? 1.0F : 0.0F;
1332 store_vector4(inst, machine, result);
1333 if (DEBUG_PROG) {
1334 printf("SGT (%g %g %g %g) = (%g %g %g %g) > (%g %g %g %g)\n",
1335 result[0], result[1], result[2], result[3],
1336 a[0], a[1], a[2], a[3],
1337 b[0], b[1], b[2], b[3]);
1338 }
1339 }
1340 break;
1341 case OPCODE_SIN:
1342 {
1343 GLfloat a[4], result[4];
1344 fetch_vector1(&inst->SrcReg[0], machine, a);
1345 result[0] = result[1] = result[2] = result[3]
1346 = (GLfloat) sin(a[0]);
1347 store_vector4(inst, machine, result);
1348 }
1349 break;
1350 case OPCODE_SLE: /* set on less or equal */
1351 {
1352 GLfloat a[4], b[4], result[4];
1353 fetch_vector4(&inst->SrcReg[0], machine, a);
1354 fetch_vector4(&inst->SrcReg[1], machine, b);
1355 result[0] = (a[0] <= b[0]) ? 1.0F : 0.0F;
1356 result[1] = (a[1] <= b[1]) ? 1.0F : 0.0F;
1357 result[2] = (a[2] <= b[2]) ? 1.0F : 0.0F;
1358 result[3] = (a[3] <= b[3]) ? 1.0F : 0.0F;
1359 store_vector4(inst, machine, result);
1360 if (DEBUG_PROG) {
1361 printf("SLE (%g %g %g %g) = (%g %g %g %g) <= (%g %g %g %g)\n",
1362 result[0], result[1], result[2], result[3],
1363 a[0], a[1], a[2], a[3],
1364 b[0], b[1], b[2], b[3]);
1365 }
1366 }
1367 break;
1368 case OPCODE_SLT: /* set on less */
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("SLT (%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_SNE: /* set on not equal */
1387 {
1388 GLfloat a[4], b[4], result[4];
1389 fetch_vector4(&inst->SrcReg[0], machine, a);
1390 fetch_vector4(&inst->SrcReg[1], machine, b);
1391 result[0] = (a[0] != b[0]) ? 1.0F : 0.0F;
1392 result[1] = (a[1] != b[1]) ? 1.0F : 0.0F;
1393 result[2] = (a[2] != b[2]) ? 1.0F : 0.0F;
1394 result[3] = (a[3] != b[3]) ? 1.0F : 0.0F;
1395 store_vector4(inst, machine, result);
1396 if (DEBUG_PROG) {
1397 printf("SNE (%g %g %g %g) = (%g %g %g %g) != (%g %g %g %g)\n",
1398 result[0], result[1], result[2], result[3],
1399 a[0], a[1], a[2], a[3],
1400 b[0], b[1], b[2], b[3]);
1401 }
1402 }
1403 break;
1404 case OPCODE_SSG: /* set sign (-1, 0 or +1) */
1405 {
1406 GLfloat a[4], result[4];
1407 fetch_vector4(&inst->SrcReg[0], machine, a);
1408 result[0] = (GLfloat) ((a[0] > 0.0F) - (a[0] < 0.0F));
1409 result[1] = (GLfloat) ((a[1] > 0.0F) - (a[1] < 0.0F));
1410 result[2] = (GLfloat) ((a[2] > 0.0F) - (a[2] < 0.0F));
1411 result[3] = (GLfloat) ((a[3] > 0.0F) - (a[3] < 0.0F));
1412 store_vector4(inst, machine, result);
1413 }
1414 break;
1415 case OPCODE_STR: /* set true, operands ignored */
1416 {
1417 static const GLfloat result[4] = { 1.0F, 1.0F, 1.0F, 1.0F };
1418 store_vector4(inst, machine, result);
1419 }
1420 break;
1421 case OPCODE_SUB:
1422 {
1423 GLfloat a[4], b[4], result[4];
1424 fetch_vector4(&inst->SrcReg[0], machine, a);
1425 fetch_vector4(&inst->SrcReg[1], machine, b);
1426 result[0] = a[0] - b[0];
1427 result[1] = a[1] - b[1];
1428 result[2] = a[2] - b[2];
1429 result[3] = a[3] - b[3];
1430 store_vector4(inst, machine, result);
1431 if (DEBUG_PROG) {
1432 printf("SUB (%g %g %g %g) = (%g %g %g %g) - (%g %g %g %g)\n",
1433 result[0], result[1], result[2], result[3],
1434 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1435 }
1436 }
1437 break;
1438 case OPCODE_SWZ: /* extended swizzle */
1439 {
1440 const struct prog_src_register *source = &inst->SrcReg[0];
1441 const GLfloat *src = get_src_register_pointer(source, machine);
1442 GLfloat result[4];
1443 GLuint i;
1444 for (i = 0; i < 4; i++) {
1445 const GLuint swz = GET_SWZ(source->Swizzle, i);
1446 if (swz == SWIZZLE_ZERO)
1447 result[i] = 0.0;
1448 else if (swz == SWIZZLE_ONE)
1449 result[i] = 1.0;
1450 else {
1451 ASSERT(swz >= 0);
1452 ASSERT(swz <= 3);
1453 result[i] = src[swz];
1454 }
1455 if (source->Negate & (1 << i))
1456 result[i] = -result[i];
1457 }
1458 store_vector4(inst, machine, result);
1459 }
1460 break;
1461 case OPCODE_TEX: /* Both ARB and NV frag prog */
1462 /* Simple texel lookup */
1463 {
1464 GLfloat texcoord[4], color[4];
1465 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1466
1467 /* For TEX, texcoord.Q should not be used and its value should not
1468 * matter (at most, we pass coord.xyz to texture3D() in GLSL).
1469 * Set Q=1 so that FetchTexelDeriv() doesn't get a garbage value
1470 * which is effectively what happens when the texcoord swizzle
1471 * is .xyzz
1472 */
1473 texcoord[3] = 1.0f;
1474
1475 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1476
1477 if (DEBUG_PROG) {
1478 printf("TEX (%g, %g, %g, %g) = texture[%d][%g, %g, %g, %g]\n",
1479 color[0], color[1], color[2], color[3],
1480 inst->TexSrcUnit,
1481 texcoord[0], texcoord[1], texcoord[2], texcoord[3]);
1482 }
1483 store_vector4(inst, machine, color);
1484 }
1485 break;
1486 case OPCODE_TXB: /* GL_ARB_fragment_program only */
1487 /* Texel lookup with LOD bias */
1488 {
1489 GLfloat texcoord[4], color[4], lodBias;
1490
1491 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1492
1493 /* texcoord[3] is the bias to add to lambda */
1494 lodBias = texcoord[3];
1495
1496 fetch_texel(ctx, machine, inst, texcoord, lodBias, color);
1497
1498 if (DEBUG_PROG) {
1499 printf("TXB (%g, %g, %g, %g) = texture[%d][%g %g %g %g]"
1500 " bias %g\n",
1501 color[0], color[1], color[2], color[3],
1502 inst->TexSrcUnit,
1503 texcoord[0],
1504 texcoord[1],
1505 texcoord[2],
1506 texcoord[3],
1507 lodBias);
1508 }
1509
1510 store_vector4(inst, machine, color);
1511 }
1512 break;
1513 case OPCODE_TXD: /* GL_NV_fragment_program only */
1514 /* Texture lookup w/ partial derivatives for LOD */
1515 {
1516 GLfloat texcoord[4], dtdx[4], dtdy[4], color[4];
1517 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1518 fetch_vector4(&inst->SrcReg[1], machine, dtdx);
1519 fetch_vector4(&inst->SrcReg[2], machine, dtdy);
1520 machine->FetchTexelDeriv(ctx, texcoord, dtdx, dtdy,
1521 0.0, /* lodBias */
1522 inst->TexSrcUnit, color);
1523 store_vector4(inst, machine, color);
1524 }
1525 break;
1526 case OPCODE_TXL:
1527 /* Texel lookup with explicit LOD */
1528 {
1529 GLfloat texcoord[4], color[4], lod;
1530
1531 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1532
1533 /* texcoord[3] is the LOD */
1534 lod = texcoord[3];
1535
1536 machine->FetchTexelLod(ctx, texcoord, lod,
1537 machine->Samplers[inst->TexSrcUnit], color);
1538
1539 store_vector4(inst, machine, color);
1540 }
1541 break;
1542 case OPCODE_TXP: /* GL_ARB_fragment_program only */
1543 /* Texture lookup w/ projective divide */
1544 {
1545 GLfloat texcoord[4], color[4];
1546
1547 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1548 /* Not so sure about this test - if texcoord[3] is
1549 * zero, we'd probably be fine except for an ASSERT in
1550 * IROUND_POS() which gets triggered by the inf values created.
1551 */
1552 if (texcoord[3] != 0.0) {
1553 texcoord[0] /= texcoord[3];
1554 texcoord[1] /= texcoord[3];
1555 texcoord[2] /= texcoord[3];
1556 }
1557
1558 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1559
1560 store_vector4(inst, machine, color);
1561 }
1562 break;
1563 case OPCODE_TXP_NV: /* GL_NV_fragment_program only */
1564 /* Texture lookup w/ projective divide, as above, but do not
1565 * do the divide by w if sampling from a cube map.
1566 */
1567 {
1568 GLfloat texcoord[4], color[4];
1569
1570 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1571 if (inst->TexSrcTarget != TEXTURE_CUBE_INDEX &&
1572 texcoord[3] != 0.0) {
1573 texcoord[0] /= texcoord[3];
1574 texcoord[1] /= texcoord[3];
1575 texcoord[2] /= texcoord[3];
1576 }
1577
1578 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1579
1580 store_vector4(inst, machine, color);
1581 }
1582 break;
1583 case OPCODE_TRUNC: /* truncate toward zero */
1584 {
1585 GLfloat a[4], result[4];
1586 fetch_vector4(&inst->SrcReg[0], machine, a);
1587 result[0] = (GLfloat) (GLint) a[0];
1588 result[1] = (GLfloat) (GLint) a[1];
1589 result[2] = (GLfloat) (GLint) a[2];
1590 result[3] = (GLfloat) (GLint) a[3];
1591 store_vector4(inst, machine, result);
1592 }
1593 break;
1594 case OPCODE_UP2H: /* unpack two 16-bit floats */
1595 {
1596 const GLuint raw = fetch_vector1ui(&inst->SrcReg[0], machine);
1597 GLfloat result[4];
1598 GLushort hx, hy;
1599 hx = raw & 0xffff;
1600 hy = raw >> 16;
1601 result[0] = result[2] = _mesa_half_to_float(hx);
1602 result[1] = result[3] = _mesa_half_to_float(hy);
1603 store_vector4(inst, machine, result);
1604 }
1605 break;
1606 case OPCODE_UP2US: /* unpack two GLushorts */
1607 {
1608 const GLuint raw = fetch_vector1ui(&inst->SrcReg[0], machine);
1609 GLfloat result[4];
1610 GLushort usx, usy;
1611 usx = raw & 0xffff;
1612 usy = raw >> 16;
1613 result[0] = result[2] = usx * (1.0f / 65535.0f);
1614 result[1] = result[3] = usy * (1.0f / 65535.0f);
1615 store_vector4(inst, machine, result);
1616 }
1617 break;
1618 case OPCODE_UP4B: /* unpack four GLbytes */
1619 {
1620 const GLuint raw = fetch_vector1ui(&inst->SrcReg[0], machine);
1621 GLfloat result[4];
1622 result[0] = (((raw >> 0) & 0xff) - 128) / 127.0F;
1623 result[1] = (((raw >> 8) & 0xff) - 128) / 127.0F;
1624 result[2] = (((raw >> 16) & 0xff) - 128) / 127.0F;
1625 result[3] = (((raw >> 24) & 0xff) - 128) / 127.0F;
1626 store_vector4(inst, machine, result);
1627 }
1628 break;
1629 case OPCODE_UP4UB: /* unpack four GLubytes */
1630 {
1631 const GLuint raw = fetch_vector1ui(&inst->SrcReg[0], machine);
1632 GLfloat result[4];
1633 result[0] = ((raw >> 0) & 0xff) / 255.0F;
1634 result[1] = ((raw >> 8) & 0xff) / 255.0F;
1635 result[2] = ((raw >> 16) & 0xff) / 255.0F;
1636 result[3] = ((raw >> 24) & 0xff) / 255.0F;
1637 store_vector4(inst, machine, result);
1638 }
1639 break;
1640 case OPCODE_XPD: /* cross product */
1641 {
1642 GLfloat a[4], b[4], result[4];
1643 fetch_vector4(&inst->SrcReg[0], machine, a);
1644 fetch_vector4(&inst->SrcReg[1], machine, b);
1645 result[0] = a[1] * b[2] - a[2] * b[1];
1646 result[1] = a[2] * b[0] - a[0] * b[2];
1647 result[2] = a[0] * b[1] - a[1] * b[0];
1648 result[3] = 1.0;
1649 store_vector4(inst, machine, result);
1650 if (DEBUG_PROG) {
1651 printf("XPD (%g %g %g %g) = (%g %g %g) X (%g %g %g)\n",
1652 result[0], result[1], result[2], result[3],
1653 a[0], a[1], a[2], b[0], b[1], b[2]);
1654 }
1655 }
1656 break;
1657 case OPCODE_X2D: /* 2-D matrix transform */
1658 {
1659 GLfloat a[4], b[4], c[4], result[4];
1660 fetch_vector4(&inst->SrcReg[0], machine, a);
1661 fetch_vector4(&inst->SrcReg[1], machine, b);
1662 fetch_vector4(&inst->SrcReg[2], machine, c);
1663 result[0] = a[0] + b[0] * c[0] + b[1] * c[1];
1664 result[1] = a[1] + b[0] * c[2] + b[1] * c[3];
1665 result[2] = a[2] + b[0] * c[0] + b[1] * c[1];
1666 result[3] = a[3] + b[0] * c[2] + b[1] * c[3];
1667 store_vector4(inst, machine, result);
1668 }
1669 break;
1670 case OPCODE_END:
1671 return GL_TRUE;
1672 default:
1673 _mesa_problem(ctx, "Bad opcode %d in _mesa_execute_program",
1674 inst->Opcode);
1675 return GL_TRUE; /* return value doesn't matter */
1676 }
1677
1678 numExec++;
1679 if (numExec > maxExec) {
1680 static GLboolean reported = GL_FALSE;
1681 if (!reported) {
1682 _mesa_problem(ctx, "Infinite loop detected in fragment program");
1683 reported = GL_TRUE;
1684 }
1685 return GL_TRUE;
1686 }
1687
1688 } /* for pc */
1689
1690 return GL_TRUE;
1691 }