projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
llvmpipe: check for texture usage in all scenes
[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 ASSERT(program->Instructions[inst->BranchTarget].Opcode
688 == OPCODE_ENDLOOP);
689 break;
690 case OPCODE_ENDLOOP:
691 /* subtract 1 here since pc is incremented by for(pc) loop */
692 ASSERT(program->Instructions[inst->BranchTarget].Opcode
693 == OPCODE_BGNLOOP);
694 pc = inst->BranchTarget - 1; /* go to matching BNGLOOP */
695 break;
696 case OPCODE_BGNSUB: /* begin subroutine */
697 break;
698 case OPCODE_ENDSUB: /* end subroutine */
699 break;
700 case OPCODE_BRA: /* branch (conditional) */
701 if (eval_condition(machine, inst)) {
702 /* take branch */
703 /* Subtract 1 here since we'll do pc++ below */
704 pc = inst->BranchTarget - 1;
705 }
706 break;
707 case OPCODE_BRK: /* break out of loop (conditional) */
708 ASSERT(program->Instructions[inst->BranchTarget].Opcode
709 == OPCODE_ENDLOOP);
710 if (eval_condition(machine, inst)) {
711 /* break out of loop */
712 /* pc++ at end of for-loop will put us after the ENDLOOP inst */
713 pc = inst->BranchTarget;
714 }
715 break;
716 case OPCODE_CONT: /* continue loop (conditional) */
717 ASSERT(program->Instructions[inst->BranchTarget].Opcode
718 == OPCODE_ENDLOOP);
719 if (eval_condition(machine, inst)) {
720 /* continue at ENDLOOP */
721 /* Subtract 1 here since we'll do pc++ at end of for-loop */
722 pc = inst->BranchTarget - 1;
723 }
724 break;
725 case OPCODE_CAL: /* Call subroutine (conditional) */
726 if (eval_condition(machine, inst)) {
727 /* call the subroutine */
728 if (machine->StackDepth >= MAX_PROGRAM_CALL_DEPTH) {
729 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
730 }
731 machine->CallStack[machine->StackDepth++] = pc + 1; /* next inst */
732 /* Subtract 1 here since we'll do pc++ at end of for-loop */
733 pc = inst->BranchTarget - 1;
734 }
735 break;
736 case OPCODE_CMP:
737 {
738 GLfloat a[4], b[4], c[4], result[4];
739 fetch_vector4(&inst->SrcReg[0], machine, a);
740 fetch_vector4(&inst->SrcReg[1], machine, b);
741 fetch_vector4(&inst->SrcReg[2], machine, c);
742 result[0] = a[0] < 0.0F ? b[0] : c[0];
743 result[1] = a[1] < 0.0F ? b[1] : c[1];
744 result[2] = a[2] < 0.0F ? b[2] : c[2];
745 result[3] = a[3] < 0.0F ? b[3] : c[3];
746 store_vector4(inst, machine, result);
747 }
748 break;
749 case OPCODE_COS:
750 {
751 GLfloat a[4], result[4];
752 fetch_vector1(&inst->SrcReg[0], machine, a);
753 result[0] = result[1] = result[2] = result[3]
754 = (GLfloat) _mesa_cos(a[0]);
755 store_vector4(inst, machine, result);
756 }
757 break;
758 case OPCODE_DDX: /* Partial derivative with respect to X */
759 {
760 GLfloat result[4];
761 fetch_vector4_deriv(ctx, &inst->SrcReg[0], machine,
762 'X', result);
763 store_vector4(inst, machine, result);
764 }
765 break;
766 case OPCODE_DDY: /* Partial derivative with respect to Y */
767 {
768 GLfloat result[4];
769 fetch_vector4_deriv(ctx, &inst->SrcReg[0], machine,
770 'Y', result);
771 store_vector4(inst, machine, result);
772 }
773 break;
774 case OPCODE_DP2:
775 {
776 GLfloat a[4], b[4], result[4];
777 fetch_vector4(&inst->SrcReg[0], machine, a);
778 fetch_vector4(&inst->SrcReg[1], machine, b);
779 result[0] = result[1] = result[2] = result[3] = DOT2(a, b);
780 store_vector4(inst, machine, result);
781 if (DEBUG_PROG) {
782 printf("DP2 %g = (%g %g) . (%g %g)\n",
783 result[0], a[0], a[1], b[0], b[1]);
784 }
785 }
786 break;
787 case OPCODE_DP2A:
788 {
789 GLfloat a[4], b[4], c, result[4];
790 fetch_vector4(&inst->SrcReg[0], machine, a);
791 fetch_vector4(&inst->SrcReg[1], machine, b);
792 fetch_vector1(&inst->SrcReg[1], machine, &c);
793 result[0] = result[1] = result[2] = result[3] = DOT2(a, b) + c;
794 store_vector4(inst, machine, result);
795 if (DEBUG_PROG) {
796 printf("DP2A %g = (%g %g) . (%g %g) + %g\n",
797 result[0], a[0], a[1], b[0], b[1], c);
798 }
799 }
800 break;
801 case OPCODE_DP3:
802 {
803 GLfloat a[4], b[4], result[4];
804 fetch_vector4(&inst->SrcReg[0], machine, a);
805 fetch_vector4(&inst->SrcReg[1], machine, b);
806 result[0] = result[1] = result[2] = result[3] = DOT3(a, b);
807 store_vector4(inst, machine, result);
808 if (DEBUG_PROG) {
809 printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
810 result[0], a[0], a[1], a[2], b[0], b[1], b[2]);
811 }
812 }
813 break;
814 case OPCODE_DP4:
815 {
816 GLfloat a[4], b[4], result[4];
817 fetch_vector4(&inst->SrcReg[0], machine, a);
818 fetch_vector4(&inst->SrcReg[1], machine, b);
819 result[0] = result[1] = result[2] = result[3] = DOT4(a, b);
820 store_vector4(inst, machine, result);
821 if (DEBUG_PROG) {
822 printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
823 result[0], a[0], a[1], a[2], a[3],
824 b[0], b[1], b[2], b[3]);
825 }
826 }
827 break;
828 case OPCODE_DPH:
829 {
830 GLfloat a[4], b[4], result[4];
831 fetch_vector4(&inst->SrcReg[0], machine, a);
832 fetch_vector4(&inst->SrcReg[1], machine, b);
833 result[0] = result[1] = result[2] = result[3] = DOT3(a, b) + b[3];
834 store_vector4(inst, machine, result);
835 }
836 break;
837 case OPCODE_DST: /* Distance vector */
838 {
839 GLfloat a[4], b[4], result[4];
840 fetch_vector4(&inst->SrcReg[0], machine, a);
841 fetch_vector4(&inst->SrcReg[1], machine, b);
842 result[0] = 1.0F;
843 result[1] = a[1] * b[1];
844 result[2] = a[2];
845 result[3] = b[3];
846 store_vector4(inst, machine, result);
847 }
848 break;
849 case OPCODE_EXP:
850 {
851 GLfloat t[4], q[4], floor_t0;
852 fetch_vector1(&inst->SrcReg[0], machine, t);
853 floor_t0 = FLOORF(t[0]);
854 if (floor_t0 > FLT_MAX_EXP) {
855 SET_POS_INFINITY(q[0]);
856 SET_POS_INFINITY(q[2]);
857 }
858 else if (floor_t0 < FLT_MIN_EXP) {
859 q[0] = 0.0F;
860 q[2] = 0.0F;
861 }
862 else {
863 q[0] = LDEXPF(1.0, (int) floor_t0);
864 /* Note: GL_NV_vertex_program expects
865 * result.z = result.x * APPX(result.y)
866 * We do what the ARB extension says.
867 */
868 q[2] = (GLfloat) _mesa_pow(2.0, t[0]);
869 }
870 q[1] = t[0] - floor_t0;
871 q[3] = 1.0F;
872 store_vector4( inst, machine, q );
873 }
874 break;
875 case OPCODE_EX2: /* Exponential base 2 */
876 {
877 GLfloat a[4], result[4], val;
878 fetch_vector1(&inst->SrcReg[0], machine, a);
879 val = (GLfloat) _mesa_pow(2.0, a[0]);
880 /*
881 if (IS_INF_OR_NAN(val))
882 val = 1.0e10;
883 */
884 result[0] = result[1] = result[2] = result[3] = val;
885 store_vector4(inst, machine, result);
886 }
887 break;
888 case OPCODE_FLR:
889 {
890 GLfloat a[4], result[4];
891 fetch_vector4(&inst->SrcReg[0], machine, a);
892 result[0] = FLOORF(a[0]);
893 result[1] = FLOORF(a[1]);
894 result[2] = FLOORF(a[2]);
895 result[3] = FLOORF(a[3]);
896 store_vector4(inst, machine, result);
897 }
898 break;
899 case OPCODE_FRC:
900 {
901 GLfloat a[4], result[4];
902 fetch_vector4(&inst->SrcReg[0], machine, a);
903 result[0] = a[0] - FLOORF(a[0]);
904 result[1] = a[1] - FLOORF(a[1]);
905 result[2] = a[2] - FLOORF(a[2]);
906 result[3] = a[3] - FLOORF(a[3]);
907 store_vector4(inst, machine, result);
908 }
909 break;
910 case OPCODE_IF:
911 {
912 GLboolean cond;
913 ASSERT(program->Instructions[inst->BranchTarget].Opcode
914 == OPCODE_ELSE ||
915 program->Instructions[inst->BranchTarget].Opcode
916 == OPCODE_ENDIF);
917 /* eval condition */
918 if (inst->SrcReg[0].File != PROGRAM_UNDEFINED) {
919 GLfloat a[4];
920 fetch_vector1(&inst->SrcReg[0], machine, a);
921 cond = (a[0] != 0.0);
922 }
923 else {
924 cond = eval_condition(machine, inst);
925 }
926 if (DEBUG_PROG) {
927 printf("IF: %d\n", cond);
928 }
929 /* do if/else */
930 if (cond) {
931 /* do if-clause (just continue execution) */
932 }
933 else {
934 /* go to the instruction after ELSE or ENDIF */
935 assert(inst->BranchTarget >= 0);
936 pc = inst->BranchTarget;
937 }
938 }
939 break;
940 case OPCODE_ELSE:
941 /* goto ENDIF */
942 ASSERT(program->Instructions[inst->BranchTarget].Opcode
943 == OPCODE_ENDIF);
944 assert(inst->BranchTarget >= 0);
945 pc = inst->BranchTarget;
946 break;
947 case OPCODE_ENDIF:
948 /* nothing */
949 break;
950 case OPCODE_KIL_NV: /* NV_f_p only (conditional) */
951 if (eval_condition(machine, inst)) {
952 return GL_FALSE;
953 }
954 break;
955 case OPCODE_KIL: /* ARB_f_p only */
956 {
957 GLfloat a[4];
958 fetch_vector4(&inst->SrcReg[0], machine, a);
959 if (DEBUG_PROG) {
960 printf("KIL if (%g %g %g %g) <= 0.0\n",
961 a[0], a[1], a[2], a[3]);
962 }
963
964 if (a[0] < 0.0F || a[1] < 0.0F || a[2] < 0.0F || a[3] < 0.0F) {
965 return GL_FALSE;
966 }
967 }
968 break;
969 case OPCODE_LG2: /* log base 2 */
970 {
971 GLfloat a[4], result[4], val;
972 fetch_vector1(&inst->SrcReg[0], machine, a);
973 /* The fast LOG2 macro doesn't meet the precision requirements.
974 */
975 if (a[0] == 0.0F) {
976 val = -FLT_MAX;
977 }
978 else {
979 val = log(a[0]) * 1.442695F;
980 }
981 result[0] = result[1] = result[2] = result[3] = val;
982 store_vector4(inst, machine, result);
983 }
984 break;
985 case OPCODE_LIT:
986 {
987 const GLfloat epsilon = 1.0F / 256.0F; /* from NV VP spec */
988 GLfloat a[4], result[4];
989 fetch_vector4(&inst->SrcReg[0], machine, a);
990 a[0] = MAX2(a[0], 0.0F);
991 a[1] = MAX2(a[1], 0.0F);
992 /* XXX ARB version clamps a[3], NV version doesn't */
993 a[3] = CLAMP(a[3], -(128.0F - epsilon), (128.0F - epsilon));
994 result[0] = 1.0F;
995 result[1] = a[0];
996 /* XXX we could probably just use pow() here */
997 if (a[0] > 0.0F) {
998 if (a[1] == 0.0 && a[3] == 0.0)
999 result[2] = 1.0;
1000 else
1001 result[2] = (GLfloat) _mesa_pow(a[1], a[3]);
1002 }
1003 else {
1004 result[2] = 0.0;
1005 }
1006 result[3] = 1.0F;
1007 store_vector4(inst, machine, result);
1008 if (DEBUG_PROG) {
1009 printf("LIT (%g %g %g %g) : (%g %g %g %g)\n",
1010 result[0], result[1], result[2], result[3],
1011 a[0], a[1], a[2], a[3]);
1012 }
1013 }
1014 break;
1015 case OPCODE_LOG:
1016 {
1017 GLfloat t[4], q[4], abs_t0;
1018 fetch_vector1(&inst->SrcReg[0], machine, t);
1019 abs_t0 = FABSF(t[0]);
1020 if (abs_t0 != 0.0F) {
1021 /* Since we really can't handle infinite values on VMS
1022 * like other OSes we'll use __MAXFLOAT to represent
1023 * infinity. This may need some tweaking.
1024 */
1025 #ifdef VMS
1026 if (abs_t0 == __MAXFLOAT)
1027 #else
1028 if (IS_INF_OR_NAN(abs_t0))
1029 #endif
1030 {
1031 SET_POS_INFINITY(q[0]);
1032 q[1] = 1.0F;
1033 SET_POS_INFINITY(q[2]);
1034 }
1035 else {
1036 int exponent;
1037 GLfloat mantissa = FREXPF(t[0], &exponent);
1038 q[0] = (GLfloat) (exponent - 1);
1039 q[1] = (GLfloat) (2.0 * mantissa); /* map [.5, 1) -> [1, 2) */
1040
1041 /* The fast LOG2 macro doesn't meet the precision
1042 * requirements.
1043 */
1044 q[2] = (log(t[0]) * 1.442695F);
1045 }
1046 }
1047 else {
1048 SET_NEG_INFINITY(q[0]);
1049 q[1] = 1.0F;
1050 SET_NEG_INFINITY(q[2]);
1051 }
1052 q[3] = 1.0;
1053 store_vector4(inst, machine, q);
1054 }
1055 break;
1056 case OPCODE_LRP:
1057 {
1058 GLfloat a[4], b[4], c[4], result[4];
1059 fetch_vector4(&inst->SrcReg[0], machine, a);
1060 fetch_vector4(&inst->SrcReg[1], machine, b);
1061 fetch_vector4(&inst->SrcReg[2], machine, c);
1062 result[0] = a[0] * b[0] + (1.0F - a[0]) * c[0];
1063 result[1] = a[1] * b[1] + (1.0F - a[1]) * c[1];
1064 result[2] = a[2] * b[2] + (1.0F - a[2]) * c[2];
1065 result[3] = a[3] * b[3] + (1.0F - a[3]) * c[3];
1066 store_vector4(inst, machine, result);
1067 if (DEBUG_PROG) {
1068 printf("LRP (%g %g %g %g) = (%g %g %g %g), "
1069 "(%g %g %g %g), (%g %g %g %g)\n",
1070 result[0], result[1], result[2], result[3],
1071 a[0], a[1], a[2], a[3],
1072 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
1073 }
1074 }
1075 break;
1076 case OPCODE_MAD:
1077 {
1078 GLfloat a[4], b[4], c[4], result[4];
1079 fetch_vector4(&inst->SrcReg[0], machine, a);
1080 fetch_vector4(&inst->SrcReg[1], machine, b);
1081 fetch_vector4(&inst->SrcReg[2], machine, c);
1082 result[0] = a[0] * b[0] + c[0];
1083 result[1] = a[1] * b[1] + c[1];
1084 result[2] = a[2] * b[2] + c[2];
1085 result[3] = a[3] * b[3] + c[3];
1086 store_vector4(inst, machine, result);
1087 if (DEBUG_PROG) {
1088 printf("MAD (%g %g %g %g) = (%g %g %g %g) * "
1089 "(%g %g %g %g) + (%g %g %g %g)\n",
1090 result[0], result[1], result[2], result[3],
1091 a[0], a[1], a[2], a[3],
1092 b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
1093 }
1094 }
1095 break;
1096 case OPCODE_MAX:
1097 {
1098 GLfloat a[4], b[4], result[4];
1099 fetch_vector4(&inst->SrcReg[0], machine, a);
1100 fetch_vector4(&inst->SrcReg[1], machine, b);
1101 result[0] = MAX2(a[0], b[0]);
1102 result[1] = MAX2(a[1], b[1]);
1103 result[2] = MAX2(a[2], b[2]);
1104 result[3] = MAX2(a[3], b[3]);
1105 store_vector4(inst, machine, result);
1106 if (DEBUG_PROG) {
1107 printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
1108 result[0], result[1], result[2], result[3],
1109 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1110 }
1111 }
1112 break;
1113 case OPCODE_MIN:
1114 {
1115 GLfloat a[4], b[4], result[4];
1116 fetch_vector4(&inst->SrcReg[0], machine, a);
1117 fetch_vector4(&inst->SrcReg[1], machine, b);
1118 result[0] = MIN2(a[0], b[0]);
1119 result[1] = MIN2(a[1], b[1]);
1120 result[2] = MIN2(a[2], b[2]);
1121 result[3] = MIN2(a[3], b[3]);
1122 store_vector4(inst, machine, result);
1123 }
1124 break;
1125 case OPCODE_MOV:
1126 {
1127 GLfloat result[4];
1128 fetch_vector4(&inst->SrcReg[0], machine, result);
1129 store_vector4(inst, machine, result);
1130 if (DEBUG_PROG) {
1131 printf("MOV (%g %g %g %g)\n",
1132 result[0], result[1], result[2], result[3]);
1133 }
1134 }
1135 break;
1136 case OPCODE_MUL:
1137 {
1138 GLfloat a[4], b[4], result[4];
1139 fetch_vector4(&inst->SrcReg[0], machine, a);
1140 fetch_vector4(&inst->SrcReg[1], machine, b);
1141 result[0] = a[0] * b[0];
1142 result[1] = a[1] * b[1];
1143 result[2] = a[2] * b[2];
1144 result[3] = a[3] * b[3];
1145 store_vector4(inst, machine, result);
1146 if (DEBUG_PROG) {
1147 printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
1148 result[0], result[1], result[2], result[3],
1149 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1150 }
1151 }
1152 break;
1153 case OPCODE_NOISE1:
1154 {
1155 GLfloat a[4], result[4];
1156 fetch_vector1(&inst->SrcReg[0], machine, a);
1157 result[0] =
1158 result[1] =
1159 result[2] =
1160 result[3] = _mesa_noise1(a[0]);
1161 store_vector4(inst, machine, result);
1162 }
1163 break;
1164 case OPCODE_NOISE2:
1165 {
1166 GLfloat a[4], result[4];
1167 fetch_vector4(&inst->SrcReg[0], machine, a);
1168 result[0] =
1169 result[1] =
1170 result[2] = result[3] = _mesa_noise2(a[0], a[1]);
1171 store_vector4(inst, machine, result);
1172 }
1173 break;
1174 case OPCODE_NOISE3:
1175 {
1176 GLfloat a[4], result[4];
1177 fetch_vector4(&inst->SrcReg[0], machine, a);
1178 result[0] =
1179 result[1] =
1180 result[2] =
1181 result[3] = _mesa_noise3(a[0], a[1], a[2]);
1182 store_vector4(inst, machine, result);
1183 }
1184 break;
1185 case OPCODE_NOISE4:
1186 {
1187 GLfloat a[4], result[4];
1188 fetch_vector4(&inst->SrcReg[0], machine, a);
1189 result[0] =
1190 result[1] =
1191 result[2] =
1192 result[3] = _mesa_noise4(a[0], a[1], a[2], a[3]);
1193 store_vector4(inst, machine, result);
1194 }
1195 break;
1196 case OPCODE_NOP:
1197 break;
1198 case OPCODE_NOT: /* bitwise NOT */
1199 {
1200 GLuint a[4], result[4];
1201 fetch_vector4ui(&inst->SrcReg[0], machine, a);
1202 result[0] = ~a[0];
1203 result[1] = ~a[1];
1204 result[2] = ~a[2];
1205 result[3] = ~a[3];
1206 store_vector4ui(inst, machine, result);
1207 }
1208 break;
1209 case OPCODE_NRM3: /* 3-component normalization */
1210 {
1211 GLfloat a[4], result[4];
1212 GLfloat tmp;
1213 fetch_vector4(&inst->SrcReg[0], machine, a);
1214 tmp = a[0] * a[0] + a[1] * a[1] + a[2] * a[2];
1215 if (tmp != 0.0F)
1216 tmp = INV_SQRTF(tmp);
1217 result[0] = tmp * a[0];
1218 result[1] = tmp * a[1];
1219 result[2] = tmp * a[2];
1220 result[3] = 0.0; /* undefined, but prevent valgrind warnings */
1221 store_vector4(inst, machine, result);
1222 }
1223 break;
1224 case OPCODE_NRM4: /* 4-component normalization */
1225 {
1226 GLfloat a[4], result[4];
1227 GLfloat tmp;
1228 fetch_vector4(&inst->SrcReg[0], machine, a);
1229 tmp = a[0] * a[0] + a[1] * a[1] + a[2] * a[2] + a[3] * a[3];
1230 if (tmp != 0.0F)
1231 tmp = INV_SQRTF(tmp);
1232 result[0] = tmp * a[0];
1233 result[1] = tmp * a[1];
1234 result[2] = tmp * a[2];
1235 result[3] = tmp * a[3];
1236 store_vector4(inst, machine, result);
1237 }
1238 break;
1239 case OPCODE_OR: /* bitwise OR */
1240 {
1241 GLuint a[4], b[4], result[4];
1242 fetch_vector4ui(&inst->SrcReg[0], machine, a);
1243 fetch_vector4ui(&inst->SrcReg[1], machine, b);
1244 result[0] = a[0] | b[0];
1245 result[1] = a[1] | b[1];
1246 result[2] = a[2] | b[2];
1247 result[3] = a[3] | b[3];
1248 store_vector4ui(inst, machine, result);
1249 }
1250 break;
1251 case OPCODE_PK2H: /* pack two 16-bit floats in one 32-bit float */
1252 {
1253 GLfloat a[4];
1254 GLuint result[4];
1255 GLhalfNV hx, hy;
1256 fetch_vector4(&inst->SrcReg[0], machine, a);
1257 hx = _mesa_float_to_half(a[0]);
1258 hy = _mesa_float_to_half(a[1]);
1259 result[0] =
1260 result[1] =
1261 result[2] =
1262 result[3] = hx | (hy << 16);
1263 store_vector4ui(inst, machine, result);
1264 }
1265 break;
1266 case OPCODE_PK2US: /* pack two GLushorts into one 32-bit float */
1267 {
1268 GLfloat a[4];
1269 GLuint result[4], usx, usy;
1270 fetch_vector4(&inst->SrcReg[0], machine, a);
1271 a[0] = CLAMP(a[0], 0.0F, 1.0F);
1272 a[1] = CLAMP(a[1], 0.0F, 1.0F);
1273 usx = IROUND(a[0] * 65535.0F);
1274 usy = IROUND(a[1] * 65535.0F);
1275 result[0] =
1276 result[1] =
1277 result[2] =
1278 result[3] = usx | (usy << 16);
1279 store_vector4ui(inst, machine, result);
1280 }
1281 break;
1282 case OPCODE_PK4B: /* pack four GLbytes into one 32-bit float */
1283 {
1284 GLfloat a[4];
1285 GLuint result[4], ubx, uby, ubz, ubw;
1286 fetch_vector4(&inst->SrcReg[0], machine, a);
1287 a[0] = CLAMP(a[0], -128.0F / 127.0F, 1.0F);
1288 a[1] = CLAMP(a[1], -128.0F / 127.0F, 1.0F);
1289 a[2] = CLAMP(a[2], -128.0F / 127.0F, 1.0F);
1290 a[3] = CLAMP(a[3], -128.0F / 127.0F, 1.0F);
1291 ubx = IROUND(127.0F * a[0] + 128.0F);
1292 uby = IROUND(127.0F * a[1] + 128.0F);
1293 ubz = IROUND(127.0F * a[2] + 128.0F);
1294 ubw = IROUND(127.0F * a[3] + 128.0F);
1295 result[0] =
1296 result[1] =
1297 result[2] =
1298 result[3] = ubx | (uby << 8) | (ubz << 16) | (ubw << 24);
1299 store_vector4ui(inst, machine, result);
1300 }
1301 break;
1302 case OPCODE_PK4UB: /* pack four GLubytes into one 32-bit float */
1303 {
1304 GLfloat a[4];
1305 GLuint result[4], ubx, uby, ubz, ubw;
1306 fetch_vector4(&inst->SrcReg[0], machine, a);
1307 a[0] = CLAMP(a[0], 0.0F, 1.0F);
1308 a[1] = CLAMP(a[1], 0.0F, 1.0F);
1309 a[2] = CLAMP(a[2], 0.0F, 1.0F);
1310 a[3] = CLAMP(a[3], 0.0F, 1.0F);
1311 ubx = IROUND(255.0F * a[0]);
1312 uby = IROUND(255.0F * a[1]);
1313 ubz = IROUND(255.0F * a[2]);
1314 ubw = IROUND(255.0F * a[3]);
1315 result[0] =
1316 result[1] =
1317 result[2] =
1318 result[3] = ubx | (uby << 8) | (ubz << 16) | (ubw << 24);
1319 store_vector4ui(inst, machine, result);
1320 }
1321 break;
1322 case OPCODE_POW:
1323 {
1324 GLfloat a[4], b[4], result[4];
1325 fetch_vector1(&inst->SrcReg[0], machine, a);
1326 fetch_vector1(&inst->SrcReg[1], machine, b);
1327 result[0] = result[1] = result[2] = result[3]
1328 = (GLfloat) _mesa_pow(a[0], b[0]);
1329 store_vector4(inst, machine, result);
1330 }
1331 break;
1332 case OPCODE_RCP:
1333 {
1334 GLfloat a[4], result[4];
1335 fetch_vector1(&inst->SrcReg[0], machine, a);
1336 if (DEBUG_PROG) {
1337 if (a[0] == 0)
1338 printf("RCP(0)\n");
1339 else if (IS_INF_OR_NAN(a[0]))
1340 printf("RCP(inf)\n");
1341 }
1342 result[0] = result[1] = result[2] = result[3] = 1.0F / a[0];
1343 store_vector4(inst, machine, result);
1344 }
1345 break;
1346 case OPCODE_RET: /* return from subroutine (conditional) */
1347 if (eval_condition(machine, inst)) {
1348 if (machine->StackDepth == 0) {
1349 return GL_TRUE; /* Per GL_NV_vertex_program2 spec */
1350 }
1351 /* subtract one because of pc++ in the for loop */
1352 pc = machine->CallStack[--machine->StackDepth] - 1;
1353 }
1354 break;
1355 case OPCODE_RFL: /* reflection vector */
1356 {
1357 GLfloat axis[4], dir[4], result[4], tmpX, tmpW;
1358 fetch_vector4(&inst->SrcReg[0], machine, axis);
1359 fetch_vector4(&inst->SrcReg[1], machine, dir);
1360 tmpW = DOT3(axis, axis);
1361 tmpX = (2.0F * DOT3(axis, dir)) / tmpW;
1362 result[0] = tmpX * axis[0] - dir[0];
1363 result[1] = tmpX * axis[1] - dir[1];
1364 result[2] = tmpX * axis[2] - dir[2];
1365 /* result[3] is never written! XXX enforce in parser! */
1366 store_vector4(inst, machine, result);
1367 }
1368 break;
1369 case OPCODE_RSQ: /* 1 / sqrt() */
1370 {
1371 GLfloat a[4], result[4];
1372 fetch_vector1(&inst->SrcReg[0], machine, a);
1373 a[0] = FABSF(a[0]);
1374 result[0] = result[1] = result[2] = result[3] = INV_SQRTF(a[0]);
1375 store_vector4(inst, machine, result);
1376 if (DEBUG_PROG) {
1377 printf("RSQ %g = 1/sqrt(|%g|)\n", result[0], a[0]);
1378 }
1379 }
1380 break;
1381 case OPCODE_SCS: /* sine and cos */
1382 {
1383 GLfloat a[4], result[4];
1384 fetch_vector1(&inst->SrcReg[0], machine, a);
1385 result[0] = (GLfloat) _mesa_cos(a[0]);
1386 result[1] = (GLfloat) _mesa_sin(a[0]);
1387 result[2] = 0.0; /* undefined! */
1388 result[3] = 0.0; /* undefined! */
1389 store_vector4(inst, machine, result);
1390 }
1391 break;
1392 case OPCODE_SEQ: /* set on 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("SEQ (%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_SFL: /* set false, operands ignored */
1411 {
1412 static const GLfloat result[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
1413 store_vector4(inst, machine, result);
1414 }
1415 break;
1416 case OPCODE_SGE: /* set on greater or equal */
1417 {
1418 GLfloat a[4], b[4], result[4];
1419 fetch_vector4(&inst->SrcReg[0], machine, a);
1420 fetch_vector4(&inst->SrcReg[1], machine, b);
1421 result[0] = (a[0] >= b[0]) ? 1.0F : 0.0F;
1422 result[1] = (a[1] >= b[1]) ? 1.0F : 0.0F;
1423 result[2] = (a[2] >= b[2]) ? 1.0F : 0.0F;
1424 result[3] = (a[3] >= b[3]) ? 1.0F : 0.0F;
1425 store_vector4(inst, machine, result);
1426 if (DEBUG_PROG) {
1427 printf("SGE (%g %g %g %g) = (%g %g %g %g) >= (%g %g %g %g)\n",
1428 result[0], result[1], result[2], result[3],
1429 a[0], a[1], a[2], a[3],
1430 b[0], b[1], b[2], b[3]);
1431 }
1432 }
1433 break;
1434 case OPCODE_SGT: /* set on greater */
1435 {
1436 GLfloat a[4], b[4], result[4];
1437 fetch_vector4(&inst->SrcReg[0], machine, a);
1438 fetch_vector4(&inst->SrcReg[1], machine, b);
1439 result[0] = (a[0] > b[0]) ? 1.0F : 0.0F;
1440 result[1] = (a[1] > b[1]) ? 1.0F : 0.0F;
1441 result[2] = (a[2] > b[2]) ? 1.0F : 0.0F;
1442 result[3] = (a[3] > b[3]) ? 1.0F : 0.0F;
1443 store_vector4(inst, machine, result);
1444 if (DEBUG_PROG) {
1445 printf("SGT (%g %g %g %g) = (%g %g %g %g) > (%g %g %g %g)\n",
1446 result[0], result[1], result[2], result[3],
1447 a[0], a[1], a[2], a[3],
1448 b[0], b[1], b[2], b[3]);
1449 }
1450 }
1451 break;
1452 case OPCODE_SIN:
1453 {
1454 GLfloat a[4], result[4];
1455 fetch_vector1(&inst->SrcReg[0], machine, a);
1456 result[0] = result[1] = result[2] = result[3]
1457 = (GLfloat) _mesa_sin(a[0]);
1458 store_vector4(inst, machine, result);
1459 }
1460 break;
1461 case OPCODE_SLE: /* set on less or equal */
1462 {
1463 GLfloat a[4], b[4], result[4];
1464 fetch_vector4(&inst->SrcReg[0], machine, a);
1465 fetch_vector4(&inst->SrcReg[1], machine, b);
1466 result[0] = (a[0] <= b[0]) ? 1.0F : 0.0F;
1467 result[1] = (a[1] <= b[1]) ? 1.0F : 0.0F;
1468 result[2] = (a[2] <= b[2]) ? 1.0F : 0.0F;
1469 result[3] = (a[3] <= b[3]) ? 1.0F : 0.0F;
1470 store_vector4(inst, machine, result);
1471 if (DEBUG_PROG) {
1472 printf("SLE (%g %g %g %g) = (%g %g %g %g) <= (%g %g %g %g)\n",
1473 result[0], result[1], result[2], result[3],
1474 a[0], a[1], a[2], a[3],
1475 b[0], b[1], b[2], b[3]);
1476 }
1477 }
1478 break;
1479 case OPCODE_SLT: /* set on less */
1480 {
1481 GLfloat a[4], b[4], result[4];
1482 fetch_vector4(&inst->SrcReg[0], machine, a);
1483 fetch_vector4(&inst->SrcReg[1], machine, b);
1484 result[0] = (a[0] < b[0]) ? 1.0F : 0.0F;
1485 result[1] = (a[1] < b[1]) ? 1.0F : 0.0F;
1486 result[2] = (a[2] < b[2]) ? 1.0F : 0.0F;
1487 result[3] = (a[3] < b[3]) ? 1.0F : 0.0F;
1488 store_vector4(inst, machine, result);
1489 if (DEBUG_PROG) {
1490 printf("SLT (%g %g %g %g) = (%g %g %g %g) < (%g %g %g %g)\n",
1491 result[0], result[1], result[2], result[3],
1492 a[0], a[1], a[2], a[3],
1493 b[0], b[1], b[2], b[3]);
1494 }
1495 }
1496 break;
1497 case OPCODE_SNE: /* set on not equal */
1498 {
1499 GLfloat a[4], b[4], result[4];
1500 fetch_vector4(&inst->SrcReg[0], machine, a);
1501 fetch_vector4(&inst->SrcReg[1], machine, b);
1502 result[0] = (a[0] != b[0]) ? 1.0F : 0.0F;
1503 result[1] = (a[1] != b[1]) ? 1.0F : 0.0F;
1504 result[2] = (a[2] != b[2]) ? 1.0F : 0.0F;
1505 result[3] = (a[3] != b[3]) ? 1.0F : 0.0F;
1506 store_vector4(inst, machine, result);
1507 if (DEBUG_PROG) {
1508 printf("SNE (%g %g %g %g) = (%g %g %g %g) != (%g %g %g %g)\n",
1509 result[0], result[1], result[2], result[3],
1510 a[0], a[1], a[2], a[3],
1511 b[0], b[1], b[2], b[3]);
1512 }
1513 }
1514 break;
1515 case OPCODE_SSG: /* set sign (-1, 0 or +1) */
1516 {
1517 GLfloat a[4], result[4];
1518 fetch_vector4(&inst->SrcReg[0], machine, a);
1519 result[0] = (GLfloat) ((a[0] > 0.0F) - (a[0] < 0.0F));
1520 result[1] = (GLfloat) ((a[1] > 0.0F) - (a[1] < 0.0F));
1521 result[2] = (GLfloat) ((a[2] > 0.0F) - (a[2] < 0.0F));
1522 result[3] = (GLfloat) ((a[3] > 0.0F) - (a[3] < 0.0F));
1523 store_vector4(inst, machine, result);
1524 }
1525 break;
1526 case OPCODE_STR: /* set true, operands ignored */
1527 {
1528 static const GLfloat result[4] = { 1.0F, 1.0F, 1.0F, 1.0F };
1529 store_vector4(inst, machine, result);
1530 }
1531 break;
1532 case OPCODE_SUB:
1533 {
1534 GLfloat a[4], b[4], result[4];
1535 fetch_vector4(&inst->SrcReg[0], machine, a);
1536 fetch_vector4(&inst->SrcReg[1], machine, b);
1537 result[0] = a[0] - b[0];
1538 result[1] = a[1] - b[1];
1539 result[2] = a[2] - b[2];
1540 result[3] = a[3] - b[3];
1541 store_vector4(inst, machine, result);
1542 if (DEBUG_PROG) {
1543 printf("SUB (%g %g %g %g) = (%g %g %g %g) - (%g %g %g %g)\n",
1544 result[0], result[1], result[2], result[3],
1545 a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1546 }
1547 }
1548 break;
1549 case OPCODE_SWZ: /* extended swizzle */
1550 {
1551 const struct prog_src_register *source = &inst->SrcReg[0];
1552 const GLfloat *src = get_src_register_pointer(source, machine);
1553 GLfloat result[4];
1554 GLuint i;
1555 for (i = 0; i < 4; i++) {
1556 const GLuint swz = GET_SWZ(source->Swizzle, i);
1557 if (swz == SWIZZLE_ZERO)
1558 result[i] = 0.0;
1559 else if (swz == SWIZZLE_ONE)
1560 result[i] = 1.0;
1561 else {
1562 ASSERT(swz >= 0);
1563 ASSERT(swz <= 3);
1564 result[i] = src[swz];
1565 }
1566 if (source->Negate & (1 << i))
1567 result[i] = -result[i];
1568 }
1569 store_vector4(inst, machine, result);
1570 }
1571 break;
1572 case OPCODE_TEX: /* Both ARB and NV frag prog */
1573 /* Simple texel lookup */
1574 {
1575 GLfloat texcoord[4], color[4];
1576 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1577
1578 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1579
1580 if (DEBUG_PROG) {
1581 printf("TEX (%g, %g, %g, %g) = texture[%d][%g, %g, %g, %g]\n",
1582 color[0], color[1], color[2], color[3],
1583 inst->TexSrcUnit,
1584 texcoord[0], texcoord[1], texcoord[2], texcoord[3]);
1585 }
1586 store_vector4(inst, machine, color);
1587 }
1588 break;
1589 case OPCODE_TXB: /* GL_ARB_fragment_program only */
1590 /* Texel lookup with LOD bias */
1591 {
1592 GLfloat texcoord[4], color[4], lodBias;
1593
1594 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1595
1596 /* texcoord[3] is the bias to add to lambda */
1597 lodBias = texcoord[3];
1598
1599 fetch_texel(ctx, machine, inst, texcoord, lodBias, color);
1600
1601 store_vector4(inst, machine, color);
1602 }
1603 break;
1604 case OPCODE_TXD: /* GL_NV_fragment_program only */
1605 /* Texture lookup w/ partial derivatives for LOD */
1606 {
1607 GLfloat texcoord[4], dtdx[4], dtdy[4], color[4];
1608 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1609 fetch_vector4(&inst->SrcReg[1], machine, dtdx);
1610 fetch_vector4(&inst->SrcReg[2], machine, dtdy);
1611 machine->FetchTexelDeriv(ctx, texcoord, dtdx, dtdy,
1612 0.0, /* lodBias */
1613 inst->TexSrcUnit, color);
1614 store_vector4(inst, machine, color);
1615 }
1616 break;
1617 case OPCODE_TXP: /* GL_ARB_fragment_program only */
1618 /* Texture lookup w/ projective divide */
1619 {
1620 GLfloat texcoord[4], color[4];
1621
1622 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1623 /* Not so sure about this test - if texcoord[3] is
1624 * zero, we'd probably be fine except for an ASSERT in
1625 * IROUND_POS() which gets triggered by the inf values created.
1626 */
1627 if (texcoord[3] != 0.0) {
1628 texcoord[0] /= texcoord[3];
1629 texcoord[1] /= texcoord[3];
1630 texcoord[2] /= texcoord[3];
1631 }
1632
1633 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1634
1635 store_vector4(inst, machine, color);
1636 }
1637 break;
1638 case OPCODE_TXP_NV: /* GL_NV_fragment_program only */
1639 /* Texture lookup w/ projective divide, as above, but do not
1640 * do the divide by w if sampling from a cube map.
1641 */
1642 {
1643 GLfloat texcoord[4], color[4];
1644
1645 fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1646 if (inst->TexSrcTarget != TEXTURE_CUBE_INDEX &&
1647 texcoord[3] != 0.0) {
1648 texcoord[0] /= texcoord[3];
1649 texcoord[1] /= texcoord[3];
1650 texcoord[2] /= texcoord[3];
1651 }
1652
1653 fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1654
1655 store_vector4(inst, machine, color);
1656 }
1657 break;
1658 case OPCODE_TRUNC: /* truncate toward zero */
1659 {
1660 GLfloat a[4], result[4];
1661 fetch_vector4(&inst->SrcReg[0], machine, a);
1662 result[0] = (GLfloat) (GLint) a[0];
1663 result[1] = (GLfloat) (GLint) a[1];
1664 result[2] = (GLfloat) (GLint) a[2];
1665 result[3] = (GLfloat) (GLint) a[3];
1666 store_vector4(inst, machine, result);
1667 }
1668 break;
1669 case OPCODE_UP2H: /* unpack two 16-bit floats */
1670 {
1671 GLfloat a[4], result[4];
1672 fi_type fi;
1673 GLhalfNV hx, hy;
1674 fetch_vector1(&inst->SrcReg[0], machine, a);
1675 fi.f = a[0];
1676 hx = fi.i & 0xffff;
1677 hy = fi.i >> 16;
1678 result[0] = result[2] = _mesa_half_to_float(hx);
1679 result[1] = result[3] = _mesa_half_to_float(hy);
1680 store_vector4(inst, machine, result);
1681 }
1682 break;
1683 case OPCODE_UP2US: /* unpack two GLushorts */
1684 {
1685 GLfloat a[4], result[4];
1686 fi_type fi;
1687 GLushort usx, usy;
1688 fetch_vector1(&inst->SrcReg[0], machine, a);
1689 fi.f = a[0];
1690 usx = fi.i & 0xffff;
1691 usy = fi.i >> 16;
1692 result[0] = result[2] = usx * (1.0f / 65535.0f);
1693 result[1] = result[3] = usy * (1.0f / 65535.0f);
1694 store_vector4(inst, machine, result);
1695 }
1696 break;
1697 case OPCODE_UP4B: /* unpack four GLbytes */
1698 {
1699 GLfloat a[4], result[4];
1700 fi_type fi;
1701 fetch_vector1(&inst->SrcReg[0], machine, a);
1702 fi.f = a[0];
1703 result[0] = (((fi.i >> 0) & 0xff) - 128) / 127.0F;
1704 result[1] = (((fi.i >> 8) & 0xff) - 128) / 127.0F;
1705 result[2] = (((fi.i >> 16) & 0xff) - 128) / 127.0F;
1706 result[3] = (((fi.i >> 24) & 0xff) - 128) / 127.0F;
1707 store_vector4(inst, machine, result);
1708 }
1709 break;
1710 case OPCODE_UP4UB: /* unpack four GLubytes */
1711 {
1712 GLfloat a[4], result[4];
1713 fi_type fi;
1714 fetch_vector1(&inst->SrcReg[0], machine, a);
1715 fi.f = a[0];
1716 result[0] = ((fi.i >> 0) & 0xff) / 255.0F;
1717 result[1] = ((fi.i >> 8) & 0xff) / 255.0F;
1718 result[2] = ((fi.i >> 16) & 0xff) / 255.0F;
1719 result[3] = ((fi.i >> 24) & 0xff) / 255.0F;
1720 store_vector4(inst, machine, result);
1721 }
1722 break;
1723 case OPCODE_XOR: /* bitwise XOR */
1724 {
1725 GLuint a[4], b[4], result[4];
1726 fetch_vector4ui(&inst->SrcReg[0], machine, a);
1727 fetch_vector4ui(&inst->SrcReg[1], machine, b);
1728 result[0] = a[0] ^ b[0];
1729 result[1] = a[1] ^ b[1];
1730 result[2] = a[2] ^ b[2];
1731 result[3] = a[3] ^ b[3];
1732 store_vector4ui(inst, machine, result);
1733 }
1734 break;
1735 case OPCODE_XPD: /* cross product */
1736 {
1737 GLfloat a[4], b[4], result[4];
1738 fetch_vector4(&inst->SrcReg[0], machine, a);
1739 fetch_vector4(&inst->SrcReg[1], machine, b);
1740 result[0] = a[1] * b[2] - a[2] * b[1];
1741 result[1] = a[2] * b[0] - a[0] * b[2];
1742 result[2] = a[0] * b[1] - a[1] * b[0];
1743 result[3] = 1.0;
1744 store_vector4(inst, machine, result);
1745 if (DEBUG_PROG) {
1746 printf("XPD (%g %g %g %g) = (%g %g %g) X (%g %g %g)\n",
1747 result[0], result[1], result[2], result[3],
1748 a[0], a[1], a[2], b[0], b[1], b[2]);
1749 }
1750 }
1751 break;
1752 case OPCODE_X2D: /* 2-D matrix transform */
1753 {
1754 GLfloat a[4], b[4], c[4], result[4];
1755 fetch_vector4(&inst->SrcReg[0], machine, a);
1756 fetch_vector4(&inst->SrcReg[1], machine, b);
1757 fetch_vector4(&inst->SrcReg[2], machine, c);
1758 result[0] = a[0] + b[0] * c[0] + b[1] * c[1];
1759 result[1] = a[1] + b[0] * c[2] + b[1] * c[3];
1760 result[2] = a[2] + b[0] * c[0] + b[1] * c[1];
1761 result[3] = a[3] + b[0] * c[2] + b[1] * c[3];
1762 store_vector4(inst, machine, result);
1763 }
1764 break;
1765 case OPCODE_PRINT:
1766 {
1767 if (inst->SrcReg[0].File != -1) {
1768 GLfloat a[4];
1769 fetch_vector4(&inst->SrcReg[0], machine, a);
1770 _mesa_printf("%s%g, %g, %g, %g\n", (const char *) inst->Data,
1771 a[0], a[1], a[2], a[3]);
1772 }
1773 else {
1774 _mesa_printf("%s\n", (const char *) inst->Data);
1775 }
1776 }
1777 break;
1778 case OPCODE_END:
1779 return GL_TRUE;
1780 default:
1781 _mesa_problem(ctx, "Bad opcode %d in _mesa_execute_program",
1782 inst->Opcode);
1783 return GL_TRUE; /* return value doesn't matter */
1784 }
1785
1786 numExec++;
1787 if (numExec > maxExec) {
1788 _mesa_problem(ctx, "Infinite loop detected in fragment program");
1789 return GL_TRUE;
1790 }
1791
1792 } /* for pc */
1793
1794 return GL_TRUE;
1795 }