1/*
2 * Mesa 3-D graphics library
3 *
4 * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included
14 * in all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25/**
26 * \file prog_execute.c
27 * Software interpreter for vertex/fragment programs.
28 * \author Brian Paul
29 */
30
31/*
32 * NOTE: we do everything in single-precision floating point; we don't
33 * currently observe the single/half/fixed-precision qualifiers.
34 *
35 */
36
37
38#include "c99_math.h"
39#include "main/errors.h"
40#include "main/glheader.h"
41#include "main/macros.h"
42#include "main/mtypes.h"
43#include "prog_execute.h"
44#include "prog_instruction.h"
45#include "prog_parameter.h"
46#include "prog_print.h"
47#include "prog_noise.h"
48
49
50/* debug predicate */
51#define DEBUG_PROG 0
52
53
54/**
55 * Set x to positive or negative infinity.
56 */
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
70#define SET_FLOAT_BITS(x, bits) ((fi_type *) (void *) &(x))->i = bits
71
72
73static const GLfloat ZeroVec[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
74
75
76/**
77 * Return a pointer to the 4-element float vector specified by the given
78 * source register.
79 */
80static inline const GLfloat *
81get_src_register_pointer(const struct prog_src_register *source,
82                         const struct gl_program_machine *machine)
83{
84   const struct gl_program *prog = machine->CurProgram;
85   GLint reg = source->Index;
86
87   if (source->RelAddr) {
88      /* add address register value to src index/offset */
89      reg += machine->AddressReg[0][0];
90      if (reg < 0) {
91         return ZeroVec;
92      }
93   }
94
95   switch (source->File) {
96   case PROGRAM_TEMPORARY:
97      if (reg >= MAX_PROGRAM_TEMPS)
98         return ZeroVec;
99      return machine->Temporaries[reg];
100
101   case PROGRAM_INPUT:
102      if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
103         if (reg >= VERT_ATTRIB_MAX)
104            return ZeroVec;
105         return machine->VertAttribs[reg];
106      }
107      else {
108         if (reg >= VARYING_SLOT_MAX)
109            return ZeroVec;
110         return machine->Attribs[reg][machine->CurElement];
111      }
112
113   case PROGRAM_OUTPUT:
114      if (reg >= MAX_PROGRAM_OUTPUTS)
115         return ZeroVec;
116      return machine->Outputs[reg];
117
118   case PROGRAM_STATE_VAR:
119      /* Fallthrough */
120   case PROGRAM_CONSTANT:
121      /* Fallthrough */
122   case PROGRAM_UNIFORM: {
123      if (reg >= (GLint) prog->Parameters->NumParameters)
124         return ZeroVec;
125
126      unsigned pvo = prog->Parameters->ParameterValueOffset[reg];
127      return (GLfloat *) prog->Parameters->ParameterValues + pvo;
128   }
129   case PROGRAM_SYSTEM_VALUE:
130      assert(reg < (GLint) ARRAY_SIZE(machine->SystemValues));
131      return machine->SystemValues[reg];
132
133   default:
134      _mesa_problem(NULL,
135         "Invalid src register file %d in get_src_register_pointer()",
136         source->File);
137      return ZeroVec;
138   }
139}
140
141
142/**
143 * Return a pointer to the 4-element float vector specified by the given
144 * destination register.
145 */
146static inline GLfloat *
147get_dst_register_pointer(const struct prog_dst_register *dest,
148                         struct gl_program_machine *machine)
149{
150   static GLfloat dummyReg[4];
151   GLint reg = dest->Index;
152
153   if (dest->RelAddr) {
154      /* add address register value to src index/offset */
155      reg += machine->AddressReg[0][0];
156      if (reg < 0) {
157         return dummyReg;
158      }
159   }
160
161   switch (dest->File) {
162   case PROGRAM_TEMPORARY:
163      if (reg >= MAX_PROGRAM_TEMPS)
164         return dummyReg;
165      return machine->Temporaries[reg];
166
167   case PROGRAM_OUTPUT:
168      if (reg >= MAX_PROGRAM_OUTPUTS)
169         return dummyReg;
170      return machine->Outputs[reg];
171
172   default:
173      _mesa_problem(NULL,
174         "Invalid dest register file %d in get_dst_register_pointer()",
175         dest->File);
176      return dummyReg;
177   }
178}
179
180
181
182/**
183 * Fetch a 4-element float vector from the given source register.
184 * Apply swizzling and negating as needed.
185 */
186static void
187fetch_vector4(const struct prog_src_register *source,
188              const struct gl_program_machine *machine, GLfloat result[4])
189{
190   const GLfloat *src = get_src_register_pointer(source, machine);
191
192   if (source->Swizzle == SWIZZLE_NOOP) {
193      /* no swizzling */
194      COPY_4V(result, src);
195   }
196   else {
197      assert(GET_SWZ(source->Swizzle, 0) <= 3);
198      assert(GET_SWZ(source->Swizzle, 1) <= 3);
199      assert(GET_SWZ(source->Swizzle, 2) <= 3);
200      assert(GET_SWZ(source->Swizzle, 3) <= 3);
201      result[0] = src[GET_SWZ(source->Swizzle, 0)];
202      result[1] = src[GET_SWZ(source->Swizzle, 1)];
203      result[2] = src[GET_SWZ(source->Swizzle, 2)];
204      result[3] = src[GET_SWZ(source->Swizzle, 3)];
205   }
206
207   if (source->Negate) {
208      assert(source->Negate == NEGATE_XYZW);
209      result[0] = -result[0];
210      result[1] = -result[1];
211      result[2] = -result[2];
212      result[3] = -result[3];
213   }
214
215#ifdef NAN_CHECK
216   assert(!IS_INF_OR_NAN(result[0]));
217   assert(!IS_INF_OR_NAN(result[0]));
218   assert(!IS_INF_OR_NAN(result[0]));
219   assert(!IS_INF_OR_NAN(result[0]));
220#endif
221}
222
223
224/**
225 * Fetch the derivative with respect to X or Y for the given register.
226 * XXX this currently only works for fragment program input attribs.
227 */
228static void
229fetch_vector4_deriv(const struct prog_src_register *source,
230                    const struct gl_program_machine *machine,
231                    char xOrY, GLfloat result[4])
232{
233   if (source->File == PROGRAM_INPUT &&
234       source->Index < (GLint) machine->NumDeriv) {
235      const GLint col = machine->CurElement;
236      const GLfloat w = machine->Attribs[VARYING_SLOT_POS][col][3];
237      const GLfloat invQ = 1.0f / w;
238      GLfloat deriv[4];
239
240      if (xOrY == 'X') {
241         deriv[0] = machine->DerivX[source->Index][0] * invQ;
242         deriv[1] = machine->DerivX[source->Index][1] * invQ;
243         deriv[2] = machine->DerivX[source->Index][2] * invQ;
244         deriv[3] = machine->DerivX[source->Index][3] * invQ;
245      }
246      else {
247         deriv[0] = machine->DerivY[source->Index][0] * invQ;
248         deriv[1] = machine->DerivY[source->Index][1] * invQ;
249         deriv[2] = machine->DerivY[source->Index][2] * invQ;
250         deriv[3] = machine->DerivY[source->Index][3] * invQ;
251      }
252
253      result[0] = deriv[GET_SWZ(source->Swizzle, 0)];
254      result[1] = deriv[GET_SWZ(source->Swizzle, 1)];
255      result[2] = deriv[GET_SWZ(source->Swizzle, 2)];
256      result[3] = deriv[GET_SWZ(source->Swizzle, 3)];
257
258      if (source->Negate) {
259         assert(source->Negate == NEGATE_XYZW);
260         result[0] = -result[0];
261         result[1] = -result[1];
262         result[2] = -result[2];
263         result[3] = -result[3];
264      }
265   }
266   else {
267      ASSIGN_4V(result, 0.0, 0.0, 0.0, 0.0);
268   }
269}
270
271
272/**
273 * As above, but only return result[0] element.
274 */
275static void
276fetch_vector1(const struct prog_src_register *source,
277              const struct gl_program_machine *machine, GLfloat result[4])
278{
279   const GLfloat *src = get_src_register_pointer(source, machine);
280
281   result[0] = src[GET_SWZ(source->Swizzle, 0)];
282
283   if (source->Negate) {
284      result[0] = -result[0];
285   }
286}
287
288
289/**
290 * Fetch texel from texture.  Use partial derivatives when possible.
291 */
292static inline void
293fetch_texel(struct gl_context *ctx,
294            const struct gl_program_machine *machine,
295            const struct prog_instruction *inst,
296            const GLfloat texcoord[4], GLfloat lodBias,
297            GLfloat color[4])
298{
299   const GLuint unit = machine->Samplers[inst->TexSrcUnit];
300
301   /* Note: we only have the right derivatives for fragment input attribs.
302    */
303   if (machine->NumDeriv > 0 &&
304       inst->SrcReg[0].File == PROGRAM_INPUT &&
305       inst->SrcReg[0].Index == VARYING_SLOT_TEX0 + inst->TexSrcUnit) {
306      /* simple texture fetch for which we should have derivatives */
307      GLuint attr = inst->SrcReg[0].Index;
308      machine->FetchTexelDeriv(ctx, texcoord,
309                               machine->DerivX[attr],
310                               machine->DerivY[attr],
311                               lodBias, unit, color);
312   }
313   else {
314      machine->FetchTexelLod(ctx, texcoord, lodBias, unit, color);
315   }
316}
317
318
319/**
320 * Store 4 floats into a register.  Observe the instructions saturate and
321 * set-condition-code flags.
322 */
323static void
324store_vector4(const struct prog_instruction *inst,
325              struct gl_program_machine *machine, const GLfloat value[4])
326{
327   const struct prog_dst_register *dstReg = &(inst->DstReg);
328   const GLboolean clamp = inst->Saturate;
329   GLuint writeMask = dstReg->WriteMask;
330   GLfloat clampedValue[4];
331   GLfloat *dst = get_dst_register_pointer(dstReg, machine);
332
333#if 0
334   if (value[0] > 1.0e10 ||
335       IS_INF_OR_NAN(value[0]) ||
336       IS_INF_OR_NAN(value[1]) ||
337       IS_INF_OR_NAN(value[2]) || IS_INF_OR_NAN(value[3]))
338      printf("store %g %g %g %g\n", value[0], value[1], value[2], value[3]);
339#endif
340
341   if (clamp) {
342      clampedValue[0] = CLAMP(value[0], 0.0F, 1.0F);
343      clampedValue[1] = CLAMP(value[1], 0.0F, 1.0F);
344      clampedValue[2] = CLAMP(value[2], 0.0F, 1.0F);
345      clampedValue[3] = CLAMP(value[3], 0.0F, 1.0F);
346      value = clampedValue;
347   }
348
349#ifdef NAN_CHECK
350   assert(!IS_INF_OR_NAN(value[0]));
351   assert(!IS_INF_OR_NAN(value[0]));
352   assert(!IS_INF_OR_NAN(value[0]));
353   assert(!IS_INF_OR_NAN(value[0]));
354#endif
355
356   if (writeMask & WRITEMASK_X)
357      dst[0] = value[0];
358   if (writeMask & WRITEMASK_Y)
359      dst[1] = value[1];
360   if (writeMask & WRITEMASK_Z)
361      dst[2] = value[2];
362   if (writeMask & WRITEMASK_W)
363      dst[3] = value[3];
364}
365
366
367/**
368 * Execute the given vertex/fragment program.
369 *
370 * \param ctx  rendering context
371 * \param program  the program to execute
372 * \param machine  machine state (must be initialized)
373 * \return GL_TRUE if program completed or GL_FALSE if program executed KIL.
374 */
375GLboolean
376_mesa_execute_program(struct gl_context * ctx,
377                      const struct gl_program *program,
378                      struct gl_program_machine *machine)
379{
380   const GLuint numInst = program->arb.NumInstructions;
381   const GLuint maxExec = 65536;
382   GLuint pc, numExec = 0;
383
384   machine->CurProgram = program;
385
386   if (DEBUG_PROG) {
387      printf("execute program %u --------------------\n", program->Id);
388   }
389
390   if (program->Target == GL_VERTEX_PROGRAM_ARB) {
391      machine->EnvParams = ctx->VertexProgram.Parameters;
392   }
393   else {
394      machine->EnvParams = ctx->FragmentProgram.Parameters;
395   }
396
397   for (pc = 0; pc < numInst; pc++) {
398      const struct prog_instruction *inst = program->arb.Instructions + pc;
399
400      if (DEBUG_PROG) {
401         _mesa_print_instruction(inst);
402      }
403
404      switch (inst->Opcode) {
405      case OPCODE_ABS:
406         {
407            GLfloat a[4], result[4];
408            fetch_vector4(&inst->SrcReg[0], machine, a);
409            result[0] = fabsf(a[0]);
410            result[1] = fabsf(a[1]);
411            result[2] = fabsf(a[2]);
412            result[3] = fabsf(a[3]);
413            store_vector4(inst, machine, result);
414         }
415         break;
416      case OPCODE_ADD:
417         {
418            GLfloat a[4], b[4], result[4];
419            fetch_vector4(&inst->SrcReg[0], machine, a);
420            fetch_vector4(&inst->SrcReg[1], machine, b);
421            result[0] = a[0] + b[0];
422            result[1] = a[1] + b[1];
423            result[2] = a[2] + b[2];
424            result[3] = a[3] + b[3];
425            store_vector4(inst, machine, result);
426            if (DEBUG_PROG) {
427               printf("ADD (%g %g %g %g) = (%g %g %g %g) + (%g %g %g %g)\n",
428                      result[0], result[1], result[2], result[3],
429                      a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
430            }
431         }
432         break;
433      case OPCODE_ARL:
434         {
435            GLfloat t[4];
436            fetch_vector4(&inst->SrcReg[0], machine, t);
437            machine->AddressReg[0][0] = IFLOOR(t[0]);
438            if (DEBUG_PROG) {
439               printf("ARL %d\n", machine->AddressReg[0][0]);
440            }
441         }
442         break;
443      case OPCODE_BGNLOOP:
444         /* no-op */
445         assert(program->arb.Instructions[inst->BranchTarget].Opcode
446                == OPCODE_ENDLOOP);
447         break;
448      case OPCODE_ENDLOOP:
449         /* subtract 1 here since pc is incremented by for(pc) loop */
450         assert(program->arb.Instructions[inst->BranchTarget].Opcode
451                == OPCODE_BGNLOOP);
452         pc = inst->BranchTarget - 1;   /* go to matching BNGLOOP */
453         break;
454      case OPCODE_BGNSUB:      /* begin subroutine */
455         break;
456      case OPCODE_ENDSUB:      /* end subroutine */
457         break;
458      case OPCODE_BRK:         /* break out of loop (conditional) */
459         assert(program->arb.Instructions[inst->BranchTarget].Opcode
460                == OPCODE_ENDLOOP);
461         /* break out of loop */
462         /* pc++ at end of for-loop will put us after the ENDLOOP inst */
463         pc = inst->BranchTarget;
464         break;
465      case OPCODE_CONT:        /* continue loop (conditional) */
466         assert(program->arb.Instructions[inst->BranchTarget].Opcode
467                == OPCODE_ENDLOOP);
468         /* continue at ENDLOOP */
469         /* Subtract 1 here since we'll do pc++ at end of for-loop */
470         pc = inst->BranchTarget - 1;
471         break;
472      case OPCODE_CAL:         /* Call subroutine (conditional) */
473         /* call the subroutine */
474         if (machine->StackDepth >= MAX_PROGRAM_CALL_DEPTH) {
475            return GL_TRUE;  /* Per GL_NV_vertex_program2 spec */
476         }
477         machine->CallStack[machine->StackDepth++] = pc + 1; /* next inst */
478         /* Subtract 1 here since we'll do pc++ at end of for-loop */
479         pc = inst->BranchTarget - 1;
480         break;
481      case OPCODE_CMP:
482         {
483            GLfloat a[4], b[4], c[4], result[4];
484            fetch_vector4(&inst->SrcReg[0], machine, a);
485            fetch_vector4(&inst->SrcReg[1], machine, b);
486            fetch_vector4(&inst->SrcReg[2], machine, c);
487            result[0] = a[0] < 0.0F ? b[0] : c[0];
488            result[1] = a[1] < 0.0F ? b[1] : c[1];
489            result[2] = a[2] < 0.0F ? b[2] : c[2];
490            result[3] = a[3] < 0.0F ? b[3] : c[3];
491            store_vector4(inst, machine, result);
492            if (DEBUG_PROG) {
493               printf("CMP (%g %g %g %g) = (%g %g %g %g) < 0 ? (%g %g %g %g) : (%g %g %g %g)\n",
494                      result[0], result[1], result[2], result[3],
495                      a[0], a[1], a[2], a[3],
496                      b[0], b[1], b[2], b[3],
497                      c[0], c[1], c[2], c[3]);
498            }
499         }
500         break;
501      case OPCODE_COS:
502         {
503            GLfloat a[4], result[4];
504            fetch_vector1(&inst->SrcReg[0], machine, a);
505            result[0] = result[1] = result[2] = result[3]
506               = cosf(a[0]);
507            store_vector4(inst, machine, result);
508         }
509         break;
510      case OPCODE_DDX:         /* Partial derivative with respect to X */
511         {
512            GLfloat result[4];
513            fetch_vector4_deriv(&inst->SrcReg[0], machine, 'X', result);
514            store_vector4(inst, machine, result);
515         }
516         break;
517      case OPCODE_DDY:         /* Partial derivative with respect to Y */
518         {
519            GLfloat result[4];
520            fetch_vector4_deriv(&inst->SrcReg[0], machine, 'Y', result);
521            store_vector4(inst, machine, result);
522         }
523         break;
524      case OPCODE_DP2:
525         {
526            GLfloat a[4], b[4], result[4];
527            fetch_vector4(&inst->SrcReg[0], machine, a);
528            fetch_vector4(&inst->SrcReg[1], machine, b);
529            result[0] = result[1] = result[2] = result[3] = DOT2(a, b);
530            store_vector4(inst, machine, result);
531            if (DEBUG_PROG) {
532               printf("DP2 %g = (%g %g) . (%g %g)\n",
533                      result[0], a[0], a[1], b[0], b[1]);
534            }
535         }
536         break;
537      case OPCODE_DP3:
538         {
539            GLfloat a[4], b[4], result[4];
540            fetch_vector4(&inst->SrcReg[0], machine, a);
541            fetch_vector4(&inst->SrcReg[1], machine, b);
542            result[0] = result[1] = result[2] = result[3] = DOT3(a, b);
543            store_vector4(inst, machine, result);
544            if (DEBUG_PROG) {
545               printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
546                      result[0], a[0], a[1], a[2], b[0], b[1], b[2]);
547            }
548         }
549         break;
550      case OPCODE_DP4:
551         {
552            GLfloat a[4], b[4], result[4];
553            fetch_vector4(&inst->SrcReg[0], machine, a);
554            fetch_vector4(&inst->SrcReg[1], machine, b);
555            result[0] = result[1] = result[2] = result[3] = DOT4(a, b);
556            store_vector4(inst, machine, result);
557            if (DEBUG_PROG) {
558               printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
559                      result[0], a[0], a[1], a[2], a[3],
560                      b[0], b[1], b[2], b[3]);
561            }
562         }
563         break;
564      case OPCODE_DPH:
565         {
566            GLfloat a[4], b[4], result[4];
567            fetch_vector4(&inst->SrcReg[0], machine, a);
568            fetch_vector4(&inst->SrcReg[1], machine, b);
569            result[0] = result[1] = result[2] = result[3] = DOT3(a, b) + b[3];
570            store_vector4(inst, machine, result);
571         }
572         break;
573      case OPCODE_DST:         /* Distance vector */
574         {
575            GLfloat a[4], b[4], result[4];
576            fetch_vector4(&inst->SrcReg[0], machine, a);
577            fetch_vector4(&inst->SrcReg[1], machine, b);
578            result[0] = 1.0F;
579            result[1] = a[1] * b[1];
580            result[2] = a[2];
581            result[3] = b[3];
582            store_vector4(inst, machine, result);
583         }
584         break;
585      case OPCODE_EXP:
586         {
587            GLfloat t[4], q[4], floor_t0;
588            fetch_vector1(&inst->SrcReg[0], machine, t);
589            floor_t0 = floorf(t[0]);
590            if (floor_t0 > FLT_MAX_EXP) {
591               SET_POS_INFINITY(q[0]);
592               SET_POS_INFINITY(q[2]);
593            }
594            else if (floor_t0 < FLT_MIN_EXP) {
595               q[0] = 0.0F;
596               q[2] = 0.0F;
597            }
598            else {
599               q[0] = ldexpf(1.0, (int) floor_t0);
600               /* Note: GL_NV_vertex_program expects
601                * result.z = result.x * APPX(result.y)
602                * We do what the ARB extension says.
603                */
604               q[2] = exp2f(t[0]);
605            }
606            q[1] = t[0] - floor_t0;
607            q[3] = 1.0F;
608            store_vector4( inst, machine, q );
609         }
610         break;
611      case OPCODE_EX2:         /* Exponential base 2 */
612         {
613            GLfloat a[4], result[4], val;
614            fetch_vector1(&inst->SrcReg[0], machine, a);
615            val = exp2f(a[0]);
616            /*
617            if (IS_INF_OR_NAN(val))
618               val = 1.0e10;
619            */
620            result[0] = result[1] = result[2] = result[3] = val;
621            store_vector4(inst, machine, result);
622         }
623         break;
624      case OPCODE_FLR:
625         {
626            GLfloat a[4], result[4];
627            fetch_vector4(&inst->SrcReg[0], machine, a);
628            result[0] = floorf(a[0]);
629            result[1] = floorf(a[1]);
630            result[2] = floorf(a[2]);
631            result[3] = floorf(a[3]);
632            store_vector4(inst, machine, result);
633         }
634         break;
635      case OPCODE_FRC:
636         {
637            GLfloat a[4], result[4];
638            fetch_vector4(&inst->SrcReg[0], machine, a);
639            result[0] = a[0] - floorf(a[0]);
640            result[1] = a[1] - floorf(a[1]);
641            result[2] = a[2] - floorf(a[2]);
642            result[3] = a[3] - floorf(a[3]);
643            store_vector4(inst, machine, result);
644         }
645         break;
646      case OPCODE_IF:
647         {
648            GLboolean cond;
649            assert(program->arb.Instructions[inst->BranchTarget].Opcode
650                   == OPCODE_ELSE ||
651                   program->arb.Instructions[inst->BranchTarget].Opcode
652                   == OPCODE_ENDIF);
653            /* eval condition */
654            GLfloat a[4];
655            fetch_vector1(&inst->SrcReg[0], machine, a);
656            cond = (a[0] != 0.0F);
657            if (DEBUG_PROG) {
658               printf("IF: %d\n", cond);
659            }
660            /* do if/else */
661            if (cond) {
662               /* do if-clause (just continue execution) */
663            }
664            else {
665               /* go to the instruction after ELSE or ENDIF */
666               assert(inst->BranchTarget >= 0);
667               pc = inst->BranchTarget;
668            }
669         }
670         break;
671      case OPCODE_ELSE:
672         /* goto ENDIF */
673         assert(program->arb.Instructions[inst->BranchTarget].Opcode
674                == OPCODE_ENDIF);
675         assert(inst->BranchTarget >= 0);
676         pc = inst->BranchTarget;
677         break;
678      case OPCODE_ENDIF:
679         /* nothing */
680         break;
681      case OPCODE_KIL:         /* ARB_f_p only */
682         {
683            GLfloat a[4];
684            fetch_vector4(&inst->SrcReg[0], machine, a);
685            if (DEBUG_PROG) {
686               printf("KIL if (%g %g %g %g) <= 0.0\n",
687                      a[0], a[1], a[2], a[3]);
688            }
689
690            if (a[0] < 0.0F || a[1] < 0.0F || a[2] < 0.0F || a[3] < 0.0F) {
691               return GL_FALSE;
692            }
693         }
694         break;
695      case OPCODE_LG2:         /* log base 2 */
696         {
697            GLfloat a[4], result[4], val;
698            fetch_vector1(&inst->SrcReg[0], machine, a);
699	    /* The fast LOG2 macro doesn't meet the precision requirements.
700	     */
701            if (a[0] == 0.0F) {
702               val = -FLT_MAX;
703            }
704            else {
705               val = logf(a[0]) * 1.442695F;
706            }
707            result[0] = result[1] = result[2] = result[3] = val;
708            store_vector4(inst, machine, result);
709         }
710         break;
711      case OPCODE_LIT:
712         {
713            const GLfloat epsilon = 1.0F / 256.0F;      /* from NV VP spec */
714            GLfloat a[4], result[4];
715            fetch_vector4(&inst->SrcReg[0], machine, a);
716            a[0] = MAX2(a[0], 0.0F);
717            a[1] = MAX2(a[1], 0.0F);
718            /* XXX ARB version clamps a[3], NV version doesn't */
719            a[3] = CLAMP(a[3], -(128.0F - epsilon), (128.0F - epsilon));
720            result[0] = 1.0F;
721            result[1] = a[0];
722            /* XXX we could probably just use pow() here */
723            if (a[0] > 0.0F) {
724               if (a[1] == 0.0F && a[3] == 0.0F)
725                  result[2] = 1.0F;
726               else
727                  result[2] = powf(a[1], a[3]);
728            }
729            else {
730               result[2] = 0.0F;
731            }
732            result[3] = 1.0F;
733            store_vector4(inst, machine, result);
734            if (DEBUG_PROG) {
735               printf("LIT (%g %g %g %g) : (%g %g %g %g)\n",
736                      result[0], result[1], result[2], result[3],
737                      a[0], a[1], a[2], a[3]);
738            }
739         }
740         break;
741      case OPCODE_LOG:
742         {
743            GLfloat t[4], q[4], abs_t0;
744            fetch_vector1(&inst->SrcReg[0], machine, t);
745            abs_t0 = fabsf(t[0]);
746            if (abs_t0 != 0.0F) {
747               if (IS_INF_OR_NAN(abs_t0))
748               {
749                  SET_POS_INFINITY(q[0]);
750                  q[1] = 1.0F;
751                  SET_POS_INFINITY(q[2]);
752               }
753               else {
754                  int exponent;
755                  GLfloat mantissa = frexpf(t[0], &exponent);
756                  q[0] = (GLfloat) (exponent - 1);
757                  q[1] = 2.0F * mantissa; /* map [.5, 1) -> [1, 2) */
758
759		  /* The fast LOG2 macro doesn't meet the precision
760		   * requirements.
761		   */
762                  q[2] = logf(t[0]) * 1.442695F;
763               }
764            }
765            else {
766               SET_NEG_INFINITY(q[0]);
767               q[1] = 1.0F;
768               SET_NEG_INFINITY(q[2]);
769            }
770            q[3] = 1.0;
771            store_vector4(inst, machine, q);
772         }
773         break;
774      case OPCODE_LRP:
775         {
776            GLfloat a[4], b[4], c[4], result[4];
777            fetch_vector4(&inst->SrcReg[0], machine, a);
778            fetch_vector4(&inst->SrcReg[1], machine, b);
779            fetch_vector4(&inst->SrcReg[2], machine, c);
780            result[0] = a[0] * b[0] + (1.0F - a[0]) * c[0];
781            result[1] = a[1] * b[1] + (1.0F - a[1]) * c[1];
782            result[2] = a[2] * b[2] + (1.0F - a[2]) * c[2];
783            result[3] = a[3] * b[3] + (1.0F - a[3]) * c[3];
784            store_vector4(inst, machine, result);
785            if (DEBUG_PROG) {
786               printf("LRP (%g %g %g %g) = (%g %g %g %g), "
787                      "(%g %g %g %g), (%g %g %g %g)\n",
788                      result[0], result[1], result[2], result[3],
789                      a[0], a[1], a[2], a[3],
790                      b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
791            }
792         }
793         break;
794      case OPCODE_MAD:
795         {
796            GLfloat a[4], b[4], c[4], result[4];
797            fetch_vector4(&inst->SrcReg[0], machine, a);
798            fetch_vector4(&inst->SrcReg[1], machine, b);
799            fetch_vector4(&inst->SrcReg[2], machine, c);
800            result[0] = a[0] * b[0] + c[0];
801            result[1] = a[1] * b[1] + c[1];
802            result[2] = a[2] * b[2] + c[2];
803            result[3] = a[3] * b[3] + c[3];
804            store_vector4(inst, machine, result);
805            if (DEBUG_PROG) {
806               printf("MAD (%g %g %g %g) = (%g %g %g %g) * "
807                      "(%g %g %g %g) + (%g %g %g %g)\n",
808                      result[0], result[1], result[2], result[3],
809                      a[0], a[1], a[2], a[3],
810                      b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
811            }
812         }
813         break;
814      case OPCODE_MAX:
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] = MAX2(a[0], b[0]);
820            result[1] = MAX2(a[1], b[1]);
821            result[2] = MAX2(a[2], b[2]);
822            result[3] = MAX2(a[3], b[3]);
823            store_vector4(inst, machine, result);
824            if (DEBUG_PROG) {
825               printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
826                      result[0], result[1], result[2], result[3],
827                      a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
828            }
829         }
830         break;
831      case OPCODE_MIN:
832         {
833            GLfloat a[4], b[4], result[4];
834            fetch_vector4(&inst->SrcReg[0], machine, a);
835            fetch_vector4(&inst->SrcReg[1], machine, b);
836            result[0] = MIN2(a[0], b[0]);
837            result[1] = MIN2(a[1], b[1]);
838            result[2] = MIN2(a[2], b[2]);
839            result[3] = MIN2(a[3], b[3]);
840            store_vector4(inst, machine, result);
841         }
842         break;
843      case OPCODE_MOV:
844         {
845            GLfloat result[4];
846            fetch_vector4(&inst->SrcReg[0], machine, result);
847            store_vector4(inst, machine, result);
848            if (DEBUG_PROG) {
849               printf("MOV (%g %g %g %g)\n",
850                      result[0], result[1], result[2], result[3]);
851            }
852         }
853         break;
854      case OPCODE_MUL:
855         {
856            GLfloat a[4], b[4], result[4];
857            fetch_vector4(&inst->SrcReg[0], machine, a);
858            fetch_vector4(&inst->SrcReg[1], machine, b);
859            result[0] = a[0] * b[0];
860            result[1] = a[1] * b[1];
861            result[2] = a[2] * b[2];
862            result[3] = a[3] * b[3];
863            store_vector4(inst, machine, result);
864            if (DEBUG_PROG) {
865               printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
866                      result[0], result[1], result[2], result[3],
867                      a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
868            }
869         }
870         break;
871      case OPCODE_NOISE1:
872         {
873            GLfloat a[4], result[4];
874            fetch_vector1(&inst->SrcReg[0], machine, a);
875            result[0] =
876               result[1] =
877               result[2] =
878               result[3] = _mesa_noise1(a[0]);
879            store_vector4(inst, machine, result);
880         }
881         break;
882      case OPCODE_NOISE2:
883         {
884            GLfloat a[4], result[4];
885            fetch_vector4(&inst->SrcReg[0], machine, a);
886            result[0] =
887               result[1] =
888               result[2] = result[3] = _mesa_noise2(a[0], a[1]);
889            store_vector4(inst, machine, result);
890         }
891         break;
892      case OPCODE_NOISE3:
893         {
894            GLfloat a[4], result[4];
895            fetch_vector4(&inst->SrcReg[0], machine, a);
896            result[0] =
897               result[1] =
898               result[2] =
899               result[3] = _mesa_noise3(a[0], a[1], a[2]);
900            store_vector4(inst, machine, result);
901         }
902         break;
903      case OPCODE_NOISE4:
904         {
905            GLfloat a[4], result[4];
906            fetch_vector4(&inst->SrcReg[0], machine, a);
907            result[0] =
908               result[1] =
909               result[2] =
910               result[3] = _mesa_noise4(a[0], a[1], a[2], a[3]);
911            store_vector4(inst, machine, result);
912         }
913         break;
914      case OPCODE_NOP:
915         break;
916      case OPCODE_POW:
917         {
918            GLfloat a[4], b[4], result[4];
919            fetch_vector1(&inst->SrcReg[0], machine, a);
920            fetch_vector1(&inst->SrcReg[1], machine, b);
921            result[0] = result[1] = result[2] = result[3]
922               = powf(a[0], b[0]);
923            store_vector4(inst, machine, result);
924         }
925         break;
926
927      case OPCODE_RCP:
928         {
929            GLfloat a[4], result[4];
930            fetch_vector1(&inst->SrcReg[0], machine, a);
931            if (DEBUG_PROG) {
932               if (a[0] == 0)
933                  printf("RCP(0)\n");
934               else if (IS_INF_OR_NAN(a[0]))
935                  printf("RCP(inf)\n");
936            }
937            result[0] = result[1] = result[2] = result[3] = 1.0F / a[0];
938            store_vector4(inst, machine, result);
939         }
940         break;
941      case OPCODE_RET:         /* return from subroutine (conditional) */
942         if (machine->StackDepth == 0) {
943            return GL_TRUE;  /* Per GL_NV_vertex_program2 spec */
944         }
945         /* subtract one because of pc++ in the for loop */
946         pc = machine->CallStack[--machine->StackDepth] - 1;
947         break;
948      case OPCODE_RSQ:         /* 1 / sqrt() */
949         {
950            GLfloat a[4], result[4];
951            fetch_vector1(&inst->SrcReg[0], machine, a);
952            a[0] = fabsf(a[0]);
953            result[0] = result[1] = result[2] = result[3] = 1.0f / sqrtf(a[0]);
954            store_vector4(inst, machine, result);
955            if (DEBUG_PROG) {
956               printf("RSQ %g = 1/sqrt(|%g|)\n", result[0], a[0]);
957            }
958         }
959         break;
960      case OPCODE_SCS:         /* sine and cos */
961         {
962            GLfloat a[4], result[4];
963            fetch_vector1(&inst->SrcReg[0], machine, a);
964            result[0] = cosf(a[0]);
965            result[1] = sinf(a[0]);
966            result[2] = 0.0F;    /* undefined! */
967            result[3] = 0.0F;    /* undefined! */
968            store_vector4(inst, machine, result);
969         }
970         break;
971      case OPCODE_SGE:         /* set on greater or equal */
972         {
973            GLfloat a[4], b[4], result[4];
974            fetch_vector4(&inst->SrcReg[0], machine, a);
975            fetch_vector4(&inst->SrcReg[1], machine, b);
976            result[0] = (a[0] >= b[0]) ? 1.0F : 0.0F;
977            result[1] = (a[1] >= b[1]) ? 1.0F : 0.0F;
978            result[2] = (a[2] >= b[2]) ? 1.0F : 0.0F;
979            result[3] = (a[3] >= b[3]) ? 1.0F : 0.0F;
980            store_vector4(inst, machine, result);
981            if (DEBUG_PROG) {
982               printf("SGE (%g %g %g %g) = (%g %g %g %g) >= (%g %g %g %g)\n",
983                      result[0], result[1], result[2], result[3],
984                      a[0], a[1], a[2], a[3],
985                      b[0], b[1], b[2], b[3]);
986            }
987         }
988         break;
989      case OPCODE_SIN:
990         {
991            GLfloat a[4], result[4];
992            fetch_vector1(&inst->SrcReg[0], machine, a);
993            result[0] = result[1] = result[2] = result[3]
994               = sinf(a[0]);
995            store_vector4(inst, machine, result);
996         }
997         break;
998      case OPCODE_SLT:         /* set on less */
999         {
1000            GLfloat a[4], b[4], result[4];
1001            fetch_vector4(&inst->SrcReg[0], machine, a);
1002            fetch_vector4(&inst->SrcReg[1], machine, b);
1003            result[0] = (a[0] < b[0]) ? 1.0F : 0.0F;
1004            result[1] = (a[1] < b[1]) ? 1.0F : 0.0F;
1005            result[2] = (a[2] < b[2]) ? 1.0F : 0.0F;
1006            result[3] = (a[3] < b[3]) ? 1.0F : 0.0F;
1007            store_vector4(inst, machine, result);
1008            if (DEBUG_PROG) {
1009               printf("SLT (%g %g %g %g) = (%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                      b[0], b[1], b[2], b[3]);
1013            }
1014         }
1015         break;
1016      case OPCODE_SSG:         /* set sign (-1, 0 or +1) */
1017         {
1018            GLfloat a[4], result[4];
1019            fetch_vector4(&inst->SrcReg[0], machine, a);
1020            result[0] = (GLfloat) ((a[0] > 0.0F) - (a[0] < 0.0F));
1021            result[1] = (GLfloat) ((a[1] > 0.0F) - (a[1] < 0.0F));
1022            result[2] = (GLfloat) ((a[2] > 0.0F) - (a[2] < 0.0F));
1023            result[3] = (GLfloat) ((a[3] > 0.0F) - (a[3] < 0.0F));
1024            store_vector4(inst, machine, result);
1025         }
1026         break;
1027      case OPCODE_SUB:
1028         {
1029            GLfloat a[4], b[4], result[4];
1030            fetch_vector4(&inst->SrcReg[0], machine, a);
1031            fetch_vector4(&inst->SrcReg[1], machine, b);
1032            result[0] = a[0] - b[0];
1033            result[1] = a[1] - b[1];
1034            result[2] = a[2] - b[2];
1035            result[3] = a[3] - b[3];
1036            store_vector4(inst, machine, result);
1037            if (DEBUG_PROG) {
1038               printf("SUB (%g %g %g %g) = (%g %g %g %g) - (%g %g %g %g)\n",
1039                      result[0], result[1], result[2], result[3],
1040                      a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1041            }
1042         }
1043         break;
1044      case OPCODE_SWZ:         /* extended swizzle */
1045         {
1046            const struct prog_src_register *source = &inst->SrcReg[0];
1047            const GLfloat *src = get_src_register_pointer(source, machine);
1048            GLfloat result[4];
1049            GLuint i;
1050            for (i = 0; i < 4; i++) {
1051               const GLuint swz = GET_SWZ(source->Swizzle, i);
1052               if (swz == SWIZZLE_ZERO)
1053                  result[i] = 0.0;
1054               else if (swz == SWIZZLE_ONE)
1055                  result[i] = 1.0;
1056               else {
1057                  assert(swz <= 3);
1058                  result[i] = src[swz];
1059               }
1060               if (source->Negate & (1 << i))
1061                  result[i] = -result[i];
1062            }
1063            store_vector4(inst, machine, result);
1064         }
1065         break;
1066      case OPCODE_TEX:         /* Both ARB and NV frag prog */
1067         /* Simple texel lookup */
1068         {
1069            GLfloat texcoord[4], color[4];
1070            fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1071
1072            /* For TEX, texcoord.Q should not be used and its value should not
1073             * matter (at most, we pass coord.xyz to texture3D() in GLSL).
1074             * Set Q=1 so that FetchTexelDeriv() doesn't get a garbage value
1075             * which is effectively what happens when the texcoord swizzle
1076             * is .xyzz
1077             */
1078            texcoord[3] = 1.0f;
1079
1080            fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1081
1082            if (DEBUG_PROG) {
1083               printf("TEX (%g, %g, %g, %g) = texture[%d][%g, %g, %g, %g]\n",
1084                      color[0], color[1], color[2], color[3],
1085                      inst->TexSrcUnit,
1086                      texcoord[0], texcoord[1], texcoord[2], texcoord[3]);
1087            }
1088            store_vector4(inst, machine, color);
1089         }
1090         break;
1091      case OPCODE_TXB:         /* GL_ARB_fragment_program only */
1092         /* Texel lookup with LOD bias */
1093         {
1094            GLfloat texcoord[4], color[4], lodBias;
1095
1096            fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1097
1098            /* texcoord[3] is the bias to add to lambda */
1099            lodBias = texcoord[3];
1100
1101            fetch_texel(ctx, machine, inst, texcoord, lodBias, color);
1102
1103            if (DEBUG_PROG) {
1104               printf("TXB (%g, %g, %g, %g) = texture[%d][%g %g %g %g]"
1105                      "  bias %g\n",
1106                      color[0], color[1], color[2], color[3],
1107                      inst->TexSrcUnit,
1108                      texcoord[0],
1109                      texcoord[1],
1110                      texcoord[2],
1111                      texcoord[3],
1112                      lodBias);
1113            }
1114
1115            store_vector4(inst, machine, color);
1116         }
1117         break;
1118      case OPCODE_TXD:
1119         /* Texture lookup w/ partial derivatives for LOD */
1120         {
1121            GLfloat texcoord[4], dtdx[4], dtdy[4], color[4];
1122            fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1123            fetch_vector4(&inst->SrcReg[1], machine, dtdx);
1124            fetch_vector4(&inst->SrcReg[2], machine, dtdy);
1125            machine->FetchTexelDeriv(ctx, texcoord, dtdx, dtdy,
1126                                     0.0, /* lodBias */
1127                                     inst->TexSrcUnit, color);
1128            store_vector4(inst, machine, color);
1129         }
1130         break;
1131      case OPCODE_TXL:
1132         /* Texel lookup with explicit LOD */
1133         {
1134            GLfloat texcoord[4], color[4], lod;
1135
1136            fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1137
1138            /* texcoord[3] is the LOD */
1139            lod = texcoord[3];
1140
1141	    machine->FetchTexelLod(ctx, texcoord, lod,
1142				   machine->Samplers[inst->TexSrcUnit], color);
1143
1144            store_vector4(inst, machine, color);
1145         }
1146         break;
1147      case OPCODE_TXP:         /* GL_ARB_fragment_program only */
1148         /* Texture lookup w/ projective divide */
1149         {
1150            GLfloat texcoord[4], color[4];
1151
1152            fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1153            /* Not so sure about this test - if texcoord[3] is
1154             * zero, we'd probably be fine except for an assert in
1155             * IROUND_POS() which gets triggered by the inf values created.
1156             */
1157            if (texcoord[3] != 0.0F) {
1158               texcoord[0] /= texcoord[3];
1159               texcoord[1] /= texcoord[3];
1160               texcoord[2] /= texcoord[3];
1161            }
1162
1163            fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1164
1165            store_vector4(inst, machine, color);
1166         }
1167         break;
1168      case OPCODE_TRUNC:       /* truncate toward zero */
1169         {
1170            GLfloat a[4], result[4];
1171            fetch_vector4(&inst->SrcReg[0], machine, a);
1172            result[0] = (GLfloat) (GLint) a[0];
1173            result[1] = (GLfloat) (GLint) a[1];
1174            result[2] = (GLfloat) (GLint) a[2];
1175            result[3] = (GLfloat) (GLint) a[3];
1176            store_vector4(inst, machine, result);
1177         }
1178         break;
1179      case OPCODE_XPD:         /* cross product */
1180         {
1181            GLfloat a[4], b[4], result[4];
1182            fetch_vector4(&inst->SrcReg[0], machine, a);
1183            fetch_vector4(&inst->SrcReg[1], machine, b);
1184            result[0] = a[1] * b[2] - a[2] * b[1];
1185            result[1] = a[2] * b[0] - a[0] * b[2];
1186            result[2] = a[0] * b[1] - a[1] * b[0];
1187            result[3] = 1.0;
1188            store_vector4(inst, machine, result);
1189            if (DEBUG_PROG) {
1190               printf("XPD (%g %g %g %g) = (%g %g %g) X (%g %g %g)\n",
1191                      result[0], result[1], result[2], result[3],
1192                      a[0], a[1], a[2], b[0], b[1], b[2]);
1193            }
1194         }
1195         break;
1196      case OPCODE_END:
1197         return GL_TRUE;
1198      default:
1199         _mesa_problem(ctx, "Bad opcode %d in _mesa_execute_program",
1200                       inst->Opcode);
1201         return GL_TRUE;        /* return value doesn't matter */
1202      }
1203
1204      numExec++;
1205      if (numExec > maxExec) {
1206	 static GLboolean reported = GL_FALSE;
1207	 if (!reported) {
1208	    _mesa_problem(ctx, "Infinite loop detected in fragment program");
1209	    reported = GL_TRUE;
1210	 }
1211         return GL_TRUE;
1212      }
1213
1214   } /* for pc */
1215
1216   return GL_TRUE;
1217}
1218