• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Mesa 3-D graphics library
3  *
4  * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included
14  * in all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  */
24 
25 /**
26  * \file prog_execute.c
27  * Software interpreter for vertex/fragment programs.
28  * \author Brian Paul
29  */
30 
31 /*
32  * NOTE: we do everything in single-precision floating point; we don't
33  * currently observe the single/half/fixed-precision qualifiers.
34  *
35  */
36 
37 
38 #include "c99_math.h"
39 #include "main/glheader.h"
40 #include "main/macros.h"
41 #include "prog_execute.h"
42 #include "prog_instruction.h"
43 #include "prog_parameter.h"
44 #include "prog_print.h"
45 #include "prog_noise.h"
46 
47 
48 /* debug predicate */
49 #define DEBUG_PROG 0
50 
51 
52 /**
53  * Set x to positive or negative infinity.
54  */
55 #define SET_POS_INFINITY(x)                  \
56    do {                                      \
57          fi_type fi;                         \
58          fi.i = 0x7F800000;                  \
59          x = fi.f;                           \
60    } while (0)
61 #define SET_NEG_INFINITY(x)                  \
62    do {                                      \
63          fi_type fi;                         \
64          fi.i = 0xFF800000;                  \
65          x = fi.f;                           \
66    } while (0)
67 
68 #define SET_FLOAT_BITS(x, bits) ((fi_type *) (void *) &(x))->i = bits
69 
70 
71 static const GLfloat ZeroVec[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
72 
73 
74 /**
75  * Return a pointer to the 4-element float vector specified by the given
76  * source register.
77  */
78 static inline const GLfloat *
get_src_register_pointer(const struct prog_src_register * source,const struct gl_program_machine * machine)79 get_src_register_pointer(const struct prog_src_register *source,
80                          const struct gl_program_machine *machine)
81 {
82    const struct gl_program *prog = machine->CurProgram;
83    GLint reg = source->Index;
84 
85    if (source->RelAddr) {
86       /* add address register value to src index/offset */
87       reg += machine->AddressReg[0][0];
88       if (reg < 0) {
89          return ZeroVec;
90       }
91    }
92 
93    switch (source->File) {
94    case PROGRAM_TEMPORARY:
95       if (reg >= MAX_PROGRAM_TEMPS)
96          return ZeroVec;
97       return machine->Temporaries[reg];
98 
99    case PROGRAM_INPUT:
100       if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
101          if (reg >= VERT_ATTRIB_MAX)
102             return ZeroVec;
103          return machine->VertAttribs[reg];
104       }
105       else {
106          if (reg >= VARYING_SLOT_MAX)
107             return ZeroVec;
108          return machine->Attribs[reg][machine->CurElement];
109       }
110 
111    case PROGRAM_OUTPUT:
112       if (reg >= MAX_PROGRAM_OUTPUTS)
113          return ZeroVec;
114       return machine->Outputs[reg];
115 
116    case PROGRAM_STATE_VAR:
117       /* Fallthrough */
118    case PROGRAM_CONSTANT:
119       /* Fallthrough */
120    case PROGRAM_UNIFORM:
121       if (reg >= (GLint) prog->Parameters->NumParameters)
122          return ZeroVec;
123       return (GLfloat *) prog->Parameters->ParameterValues[reg];
124 
125    case PROGRAM_SYSTEM_VALUE:
126       assert(reg < (GLint) ARRAY_SIZE(machine->SystemValues));
127       return machine->SystemValues[reg];
128 
129    default:
130       _mesa_problem(NULL,
131          "Invalid src register file %d in get_src_register_pointer()",
132          source->File);
133       return ZeroVec;
134    }
135 }
136 
137 
138 /**
139  * Return a pointer to the 4-element float vector specified by the given
140  * destination register.
141  */
142 static inline GLfloat *
get_dst_register_pointer(const struct prog_dst_register * dest,struct gl_program_machine * machine)143 get_dst_register_pointer(const struct prog_dst_register *dest,
144                          struct gl_program_machine *machine)
145 {
146    static GLfloat dummyReg[4];
147    GLint reg = dest->Index;
148 
149    if (dest->RelAddr) {
150       /* add address register value to src index/offset */
151       reg += machine->AddressReg[0][0];
152       if (reg < 0) {
153          return dummyReg;
154       }
155    }
156 
157    switch (dest->File) {
158    case PROGRAM_TEMPORARY:
159       if (reg >= MAX_PROGRAM_TEMPS)
160          return dummyReg;
161       return machine->Temporaries[reg];
162 
163    case PROGRAM_OUTPUT:
164       if (reg >= MAX_PROGRAM_OUTPUTS)
165          return dummyReg;
166       return machine->Outputs[reg];
167 
168    default:
169       _mesa_problem(NULL,
170          "Invalid dest register file %d in get_dst_register_pointer()",
171          dest->File);
172       return dummyReg;
173    }
174 }
175 
176 
177 
178 /**
179  * Fetch a 4-element float vector from the given source register.
180  * Apply swizzling and negating as needed.
181  */
182 static void
fetch_vector4(const struct prog_src_register * source,const struct gl_program_machine * machine,GLfloat result[4])183 fetch_vector4(const struct prog_src_register *source,
184               const struct gl_program_machine *machine, GLfloat result[4])
185 {
186    const GLfloat *src = get_src_register_pointer(source, machine);
187 
188    if (source->Swizzle == SWIZZLE_NOOP) {
189       /* no swizzling */
190       COPY_4V(result, src);
191    }
192    else {
193       assert(GET_SWZ(source->Swizzle, 0) <= 3);
194       assert(GET_SWZ(source->Swizzle, 1) <= 3);
195       assert(GET_SWZ(source->Swizzle, 2) <= 3);
196       assert(GET_SWZ(source->Swizzle, 3) <= 3);
197       result[0] = src[GET_SWZ(source->Swizzle, 0)];
198       result[1] = src[GET_SWZ(source->Swizzle, 1)];
199       result[2] = src[GET_SWZ(source->Swizzle, 2)];
200       result[3] = src[GET_SWZ(source->Swizzle, 3)];
201    }
202 
203    if (source->Negate) {
204       assert(source->Negate == NEGATE_XYZW);
205       result[0] = -result[0];
206       result[1] = -result[1];
207       result[2] = -result[2];
208       result[3] = -result[3];
209    }
210 
211 #ifdef NAN_CHECK
212    assert(!IS_INF_OR_NAN(result[0]));
213    assert(!IS_INF_OR_NAN(result[0]));
214    assert(!IS_INF_OR_NAN(result[0]));
215    assert(!IS_INF_OR_NAN(result[0]));
216 #endif
217 }
218 
219 
220 /**
221  * Fetch the derivative with respect to X or Y for the given register.
222  * XXX this currently only works for fragment program input attribs.
223  */
224 static void
fetch_vector4_deriv(const struct prog_src_register * source,const struct gl_program_machine * machine,char xOrY,GLfloat result[4])225 fetch_vector4_deriv(const struct prog_src_register *source,
226                     const struct gl_program_machine *machine,
227                     char xOrY, GLfloat result[4])
228 {
229    if (source->File == PROGRAM_INPUT &&
230        source->Index < (GLint) machine->NumDeriv) {
231       const GLint col = machine->CurElement;
232       const GLfloat w = machine->Attribs[VARYING_SLOT_POS][col][3];
233       const GLfloat invQ = 1.0f / w;
234       GLfloat deriv[4];
235 
236       if (xOrY == 'X') {
237          deriv[0] = machine->DerivX[source->Index][0] * invQ;
238          deriv[1] = machine->DerivX[source->Index][1] * invQ;
239          deriv[2] = machine->DerivX[source->Index][2] * invQ;
240          deriv[3] = machine->DerivX[source->Index][3] * invQ;
241       }
242       else {
243          deriv[0] = machine->DerivY[source->Index][0] * invQ;
244          deriv[1] = machine->DerivY[source->Index][1] * invQ;
245          deriv[2] = machine->DerivY[source->Index][2] * invQ;
246          deriv[3] = machine->DerivY[source->Index][3] * invQ;
247       }
248 
249       result[0] = deriv[GET_SWZ(source->Swizzle, 0)];
250       result[1] = deriv[GET_SWZ(source->Swizzle, 1)];
251       result[2] = deriv[GET_SWZ(source->Swizzle, 2)];
252       result[3] = deriv[GET_SWZ(source->Swizzle, 3)];
253 
254       if (source->Negate) {
255          assert(source->Negate == NEGATE_XYZW);
256          result[0] = -result[0];
257          result[1] = -result[1];
258          result[2] = -result[2];
259          result[3] = -result[3];
260       }
261    }
262    else {
263       ASSIGN_4V(result, 0.0, 0.0, 0.0, 0.0);
264    }
265 }
266 
267 
268 /**
269  * As above, but only return result[0] element.
270  */
271 static void
fetch_vector1(const struct prog_src_register * source,const struct gl_program_machine * machine,GLfloat result[4])272 fetch_vector1(const struct prog_src_register *source,
273               const struct gl_program_machine *machine, GLfloat result[4])
274 {
275    const GLfloat *src = get_src_register_pointer(source, machine);
276 
277    result[0] = src[GET_SWZ(source->Swizzle, 0)];
278 
279    if (source->Negate) {
280       result[0] = -result[0];
281    }
282 }
283 
284 
285 /**
286  * Fetch texel from texture.  Use partial derivatives when possible.
287  */
288 static inline void
fetch_texel(struct gl_context * ctx,const struct gl_program_machine * machine,const struct prog_instruction * inst,const GLfloat texcoord[4],GLfloat lodBias,GLfloat color[4])289 fetch_texel(struct gl_context *ctx,
290             const struct gl_program_machine *machine,
291             const struct prog_instruction *inst,
292             const GLfloat texcoord[4], GLfloat lodBias,
293             GLfloat color[4])
294 {
295    const GLuint unit = machine->Samplers[inst->TexSrcUnit];
296 
297    /* Note: we only have the right derivatives for fragment input attribs.
298     */
299    if (machine->NumDeriv > 0 &&
300        inst->SrcReg[0].File == PROGRAM_INPUT &&
301        inst->SrcReg[0].Index == VARYING_SLOT_TEX0 + inst->TexSrcUnit) {
302       /* simple texture fetch for which we should have derivatives */
303       GLuint attr = inst->SrcReg[0].Index;
304       machine->FetchTexelDeriv(ctx, texcoord,
305                                machine->DerivX[attr],
306                                machine->DerivY[attr],
307                                lodBias, unit, color);
308    }
309    else {
310       machine->FetchTexelLod(ctx, texcoord, lodBias, unit, color);
311    }
312 }
313 
314 
315 /**
316  * Store 4 floats into a register.  Observe the instructions saturate and
317  * set-condition-code flags.
318  */
319 static void
store_vector4(const struct prog_instruction * inst,struct gl_program_machine * machine,const GLfloat value[4])320 store_vector4(const struct prog_instruction *inst,
321               struct gl_program_machine *machine, const GLfloat value[4])
322 {
323    const struct prog_dst_register *dstReg = &(inst->DstReg);
324    const GLboolean clamp = inst->Saturate;
325    GLuint writeMask = dstReg->WriteMask;
326    GLfloat clampedValue[4];
327    GLfloat *dst = get_dst_register_pointer(dstReg, machine);
328 
329 #if 0
330    if (value[0] > 1.0e10 ||
331        IS_INF_OR_NAN(value[0]) ||
332        IS_INF_OR_NAN(value[1]) ||
333        IS_INF_OR_NAN(value[2]) || IS_INF_OR_NAN(value[3]))
334       printf("store %g %g %g %g\n", value[0], value[1], value[2], value[3]);
335 #endif
336 
337    if (clamp) {
338       clampedValue[0] = CLAMP(value[0], 0.0F, 1.0F);
339       clampedValue[1] = CLAMP(value[1], 0.0F, 1.0F);
340       clampedValue[2] = CLAMP(value[2], 0.0F, 1.0F);
341       clampedValue[3] = CLAMP(value[3], 0.0F, 1.0F);
342       value = clampedValue;
343    }
344 
345 #ifdef NAN_CHECK
346    assert(!IS_INF_OR_NAN(value[0]));
347    assert(!IS_INF_OR_NAN(value[0]));
348    assert(!IS_INF_OR_NAN(value[0]));
349    assert(!IS_INF_OR_NAN(value[0]));
350 #endif
351 
352    if (writeMask & WRITEMASK_X)
353       dst[0] = value[0];
354    if (writeMask & WRITEMASK_Y)
355       dst[1] = value[1];
356    if (writeMask & WRITEMASK_Z)
357       dst[2] = value[2];
358    if (writeMask & WRITEMASK_W)
359       dst[3] = value[3];
360 }
361 
362 
363 /**
364  * Execute the given vertex/fragment program.
365  *
366  * \param ctx  rendering context
367  * \param program  the program to execute
368  * \param machine  machine state (must be initialized)
369  * \return GL_TRUE if program completed or GL_FALSE if program executed KIL.
370  */
371 GLboolean
_mesa_execute_program(struct gl_context * ctx,const struct gl_program * program,struct gl_program_machine * machine)372 _mesa_execute_program(struct gl_context * ctx,
373                       const struct gl_program *program,
374                       struct gl_program_machine *machine)
375 {
376    const GLuint numInst = program->arb.NumInstructions;
377    const GLuint maxExec = 65536;
378    GLuint pc, numExec = 0;
379 
380    machine->CurProgram = program;
381 
382    if (DEBUG_PROG) {
383       printf("execute program %u --------------------\n", program->Id);
384    }
385 
386    if (program->Target == GL_VERTEX_PROGRAM_ARB) {
387       machine->EnvParams = ctx->VertexProgram.Parameters;
388    }
389    else {
390       machine->EnvParams = ctx->FragmentProgram.Parameters;
391    }
392 
393    for (pc = 0; pc < numInst; pc++) {
394       const struct prog_instruction *inst = program->arb.Instructions + pc;
395 
396       if (DEBUG_PROG) {
397          _mesa_print_instruction(inst);
398       }
399 
400       switch (inst->Opcode) {
401       case OPCODE_ABS:
402          {
403             GLfloat a[4], result[4];
404             fetch_vector4(&inst->SrcReg[0], machine, a);
405             result[0] = fabsf(a[0]);
406             result[1] = fabsf(a[1]);
407             result[2] = fabsf(a[2]);
408             result[3] = fabsf(a[3]);
409             store_vector4(inst, machine, result);
410          }
411          break;
412       case OPCODE_ADD:
413          {
414             GLfloat a[4], b[4], result[4];
415             fetch_vector4(&inst->SrcReg[0], machine, a);
416             fetch_vector4(&inst->SrcReg[1], machine, b);
417             result[0] = a[0] + b[0];
418             result[1] = a[1] + b[1];
419             result[2] = a[2] + b[2];
420             result[3] = a[3] + b[3];
421             store_vector4(inst, machine, result);
422             if (DEBUG_PROG) {
423                printf("ADD (%g %g %g %g) = (%g %g %g %g) + (%g %g %g %g)\n",
424                       result[0], result[1], result[2], result[3],
425                       a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
426             }
427          }
428          break;
429       case OPCODE_ARL:
430          {
431             GLfloat t[4];
432             fetch_vector4(&inst->SrcReg[0], machine, t);
433             machine->AddressReg[0][0] = IFLOOR(t[0]);
434             if (DEBUG_PROG) {
435                printf("ARL %d\n", machine->AddressReg[0][0]);
436             }
437          }
438          break;
439       case OPCODE_BGNLOOP:
440          /* no-op */
441          assert(program->arb.Instructions[inst->BranchTarget].Opcode
442                 == OPCODE_ENDLOOP);
443          break;
444       case OPCODE_ENDLOOP:
445          /* subtract 1 here since pc is incremented by for(pc) loop */
446          assert(program->arb.Instructions[inst->BranchTarget].Opcode
447                 == OPCODE_BGNLOOP);
448          pc = inst->BranchTarget - 1;   /* go to matching BNGLOOP */
449          break;
450       case OPCODE_BGNSUB:      /* begin subroutine */
451          break;
452       case OPCODE_ENDSUB:      /* end subroutine */
453          break;
454       case OPCODE_BRK:         /* break out of loop (conditional) */
455          assert(program->arb.Instructions[inst->BranchTarget].Opcode
456                 == OPCODE_ENDLOOP);
457          /* break out of loop */
458          /* pc++ at end of for-loop will put us after the ENDLOOP inst */
459          pc = inst->BranchTarget;
460          break;
461       case OPCODE_CONT:        /* continue loop (conditional) */
462          assert(program->arb.Instructions[inst->BranchTarget].Opcode
463                 == OPCODE_ENDLOOP);
464          /* continue at ENDLOOP */
465          /* Subtract 1 here since we'll do pc++ at end of for-loop */
466          pc = inst->BranchTarget - 1;
467          break;
468       case OPCODE_CAL:         /* Call subroutine (conditional) */
469          /* call the subroutine */
470          if (machine->StackDepth >= MAX_PROGRAM_CALL_DEPTH) {
471             return GL_TRUE;  /* Per GL_NV_vertex_program2 spec */
472          }
473          machine->CallStack[machine->StackDepth++] = pc + 1; /* next inst */
474          /* Subtract 1 here since we'll do pc++ at end of for-loop */
475          pc = inst->BranchTarget - 1;
476          break;
477       case OPCODE_CMP:
478          {
479             GLfloat a[4], b[4], c[4], result[4];
480             fetch_vector4(&inst->SrcReg[0], machine, a);
481             fetch_vector4(&inst->SrcReg[1], machine, b);
482             fetch_vector4(&inst->SrcReg[2], machine, c);
483             result[0] = a[0] < 0.0F ? b[0] : c[0];
484             result[1] = a[1] < 0.0F ? b[1] : c[1];
485             result[2] = a[2] < 0.0F ? b[2] : c[2];
486             result[3] = a[3] < 0.0F ? b[3] : c[3];
487             store_vector4(inst, machine, result);
488             if (DEBUG_PROG) {
489                printf("CMP (%g %g %g %g) = (%g %g %g %g) < 0 ? (%g %g %g %g) : (%g %g %g %g)\n",
490                       result[0], result[1], result[2], result[3],
491                       a[0], a[1], a[2], a[3],
492                       b[0], b[1], b[2], b[3],
493                       c[0], c[1], c[2], c[3]);
494             }
495          }
496          break;
497       case OPCODE_COS:
498          {
499             GLfloat a[4], result[4];
500             fetch_vector1(&inst->SrcReg[0], machine, a);
501             result[0] = result[1] = result[2] = result[3]
502                = cosf(a[0]);
503             store_vector4(inst, machine, result);
504          }
505          break;
506       case OPCODE_DDX:         /* Partial derivative with respect to X */
507          {
508             GLfloat result[4];
509             fetch_vector4_deriv(&inst->SrcReg[0], machine, 'X', result);
510             store_vector4(inst, machine, result);
511          }
512          break;
513       case OPCODE_DDY:         /* Partial derivative with respect to Y */
514          {
515             GLfloat result[4];
516             fetch_vector4_deriv(&inst->SrcReg[0], machine, 'Y', result);
517             store_vector4(inst, machine, result);
518          }
519          break;
520       case OPCODE_DP2:
521          {
522             GLfloat a[4], b[4], result[4];
523             fetch_vector4(&inst->SrcReg[0], machine, a);
524             fetch_vector4(&inst->SrcReg[1], machine, b);
525             result[0] = result[1] = result[2] = result[3] = DOT2(a, b);
526             store_vector4(inst, machine, result);
527             if (DEBUG_PROG) {
528                printf("DP2 %g = (%g %g) . (%g %g)\n",
529                       result[0], a[0], a[1], b[0], b[1]);
530             }
531          }
532          break;
533       case OPCODE_DP3:
534          {
535             GLfloat a[4], b[4], result[4];
536             fetch_vector4(&inst->SrcReg[0], machine, a);
537             fetch_vector4(&inst->SrcReg[1], machine, b);
538             result[0] = result[1] = result[2] = result[3] = DOT3(a, b);
539             store_vector4(inst, machine, result);
540             if (DEBUG_PROG) {
541                printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
542                       result[0], a[0], a[1], a[2], b[0], b[1], b[2]);
543             }
544          }
545          break;
546       case OPCODE_DP4:
547          {
548             GLfloat a[4], b[4], result[4];
549             fetch_vector4(&inst->SrcReg[0], machine, a);
550             fetch_vector4(&inst->SrcReg[1], machine, b);
551             result[0] = result[1] = result[2] = result[3] = DOT4(a, b);
552             store_vector4(inst, machine, result);
553             if (DEBUG_PROG) {
554                printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
555                       result[0], a[0], a[1], a[2], a[3],
556                       b[0], b[1], b[2], b[3]);
557             }
558          }
559          break;
560       case OPCODE_DPH:
561          {
562             GLfloat a[4], b[4], result[4];
563             fetch_vector4(&inst->SrcReg[0], machine, a);
564             fetch_vector4(&inst->SrcReg[1], machine, b);
565             result[0] = result[1] = result[2] = result[3] = DOT3(a, b) + b[3];
566             store_vector4(inst, machine, result);
567          }
568          break;
569       case OPCODE_DST:         /* Distance vector */
570          {
571             GLfloat a[4], b[4], result[4];
572             fetch_vector4(&inst->SrcReg[0], machine, a);
573             fetch_vector4(&inst->SrcReg[1], machine, b);
574             result[0] = 1.0F;
575             result[1] = a[1] * b[1];
576             result[2] = a[2];
577             result[3] = b[3];
578             store_vector4(inst, machine, result);
579          }
580          break;
581       case OPCODE_EXP:
582          {
583             GLfloat t[4], q[4], floor_t0;
584             fetch_vector1(&inst->SrcReg[0], machine, t);
585             floor_t0 = floorf(t[0]);
586             if (floor_t0 > FLT_MAX_EXP) {
587                SET_POS_INFINITY(q[0]);
588                SET_POS_INFINITY(q[2]);
589             }
590             else if (floor_t0 < FLT_MIN_EXP) {
591                q[0] = 0.0F;
592                q[2] = 0.0F;
593             }
594             else {
595                q[0] = ldexpf(1.0, (int) floor_t0);
596                /* Note: GL_NV_vertex_program expects
597                 * result.z = result.x * APPX(result.y)
598                 * We do what the ARB extension says.
599                 */
600                q[2] = exp2f(t[0]);
601             }
602             q[1] = t[0] - floor_t0;
603             q[3] = 1.0F;
604             store_vector4( inst, machine, q );
605          }
606          break;
607       case OPCODE_EX2:         /* Exponential base 2 */
608          {
609             GLfloat a[4], result[4], val;
610             fetch_vector1(&inst->SrcReg[0], machine, a);
611             val = exp2f(a[0]);
612             /*
613             if (IS_INF_OR_NAN(val))
614                val = 1.0e10;
615             */
616             result[0] = result[1] = result[2] = result[3] = val;
617             store_vector4(inst, machine, result);
618          }
619          break;
620       case OPCODE_FLR:
621          {
622             GLfloat a[4], result[4];
623             fetch_vector4(&inst->SrcReg[0], machine, a);
624             result[0] = floorf(a[0]);
625             result[1] = floorf(a[1]);
626             result[2] = floorf(a[2]);
627             result[3] = floorf(a[3]);
628             store_vector4(inst, machine, result);
629          }
630          break;
631       case OPCODE_FRC:
632          {
633             GLfloat a[4], result[4];
634             fetch_vector4(&inst->SrcReg[0], machine, a);
635             result[0] = a[0] - floorf(a[0]);
636             result[1] = a[1] - floorf(a[1]);
637             result[2] = a[2] - floorf(a[2]);
638             result[3] = a[3] - floorf(a[3]);
639             store_vector4(inst, machine, result);
640          }
641          break;
642       case OPCODE_IF:
643          {
644             GLboolean cond;
645             assert(program->arb.Instructions[inst->BranchTarget].Opcode
646                    == OPCODE_ELSE ||
647                    program->arb.Instructions[inst->BranchTarget].Opcode
648                    == OPCODE_ENDIF);
649             /* eval condition */
650             GLfloat a[4];
651             fetch_vector1(&inst->SrcReg[0], machine, a);
652             cond = (a[0] != 0.0F);
653             if (DEBUG_PROG) {
654                printf("IF: %d\n", cond);
655             }
656             /* do if/else */
657             if (cond) {
658                /* do if-clause (just continue execution) */
659             }
660             else {
661                /* go to the instruction after ELSE or ENDIF */
662                assert(inst->BranchTarget >= 0);
663                pc = inst->BranchTarget;
664             }
665          }
666          break;
667       case OPCODE_ELSE:
668          /* goto ENDIF */
669          assert(program->arb.Instructions[inst->BranchTarget].Opcode
670                 == OPCODE_ENDIF);
671          assert(inst->BranchTarget >= 0);
672          pc = inst->BranchTarget;
673          break;
674       case OPCODE_ENDIF:
675          /* nothing */
676          break;
677       case OPCODE_KIL:         /* ARB_f_p only */
678          {
679             GLfloat a[4];
680             fetch_vector4(&inst->SrcReg[0], machine, a);
681             if (DEBUG_PROG) {
682                printf("KIL if (%g %g %g %g) <= 0.0\n",
683                       a[0], a[1], a[2], a[3]);
684             }
685 
686             if (a[0] < 0.0F || a[1] < 0.0F || a[2] < 0.0F || a[3] < 0.0F) {
687                return GL_FALSE;
688             }
689          }
690          break;
691       case OPCODE_LG2:         /* log base 2 */
692          {
693             GLfloat a[4], result[4], val;
694             fetch_vector1(&inst->SrcReg[0], machine, a);
695 	    /* The fast LOG2 macro doesn't meet the precision requirements.
696 	     */
697             if (a[0] == 0.0F) {
698                val = -FLT_MAX;
699             }
700             else {
701                val = logf(a[0]) * 1.442695F;
702             }
703             result[0] = result[1] = result[2] = result[3] = val;
704             store_vector4(inst, machine, result);
705          }
706          break;
707       case OPCODE_LIT:
708          {
709             const GLfloat epsilon = 1.0F / 256.0F;      /* from NV VP spec */
710             GLfloat a[4], result[4];
711             fetch_vector4(&inst->SrcReg[0], machine, a);
712             a[0] = MAX2(a[0], 0.0F);
713             a[1] = MAX2(a[1], 0.0F);
714             /* XXX ARB version clamps a[3], NV version doesn't */
715             a[3] = CLAMP(a[3], -(128.0F - epsilon), (128.0F - epsilon));
716             result[0] = 1.0F;
717             result[1] = a[0];
718             /* XXX we could probably just use pow() here */
719             if (a[0] > 0.0F) {
720                if (a[1] == 0.0F && a[3] == 0.0F)
721                   result[2] = 1.0F;
722                else
723                   result[2] = powf(a[1], a[3]);
724             }
725             else {
726                result[2] = 0.0F;
727             }
728             result[3] = 1.0F;
729             store_vector4(inst, machine, result);
730             if (DEBUG_PROG) {
731                printf("LIT (%g %g %g %g) : (%g %g %g %g)\n",
732                       result[0], result[1], result[2], result[3],
733                       a[0], a[1], a[2], a[3]);
734             }
735          }
736          break;
737       case OPCODE_LOG:
738          {
739             GLfloat t[4], q[4], abs_t0;
740             fetch_vector1(&inst->SrcReg[0], machine, t);
741             abs_t0 = fabsf(t[0]);
742             if (abs_t0 != 0.0F) {
743                if (IS_INF_OR_NAN(abs_t0))
744                {
745                   SET_POS_INFINITY(q[0]);
746                   q[1] = 1.0F;
747                   SET_POS_INFINITY(q[2]);
748                }
749                else {
750                   int exponent;
751                   GLfloat mantissa = frexpf(t[0], &exponent);
752                   q[0] = (GLfloat) (exponent - 1);
753                   q[1] = 2.0F * mantissa; /* map [.5, 1) -> [1, 2) */
754 
755 		  /* The fast LOG2 macro doesn't meet the precision
756 		   * requirements.
757 		   */
758                   q[2] = logf(t[0]) * 1.442695F;
759                }
760             }
761             else {
762                SET_NEG_INFINITY(q[0]);
763                q[1] = 1.0F;
764                SET_NEG_INFINITY(q[2]);
765             }
766             q[3] = 1.0;
767             store_vector4(inst, machine, q);
768          }
769          break;
770       case OPCODE_LRP:
771          {
772             GLfloat a[4], b[4], c[4], result[4];
773             fetch_vector4(&inst->SrcReg[0], machine, a);
774             fetch_vector4(&inst->SrcReg[1], machine, b);
775             fetch_vector4(&inst->SrcReg[2], machine, c);
776             result[0] = a[0] * b[0] + (1.0F - a[0]) * c[0];
777             result[1] = a[1] * b[1] + (1.0F - a[1]) * c[1];
778             result[2] = a[2] * b[2] + (1.0F - a[2]) * c[2];
779             result[3] = a[3] * b[3] + (1.0F - a[3]) * c[3];
780             store_vector4(inst, machine, result);
781             if (DEBUG_PROG) {
782                printf("LRP (%g %g %g %g) = (%g %g %g %g), "
783                       "(%g %g %g %g), (%g %g %g %g)\n",
784                       result[0], result[1], result[2], result[3],
785                       a[0], a[1], a[2], a[3],
786                       b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
787             }
788          }
789          break;
790       case OPCODE_MAD:
791          {
792             GLfloat a[4], b[4], c[4], result[4];
793             fetch_vector4(&inst->SrcReg[0], machine, a);
794             fetch_vector4(&inst->SrcReg[1], machine, b);
795             fetch_vector4(&inst->SrcReg[2], machine, c);
796             result[0] = a[0] * b[0] + c[0];
797             result[1] = a[1] * b[1] + c[1];
798             result[2] = a[2] * b[2] + c[2];
799             result[3] = a[3] * b[3] + c[3];
800             store_vector4(inst, machine, result);
801             if (DEBUG_PROG) {
802                printf("MAD (%g %g %g %g) = (%g %g %g %g) * "
803                       "(%g %g %g %g) + (%g %g %g %g)\n",
804                       result[0], result[1], result[2], result[3],
805                       a[0], a[1], a[2], a[3],
806                       b[0], b[1], b[2], b[3], c[0], c[1], c[2], c[3]);
807             }
808          }
809          break;
810       case OPCODE_MAX:
811          {
812             GLfloat a[4], b[4], result[4];
813             fetch_vector4(&inst->SrcReg[0], machine, a);
814             fetch_vector4(&inst->SrcReg[1], machine, b);
815             result[0] = MAX2(a[0], b[0]);
816             result[1] = MAX2(a[1], b[1]);
817             result[2] = MAX2(a[2], b[2]);
818             result[3] = MAX2(a[3], b[3]);
819             store_vector4(inst, machine, result);
820             if (DEBUG_PROG) {
821                printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
822                       result[0], result[1], result[2], result[3],
823                       a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
824             }
825          }
826          break;
827       case OPCODE_MIN:
828          {
829             GLfloat a[4], b[4], result[4];
830             fetch_vector4(&inst->SrcReg[0], machine, a);
831             fetch_vector4(&inst->SrcReg[1], machine, b);
832             result[0] = MIN2(a[0], b[0]);
833             result[1] = MIN2(a[1], b[1]);
834             result[2] = MIN2(a[2], b[2]);
835             result[3] = MIN2(a[3], b[3]);
836             store_vector4(inst, machine, result);
837          }
838          break;
839       case OPCODE_MOV:
840          {
841             GLfloat result[4];
842             fetch_vector4(&inst->SrcReg[0], machine, result);
843             store_vector4(inst, machine, result);
844             if (DEBUG_PROG) {
845                printf("MOV (%g %g %g %g)\n",
846                       result[0], result[1], result[2], result[3]);
847             }
848          }
849          break;
850       case OPCODE_MUL:
851          {
852             GLfloat a[4], b[4], result[4];
853             fetch_vector4(&inst->SrcReg[0], machine, a);
854             fetch_vector4(&inst->SrcReg[1], machine, b);
855             result[0] = a[0] * b[0];
856             result[1] = a[1] * b[1];
857             result[2] = a[2] * b[2];
858             result[3] = a[3] * b[3];
859             store_vector4(inst, machine, result);
860             if (DEBUG_PROG) {
861                printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
862                       result[0], result[1], result[2], result[3],
863                       a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
864             }
865          }
866          break;
867       case OPCODE_NOISE1:
868          {
869             GLfloat a[4], result[4];
870             fetch_vector1(&inst->SrcReg[0], machine, a);
871             result[0] =
872                result[1] =
873                result[2] =
874                result[3] = _mesa_noise1(a[0]);
875             store_vector4(inst, machine, result);
876          }
877          break;
878       case OPCODE_NOISE2:
879          {
880             GLfloat a[4], result[4];
881             fetch_vector4(&inst->SrcReg[0], machine, a);
882             result[0] =
883                result[1] =
884                result[2] = result[3] = _mesa_noise2(a[0], a[1]);
885             store_vector4(inst, machine, result);
886          }
887          break;
888       case OPCODE_NOISE3:
889          {
890             GLfloat a[4], result[4];
891             fetch_vector4(&inst->SrcReg[0], machine, a);
892             result[0] =
893                result[1] =
894                result[2] =
895                result[3] = _mesa_noise3(a[0], a[1], a[2]);
896             store_vector4(inst, machine, result);
897          }
898          break;
899       case OPCODE_NOISE4:
900          {
901             GLfloat a[4], result[4];
902             fetch_vector4(&inst->SrcReg[0], machine, a);
903             result[0] =
904                result[1] =
905                result[2] =
906                result[3] = _mesa_noise4(a[0], a[1], a[2], a[3]);
907             store_vector4(inst, machine, result);
908          }
909          break;
910       case OPCODE_NOP:
911          break;
912       case OPCODE_POW:
913          {
914             GLfloat a[4], b[4], result[4];
915             fetch_vector1(&inst->SrcReg[0], machine, a);
916             fetch_vector1(&inst->SrcReg[1], machine, b);
917             result[0] = result[1] = result[2] = result[3]
918                = powf(a[0], b[0]);
919             store_vector4(inst, machine, result);
920          }
921          break;
922 
923       case OPCODE_RCP:
924          {
925             GLfloat a[4], result[4];
926             fetch_vector1(&inst->SrcReg[0], machine, a);
927             if (DEBUG_PROG) {
928                if (a[0] == 0)
929                   printf("RCP(0)\n");
930                else if (IS_INF_OR_NAN(a[0]))
931                   printf("RCP(inf)\n");
932             }
933             result[0] = result[1] = result[2] = result[3] = 1.0F / a[0];
934             store_vector4(inst, machine, result);
935          }
936          break;
937       case OPCODE_RET:         /* return from subroutine (conditional) */
938          if (machine->StackDepth == 0) {
939             return GL_TRUE;  /* Per GL_NV_vertex_program2 spec */
940          }
941          /* subtract one because of pc++ in the for loop */
942          pc = machine->CallStack[--machine->StackDepth] - 1;
943          break;
944       case OPCODE_RSQ:         /* 1 / sqrt() */
945          {
946             GLfloat a[4], result[4];
947             fetch_vector1(&inst->SrcReg[0], machine, a);
948             a[0] = fabsf(a[0]);
949             result[0] = result[1] = result[2] = result[3] = 1.0f / sqrtf(a[0]);
950             store_vector4(inst, machine, result);
951             if (DEBUG_PROG) {
952                printf("RSQ %g = 1/sqrt(|%g|)\n", result[0], a[0]);
953             }
954          }
955          break;
956       case OPCODE_SCS:         /* sine and cos */
957          {
958             GLfloat a[4], result[4];
959             fetch_vector1(&inst->SrcReg[0], machine, a);
960             result[0] = cosf(a[0]);
961             result[1] = sinf(a[0]);
962             result[2] = 0.0F;    /* undefined! */
963             result[3] = 0.0F;    /* undefined! */
964             store_vector4(inst, machine, result);
965          }
966          break;
967       case OPCODE_SGE:         /* set on greater or equal */
968          {
969             GLfloat a[4], b[4], result[4];
970             fetch_vector4(&inst->SrcReg[0], machine, a);
971             fetch_vector4(&inst->SrcReg[1], machine, b);
972             result[0] = (a[0] >= b[0]) ? 1.0F : 0.0F;
973             result[1] = (a[1] >= b[1]) ? 1.0F : 0.0F;
974             result[2] = (a[2] >= b[2]) ? 1.0F : 0.0F;
975             result[3] = (a[3] >= b[3]) ? 1.0F : 0.0F;
976             store_vector4(inst, machine, result);
977             if (DEBUG_PROG) {
978                printf("SGE (%g %g %g %g) = (%g %g %g %g) >= (%g %g %g %g)\n",
979                       result[0], result[1], result[2], result[3],
980                       a[0], a[1], a[2], a[3],
981                       b[0], b[1], b[2], b[3]);
982             }
983          }
984          break;
985       case OPCODE_SIN:
986          {
987             GLfloat a[4], result[4];
988             fetch_vector1(&inst->SrcReg[0], machine, a);
989             result[0] = result[1] = result[2] = result[3]
990                = sinf(a[0]);
991             store_vector4(inst, machine, result);
992          }
993          break;
994       case OPCODE_SLT:         /* set on less */
995          {
996             GLfloat a[4], b[4], result[4];
997             fetch_vector4(&inst->SrcReg[0], machine, a);
998             fetch_vector4(&inst->SrcReg[1], machine, b);
999             result[0] = (a[0] < b[0]) ? 1.0F : 0.0F;
1000             result[1] = (a[1] < b[1]) ? 1.0F : 0.0F;
1001             result[2] = (a[2] < b[2]) ? 1.0F : 0.0F;
1002             result[3] = (a[3] < b[3]) ? 1.0F : 0.0F;
1003             store_vector4(inst, machine, result);
1004             if (DEBUG_PROG) {
1005                printf("SLT (%g %g %g %g) = (%g %g %g %g) < (%g %g %g %g)\n",
1006                       result[0], result[1], result[2], result[3],
1007                       a[0], a[1], a[2], a[3],
1008                       b[0], b[1], b[2], b[3]);
1009             }
1010          }
1011          break;
1012       case OPCODE_SSG:         /* set sign (-1, 0 or +1) */
1013          {
1014             GLfloat a[4], result[4];
1015             fetch_vector4(&inst->SrcReg[0], machine, a);
1016             result[0] = (GLfloat) ((a[0] > 0.0F) - (a[0] < 0.0F));
1017             result[1] = (GLfloat) ((a[1] > 0.0F) - (a[1] < 0.0F));
1018             result[2] = (GLfloat) ((a[2] > 0.0F) - (a[2] < 0.0F));
1019             result[3] = (GLfloat) ((a[3] > 0.0F) - (a[3] < 0.0F));
1020             store_vector4(inst, machine, result);
1021          }
1022          break;
1023       case OPCODE_SUB:
1024          {
1025             GLfloat a[4], b[4], result[4];
1026             fetch_vector4(&inst->SrcReg[0], machine, a);
1027             fetch_vector4(&inst->SrcReg[1], machine, b);
1028             result[0] = a[0] - b[0];
1029             result[1] = a[1] - b[1];
1030             result[2] = a[2] - b[2];
1031             result[3] = a[3] - b[3];
1032             store_vector4(inst, machine, result);
1033             if (DEBUG_PROG) {
1034                printf("SUB (%g %g %g %g) = (%g %g %g %g) - (%g %g %g %g)\n",
1035                       result[0], result[1], result[2], result[3],
1036                       a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
1037             }
1038          }
1039          break;
1040       case OPCODE_SWZ:         /* extended swizzle */
1041          {
1042             const struct prog_src_register *source = &inst->SrcReg[0];
1043             const GLfloat *src = get_src_register_pointer(source, machine);
1044             GLfloat result[4];
1045             GLuint i;
1046             for (i = 0; i < 4; i++) {
1047                const GLuint swz = GET_SWZ(source->Swizzle, i);
1048                if (swz == SWIZZLE_ZERO)
1049                   result[i] = 0.0;
1050                else if (swz == SWIZZLE_ONE)
1051                   result[i] = 1.0;
1052                else {
1053                   assert(swz <= 3);
1054                   result[i] = src[swz];
1055                }
1056                if (source->Negate & (1 << i))
1057                   result[i] = -result[i];
1058             }
1059             store_vector4(inst, machine, result);
1060          }
1061          break;
1062       case OPCODE_TEX:         /* Both ARB and NV frag prog */
1063          /* Simple texel lookup */
1064          {
1065             GLfloat texcoord[4], color[4];
1066             fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1067 
1068             /* For TEX, texcoord.Q should not be used and its value should not
1069              * matter (at most, we pass coord.xyz to texture3D() in GLSL).
1070              * Set Q=1 so that FetchTexelDeriv() doesn't get a garbage value
1071              * which is effectively what happens when the texcoord swizzle
1072              * is .xyzz
1073              */
1074             texcoord[3] = 1.0f;
1075 
1076             fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1077 
1078             if (DEBUG_PROG) {
1079                printf("TEX (%g, %g, %g, %g) = texture[%d][%g, %g, %g, %g]\n",
1080                       color[0], color[1], color[2], color[3],
1081                       inst->TexSrcUnit,
1082                       texcoord[0], texcoord[1], texcoord[2], texcoord[3]);
1083             }
1084             store_vector4(inst, machine, color);
1085          }
1086          break;
1087       case OPCODE_TXB:         /* GL_ARB_fragment_program only */
1088          /* Texel lookup with LOD bias */
1089          {
1090             GLfloat texcoord[4], color[4], lodBias;
1091 
1092             fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1093 
1094             /* texcoord[3] is the bias to add to lambda */
1095             lodBias = texcoord[3];
1096 
1097             fetch_texel(ctx, machine, inst, texcoord, lodBias, color);
1098 
1099             if (DEBUG_PROG) {
1100                printf("TXB (%g, %g, %g, %g) = texture[%d][%g %g %g %g]"
1101                       "  bias %g\n",
1102                       color[0], color[1], color[2], color[3],
1103                       inst->TexSrcUnit,
1104                       texcoord[0],
1105                       texcoord[1],
1106                       texcoord[2],
1107                       texcoord[3],
1108                       lodBias);
1109             }
1110 
1111             store_vector4(inst, machine, color);
1112          }
1113          break;
1114       case OPCODE_TXD:
1115          /* Texture lookup w/ partial derivatives for LOD */
1116          {
1117             GLfloat texcoord[4], dtdx[4], dtdy[4], color[4];
1118             fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1119             fetch_vector4(&inst->SrcReg[1], machine, dtdx);
1120             fetch_vector4(&inst->SrcReg[2], machine, dtdy);
1121             machine->FetchTexelDeriv(ctx, texcoord, dtdx, dtdy,
1122                                      0.0, /* lodBias */
1123                                      inst->TexSrcUnit, color);
1124             store_vector4(inst, machine, color);
1125          }
1126          break;
1127       case OPCODE_TXL:
1128          /* Texel lookup with explicit LOD */
1129          {
1130             GLfloat texcoord[4], color[4], lod;
1131 
1132             fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1133 
1134             /* texcoord[3] is the LOD */
1135             lod = texcoord[3];
1136 
1137 	    machine->FetchTexelLod(ctx, texcoord, lod,
1138 				   machine->Samplers[inst->TexSrcUnit], color);
1139 
1140             store_vector4(inst, machine, color);
1141          }
1142          break;
1143       case OPCODE_TXP:         /* GL_ARB_fragment_program only */
1144          /* Texture lookup w/ projective divide */
1145          {
1146             GLfloat texcoord[4], color[4];
1147 
1148             fetch_vector4(&inst->SrcReg[0], machine, texcoord);
1149             /* Not so sure about this test - if texcoord[3] is
1150              * zero, we'd probably be fine except for an assert in
1151              * IROUND_POS() which gets triggered by the inf values created.
1152              */
1153             if (texcoord[3] != 0.0F) {
1154                texcoord[0] /= texcoord[3];
1155                texcoord[1] /= texcoord[3];
1156                texcoord[2] /= texcoord[3];
1157             }
1158 
1159             fetch_texel(ctx, machine, inst, texcoord, 0.0, color);
1160 
1161             store_vector4(inst, machine, color);
1162          }
1163          break;
1164       case OPCODE_TRUNC:       /* truncate toward zero */
1165          {
1166             GLfloat a[4], result[4];
1167             fetch_vector4(&inst->SrcReg[0], machine, a);
1168             result[0] = (GLfloat) (GLint) a[0];
1169             result[1] = (GLfloat) (GLint) a[1];
1170             result[2] = (GLfloat) (GLint) a[2];
1171             result[3] = (GLfloat) (GLint) a[3];
1172             store_vector4(inst, machine, result);
1173          }
1174          break;
1175       case OPCODE_XPD:         /* cross product */
1176          {
1177             GLfloat a[4], b[4], result[4];
1178             fetch_vector4(&inst->SrcReg[0], machine, a);
1179             fetch_vector4(&inst->SrcReg[1], machine, b);
1180             result[0] = a[1] * b[2] - a[2] * b[1];
1181             result[1] = a[2] * b[0] - a[0] * b[2];
1182             result[2] = a[0] * b[1] - a[1] * b[0];
1183             result[3] = 1.0;
1184             store_vector4(inst, machine, result);
1185             if (DEBUG_PROG) {
1186                printf("XPD (%g %g %g %g) = (%g %g %g) X (%g %g %g)\n",
1187                       result[0], result[1], result[2], result[3],
1188                       a[0], a[1], a[2], b[0], b[1], b[2]);
1189             }
1190          }
1191          break;
1192       case OPCODE_END:
1193          return GL_TRUE;
1194       default:
1195          _mesa_problem(ctx, "Bad opcode %d in _mesa_execute_program",
1196                        inst->Opcode);
1197          return GL_TRUE;        /* return value doesn't matter */
1198       }
1199 
1200       numExec++;
1201       if (numExec > maxExec) {
1202 	 static GLboolean reported = GL_FALSE;
1203 	 if (!reported) {
1204 	    _mesa_problem(ctx, "Infinite loop detected in fragment program");
1205 	    reported = GL_TRUE;
1206 	 }
1207          return GL_TRUE;
1208       }
1209 
1210    } /* for pc */
1211 
1212    return GL_TRUE;
1213 }
1214