projects / mesa.git / blob
? search:


e55fe5c851a01c611342504b45e1ccb4d6bb3b2c
[mesa.git] / src / mesa / swrast / s_nvfragprog.c
1 /*
2 * Mesa 3-D graphics library
3 * Version: 6.5.2
4 *
5 * Copyright (C) 1999-2006 Brian Paul All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
21 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
22 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 */
24
25 /*
26 * Regarding GL_NV_fragment_program:
27 *
28 * Portions of this software may use or implement intellectual
29 * property owned and licensed by NVIDIA Corporation. NVIDIA disclaims
30 * any and all warranties with respect to such intellectual property,
31 * including any use thereof or modifications thereto.
32 */
33
34 #include "glheader.h"
35 #include "colormac.h"
36 #include "context.h"
37 #include "program_instruction.h"
38 #include "program.h"
39
40 #include "s_nvfragprog.h"
41 #include "s_span.h"
42
43
44 /* if 1, print some debugging info */
45 #define DEBUG_FRAG 0
46
47
48 /**
49 * Virtual machine state used during execution of a fragment programs.
50 */
51 struct fp_machine
52 {
53 GLfloat Temporaries[MAX_NV_FRAGMENT_PROGRAM_TEMPS][4];
54 GLfloat Inputs[MAX_NV_FRAGMENT_PROGRAM_INPUTS][4];
55 GLfloat Outputs[MAX_NV_FRAGMENT_PROGRAM_OUTPUTS][4];
56 GLuint CondCodes[4]; /**< COND_* value for x/y/z/w */
57 };
58
59
60 #if FEATURE_MESA_program_debug
61 static struct fp_machine *CurrentMachine = NULL;
62
63 /**
64 * For GL_MESA_program_debug.
65 * Return current value (4*GLfloat) of a fragment program register.
66 * Called via ctx->Driver.GetFragmentProgramRegister().
67 */
68 void
69 _swrast_get_program_register(GLcontext *ctx, enum register_file file,
70 GLuint index, GLfloat val[4])
71 {
72 if (CurrentMachine) {
73 switch (file) {
74 case PROGRAM_INPUT:
75 COPY_4V(val, CurrentMachine->Inputs[index]);
76 break;
77 case PROGRAM_OUTPUT:
78 COPY_4V(val, CurrentMachine->Outputs[index]);
79 break;
80 case PROGRAM_TEMPORARY:
81 COPY_4V(val, CurrentMachine->Temporaries[index]);
82 break;
83 default:
84 _mesa_problem(NULL,
85 "bad register file in _swrast_get_program_register");
86 }
87 }
88 }
89 #endif /* FEATURE_MESA_program_debug */
90
91
92 /**
93 * Fetch a texel.
94 */
95 static void
96 fetch_texel( GLcontext *ctx, const GLfloat texcoord[4], GLfloat lambda,
97 GLuint unit, GLfloat color[4] )
98 {
99 GLchan rgba[4];
100 SWcontext *swrast = SWRAST_CONTEXT(ctx);
101
102 /* XXX use a float-valued TextureSample routine here!!! */
103 swrast->TextureSample[unit](ctx, ctx->Texture.Unit[unit]._Current,
104 1, (const GLfloat (*)[4]) texcoord,
105 &lambda, &rgba);
106 color[0] = CHAN_TO_FLOAT(rgba[0]);
107 color[1] = CHAN_TO_FLOAT(rgba[1]);
108 color[2] = CHAN_TO_FLOAT(rgba[2]);
109 color[3] = CHAN_TO_FLOAT(rgba[3]);
110 }
111
112
113 /**
114 * Fetch a texel with the given partial derivatives to compute a level
115 * of detail in the mipmap.
116 */
117 static void
118 fetch_texel_deriv( GLcontext *ctx, const GLfloat texcoord[4],
119 const GLfloat texdx[4], const GLfloat texdy[4],
120 GLuint unit, GLfloat color[4] )
121 {
122 SWcontext *swrast = SWRAST_CONTEXT(ctx);
123 const struct gl_texture_object *texObj = ctx->Texture.Unit[unit]._Current;
124 const struct gl_texture_image *texImg = texObj->Image[0][texObj->BaseLevel];
125 const GLfloat texW = (GLfloat) texImg->WidthScale;
126 const GLfloat texH = (GLfloat) texImg->HeightScale;
127 GLchan rgba[4];
128
129 GLfloat lambda = _swrast_compute_lambda(texdx[0], texdy[0], /* ds/dx, ds/dy */
130 texdx[1], texdy[1], /* dt/dx, dt/dy */
131 texdx[3], texdy[2], /* dq/dx, dq/dy */
132 texW, texH,
133 texcoord[0], texcoord[1], texcoord[3],
134 1.0F / texcoord[3]);
135
136 swrast->TextureSample[unit](ctx, ctx->Texture.Unit[unit]._Current,
137 1, (const GLfloat (*)[4]) texcoord,
138 &lambda, &rgba);
139 color[0] = CHAN_TO_FLOAT(rgba[0]);
140 color[1] = CHAN_TO_FLOAT(rgba[1]);
141 color[2] = CHAN_TO_FLOAT(rgba[2]);
142 color[3] = CHAN_TO_FLOAT(rgba[3]);
143 }
144
145
146 /**
147 * Return a pointer to the 4-element float vector specified by the given
148 * source register.
149 */
150 static INLINE const GLfloat *
151 get_register_pointer( GLcontext *ctx,
152 const struct prog_src_register *source,
153 const struct fp_machine *machine,
154 const struct gl_fragment_program *program )
155 {
156 switch (source->File) {
157 case PROGRAM_TEMPORARY:
158 ASSERT(source->Index < MAX_NV_FRAGMENT_PROGRAM_TEMPS);
159 return machine->Temporaries[source->Index];
160 case PROGRAM_INPUT:
161 ASSERT(source->Index < MAX_NV_FRAGMENT_PROGRAM_INPUTS);
162 return machine->Inputs[source->Index];
163 case PROGRAM_OUTPUT:
164 /* This is only for PRINT */
165 ASSERT(source->Index < MAX_NV_FRAGMENT_PROGRAM_OUTPUTS);
166 return machine->Outputs[source->Index];
167 case PROGRAM_LOCAL_PARAM:
168 ASSERT(source->Index < MAX_PROGRAM_LOCAL_PARAMS);
169 return program->Base.LocalParams[source->Index];
170 case PROGRAM_ENV_PARAM:
171 ASSERT(source->Index < MAX_NV_FRAGMENT_PROGRAM_PARAMS);
172 return ctx->FragmentProgram.Parameters[source->Index];
173 case PROGRAM_STATE_VAR:
174 /* Fallthrough */
175 case PROGRAM_NAMED_PARAM:
176 ASSERT(source->Index < (GLint) program->Base.Parameters->NumParameters);
177 return program->Base.Parameters->ParameterValues[source->Index];
178 default:
179 _mesa_problem(ctx, "Invalid input register file %d in fetch_vector4",
180 source->File);
181 return NULL;
182 }
183 }
184
185
186 /**
187 * Fetch a 4-element float vector from the given source register.
188 * Apply swizzling and negating as needed.
189 */
190 static void
191 fetch_vector4( GLcontext *ctx,
192 const struct prog_src_register *source,
193 const struct fp_machine *machine,
194 const struct gl_fragment_program *program,
195 GLfloat result[4] )
196 {
197 const GLfloat *src = get_register_pointer(ctx, source, machine, program);
198 ASSERT(src);
199
200 if (source->Swizzle == MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y,
201 SWIZZLE_Z, SWIZZLE_W)) {
202 /* no swizzling */
203 COPY_4V(result, src);
204 }
205 else {
206 result[0] = src[GET_SWZ(source->Swizzle, 0)];
207 result[1] = src[GET_SWZ(source->Swizzle, 1)];
208 result[2] = src[GET_SWZ(source->Swizzle, 2)];
209 result[3] = src[GET_SWZ(source->Swizzle, 3)];
210 }
211
212 if (source->NegateBase) {
213 result[0] = -result[0];
214 result[1] = -result[1];
215 result[2] = -result[2];
216 result[3] = -result[3];
217 }
218 if (source->Abs) {
219 result[0] = FABSF(result[0]);
220 result[1] = FABSF(result[1]);
221 result[2] = FABSF(result[2]);
222 result[3] = FABSF(result[3]);
223 }
224 if (source->NegateAbs) {
225 result[0] = -result[0];
226 result[1] = -result[1];
227 result[2] = -result[2];
228 result[3] = -result[3];
229 }
230 }
231
232
233 /**
234 * Fetch the derivative with respect to X for the given register.
235 * \return GL_TRUE if it was easily computed or GL_FALSE if we
236 * need to execute another instance of the program (ugh)!
237 */
238 static GLboolean
239 fetch_vector4_deriv( GLcontext *ctx,
240 const struct prog_src_register *source,
241 const SWspan *span,
242 char xOrY, GLint column, GLfloat result[4] )
243 {
244 GLfloat src[4];
245
246 ASSERT(xOrY == 'X' || xOrY == 'Y');
247
248 switch (source->Index) {
249 case FRAG_ATTRIB_WPOS:
250 if (xOrY == 'X') {
251 src[0] = 1.0;
252 src[1] = 0.0;
253 src[2] = span->dzdx / ctx->DrawBuffer->_DepthMaxF;
254 src[3] = span->dwdx;
255 }
256 else {
257 src[0] = 0.0;
258 src[1] = 1.0;
259 src[2] = span->dzdy / ctx->DrawBuffer->_DepthMaxF;
260 src[3] = span->dwdy;
261 }
262 break;
263 case FRAG_ATTRIB_COL0:
264 if (xOrY == 'X') {
265 src[0] = span->drdx * (1.0F / CHAN_MAXF);
266 src[1] = span->dgdx * (1.0F / CHAN_MAXF);
267 src[2] = span->dbdx * (1.0F / CHAN_MAXF);
268 src[3] = span->dadx * (1.0F / CHAN_MAXF);
269 }
270 else {
271 src[0] = span->drdy * (1.0F / CHAN_MAXF);
272 src[1] = span->dgdy * (1.0F / CHAN_MAXF);
273 src[2] = span->dbdy * (1.0F / CHAN_MAXF);
274 src[3] = span->dady * (1.0F / CHAN_MAXF);
275 }
276 break;
277 case FRAG_ATTRIB_COL1:
278 if (xOrY == 'X') {
279 src[0] = span->dsrdx * (1.0F / CHAN_MAXF);
280 src[1] = span->dsgdx * (1.0F / CHAN_MAXF);
281 src[2] = span->dsbdx * (1.0F / CHAN_MAXF);
282 src[3] = 0.0; /* XXX need this */
283 }
284 else {
285 src[0] = span->dsrdy * (1.0F / CHAN_MAXF);
286 src[1] = span->dsgdy * (1.0F / CHAN_MAXF);
287 src[2] = span->dsbdy * (1.0F / CHAN_MAXF);
288 src[3] = 0.0; /* XXX need this */
289 }
290 break;
291 case FRAG_ATTRIB_FOGC:
292 if (xOrY == 'X') {
293 src[0] = span->dfogdx;
294 src[1] = 0.0;
295 src[2] = 0.0;
296 src[3] = 0.0;
297 }
298 else {
299 src[0] = span->dfogdy;
300 src[1] = 0.0;
301 src[2] = 0.0;
302 src[3] = 0.0;
303 }
304 break;
305 case FRAG_ATTRIB_TEX0:
306 case FRAG_ATTRIB_TEX1:
307 case FRAG_ATTRIB_TEX2:
308 case FRAG_ATTRIB_TEX3:
309 case FRAG_ATTRIB_TEX4:
310 case FRAG_ATTRIB_TEX5:
311 case FRAG_ATTRIB_TEX6:
312 case FRAG_ATTRIB_TEX7:
313 if (xOrY == 'X') {
314 const GLuint u = source->Index - FRAG_ATTRIB_TEX0;
315 /* this is a little tricky - I think I've got it right */
316 const GLfloat invQ = 1.0f / (span->tex[u][3]
317 + span->texStepX[u][3] * column);
318 src[0] = span->texStepX[u][0] * invQ;
319 src[1] = span->texStepX[u][1] * invQ;
320 src[2] = span->texStepX[u][2] * invQ;
321 src[3] = span->texStepX[u][3] * invQ;
322 }
323 else {
324 const GLuint u = source->Index - FRAG_ATTRIB_TEX0;
325 /* Tricky, as above, but in Y direction */
326 const GLfloat invQ = 1.0f / (span->tex[u][3] + span->texStepY[u][3]);
327 src[0] = span->texStepY[u][0] * invQ;
328 src[1] = span->texStepY[u][1] * invQ;
329 src[2] = span->texStepY[u][2] * invQ;
330 src[3] = span->texStepY[u][3] * invQ;
331 }
332 break;
333 default:
334 return GL_FALSE;
335 }
336
337 result[0] = src[GET_SWZ(source->Swizzle, 0)];
338 result[1] = src[GET_SWZ(source->Swizzle, 1)];
339 result[2] = src[GET_SWZ(source->Swizzle, 2)];
340 result[3] = src[GET_SWZ(source->Swizzle, 3)];
341
342 if (source->NegateBase) {
343 result[0] = -result[0];
344 result[1] = -result[1];
345 result[2] = -result[2];
346 result[3] = -result[3];
347 }
348 if (source->Abs) {
349 result[0] = FABSF(result[0]);
350 result[1] = FABSF(result[1]);
351 result[2] = FABSF(result[2]);
352 result[3] = FABSF(result[3]);
353 }
354 if (source->NegateAbs) {
355 result[0] = -result[0];
356 result[1] = -result[1];
357 result[2] = -result[2];
358 result[3] = -result[3];
359 }
360 return GL_TRUE;
361 }
362
363
364 /**
365 * As above, but only return result[0] element.
366 */
367 static void
368 fetch_vector1( GLcontext *ctx,
369 const struct prog_src_register *source,
370 const struct fp_machine *machine,
371 const struct gl_fragment_program *program,
372 GLfloat result[4] )
373 {
374 const GLfloat *src = get_register_pointer(ctx, source, machine, program);
375 ASSERT(src);
376
377 result[0] = src[GET_SWZ(source->Swizzle, 0)];
378
379 if (source->NegateBase) {
380 result[0] = -result[0];
381 }
382 if (source->Abs) {
383 result[0] = FABSF(result[0]);
384 }
385 if (source->NegateAbs) {
386 result[0] = -result[0];
387 }
388 }
389
390
391 /**
392 * Test value against zero and return GT, LT, EQ or UN if NaN.
393 */
394 static INLINE GLuint
395 generate_cc( float value )
396 {
397 if (value != value)
398 return COND_UN; /* NaN */
399 if (value > 0.0F)
400 return COND_GT;
401 if (value < 0.0F)
402 return COND_LT;
403 return COND_EQ;
404 }
405
406
407 /**
408 * Test if the ccMaskRule is satisfied by the given condition code.
409 * Used to mask destination writes according to the current condition code.
410 */
411 static INLINE GLboolean
412 test_cc(GLuint condCode, GLuint ccMaskRule)
413 {
414 switch (ccMaskRule) {
415 case COND_EQ: return (condCode == COND_EQ);
416 case COND_NE: return (condCode != COND_EQ);
417 case COND_LT: return (condCode == COND_LT);
418 case COND_GE: return (condCode == COND_GT || condCode == COND_EQ);
419 case COND_LE: return (condCode == COND_LT || condCode == COND_EQ);
420 case COND_GT: return (condCode == COND_GT);
421 case COND_TR: return GL_TRUE;
422 case COND_FL: return GL_FALSE;
423 default: return GL_TRUE;
424 }
425 }
426
427
428 /**
429 * Store 4 floats into a register. Observe the instructions saturate and
430 * set-condition-code flags.
431 */
432 static void
433 store_vector4( const struct prog_instruction *inst,
434 struct fp_machine *machine,
435 const GLfloat value[4] )
436 {
437 const struct prog_dst_register *dest = &(inst->DstReg);
438 const GLboolean clamp = inst->SaturateMode == SATURATE_ZERO_ONE;
439 GLfloat *dstReg;
440 GLfloat dummyReg[4];
441 GLfloat clampedValue[4];
442 GLuint writeMask = dest->WriteMask;
443
444 switch (dest->File) {
445 case PROGRAM_OUTPUT:
446 dstReg = machine->Outputs[dest->Index];
447 break;
448 case PROGRAM_TEMPORARY:
449 dstReg = machine->Temporaries[dest->Index];
450 break;
451 case PROGRAM_WRITE_ONLY:
452 dstReg = dummyReg;
453 return;
454 default:
455 _mesa_problem(NULL, "bad register file in store_vector4(fp)");
456 return;
457 }
458
459 #if DEBUG_FRAG
460 if (value[0] > 1.0e10 ||
461 IS_INF_OR_NAN(value[0]) ||
462 IS_INF_OR_NAN(value[1]) ||
463 IS_INF_OR_NAN(value[2]) ||
464 IS_INF_OR_NAN(value[3]) )
465 printf("store %g %g %g %g\n", value[0], value[1], value[2], value[3]);
466 #endif
467
468 if (clamp) {
469 clampedValue[0] = CLAMP(value[0], 0.0F, 1.0F);
470 clampedValue[1] = CLAMP(value[1], 0.0F, 1.0F);
471 clampedValue[2] = CLAMP(value[2], 0.0F, 1.0F);
472 clampedValue[3] = CLAMP(value[3], 0.0F, 1.0F);
473 value = clampedValue;
474 }
475
476 if (dest->CondMask != COND_TR) {
477 /* condition codes may turn off some writes */
478 if (writeMask & WRITEMASK_X) {
479 if (!test_cc(machine->CondCodes[GET_SWZ(dest->CondSwizzle, 0)],
480 dest->CondMask))
481 writeMask &= ~WRITEMASK_X;
482 }
483 if (writeMask & WRITEMASK_Y) {
484 if (!test_cc(machine->CondCodes[GET_SWZ(dest->CondSwizzle, 1)],
485 dest->CondMask))
486 writeMask &= ~WRITEMASK_Y;
487 }
488 if (writeMask & WRITEMASK_Z) {
489 if (!test_cc(machine->CondCodes[GET_SWZ(dest->CondSwizzle, 2)],
490 dest->CondMask))
491 writeMask &= ~WRITEMASK_Z;
492 }
493 if (writeMask & WRITEMASK_W) {
494 if (!test_cc(machine->CondCodes[GET_SWZ(dest->CondSwizzle, 3)],
495 dest->CondMask))
496 writeMask &= ~WRITEMASK_W;
497 }
498 }
499
500 if (writeMask & WRITEMASK_X)
501 dstReg[0] = value[0];
502 if (writeMask & WRITEMASK_Y)
503 dstReg[1] = value[1];
504 if (writeMask & WRITEMASK_Z)
505 dstReg[2] = value[2];
506 if (writeMask & WRITEMASK_W)
507 dstReg[3] = value[3];
508
509 if (inst->CondUpdate) {
510 if (writeMask & WRITEMASK_X)
511 machine->CondCodes[0] = generate_cc(value[0]);
512 if (writeMask & WRITEMASK_Y)
513 machine->CondCodes[1] = generate_cc(value[1]);
514 if (writeMask & WRITEMASK_Z)
515 machine->CondCodes[2] = generate_cc(value[2]);
516 if (writeMask & WRITEMASK_W)
517 machine->CondCodes[3] = generate_cc(value[3]);
518 }
519 }
520
521
522 /**
523 * Initialize a new machine state instance from an existing one, adding
524 * the partial derivatives onto the input registers.
525 * Used to implement DDX and DDY instructions in non-trivial cases.
526 */
527 static void
528 init_machine_deriv( GLcontext *ctx,
529 const struct fp_machine *machine,
530 const struct gl_fragment_program *program,
531 const SWspan *span, char xOrY,
532 struct fp_machine *dMachine )
533 {
534 GLuint u;
535
536 ASSERT(xOrY == 'X' || xOrY == 'Y');
537
538 /* copy existing machine */
539 _mesa_memcpy(dMachine, machine, sizeof(struct fp_machine));
540
541 if (program->Base.Target == GL_FRAGMENT_PROGRAM_NV) {
542 /* Clear temporary registers (undefined for ARB_f_p) */
543 _mesa_bzero( (void*) machine->Temporaries,
544 MAX_NV_FRAGMENT_PROGRAM_TEMPS * 4 * sizeof(GLfloat));
545 }
546
547 /* Add derivatives */
548 if (program->Base.InputsRead & (1 << FRAG_ATTRIB_WPOS)) {
549 GLfloat *wpos = (GLfloat*) machine->Inputs[FRAG_ATTRIB_WPOS];
550 if (xOrY == 'X') {
551 wpos[0] += 1.0F;
552 wpos[1] += 0.0F;
553 wpos[2] += span->dzdx;
554 wpos[3] += span->dwdx;
555 }
556 else {
557 wpos[0] += 0.0F;
558 wpos[1] += 1.0F;
559 wpos[2] += span->dzdy;
560 wpos[3] += span->dwdy;
561 }
562 }
563 if (program->Base.InputsRead & (1 << FRAG_ATTRIB_COL0)) {
564 GLfloat *col0 = (GLfloat*) machine->Inputs[FRAG_ATTRIB_COL0];
565 if (xOrY == 'X') {
566 col0[0] += span->drdx * (1.0F / CHAN_MAXF);
567 col0[1] += span->dgdx * (1.0F / CHAN_MAXF);
568 col0[2] += span->dbdx * (1.0F / CHAN_MAXF);
569 col0[3] += span->dadx * (1.0F / CHAN_MAXF);
570 }
571 else {
572 col0[0] += span->drdy * (1.0F / CHAN_MAXF);
573 col0[1] += span->dgdy * (1.0F / CHAN_MAXF);
574 col0[2] += span->dbdy * (1.0F / CHAN_MAXF);
575 col0[3] += span->dady * (1.0F / CHAN_MAXF);
576 }
577 }
578 if (program->Base.InputsRead & (1 << FRAG_ATTRIB_COL1)) {
579 GLfloat *col1 = (GLfloat*) machine->Inputs[FRAG_ATTRIB_COL1];
580 if (xOrY == 'X') {
581 col1[0] += span->dsrdx * (1.0F / CHAN_MAXF);
582 col1[1] += span->dsgdx * (1.0F / CHAN_MAXF);
583 col1[2] += span->dsbdx * (1.0F / CHAN_MAXF);
584 col1[3] += 0.0; /*XXX fix */
585 }
586 else {
587 col1[0] += span->dsrdy * (1.0F / CHAN_MAXF);
588 col1[1] += span->dsgdy * (1.0F / CHAN_MAXF);
589 col1[2] += span->dsbdy * (1.0F / CHAN_MAXF);
590 col1[3] += 0.0; /*XXX fix */
591 }
592 }
593 if (program->Base.InputsRead & (1 << FRAG_ATTRIB_FOGC)) {
594 GLfloat *fogc = (GLfloat*) machine->Inputs[FRAG_ATTRIB_FOGC];
595 if (xOrY == 'X') {
596 fogc[0] += span->dfogdx;
597 }
598 else {
599 fogc[0] += span->dfogdy;
600 }
601 }
602 for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
603 if (program->Base.InputsRead & (1 << (FRAG_ATTRIB_TEX0 + u))) {
604 GLfloat *tex = (GLfloat*) machine->Inputs[FRAG_ATTRIB_TEX0 + u];
605 /* XXX perspective-correct interpolation */
606 if (xOrY == 'X') {
607 tex[0] += span->texStepX[u][0];
608 tex[1] += span->texStepX[u][1];
609 tex[2] += span->texStepX[u][2];
610 tex[3] += span->texStepX[u][3];
611 }
612 else {
613 tex[0] += span->texStepY[u][0];
614 tex[1] += span->texStepY[u][1];
615 tex[2] += span->texStepY[u][2];
616 tex[3] += span->texStepY[u][3];
617 }
618 }
619 }
620
621 /* init condition codes */
622 dMachine->CondCodes[0] = COND_EQ;
623 dMachine->CondCodes[1] = COND_EQ;
624 dMachine->CondCodes[2] = COND_EQ;
625 dMachine->CondCodes[3] = COND_EQ;
626 }
627
628
629 /**
630 * Execute the given vertex program.
631 * NOTE: we do everything in single-precision floating point; we don't
632 * currently observe the single/half/fixed-precision qualifiers.
633 * \param ctx - rendering context
634 * \param program - the fragment program to execute
635 * \param machine - machine state (register file)
636 * \param maxInst - max number of instructions to execute
637 * \return GL_TRUE if program completed or GL_FALSE if program executed KIL.
638 */
639 static GLboolean
640 execute_program( GLcontext *ctx,
641 const struct gl_fragment_program *program, GLuint maxInst,
642 struct fp_machine *machine, const SWspan *span,
643 GLuint column )
644 {
645 GLuint pc;
646
647 #if DEBUG_FRAG
648 printf("execute fragment program --------------------\n");
649 #endif
650
651 for (pc = 0; pc < maxInst; pc++) {
652 const struct prog_instruction *inst = program->Base.Instructions + pc;
653
654 if (ctx->FragmentProgram.CallbackEnabled &&
655 ctx->FragmentProgram.Callback) {
656 ctx->FragmentProgram.CurrentPosition = inst->StringPos;
657 ctx->FragmentProgram.Callback(program->Base.Target,
658 ctx->FragmentProgram.CallbackData);
659 }
660
661 switch (inst->Opcode) {
662 case OPCODE_ABS:
663 {
664 GLfloat a[4], result[4];
665 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
666 result[0] = FABSF(a[0]);
667 result[1] = FABSF(a[1]);
668 result[2] = FABSF(a[2]);
669 result[3] = FABSF(a[3]);
670 store_vector4( inst, machine, result );
671 }
672 break;
673 case OPCODE_ADD:
674 {
675 GLfloat a[4], b[4], result[4];
676 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
677 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
678 result[0] = a[0] + b[0];
679 result[1] = a[1] + b[1];
680 result[2] = a[2] + b[2];
681 result[3] = a[3] + b[3];
682 store_vector4( inst, machine, result );
683 }
684 break;
685 case OPCODE_CMP:
686 {
687 GLfloat a[4], b[4], c[4], result[4];
688 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
689 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
690 fetch_vector4( ctx, &inst->SrcReg[2], machine, program, c );
691 result[0] = a[0] < 0.0F ? b[0] : c[0];
692 result[1] = a[1] < 0.0F ? b[1] : c[1];
693 result[2] = a[2] < 0.0F ? b[2] : c[2];
694 result[3] = a[3] < 0.0F ? b[3] : c[3];
695 store_vector4( inst, machine, result );
696 }
697 break;
698 case OPCODE_COS:
699 {
700 GLfloat a[4], result[4];
701 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
702 result[0] = result[1] = result[2] = result[3]
703 = (GLfloat) _mesa_cos(a[0]);
704 store_vector4( inst, machine, result );
705 }
706 break;
707 case OPCODE_DDX: /* Partial derivative with respect to X */
708 {
709 GLfloat a[4], aNext[4], result[4];
710 struct fp_machine dMachine;
711 if (!fetch_vector4_deriv(ctx, &inst->SrcReg[0], span, 'X',
712 column, result)) {
713 /* This is tricky. Make a copy of the current machine state,
714 * increment the input registers by the dx or dy partial
715 * derivatives, then re-execute the program up to the
716 * preceeding instruction, then fetch the source register.
717 * Finally, find the difference in the register values for
718 * the original and derivative runs.
719 */
720 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a);
721 init_machine_deriv(ctx, machine, program, span,
722 'X', &dMachine);
723 execute_program(ctx, program, pc, &dMachine, span, column);
724 fetch_vector4( ctx, &inst->SrcReg[0], &dMachine, program, aNext );
725 result[0] = aNext[0] - a[0];
726 result[1] = aNext[1] - a[1];
727 result[2] = aNext[2] - a[2];
728 result[3] = aNext[3] - a[3];
729 }
730 store_vector4( inst, machine, result );
731 }
732 break;
733 case OPCODE_DDY: /* Partial derivative with respect to Y */
734 {
735 GLfloat a[4], aNext[4], result[4];
736 struct fp_machine dMachine;
737 if (!fetch_vector4_deriv(ctx, &inst->SrcReg[0], span, 'Y',
738 column, result)) {
739 init_machine_deriv(ctx, machine, program, span,
740 'Y', &dMachine);
741 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a);
742 execute_program(ctx, program, pc, &dMachine, span, column);
743 fetch_vector4( ctx, &inst->SrcReg[0], &dMachine, program, aNext );
744 result[0] = aNext[0] - a[0];
745 result[1] = aNext[1] - a[1];
746 result[2] = aNext[2] - a[2];
747 result[3] = aNext[3] - a[3];
748 }
749 store_vector4( inst, machine, result );
750 }
751 break;
752 case OPCODE_DP3:
753 {
754 GLfloat a[4], b[4], result[4];
755 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
756 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
757 result[0] = result[1] = result[2] = result[3] = DOT3(a, b);
758 store_vector4( inst, machine, result );
759 #if DEBUG_FRAG
760 printf("DP3 %g = (%g %g %g) . (%g %g %g)\n",
761 result[0], a[0], a[1], a[2], b[0], b[1], b[2]);
762 #endif
763 }
764 break;
765 case OPCODE_DP4:
766 {
767 GLfloat a[4], b[4], result[4];
768 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
769 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
770 result[0] = result[1] = result[2] = result[3] = DOT4(a,b);
771 store_vector4( inst, machine, result );
772 #if DEBUG_FRAG
773 printf("DP4 %g = (%g, %g %g %g) . (%g, %g %g %g)\n",
774 result[0], a[0], a[1], a[2], a[3], b[0], b[1], b[2], b[3]);
775 #endif
776 }
777 break;
778 case OPCODE_DPH:
779 {
780 GLfloat a[4], b[4], result[4];
781 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
782 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
783 result[0] = result[1] = result[2] = result[3] =
784 a[0] * b[0] + a[1] * b[1] + a[2] * b[2] + b[3];
785 store_vector4( inst, machine, result );
786 }
787 break;
788 case OPCODE_DST: /* Distance vector */
789 {
790 GLfloat a[4], b[4], result[4];
791 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
792 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
793 result[0] = 1.0F;
794 result[1] = a[1] * b[1];
795 result[2] = a[2];
796 result[3] = b[3];
797 store_vector4( inst, machine, result );
798 }
799 break;
800 case OPCODE_EX2: /* Exponential base 2 */
801 {
802 GLfloat a[4], result[4];
803 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
804 result[0] = result[1] = result[2] = result[3] =
805 (GLfloat) _mesa_pow(2.0, a[0]);
806 store_vector4( inst, machine, result );
807 }
808 break;
809 case OPCODE_FLR:
810 {
811 GLfloat a[4], result[4];
812 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
813 result[0] = FLOORF(a[0]);
814 result[1] = FLOORF(a[1]);
815 result[2] = FLOORF(a[2]);
816 result[3] = FLOORF(a[3]);
817 store_vector4( inst, machine, result );
818 }
819 break;
820 case OPCODE_FRC:
821 {
822 GLfloat a[4], result[4];
823 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
824 result[0] = a[0] - FLOORF(a[0]);
825 result[1] = a[1] - FLOORF(a[1]);
826 result[2] = a[2] - FLOORF(a[2]);
827 result[3] = a[3] - FLOORF(a[3]);
828 store_vector4( inst, machine, result );
829 }
830 break;
831 case OPCODE_KIL_NV: /* NV_f_p only */
832 {
833 const GLuint swizzle = inst->DstReg.CondSwizzle;
834 const GLuint condMask = inst->DstReg.CondMask;
835 if (test_cc(machine->CondCodes[GET_SWZ(swizzle, 0)], condMask) ||
836 test_cc(machine->CondCodes[GET_SWZ(swizzle, 1)], condMask) ||
837 test_cc(machine->CondCodes[GET_SWZ(swizzle, 2)], condMask) ||
838 test_cc(machine->CondCodes[GET_SWZ(swizzle, 3)], condMask)) {
839 return GL_FALSE;
840 }
841 }
842 break;
843 case OPCODE_KIL: /* ARB_f_p only */
844 {
845 GLfloat a[4];
846 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
847 if (a[0] < 0.0F || a[1] < 0.0F || a[2] < 0.0F || a[3] < 0.0F) {
848 return GL_FALSE;
849 }
850 }
851 break;
852 case OPCODE_LG2: /* log base 2 */
853 {
854 GLfloat a[4], result[4];
855 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
856 result[0] = result[1] = result[2] = result[3] = LOG2(a[0]);
857 store_vector4( inst, machine, result );
858 }
859 break;
860 case OPCODE_LIT:
861 {
862 const GLfloat epsilon = 1.0F / 256.0F; /* from NV VP spec */
863 GLfloat a[4], result[4];
864 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
865 a[0] = MAX2(a[0], 0.0F);
866 a[1] = MAX2(a[1], 0.0F);
867 /* XXX ARB version clamps a[3], NV version doesn't */
868 a[3] = CLAMP(a[3], -(128.0F - epsilon), (128.0F - epsilon));
869 result[0] = 1.0F;
870 result[1] = a[0];
871 /* XXX we could probably just use pow() here */
872 if (a[0] > 0.0F) {
873 if (a[1] == 0.0 && a[3] == 0.0)
874 result[2] = 1.0;
875 else
876 result[2] = EXPF(a[3] * LOGF(a[1]));
877 }
878 else {
879 result[2] = 0.0;
880 }
881 result[3] = 1.0F;
882 store_vector4( inst, machine, result );
883 }
884 break;
885 case OPCODE_LRP:
886 {
887 GLfloat a[4], b[4], c[4], result[4];
888 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
889 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
890 fetch_vector4( ctx, &inst->SrcReg[2], machine, program, c );
891 result[0] = a[0] * b[0] + (1.0F - a[0]) * c[0];
892 result[1] = a[1] * b[1] + (1.0F - a[1]) * c[1];
893 result[2] = a[2] * b[2] + (1.0F - a[2]) * c[2];
894 result[3] = a[3] * b[3] + (1.0F - a[3]) * c[3];
895 store_vector4( inst, machine, result );
896 }
897 break;
898 case OPCODE_MAD:
899 {
900 GLfloat a[4], b[4], c[4], result[4];
901 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
902 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
903 fetch_vector4( ctx, &inst->SrcReg[2], machine, program, c );
904 result[0] = a[0] * b[0] + c[0];
905 result[1] = a[1] * b[1] + c[1];
906 result[2] = a[2] * b[2] + c[2];
907 result[3] = a[3] * b[3] + c[3];
908 store_vector4( inst, machine, result );
909 }
910 break;
911 case OPCODE_MAX:
912 {
913 GLfloat a[4], b[4], result[4];
914 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
915 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
916 result[0] = MAX2(a[0], b[0]);
917 result[1] = MAX2(a[1], b[1]);
918 result[2] = MAX2(a[2], b[2]);
919 result[3] = MAX2(a[3], b[3]);
920 store_vector4( inst, machine, result );
921 #if DEBUG_FRAG
922 printf("MAX (%g %g %g %g) = (%g %g %g %g), (%g %g %g %g)\n",
923 result[0], result[1], result[2], result[3],
924 a[0], a[1], a[2], a[3],
925 b[0], b[1], b[2], b[3]);
926 #endif
927 }
928 break;
929 case OPCODE_MIN:
930 {
931 GLfloat a[4], b[4], result[4];
932 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
933 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
934 result[0] = MIN2(a[0], b[0]);
935 result[1] = MIN2(a[1], b[1]);
936 result[2] = MIN2(a[2], b[2]);
937 result[3] = MIN2(a[3], b[3]);
938 store_vector4( inst, machine, result );
939 }
940 break;
941 case OPCODE_MOV:
942 {
943 GLfloat result[4];
944 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, result );
945 store_vector4( inst, machine, result );
946 #if DEBUG_FRAG
947 printf("MOV (%g %g %g %g)\n",
948 result[0], result[1], result[2], result[3]);
949 #endif
950 }
951 break;
952 case OPCODE_MUL:
953 {
954 GLfloat a[4], b[4], result[4];
955 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
956 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
957 result[0] = a[0] * b[0];
958 result[1] = a[1] * b[1];
959 result[2] = a[2] * b[2];
960 result[3] = a[3] * b[3];
961 store_vector4( inst, machine, result );
962 #if DEBUG_FRAG
963 printf("MUL (%g %g %g %g) = (%g %g %g %g) * (%g %g %g %g)\n",
964 result[0], result[1], result[2], result[3],
965 a[0], a[1], a[2], a[3],
966 b[0], b[1], b[2], b[3]);
967 #endif
968 }
969 break;
970 case OPCODE_PK2H: /* pack two 16-bit floats in one 32-bit float */
971 {
972 GLfloat a[4], result[4];
973 GLhalfNV hx, hy;
974 GLuint *rawResult = (GLuint *) result;
975 GLuint twoHalves;
976 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
977 hx = _mesa_float_to_half(a[0]);
978 hy = _mesa_float_to_half(a[1]);
979 twoHalves = hx | (hy << 16);
980 rawResult[0] = rawResult[1] = rawResult[2] = rawResult[3]
981 = twoHalves;
982 store_vector4( inst, machine, result );
983 }
984 break;
985 case OPCODE_PK2US: /* pack two GLushorts into one 32-bit float */
986 {
987 GLfloat a[4], result[4];
988 GLuint usx, usy, *rawResult = (GLuint *) result;
989 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
990 a[0] = CLAMP(a[0], 0.0F, 1.0F);
991 a[1] = CLAMP(a[1], 0.0F, 1.0F);
992 usx = IROUND(a[0] * 65535.0F);
993 usy = IROUND(a[1] * 65535.0F);
994 rawResult[0] = rawResult[1] = rawResult[2] = rawResult[3]
995 = usx | (usy << 16);
996 store_vector4( inst, machine, result );
997 }
998 break;
999 case OPCODE_PK4B: /* pack four GLbytes into one 32-bit float */
1000 {
1001 GLfloat a[4], result[4];
1002 GLuint ubx, uby, ubz, ubw, *rawResult = (GLuint *) result;
1003 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1004 a[0] = CLAMP(a[0], -128.0F / 127.0F, 1.0F);
1005 a[1] = CLAMP(a[1], -128.0F / 127.0F, 1.0F);
1006 a[2] = CLAMP(a[2], -128.0F / 127.0F, 1.0F);
1007 a[3] = CLAMP(a[3], -128.0F / 127.0F, 1.0F);
1008 ubx = IROUND(127.0F * a[0] + 128.0F);
1009 uby = IROUND(127.0F * a[1] + 128.0F);
1010 ubz = IROUND(127.0F * a[2] + 128.0F);
1011 ubw = IROUND(127.0F * a[3] + 128.0F);
1012 rawResult[0] = rawResult[1] = rawResult[2] = rawResult[3]
1013 = ubx | (uby << 8) | (ubz << 16) | (ubw << 24);
1014 store_vector4( inst, machine, result );
1015 }
1016 break;
1017 case OPCODE_PK4UB: /* pack four GLubytes into one 32-bit float */
1018 {
1019 GLfloat a[4], result[4];
1020 GLuint ubx, uby, ubz, ubw, *rawResult = (GLuint *) result;
1021 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1022 a[0] = CLAMP(a[0], 0.0F, 1.0F);
1023 a[1] = CLAMP(a[1], 0.0F, 1.0F);
1024 a[2] = CLAMP(a[2], 0.0F, 1.0F);
1025 a[3] = CLAMP(a[3], 0.0F, 1.0F);
1026 ubx = IROUND(255.0F * a[0]);
1027 uby = IROUND(255.0F * a[1]);
1028 ubz = IROUND(255.0F * a[2]);
1029 ubw = IROUND(255.0F * a[3]);
1030 rawResult[0] = rawResult[1] = rawResult[2] = rawResult[3]
1031 = ubx | (uby << 8) | (ubz << 16) | (ubw << 24);
1032 store_vector4( inst, machine, result );
1033 }
1034 break;
1035 case OPCODE_POW:
1036 {
1037 GLfloat a[4], b[4], result[4];
1038 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1039 fetch_vector1( ctx, &inst->SrcReg[1], machine, program, b );
1040 result[0] = result[1] = result[2] = result[3]
1041 = (GLfloat)_mesa_pow(a[0], b[0]);
1042 store_vector4( inst, machine, result );
1043 }
1044 break;
1045 case OPCODE_RCP:
1046 {
1047 GLfloat a[4], result[4];
1048 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1049 #if DEBUG_FRAG
1050 if (a[0] == 0)
1051 printf("RCP(0)\n");
1052 else if (IS_INF_OR_NAN(a[0]))
1053 printf("RCP(inf)\n");
1054 #endif
1055 result[0] = result[1] = result[2] = result[3] = 1.0F / a[0];
1056 store_vector4( inst, machine, result );
1057 }
1058 break;
1059 case OPCODE_RFL: /* reflection vector */
1060 {
1061 GLfloat axis[4], dir[4], result[4], tmpX, tmpW;
1062 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, axis );
1063 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, dir );
1064 tmpW = DOT3(axis, axis);
1065 tmpX = (2.0F * DOT3(axis, dir)) / tmpW;
1066 result[0] = tmpX * axis[0] - dir[0];
1067 result[1] = tmpX * axis[1] - dir[1];
1068 result[2] = tmpX * axis[2] - dir[2];
1069 /* result[3] is never written! XXX enforce in parser! */
1070 store_vector4( inst, machine, result );
1071 }
1072 break;
1073 case OPCODE_RSQ: /* 1 / sqrt() */
1074 {
1075 GLfloat a[4], result[4];
1076 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1077 a[0] = FABSF(a[0]);
1078 result[0] = result[1] = result[2] = result[3] = INV_SQRTF(a[0]);
1079 store_vector4( inst, machine, result );
1080 #if DEBUG_FRAG
1081 printf("RSQ %g = 1/sqrt(|%g|)\n", result[0], a[0]);
1082 #endif
1083 }
1084 break;
1085 case OPCODE_SCS: /* sine and cos */
1086 {
1087 GLfloat a[4], result[4];
1088 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1089 result[0] = (GLfloat)_mesa_cos(a[0]);
1090 result[1] = (GLfloat)_mesa_sin(a[0]);
1091 result[2] = 0.0; /* undefined! */
1092 result[3] = 0.0; /* undefined! */
1093 store_vector4( inst, machine, result );
1094 }
1095 break;
1096 case OPCODE_SEQ: /* set on equal */
1097 {
1098 GLfloat a[4], b[4], result[4];
1099 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1100 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1101 result[0] = (a[0] == b[0]) ? 1.0F : 0.0F;
1102 result[1] = (a[1] == b[1]) ? 1.0F : 0.0F;
1103 result[2] = (a[2] == b[2]) ? 1.0F : 0.0F;
1104 result[3] = (a[3] == b[3]) ? 1.0F : 0.0F;
1105 store_vector4( inst, machine, result );
1106 }
1107 break;
1108 case OPCODE_SFL: /* set false, operands ignored */
1109 {
1110 static const GLfloat result[4] = { 0.0F, 0.0F, 0.0F, 0.0F };
1111 store_vector4( inst, machine, result );
1112 }
1113 break;
1114 case OPCODE_SGE: /* set on greater or equal */
1115 {
1116 GLfloat a[4], b[4], result[4];
1117 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1118 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1119 result[0] = (a[0] >= b[0]) ? 1.0F : 0.0F;
1120 result[1] = (a[1] >= b[1]) ? 1.0F : 0.0F;
1121 result[2] = (a[2] >= b[2]) ? 1.0F : 0.0F;
1122 result[3] = (a[3] >= b[3]) ? 1.0F : 0.0F;
1123 store_vector4( inst, machine, result );
1124 }
1125 break;
1126 case OPCODE_SGT: /* set on greater */
1127 {
1128 GLfloat a[4], b[4], result[4];
1129 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1130 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1131 result[0] = (a[0] > b[0]) ? 1.0F : 0.0F;
1132 result[1] = (a[1] > b[1]) ? 1.0F : 0.0F;
1133 result[2] = (a[2] > b[2]) ? 1.0F : 0.0F;
1134 result[3] = (a[3] > b[3]) ? 1.0F : 0.0F;
1135 store_vector4( inst, machine, result );
1136 }
1137 break;
1138 case OPCODE_SIN:
1139 {
1140 GLfloat a[4], result[4];
1141 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1142 result[0] = result[1] = result[2] = result[3]
1143 = (GLfloat) _mesa_sin(a[0]);
1144 store_vector4( inst, machine, result );
1145 }
1146 break;
1147 case OPCODE_SLE: /* set on less or equal */
1148 {
1149 GLfloat a[4], b[4], result[4];
1150 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1151 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1152 result[0] = (a[0] <= b[0]) ? 1.0F : 0.0F;
1153 result[1] = (a[1] <= b[1]) ? 1.0F : 0.0F;
1154 result[2] = (a[2] <= b[2]) ? 1.0F : 0.0F;
1155 result[3] = (a[3] <= b[3]) ? 1.0F : 0.0F;
1156 store_vector4( inst, machine, result );
1157 }
1158 break;
1159 case OPCODE_SLT: /* set on less */
1160 {
1161 GLfloat a[4], b[4], result[4];
1162 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1163 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1164 result[0] = (a[0] < b[0]) ? 1.0F : 0.0F;
1165 result[1] = (a[1] < b[1]) ? 1.0F : 0.0F;
1166 result[2] = (a[2] < b[2]) ? 1.0F : 0.0F;
1167 result[3] = (a[3] < b[3]) ? 1.0F : 0.0F;
1168 store_vector4( inst, machine, result );
1169 }
1170 break;
1171 case OPCODE_SNE: /* set on not equal */
1172 {
1173 GLfloat a[4], b[4], result[4];
1174 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1175 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1176 result[0] = (a[0] != b[0]) ? 1.0F : 0.0F;
1177 result[1] = (a[1] != b[1]) ? 1.0F : 0.0F;
1178 result[2] = (a[2] != b[2]) ? 1.0F : 0.0F;
1179 result[3] = (a[3] != b[3]) ? 1.0F : 0.0F;
1180 store_vector4( inst, machine, result );
1181 }
1182 break;
1183 case OPCODE_STR: /* set true, operands ignored */
1184 {
1185 static const GLfloat result[4] = { 1.0F, 1.0F, 1.0F, 1.0F };
1186 store_vector4( inst, machine, result );
1187 }
1188 break;
1189 case OPCODE_SUB:
1190 {
1191 GLfloat a[4], b[4], result[4];
1192 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1193 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1194 result[0] = a[0] - b[0];
1195 result[1] = a[1] - b[1];
1196 result[2] = a[2] - b[2];
1197 result[3] = a[3] - b[3];
1198 store_vector4( inst, machine, result );
1199 }
1200 break;
1201 case OPCODE_SWZ: /* extended swizzle */
1202 {
1203 const struct prog_src_register *source = &inst->SrcReg[0];
1204 const GLfloat *src = get_register_pointer(ctx, source,
1205 machine, program);
1206 GLfloat result[4];
1207 GLuint i;
1208 for (i = 0; i < 4; i++) {
1209 const GLuint swz = GET_SWZ(source->Swizzle, i);
1210 if (swz == SWIZZLE_ZERO)
1211 result[i] = 0.0;
1212 else if (swz == SWIZZLE_ONE)
1213 result[i] = 1.0;
1214 else {
1215 ASSERT(swz >= 0);
1216 ASSERT(swz <= 3);
1217 result[i] = src[swz];
1218 }
1219 if (source->NegateBase & (1 << i))
1220 result[i] = -result[i];
1221 }
1222 store_vector4( inst, machine, result );
1223 }
1224 break;
1225 case OPCODE_TEX: /* Both ARB and NV frag prog */
1226 /* Texel lookup */
1227 {
1228 GLfloat texcoord[4], color[4];
1229 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, texcoord );
1230 /* Note: we pass 0 for LOD. The ARB extension requires it
1231 * while the NV extension says it's implementation dependant.
1232 */
1233 /* KW: Previously lambda was passed as zero, but I
1234 * believe this is incorrect, the spec seems to
1235 * indicate rather that lambda should not be
1236 * changed/biased, unlike TXB where texcoord[3] is
1237 * added to the lambda calculations. The lambda should
1238 * still be calculated normally for TEX & TXP though,
1239 * not set to zero. Otherwise it's very difficult to
1240 * implement normal GL semantics through the fragment
1241 * shader.
1242 */
1243 fetch_texel( ctx, texcoord,
1244 span->array->lambda[inst->TexSrcUnit][column],
1245 inst->TexSrcUnit, color );
1246 #if DEBUG_FRAG
1247 if (color[3])
1248 printf("color[3] = %f\n", color[3]);
1249 #endif
1250 store_vector4( inst, machine, color );
1251 }
1252 break;
1253 case OPCODE_TXB: /* GL_ARB_fragment_program only */
1254 /* Texel lookup with LOD bias */
1255 {
1256 GLfloat texcoord[4], color[4], bias, lambda;
1257
1258 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, texcoord );
1259 /* texcoord[3] is the bias to add to lambda */
1260 bias = ctx->Texture.Unit[inst->TexSrcUnit].LodBias
1261 + ctx->Texture.Unit[inst->TexSrcUnit]._Current->LodBias
1262 + texcoord[3];
1263 lambda = span->array->lambda[inst->TexSrcUnit][column] + bias;
1264 fetch_texel( ctx, texcoord, lambda,
1265 inst->TexSrcUnit, color );
1266 store_vector4( inst, machine, color );
1267 }
1268 break;
1269 case OPCODE_TXD: /* GL_NV_fragment_program only */
1270 /* Texture lookup w/ partial derivatives for LOD */
1271 {
1272 GLfloat texcoord[4], dtdx[4], dtdy[4], color[4];
1273 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, texcoord );
1274 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, dtdx );
1275 fetch_vector4( ctx, &inst->SrcReg[2], machine, program, dtdy );
1276 fetch_texel_deriv( ctx, texcoord, dtdx, dtdy, inst->TexSrcUnit,
1277 color );
1278 store_vector4( inst, machine, color );
1279 }
1280 break;
1281 case OPCODE_TXP: /* GL_ARB_fragment_program only */
1282 /* Texture lookup w/ projective divide */
1283 {
1284 GLfloat texcoord[4], color[4];
1285 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, texcoord );
1286 /* Not so sure about this test - if texcoord[3] is
1287 * zero, we'd probably be fine except for an ASSERT in
1288 * IROUND_POS() which gets triggered by the inf values created.
1289 */
1290 if (texcoord[3] != 0.0) {
1291 texcoord[0] /= texcoord[3];
1292 texcoord[1] /= texcoord[3];
1293 texcoord[2] /= texcoord[3];
1294 }
1295 /* KW: Previously lambda was passed as zero, but I
1296 * believe this is incorrect, the spec seems to
1297 * indicate rather that lambda should not be
1298 * changed/biased, unlike TXB where texcoord[3] is
1299 * added to the lambda calculations. The lambda should
1300 * still be calculated normally for TEX & TXP though,
1301 * not set to zero.
1302 */
1303 fetch_texel( ctx, texcoord,
1304 span->array->lambda[inst->TexSrcUnit][column],
1305 inst->TexSrcUnit, color );
1306 store_vector4( inst, machine, color );
1307 }
1308 break;
1309 case OPCODE_TXP_NV: /* GL_NV_fragment_program only */
1310 /* Texture lookup w/ projective divide */
1311 {
1312 GLfloat texcoord[4], color[4];
1313 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, texcoord );
1314 if (inst->TexSrcTarget != TEXTURE_CUBE_INDEX &&
1315 texcoord[3] != 0.0) {
1316 texcoord[0] /= texcoord[3];
1317 texcoord[1] /= texcoord[3];
1318 texcoord[2] /= texcoord[3];
1319 }
1320 fetch_texel( ctx, texcoord,
1321 span->array->lambda[inst->TexSrcUnit][column],
1322 inst->TexSrcUnit, color );
1323 store_vector4( inst, machine, color );
1324 }
1325 break;
1326 case OPCODE_UP2H: /* unpack two 16-bit floats */
1327 {
1328 GLfloat a[4], result[4];
1329 const GLuint *rawBits = (const GLuint *) a;
1330 GLhalfNV hx, hy;
1331 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1332 hx = rawBits[0] & 0xffff;
1333 hy = rawBits[0] >> 16;
1334 result[0] = result[2] = _mesa_half_to_float(hx);
1335 result[1] = result[3] = _mesa_half_to_float(hy);
1336 store_vector4( inst, machine, result );
1337 }
1338 break;
1339 case OPCODE_UP2US: /* unpack two GLushorts */
1340 {
1341 GLfloat a[4], result[4];
1342 const GLuint *rawBits = (const GLuint *) a;
1343 GLushort usx, usy;
1344 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1345 usx = rawBits[0] & 0xffff;
1346 usy = rawBits[0] >> 16;
1347 result[0] = result[2] = usx * (1.0f / 65535.0f);
1348 result[1] = result[3] = usy * (1.0f / 65535.0f);
1349 store_vector4( inst, machine, result );
1350 }
1351 break;
1352 case OPCODE_UP4B: /* unpack four GLbytes */
1353 {
1354 GLfloat a[4], result[4];
1355 const GLuint *rawBits = (const GLuint *) a;
1356 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1357 result[0] = (((rawBits[0] >> 0) & 0xff) - 128) / 127.0F;
1358 result[1] = (((rawBits[0] >> 8) & 0xff) - 128) / 127.0F;
1359 result[2] = (((rawBits[0] >> 16) & 0xff) - 128) / 127.0F;
1360 result[3] = (((rawBits[0] >> 24) & 0xff) - 128) / 127.0F;
1361 store_vector4( inst, machine, result );
1362 }
1363 break;
1364 case OPCODE_UP4UB: /* unpack four GLubytes */
1365 {
1366 GLfloat a[4], result[4];
1367 const GLuint *rawBits = (const GLuint *) a;
1368 fetch_vector1( ctx, &inst->SrcReg[0], machine, program, a );
1369 result[0] = ((rawBits[0] >> 0) & 0xff) / 255.0F;
1370 result[1] = ((rawBits[0] >> 8) & 0xff) / 255.0F;
1371 result[2] = ((rawBits[0] >> 16) & 0xff) / 255.0F;
1372 result[3] = ((rawBits[0] >> 24) & 0xff) / 255.0F;
1373 store_vector4( inst, machine, result );
1374 }
1375 break;
1376 case OPCODE_XPD: /* cross product */
1377 {
1378 GLfloat a[4], b[4], result[4];
1379 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1380 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1381 result[0] = a[1] * b[2] - a[2] * b[1];
1382 result[1] = a[2] * b[0] - a[0] * b[2];
1383 result[2] = a[0] * b[1] - a[1] * b[0];
1384 result[3] = 1.0;
1385 store_vector4( inst, machine, result );
1386 }
1387 break;
1388 case OPCODE_X2D: /* 2-D matrix transform */
1389 {
1390 GLfloat a[4], b[4], c[4], result[4];
1391 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a );
1392 fetch_vector4( ctx, &inst->SrcReg[1], machine, program, b );
1393 fetch_vector4( ctx, &inst->SrcReg[2], machine, program, c );
1394 result[0] = a[0] + b[0] * c[0] + b[1] * c[1];
1395 result[1] = a[1] + b[0] * c[2] + b[1] * c[3];
1396 result[2] = a[2] + b[0] * c[0] + b[1] * c[1];
1397 result[3] = a[3] + b[0] * c[2] + b[1] * c[3];
1398 store_vector4( inst, machine, result );
1399 }
1400 break;
1401 case OPCODE_PRINT:
1402 {
1403 if (inst->SrcReg[0].File != -1) {
1404 GLfloat a[4];
1405 fetch_vector4( ctx, &inst->SrcReg[0], machine, program, a);
1406 _mesa_printf("%s%g, %g, %g, %g\n", (const char *) inst->Data,
1407 a[0], a[1], a[2], a[3]);
1408 }
1409 else {
1410 _mesa_printf("%s\n", (const char *) inst->Data);
1411 }
1412 }
1413 break;
1414 case OPCODE_END:
1415 return GL_TRUE;
1416 default:
1417 _mesa_problem(ctx, "Bad opcode %d in _mesa_exec_fragment_program",
1418 inst->Opcode);
1419 return GL_TRUE; /* return value doesn't matter */
1420 }
1421 }
1422 return GL_TRUE;
1423 }
1424
1425
1426 /**
1427 * Initialize the virtual fragment program machine state prior to running
1428 * fragment program on a fragment. This involves initializing the input
1429 * registers, condition codes, etc.
1430 * \param machine the virtual machine state to init
1431 * \param program the fragment program we're about to run
1432 * \param span the span of pixels we'll operate on
1433 * \param col which element (column) of the span we'll operate on
1434 */
1435 static void
1436 init_machine( GLcontext *ctx, struct fp_machine *machine,
1437 const struct gl_fragment_program *program,
1438 const SWspan *span, GLuint col )
1439 {
1440 GLuint inputsRead = program->Base.InputsRead;
1441 GLuint u;
1442
1443 if (ctx->FragmentProgram.CallbackEnabled)
1444 inputsRead = ~0;
1445
1446 if (program->Base.Target == GL_FRAGMENT_PROGRAM_NV) {
1447 /* Clear temporary registers (undefined for ARB_f_p) */
1448 _mesa_bzero(machine->Temporaries,
1449 MAX_NV_FRAGMENT_PROGRAM_TEMPS * 4 * sizeof(GLfloat));
1450 }
1451
1452 /* Load input registers */
1453 if (inputsRead & (1 << FRAG_ATTRIB_WPOS)) {
1454 GLfloat *wpos = machine->Inputs[FRAG_ATTRIB_WPOS];
1455 ASSERT(span->arrayMask & SPAN_Z);
1456 if (span->arrayMask & SPAN_XY) {
1457 wpos[0] = (GLfloat) span->array->x[col];
1458 wpos[1] = (GLfloat) span->array->y[col];
1459 }
1460 else {
1461 wpos[0] = (GLfloat) span->x + col;
1462 wpos[1] = (GLfloat) span->y;
1463 }
1464 wpos[2] = (GLfloat) span->array->z[col] / ctx->DrawBuffer->_DepthMaxF;
1465 wpos[3] = span->w + col * span->dwdx;
1466 }
1467 if (inputsRead & (1 << FRAG_ATTRIB_COL0)) {
1468 ASSERT(span->arrayMask & SPAN_RGBA);
1469 COPY_4V(machine->Inputs[FRAG_ATTRIB_COL0],
1470 span->array->color.sz4.rgba[col]);
1471 }
1472 if (inputsRead & (1 << FRAG_ATTRIB_COL1)) {
1473 ASSERT(span->arrayMask & SPAN_SPEC);
1474 COPY_4V(machine->Inputs[FRAG_ATTRIB_COL1],
1475 span->array->color.sz4.spec[col]);
1476 }
1477 if (inputsRead & (1 << FRAG_ATTRIB_FOGC)) {
1478 GLfloat *fogc = machine->Inputs[FRAG_ATTRIB_FOGC];
1479 ASSERT(span->arrayMask & SPAN_FOG);
1480 fogc[0] = span->array->fog[col];
1481 fogc[1] = 0.0F;
1482 fogc[2] = 0.0F;
1483 fogc[3] = 0.0F;
1484 }
1485 for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
1486 if (inputsRead & (1 << (FRAG_ATTRIB_TEX0 + u))) {
1487 GLfloat *tex = machine->Inputs[FRAG_ATTRIB_TEX0 + u];
1488 /*ASSERT(ctx->Texture._EnabledCoordUnits & (1 << u));*/
1489 COPY_4V(tex, span->array->texcoords[u][col]);
1490 /*ASSERT(tex[0] != 0 || tex[1] != 0 || tex[2] != 0);*/
1491 }
1492 }
1493
1494 /* init condition codes */
1495 machine->CondCodes[0] = COND_EQ;
1496 machine->CondCodes[1] = COND_EQ;
1497 machine->CondCodes[2] = COND_EQ;
1498 machine->CondCodes[3] = COND_EQ;
1499 }
1500
1501
1502 /**
1503 * Run fragment program on the pixels in span from 'start' to 'end' - 1.
1504 */
1505 static void
1506 run_program(GLcontext *ctx, SWspan *span, GLuint start, GLuint end)
1507 {
1508 const struct gl_fragment_program *program = ctx->FragmentProgram._Current;
1509 struct fp_machine machine;
1510 GLuint i;
1511
1512 CurrentMachine = &machine;
1513
1514 for (i = start; i < end; i++) {
1515 if (span->array->mask[i]) {
1516 init_machine(ctx, &machine, program, span, i);
1517
1518 if (execute_program(ctx, program, ~0, &machine, span, i)) {
1519 /* Store result color */
1520 COPY_4V(span->array->color.sz4.rgba[i],
1521 machine.Outputs[FRAG_RESULT_COLR]);
1522
1523 /* Store result depth/z */
1524 if (program->Base.OutputsWritten & (1 << FRAG_RESULT_DEPR)) {
1525 const GLfloat depth = machine.Outputs[FRAG_RESULT_DEPR][2];
1526 if (depth <= 0.0)
1527 span->array->z[i] = 0;
1528 else if (depth >= 1.0)
1529 span->array->z[i] = ctx->DrawBuffer->_DepthMax;
1530 else
1531 span->array->z[i] = IROUND(depth * ctx->DrawBuffer->_DepthMaxF);
1532 }
1533 }
1534 else {
1535 /* killed fragment */
1536 span->array->mask[i] = GL_FALSE;
1537 span->writeAll = GL_FALSE;
1538 }
1539 }
1540 }
1541
1542 CurrentMachine = NULL;
1543 }
1544
1545
1546 /**
1547 * Execute the current fragment program for all the fragments
1548 * in the given span.
1549 */
1550 void
1551 _swrast_exec_fragment_program( GLcontext *ctx, SWspan *span )
1552 {
1553 const struct gl_fragment_program *program = ctx->FragmentProgram._Current;
1554
1555 /* incoming colors should be floats */
1556 ASSERT(span->array->ChanType == GL_FLOAT);
1557
1558 ctx->_CurrentProgram = GL_FRAGMENT_PROGRAM_ARB; /* or NV, doesn't matter */
1559
1560 if (program->Base.Parameters) {
1561 _mesa_load_state_parameters(ctx, program->Base.Parameters);
1562 }
1563
1564 run_program(ctx, span, 0, span->end);
1565
1566 if (program->Base.OutputsWritten & (1 << FRAG_RESULT_DEPR)) {
1567 span->interpMask &= ~SPAN_Z;
1568 span->arrayMask |= SPAN_Z;
1569 }
1570
1571 ctx->_CurrentProgram = 0;
1572 }
1573