projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge remote branch 'origin/master' into nv50-compiler
[mesa.git] / src / mesa / drivers / dri / i915 / i915_fragprog.c
1 /**************************************************************************
2 *
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * 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
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 #include "main/glheader.h"
29 #include "main/macros.h"
30 #include "main/enums.h"
31
32 #include "program/prog_instruction.h"
33 #include "program/prog_parameter.h"
34 #include "program/program.h"
35 #include "program/programopt.h"
36 #include "program/prog_print.h"
37
38 #include "tnl/tnl.h"
39 #include "tnl/t_context.h"
40
41 #include "intel_batchbuffer.h"
42
43 #include "i915_reg.h"
44 #include "i915_context.h"
45 #include "i915_program.h"
46
47 static const GLfloat sin_quad_constants[2][4] = {
48 {
49 2.0,
50 -1.0,
51 .5,
52 .75
53 },
54 {
55 4.0,
56 -4.0,
57 1.0 / (2.0 * M_PI),
58 .2225
59 }
60 };
61
62 static const GLfloat sin_constants[4] = { 1.0,
63 -1.0 / (3 * 2 * 1),
64 1.0 / (5 * 4 * 3 * 2 * 1),
65 -1.0 / (7 * 6 * 5 * 4 * 3 * 2 * 1)
66 };
67
68 /* 1, -1/2!, 1/4!, -1/6! */
69 static const GLfloat cos_constants[4] = { 1.0,
70 -1.0 / (2 * 1),
71 1.0 / (4 * 3 * 2 * 1),
72 -1.0 / (6 * 5 * 4 * 3 * 2 * 1)
73 };
74
75 /**
76 * Retrieve a ureg for the given source register. Will emit
77 * constants, apply swizzling and negation as needed.
78 */
79 static GLuint
80 src_vector(struct i915_fragment_program *p,
81 const struct prog_src_register *source,
82 const struct gl_fragment_program *program)
83 {
84 GLuint src;
85
86 switch (source->File) {
87
88 /* Registers:
89 */
90 case PROGRAM_TEMPORARY:
91 if (source->Index >= I915_MAX_TEMPORARY) {
92 i915_program_error(p, "Exceeded max temporary reg: %d/%d",
93 source->Index, I915_MAX_TEMPORARY);
94 return 0;
95 }
96 src = UREG(REG_TYPE_R, source->Index);
97 break;
98 case PROGRAM_INPUT:
99 switch (source->Index) {
100 case FRAG_ATTRIB_WPOS:
101 src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
102 break;
103 case FRAG_ATTRIB_COL0:
104 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
105 break;
106 case FRAG_ATTRIB_COL1:
107 src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
108 src = swizzle(src, X, Y, Z, ONE);
109 break;
110 case FRAG_ATTRIB_FOGC:
111 src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
112 src = swizzle(src, W, ZERO, ZERO, ONE);
113 break;
114 case FRAG_ATTRIB_TEX0:
115 case FRAG_ATTRIB_TEX1:
116 case FRAG_ATTRIB_TEX2:
117 case FRAG_ATTRIB_TEX3:
118 case FRAG_ATTRIB_TEX4:
119 case FRAG_ATTRIB_TEX5:
120 case FRAG_ATTRIB_TEX6:
121 case FRAG_ATTRIB_TEX7:
122 src = i915_emit_decl(p, REG_TYPE_T,
123 T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
124 D0_CHANNEL_ALL);
125 break;
126
127 case FRAG_ATTRIB_VAR0:
128 case FRAG_ATTRIB_VAR0 + 1:
129 case FRAG_ATTRIB_VAR0 + 2:
130 case FRAG_ATTRIB_VAR0 + 3:
131 case FRAG_ATTRIB_VAR0 + 4:
132 case FRAG_ATTRIB_VAR0 + 5:
133 case FRAG_ATTRIB_VAR0 + 6:
134 case FRAG_ATTRIB_VAR0 + 7:
135 src = i915_emit_decl(p, REG_TYPE_T,
136 T_TEX0 + (source->Index - FRAG_ATTRIB_VAR0),
137 D0_CHANNEL_ALL);
138 break;
139
140 default:
141 i915_program_error(p, "Bad source->Index: %d", source->Index);
142 return 0;
143 }
144 break;
145
146 case PROGRAM_OUTPUT:
147 switch (source->Index) {
148 case FRAG_RESULT_COLOR:
149 src = UREG(REG_TYPE_OC, 0);
150 break;
151 case FRAG_RESULT_DEPTH:
152 src = UREG(REG_TYPE_OD, 0);
153 break;
154 default:
155 i915_program_error(p, "Bad source->Index: %d", source->Index);
156 return 0;
157 }
158 break;
159
160 /* Various paramters and env values. All emitted to
161 * hardware as program constants.
162 */
163 case PROGRAM_LOCAL_PARAM:
164 src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
165 break;
166
167 case PROGRAM_ENV_PARAM:
168 src =
169 i915_emit_param4fv(p,
170 p->ctx->FragmentProgram.Parameters[source->
171 Index]);
172 break;
173
174 case PROGRAM_CONSTANT:
175 case PROGRAM_STATE_VAR:
176 case PROGRAM_NAMED_PARAM:
177 case PROGRAM_UNIFORM:
178 src =
179 i915_emit_param4fv(p,
180 program->Base.Parameters->ParameterValues[source->
181 Index]);
182 break;
183
184 default:
185 i915_program_error(p, "Bad source->File: %d", source->File);
186 return 0;
187 }
188
189 src = swizzle(src,
190 GET_SWZ(source->Swizzle, 0),
191 GET_SWZ(source->Swizzle, 1),
192 GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
193
194 if (source->Negate)
195 src = negate(src,
196 GET_BIT(source->Negate, 0),
197 GET_BIT(source->Negate, 1),
198 GET_BIT(source->Negate, 2),
199 GET_BIT(source->Negate, 3));
200
201 return src;
202 }
203
204
205 static GLuint
206 get_result_vector(struct i915_fragment_program *p,
207 const struct prog_instruction *inst)
208 {
209 switch (inst->DstReg.File) {
210 case PROGRAM_OUTPUT:
211 switch (inst->DstReg.Index) {
212 case FRAG_RESULT_COLOR:
213 return UREG(REG_TYPE_OC, 0);
214 case FRAG_RESULT_DEPTH:
215 p->depth_written = 1;
216 return UREG(REG_TYPE_OD, 0);
217 default:
218 i915_program_error(p, "Bad inst->DstReg.Index: %d",
219 inst->DstReg.Index);
220 return 0;
221 }
222 case PROGRAM_TEMPORARY:
223 return UREG(REG_TYPE_R, inst->DstReg.Index);
224 default:
225 i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
226 return 0;
227 }
228 }
229
230 static GLuint
231 get_result_flags(const struct prog_instruction *inst)
232 {
233 GLuint flags = 0;
234
235 if (inst->SaturateMode == SATURATE_ZERO_ONE)
236 flags |= A0_DEST_SATURATE;
237 if (inst->DstReg.WriteMask & WRITEMASK_X)
238 flags |= A0_DEST_CHANNEL_X;
239 if (inst->DstReg.WriteMask & WRITEMASK_Y)
240 flags |= A0_DEST_CHANNEL_Y;
241 if (inst->DstReg.WriteMask & WRITEMASK_Z)
242 flags |= A0_DEST_CHANNEL_Z;
243 if (inst->DstReg.WriteMask & WRITEMASK_W)
244 flags |= A0_DEST_CHANNEL_W;
245
246 return flags;
247 }
248
249 static GLuint
250 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
251 {
252 switch (bit) {
253 case TEXTURE_1D_INDEX:
254 return D0_SAMPLE_TYPE_2D;
255 case TEXTURE_2D_INDEX:
256 return D0_SAMPLE_TYPE_2D;
257 case TEXTURE_RECT_INDEX:
258 return D0_SAMPLE_TYPE_2D;
259 case TEXTURE_3D_INDEX:
260 return D0_SAMPLE_TYPE_VOLUME;
261 case TEXTURE_CUBE_INDEX:
262 return D0_SAMPLE_TYPE_CUBE;
263 default:
264 i915_program_error(p, "TexSrcBit: %d", bit);
265 return 0;
266 }
267 }
268
269 #define EMIT_TEX( OP ) \
270 do { \
271 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
272 const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current; \
273 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
274 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
275 unit, dim); \
276 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
277 /* Texel lookup */ \
278 \
279 i915_emit_texld( p, get_live_regs(p, inst), \
280 get_result_vector( p, inst ), \
281 get_result_flags( inst ), \
282 sampler, \
283 coord, \
284 OP); \
285 } while (0)
286
287 #define EMIT_ARITH( OP, N ) \
288 do { \
289 i915_emit_arith( p, \
290 OP, \
291 get_result_vector( p, inst ), \
292 get_result_flags( inst ), 0, \
293 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
294 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
295 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
296 } while (0)
297
298 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
299 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
300 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
301
302 /*
303 * TODO: consider moving this into core
304 */
305 static void calc_live_regs( struct i915_fragment_program *p )
306 {
307 const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
308 GLuint regsUsed = 0xffff0000;
309 GLint i;
310
311 for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
312 struct prog_instruction *inst = &program->Base.Instructions[i];
313 int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
314 int a;
315
316 /* Register is written to: unmark as live for this and preceeding ops */
317 if (inst->DstReg.File == PROGRAM_TEMPORARY)
318 regsUsed &= ~(1 << inst->DstReg.Index);
319
320 for (a = 0; a < opArgs; a++) {
321 /* Register is read from: mark as live for this and preceeding ops */
322 if (inst->SrcReg[a].File == PROGRAM_TEMPORARY)
323 regsUsed |= 1 << inst->SrcReg[a].Index;
324 }
325
326 p->usedRegs[i] = regsUsed;
327 }
328 }
329
330 static GLuint get_live_regs( struct i915_fragment_program *p,
331 const struct prog_instruction *inst )
332 {
333 const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
334 GLuint nr = inst - program->Base.Instructions;
335
336 return p->usedRegs[nr];
337 }
338
339
340 /* Possible concerns:
341 *
342 * SIN, COS -- could use another taylor step?
343 * LIT -- results seem a little different to sw mesa
344 * LOG -- different to mesa on negative numbers, but this is conformant.
345 *
346 * Parse failures -- Mesa doesn't currently give a good indication
347 * internally whether a particular program string parsed or not. This
348 * can lead to confusion -- hopefully we cope with it ok now.
349 *
350 */
351 static void
352 upload_program(struct i915_fragment_program *p)
353 {
354 const struct gl_fragment_program *program =
355 p->ctx->FragmentProgram._Current;
356 const struct prog_instruction *inst = program->Base.Instructions;
357
358 if (INTEL_DEBUG & DEBUG_WM)
359 _mesa_print_program(&program->Base);
360
361 /* Is this a parse-failed program? Ensure a valid program is
362 * loaded, as the flagging of an error isn't sufficient to stop
363 * this being uploaded to hardware.
364 */
365 if (inst[0].Opcode == OPCODE_END) {
366 GLuint tmp = i915_get_utemp(p);
367 i915_emit_arith(p,
368 A0_MOV,
369 UREG(REG_TYPE_OC, 0),
370 A0_DEST_CHANNEL_ALL, 0,
371 swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
372 return;
373 }
374
375 if (program->Base.NumInstructions > I915_MAX_INSN) {
376 i915_program_error(p, "Exceeded max instructions (%d out of %d)",
377 program->Base.NumInstructions, I915_MAX_INSN);
378 return;
379 }
380
381 /* Not always needed:
382 */
383 calc_live_regs(p);
384
385 while (1) {
386 GLuint src0, src1, src2, flags;
387 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
388
389 switch (inst->Opcode) {
390 case OPCODE_ABS:
391 src0 = src_vector(p, &inst->SrcReg[0], program);
392 i915_emit_arith(p,
393 A0_MAX,
394 get_result_vector(p, inst),
395 get_result_flags(inst), 0,
396 src0, negate(src0, 1, 1, 1, 1), 0);
397 break;
398
399 case OPCODE_ADD:
400 EMIT_2ARG_ARITH(A0_ADD);
401 break;
402
403 case OPCODE_CMP:
404 src0 = src_vector(p, &inst->SrcReg[0], program);
405 src1 = src_vector(p, &inst->SrcReg[1], program);
406 src2 = src_vector(p, &inst->SrcReg[2], program);
407 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
408 break;
409
410 case OPCODE_COS:
411 src0 = src_vector(p, &inst->SrcReg[0], program);
412 tmp = i915_get_utemp(p);
413 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
414 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
415
416 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
417 i915_emit_arith(p,
418 A0_MAD,
419 tmp, A0_DEST_CHANNEL_X, 0,
420 src0,
421 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
422 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
423
424 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
425
426 i915_emit_arith(p,
427 A0_MAD,
428 tmp, A0_DEST_CHANNEL_X, 0,
429 tmp,
430 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
431 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
432
433 /* Compute COS with the same calculation used for SIN, but a
434 * different source range has been mapped to [-1,1] this time.
435 */
436
437 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
438 i915_emit_arith(p,
439 A0_MAX,
440 tmp, A0_DEST_CHANNEL_Y, 0,
441 swizzle(tmp, ZERO, X, ZERO, ZERO),
442 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
443 0);
444
445 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
446 i915_emit_arith(p,
447 A0_MUL,
448 tmp, A0_DEST_CHANNEL_Y, 0,
449 swizzle(tmp, ZERO, X, ZERO, ZERO),
450 tmp,
451 0);
452
453 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
454 i915_emit_arith(p,
455 A0_DP3,
456 tmp, A0_DEST_CHANNEL_X, 0,
457 tmp,
458 swizzle(consts1, X, Y, ZERO, ZERO),
459 0);
460
461 /* tmp.x now contains a first approximation (y). Now, weight it
462 * against tmp.y**2 to get closer.
463 */
464 i915_emit_arith(p,
465 A0_MAX,
466 tmp, A0_DEST_CHANNEL_Y, 0,
467 swizzle(tmp, ZERO, X, ZERO, ZERO),
468 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
469 0);
470
471 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
472 i915_emit_arith(p,
473 A0_MAD,
474 tmp, A0_DEST_CHANNEL_Y, 0,
475 swizzle(tmp, ZERO, X, ZERO, ZERO),
476 swizzle(tmp, ZERO, Y, ZERO, ZERO),
477 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
478
479 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
480 i915_emit_arith(p,
481 A0_MAD,
482 get_result_vector(p, inst),
483 get_result_flags(inst), 0,
484 swizzle(consts1, W, W, W, W),
485 swizzle(tmp, Y, Y, Y, Y),
486 swizzle(tmp, X, X, X, X));
487 break;
488
489 case OPCODE_DP2:
490 src0 = src_vector(p, &inst->SrcReg[0], program);
491 src1 = src_vector(p, &inst->SrcReg[1], program);
492 i915_emit_arith(p,
493 A0_DP3,
494 get_result_vector(p, inst),
495 get_result_flags(inst), 0,
496 swizzle(src0, X, Y, ZERO, ZERO),
497 swizzle(src1, X, Y, ZERO, ZERO),
498 0);
499 break;
500
501 case OPCODE_DP3:
502 EMIT_2ARG_ARITH(A0_DP3);
503 break;
504
505 case OPCODE_DP4:
506 EMIT_2ARG_ARITH(A0_DP4);
507 break;
508
509 case OPCODE_DPH:
510 src0 = src_vector(p, &inst->SrcReg[0], program);
511 src1 = src_vector(p, &inst->SrcReg[1], program);
512
513 i915_emit_arith(p,
514 A0_DP4,
515 get_result_vector(p, inst),
516 get_result_flags(inst), 0,
517 swizzle(src0, X, Y, Z, ONE), src1, 0);
518 break;
519
520 case OPCODE_DST:
521 src0 = src_vector(p, &inst->SrcReg[0], program);
522 src1 = src_vector(p, &inst->SrcReg[1], program);
523
524 /* result[0] = 1 * 1;
525 * result[1] = a[1] * b[1];
526 * result[2] = a[2] * 1;
527 * result[3] = 1 * b[3];
528 */
529 i915_emit_arith(p,
530 A0_MUL,
531 get_result_vector(p, inst),
532 get_result_flags(inst), 0,
533 swizzle(src0, ONE, Y, Z, ONE),
534 swizzle(src1, ONE, Y, ONE, W), 0);
535 break;
536
537 case OPCODE_EX2:
538 src0 = src_vector(p, &inst->SrcReg[0], program);
539
540 i915_emit_arith(p,
541 A0_EXP,
542 get_result_vector(p, inst),
543 get_result_flags(inst), 0,
544 swizzle(src0, X, X, X, X), 0, 0);
545 break;
546
547 case OPCODE_FLR:
548 EMIT_1ARG_ARITH(A0_FLR);
549 break;
550
551 case OPCODE_TRUNC:
552 EMIT_1ARG_ARITH(A0_TRC);
553 break;
554
555 case OPCODE_FRC:
556 EMIT_1ARG_ARITH(A0_FRC);
557 break;
558
559 case OPCODE_KIL:
560 src0 = src_vector(p, &inst->SrcReg[0], program);
561 tmp = i915_get_utemp(p);
562
563 i915_emit_texld(p, get_live_regs(p, inst),
564 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
565 0, src0, T0_TEXKILL);
566 break;
567
568 case OPCODE_KIL_NV:
569 if (inst->DstReg.CondMask == COND_TR) {
570 tmp = i915_get_utemp(p);
571
572 i915_emit_texld(p, get_live_regs(p, inst),
573 tmp, A0_DEST_CHANNEL_ALL,
574 0, /* use a dummy dest reg */
575 swizzle(tmp, ONE, ONE, ONE, ONE), /* always */
576 T0_TEXKILL);
577 } else {
578 p->error = 1;
579 i915_program_error(p, "Unsupported KIL_NV condition code: %d",
580 inst->DstReg.CondMask);
581 }
582 break;
583
584 case OPCODE_LG2:
585 src0 = src_vector(p, &inst->SrcReg[0], program);
586
587 i915_emit_arith(p,
588 A0_LOG,
589 get_result_vector(p, inst),
590 get_result_flags(inst), 0,
591 swizzle(src0, X, X, X, X), 0, 0);
592 break;
593
594 case OPCODE_LIT:
595 src0 = src_vector(p, &inst->SrcReg[0], program);
596 tmp = i915_get_utemp(p);
597
598 /* tmp = max( a.xyzw, a.00zw )
599 * XXX: Clamp tmp.w to -128..128
600 * tmp.y = log(tmp.y)
601 * tmp.y = tmp.w * tmp.y
602 * tmp.y = exp(tmp.y)
603 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
604 */
605 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
606 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
607
608 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
609 swizzle(tmp, Y, Y, Y, Y), 0, 0);
610
611 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
612 swizzle(tmp, ZERO, Y, ZERO, ZERO),
613 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
614
615 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
616 swizzle(tmp, Y, Y, Y, Y), 0, 0);
617
618 i915_emit_arith(p, A0_CMP,
619 get_result_vector(p, inst),
620 get_result_flags(inst), 0,
621 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
622 swizzle(tmp, ONE, X, ZERO, ONE),
623 swizzle(tmp, ONE, X, Y, ONE));
624
625 break;
626
627 case OPCODE_LRP:
628 src0 = src_vector(p, &inst->SrcReg[0], program);
629 src1 = src_vector(p, &inst->SrcReg[1], program);
630 src2 = src_vector(p, &inst->SrcReg[2], program);
631 flags = get_result_flags(inst);
632 tmp = i915_get_utemp(p);
633
634 /* b*a + c*(1-a)
635 *
636 * b*a + c - ca
637 *
638 * tmp = b*a + c,
639 * result = (-c)*a + tmp
640 */
641 i915_emit_arith(p, A0_MAD, tmp,
642 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
643
644 i915_emit_arith(p, A0_MAD,
645 get_result_vector(p, inst),
646 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
647 break;
648
649 case OPCODE_MAD:
650 EMIT_3ARG_ARITH(A0_MAD);
651 break;
652
653 case OPCODE_MAX:
654 EMIT_2ARG_ARITH(A0_MAX);
655 break;
656
657 case OPCODE_MIN:
658 src0 = src_vector(p, &inst->SrcReg[0], program);
659 src1 = src_vector(p, &inst->SrcReg[1], program);
660 tmp = i915_get_utemp(p);
661 flags = get_result_flags(inst);
662
663 i915_emit_arith(p,
664 A0_MAX,
665 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
666 negate(src0, 1, 1, 1, 1),
667 negate(src1, 1, 1, 1, 1), 0);
668
669 i915_emit_arith(p,
670 A0_MOV,
671 get_result_vector(p, inst),
672 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
673 break;
674
675 case OPCODE_MOV:
676 EMIT_1ARG_ARITH(A0_MOV);
677 break;
678
679 case OPCODE_MUL:
680 EMIT_2ARG_ARITH(A0_MUL);
681 break;
682
683 case OPCODE_NOISE1:
684 case OPCODE_NOISE2:
685 case OPCODE_NOISE3:
686 case OPCODE_NOISE4:
687 /* Don't implement noise because we just don't have the instructions
688 * to spare. We aren't the first vendor to do so.
689 */
690 i915_program_error(p, "Stubbed-out noise functions");
691 i915_emit_arith(p,
692 A0_MOV,
693 get_result_vector(p, inst),
694 get_result_flags(inst), 0,
695 swizzle(tmp, ZERO, ZERO, ZERO, ZERO), 0, 0);
696 break;
697
698 case OPCODE_POW:
699 src0 = src_vector(p, &inst->SrcReg[0], program);
700 src1 = src_vector(p, &inst->SrcReg[1], program);
701 tmp = i915_get_utemp(p);
702 flags = get_result_flags(inst);
703
704 /* XXX: masking on intermediate values, here and elsewhere.
705 */
706 i915_emit_arith(p,
707 A0_LOG,
708 tmp, A0_DEST_CHANNEL_X, 0,
709 swizzle(src0, X, X, X, X), 0, 0);
710
711 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
712
713
714 i915_emit_arith(p,
715 A0_EXP,
716 get_result_vector(p, inst),
717 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
718
719 break;
720
721 case OPCODE_RCP:
722 src0 = src_vector(p, &inst->SrcReg[0], program);
723
724 i915_emit_arith(p,
725 A0_RCP,
726 get_result_vector(p, inst),
727 get_result_flags(inst), 0,
728 swizzle(src0, X, X, X, X), 0, 0);
729 break;
730
731 case OPCODE_RSQ:
732
733 src0 = src_vector(p, &inst->SrcReg[0], program);
734
735 i915_emit_arith(p,
736 A0_RSQ,
737 get_result_vector(p, inst),
738 get_result_flags(inst), 0,
739 swizzle(src0, X, X, X, X), 0, 0);
740 break;
741
742 case OPCODE_SCS:
743 src0 = src_vector(p, &inst->SrcReg[0], program);
744 tmp = i915_get_utemp(p);
745
746 /*
747 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
748 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
749 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
750 * scs.x = DP4 t1, sin_constants
751 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
752 * scs.y = DP4 t1, cos_constants
753 */
754 i915_emit_arith(p,
755 A0_MUL,
756 tmp, A0_DEST_CHANNEL_XY, 0,
757 swizzle(src0, X, X, ONE, ONE),
758 swizzle(src0, X, ONE, ONE, ONE), 0);
759
760 i915_emit_arith(p,
761 A0_MUL,
762 tmp, A0_DEST_CHANNEL_ALL, 0,
763 swizzle(tmp, X, Y, X, Y),
764 swizzle(tmp, X, X, ONE, ONE), 0);
765
766 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
767 GLuint tmp1;
768
769 if (inst->DstReg.WriteMask & WRITEMASK_X)
770 tmp1 = i915_get_utemp(p);
771 else
772 tmp1 = tmp;
773
774 i915_emit_arith(p,
775 A0_MUL,
776 tmp1, A0_DEST_CHANNEL_ALL, 0,
777 swizzle(tmp, X, Y, Y, W),
778 swizzle(tmp, X, Z, ONE, ONE), 0);
779
780 i915_emit_arith(p,
781 A0_DP4,
782 get_result_vector(p, inst),
783 A0_DEST_CHANNEL_Y, 0,
784 swizzle(tmp1, W, Z, Y, X),
785 i915_emit_const4fv(p, sin_constants), 0);
786 }
787
788 if (inst->DstReg.WriteMask & WRITEMASK_X) {
789 i915_emit_arith(p,
790 A0_MUL,
791 tmp, A0_DEST_CHANNEL_XYZ, 0,
792 swizzle(tmp, X, X, Z, ONE),
793 swizzle(tmp, Z, ONE, ONE, ONE), 0);
794
795 i915_emit_arith(p,
796 A0_DP4,
797 get_result_vector(p, inst),
798 A0_DEST_CHANNEL_X, 0,
799 swizzle(tmp, ONE, Z, Y, X),
800 i915_emit_const4fv(p, cos_constants), 0);
801 }
802 break;
803
804 case OPCODE_SEQ:
805 tmp = i915_get_utemp(p);
806 flags = get_result_flags(inst);
807 dst = get_result_vector(p, inst);
808
809 /* dst = src1 >= src2 */
810 i915_emit_arith(p,
811 A0_SGE,
812 dst,
813 flags, 0,
814 src_vector(p, &inst->SrcReg[0], program),
815 src_vector(p, &inst->SrcReg[1], program),
816 0);
817 /* tmp = src1 <= src2 */
818 i915_emit_arith(p,
819 A0_SGE,
820 tmp,
821 flags, 0,
822 negate(src_vector(p, &inst->SrcReg[0], program),
823 1, 1, 1, 1),
824 negate(src_vector(p, &inst->SrcReg[1], program),
825 1, 1, 1, 1),
826 0);
827 /* dst = tmp && dst */
828 i915_emit_arith(p,
829 A0_MUL,
830 dst,
831 flags, 0,
832 dst,
833 tmp,
834 0);
835 break;
836
837 case OPCODE_SIN:
838 src0 = src_vector(p, &inst->SrcReg[0], program);
839 tmp = i915_get_utemp(p);
840 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
841 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
842
843 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
844 i915_emit_arith(p,
845 A0_MAD,
846 tmp, A0_DEST_CHANNEL_X, 0,
847 src0,
848 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
849 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
850
851 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
852
853 i915_emit_arith(p,
854 A0_MAD,
855 tmp, A0_DEST_CHANNEL_X, 0,
856 tmp,
857 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
858 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
859
860 /* Compute sin using a quadratic and quartic. It gives continuity
861 * that repeating the Taylor series lacks every 2*pi, and has
862 * reduced error.
863 *
864 * The idea was described at:
865 * http://www.devmaster.net/forums/showthread.php?t=5784
866 */
867
868 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
869 i915_emit_arith(p,
870 A0_MAX,
871 tmp, A0_DEST_CHANNEL_Y, 0,
872 swizzle(tmp, ZERO, X, ZERO, ZERO),
873 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
874 0);
875
876 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
877 i915_emit_arith(p,
878 A0_MUL,
879 tmp, A0_DEST_CHANNEL_Y, 0,
880 swizzle(tmp, ZERO, X, ZERO, ZERO),
881 tmp,
882 0);
883
884 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
885 i915_emit_arith(p,
886 A0_DP3,
887 tmp, A0_DEST_CHANNEL_X, 0,
888 tmp,
889 swizzle(consts1, X, Y, ZERO, ZERO),
890 0);
891
892 /* tmp.x now contains a first approximation (y). Now, weight it
893 * against tmp.y**2 to get closer.
894 */
895 i915_emit_arith(p,
896 A0_MAX,
897 tmp, A0_DEST_CHANNEL_Y, 0,
898 swizzle(tmp, ZERO, X, ZERO, ZERO),
899 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
900 0);
901
902 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
903 i915_emit_arith(p,
904 A0_MAD,
905 tmp, A0_DEST_CHANNEL_Y, 0,
906 swizzle(tmp, ZERO, X, ZERO, ZERO),
907 swizzle(tmp, ZERO, Y, ZERO, ZERO),
908 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
909
910 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
911 i915_emit_arith(p,
912 A0_MAD,
913 get_result_vector(p, inst),
914 get_result_flags(inst), 0,
915 swizzle(consts1, W, W, W, W),
916 swizzle(tmp, Y, Y, Y, Y),
917 swizzle(tmp, X, X, X, X));
918
919 break;
920
921 case OPCODE_SGE:
922 EMIT_2ARG_ARITH(A0_SGE);
923 break;
924
925 case OPCODE_SGT:
926 i915_emit_arith(p,
927 A0_SLT,
928 get_result_vector( p, inst ),
929 get_result_flags( inst ), 0,
930 negate(src_vector( p, &inst->SrcReg[0], program),
931 1, 1, 1, 1),
932 negate(src_vector( p, &inst->SrcReg[1], program),
933 1, 1, 1, 1),
934 0);
935 break;
936
937 case OPCODE_SLE:
938 i915_emit_arith(p,
939 A0_SGE,
940 get_result_vector( p, inst ),
941 get_result_flags( inst ), 0,
942 negate(src_vector( p, &inst->SrcReg[0], program),
943 1, 1, 1, 1),
944 negate(src_vector( p, &inst->SrcReg[1], program),
945 1, 1, 1, 1),
946 0);
947 break;
948
949 case OPCODE_SLT:
950 EMIT_2ARG_ARITH(A0_SLT);
951 break;
952
953 case OPCODE_SNE:
954 tmp = i915_get_utemp(p);
955 flags = get_result_flags(inst);
956 dst = get_result_vector(p, inst);
957
958 /* dst = src1 < src2 */
959 i915_emit_arith(p,
960 A0_SLT,
961 dst,
962 flags, 0,
963 src_vector(p, &inst->SrcReg[0], program),
964 src_vector(p, &inst->SrcReg[1], program),
965 0);
966 /* tmp = src1 > src2 */
967 i915_emit_arith(p,
968 A0_SLT,
969 tmp,
970 flags, 0,
971 negate(src_vector(p, &inst->SrcReg[0], program),
972 1, 1, 1, 1),
973 negate(src_vector(p, &inst->SrcReg[1], program),
974 1, 1, 1, 1),
975 0);
976 /* dst = tmp || dst */
977 i915_emit_arith(p,
978 A0_ADD,
979 dst,
980 flags | A0_DEST_SATURATE, 0,
981 dst,
982 tmp,
983 0);
984 break;
985
986 case OPCODE_SSG:
987 dst = get_result_vector(p, inst);
988 flags = get_result_flags(inst);
989 src0 = src_vector(p, &inst->SrcReg[0], program);
990 tmp = i915_get_utemp(p);
991
992 /* tmp = (src < 0.0) */
993 i915_emit_arith(p,
994 A0_SLT,
995 tmp,
996 flags, 0,
997 src0,
998 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
999 0);
1000
1001 /* dst = (0.0 < src) */
1002 i915_emit_arith(p,
1003 A0_SLT,
1004 dst,
1005 flags, 0,
1006 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1007 src0,
1008 0);
1009
1010 /* dst = (src > 0.0) - (src < 0.0) */
1011 i915_emit_arith(p,
1012 A0_ADD,
1013 dst,
1014 flags, 0,
1015 dst,
1016 negate(tmp, 1, 1, 1, 1),
1017 0);
1018
1019 break;
1020
1021 case OPCODE_SUB:
1022 src0 = src_vector(p, &inst->SrcReg[0], program);
1023 src1 = src_vector(p, &inst->SrcReg[1], program);
1024
1025 i915_emit_arith(p,
1026 A0_ADD,
1027 get_result_vector(p, inst),
1028 get_result_flags(inst), 0,
1029 src0, negate(src1, 1, 1, 1, 1), 0);
1030 break;
1031
1032 case OPCODE_SWZ:
1033 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
1034 break;
1035
1036 case OPCODE_TEX:
1037 EMIT_TEX(T0_TEXLD);
1038 break;
1039
1040 case OPCODE_TXB:
1041 EMIT_TEX(T0_TEXLDB);
1042 break;
1043
1044 case OPCODE_TXP:
1045 EMIT_TEX(T0_TEXLDP);
1046 break;
1047
1048 case OPCODE_XPD:
1049 /* Cross product:
1050 * result.x = src0.y * src1.z - src0.z * src1.y;
1051 * result.y = src0.z * src1.x - src0.x * src1.z;
1052 * result.z = src0.x * src1.y - src0.y * src1.x;
1053 * result.w = undef;
1054 */
1055 src0 = src_vector(p, &inst->SrcReg[0], program);
1056 src1 = src_vector(p, &inst->SrcReg[1], program);
1057 tmp = i915_get_utemp(p);
1058
1059 i915_emit_arith(p,
1060 A0_MUL,
1061 tmp, A0_DEST_CHANNEL_ALL, 0,
1062 swizzle(src0, Z, X, Y, ONE),
1063 swizzle(src1, Y, Z, X, ONE), 0);
1064
1065 i915_emit_arith(p,
1066 A0_MAD,
1067 get_result_vector(p, inst),
1068 get_result_flags(inst), 0,
1069 swizzle(src0, Y, Z, X, ONE),
1070 swizzle(src1, Z, X, Y, ONE),
1071 negate(tmp, 1, 1, 1, 0));
1072 break;
1073
1074 case OPCODE_END:
1075 return;
1076
1077 case OPCODE_BGNLOOP:
1078 case OPCODE_BGNSUB:
1079 case OPCODE_BRA:
1080 case OPCODE_BRK:
1081 case OPCODE_CAL:
1082 case OPCODE_CONT:
1083 case OPCODE_DDX:
1084 case OPCODE_DDY:
1085 case OPCODE_ELSE:
1086 case OPCODE_ENDIF:
1087 case OPCODE_ENDLOOP:
1088 case OPCODE_ENDSUB:
1089 case OPCODE_IF:
1090 case OPCODE_RET:
1091 p->error = 1;
1092 i915_program_error(p, "Unsupported opcode: %s",
1093 _mesa_opcode_string(inst->Opcode));
1094 return;
1095
1096 case OPCODE_EXP:
1097 case OPCODE_LOG:
1098 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1099 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1100 * Instead, it translates to EX2 or LG2.
1101 */
1102 case OPCODE_TXD:
1103 case OPCODE_TXL:
1104 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1105 * only NV_vp/fp appears to emit them.
1106 */
1107 default:
1108 i915_program_error(p, "bad opcode: %s",
1109 _mesa_opcode_string(inst->Opcode));
1110 return;
1111 }
1112
1113 inst++;
1114 i915_release_utemps(p);
1115 }
1116 }
1117
1118 /* Rather than trying to intercept and jiggle depth writes during
1119 * emit, just move the value into its correct position at the end of
1120 * the program:
1121 */
1122 static void
1123 fixup_depth_write(struct i915_fragment_program *p)
1124 {
1125 if (p->depth_written) {
1126 GLuint depth = UREG(REG_TYPE_OD, 0);
1127
1128 i915_emit_arith(p,
1129 A0_MOV,
1130 depth, A0_DEST_CHANNEL_W, 0,
1131 swizzle(depth, X, Y, Z, Z), 0, 0);
1132 }
1133 }
1134
1135
1136 static void
1137 check_wpos(struct i915_fragment_program *p)
1138 {
1139 GLuint inputs = p->FragProg.Base.InputsRead;
1140 GLint i;
1141
1142 p->wpos_tex = -1;
1143
1144 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1145 if (inputs & (FRAG_BIT_TEX(i) | FRAG_BIT_VAR(i)))
1146 continue;
1147 else if (inputs & FRAG_BIT_WPOS) {
1148 p->wpos_tex = i;
1149 inputs &= ~FRAG_BIT_WPOS;
1150 }
1151 }
1152
1153 if (inputs & FRAG_BIT_WPOS) {
1154 i915_program_error(p, "No free texcoord for wpos value");
1155 }
1156 }
1157
1158
1159 static void
1160 translate_program(struct i915_fragment_program *p)
1161 {
1162 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1163
1164 if (INTEL_DEBUG & DEBUG_WM) {
1165 printf("fp:\n");
1166 _mesa_print_program(&p->ctx->FragmentProgram._Current->Base);
1167 printf("\n");
1168 }
1169
1170 i915_init_program(i915, p);
1171 check_wpos(p);
1172 upload_program(p);
1173 fixup_depth_write(p);
1174 i915_fini_program(p);
1175
1176 if (INTEL_DEBUG & DEBUG_WM) {
1177 printf("i915:\n");
1178 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1179 }
1180
1181 p->translated = 1;
1182 }
1183
1184
1185 static void
1186 track_params(struct i915_fragment_program *p)
1187 {
1188 GLint i;
1189
1190 if (p->nr_params)
1191 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1192
1193 for (i = 0; i < p->nr_params; i++) {
1194 GLint reg = p->param[i].reg;
1195 COPY_4V(p->constant[reg], p->param[i].values);
1196 }
1197
1198 p->params_uptodate = 1;
1199 p->on_hardware = 0; /* overkill */
1200 }
1201
1202
1203 static void
1204 i915BindProgram(GLcontext * ctx, GLenum target, struct gl_program *prog)
1205 {
1206 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1207 struct i915_context *i915 = I915_CONTEXT(ctx);
1208 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1209
1210 if (i915->current_program == p)
1211 return;
1212
1213 if (i915->current_program) {
1214 i915->current_program->on_hardware = 0;
1215 i915->current_program->params_uptodate = 0;
1216 }
1217
1218 i915->current_program = p;
1219
1220 assert(p->on_hardware == 0);
1221 assert(p->params_uptodate == 0);
1222
1223 }
1224 }
1225
1226 static struct gl_program *
1227 i915NewProgram(GLcontext * ctx, GLenum target, GLuint id)
1228 {
1229 switch (target) {
1230 case GL_VERTEX_PROGRAM_ARB:
1231 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1232 target, id);
1233
1234 case GL_FRAGMENT_PROGRAM_ARB:{
1235 struct i915_fragment_program *prog =
1236 CALLOC_STRUCT(i915_fragment_program);
1237 if (prog) {
1238 i915_init_program(I915_CONTEXT(ctx), prog);
1239
1240 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1241 target, id);
1242 }
1243 else
1244 return NULL;
1245 }
1246
1247 default:
1248 /* Just fallback:
1249 */
1250 return _mesa_new_program(ctx, target, id);
1251 }
1252 }
1253
1254 static void
1255 i915DeleteProgram(GLcontext * ctx, struct gl_program *prog)
1256 {
1257 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1258 struct i915_context *i915 = I915_CONTEXT(ctx);
1259 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1260
1261 if (i915->current_program == p)
1262 i915->current_program = 0;
1263 }
1264
1265 _mesa_delete_program(ctx, prog);
1266 }
1267
1268
1269 static GLboolean
1270 i915IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
1271 {
1272 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1273 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1274
1275 if (!p->translated)
1276 translate_program(p);
1277
1278 return !p->error;
1279 }
1280 else
1281 return GL_TRUE;
1282 }
1283
1284 static GLboolean
1285 i915ProgramStringNotify(GLcontext * ctx,
1286 GLenum target, struct gl_program *prog)
1287 {
1288 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1289 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1290 p->translated = 0;
1291
1292 /* Hack: make sure fog is correctly enabled according to this
1293 * fragment program's fog options.
1294 */
1295 if (p->FragProg.FogOption) {
1296 /* add extra instructions to do fog, then turn off FogOption field */
1297 _mesa_append_fog_code(ctx, &p->FragProg);
1298 p->FragProg.FogOption = GL_NONE;
1299 }
1300 }
1301
1302 (void) _tnl_program_string(ctx, target, prog);
1303
1304 /* XXX check if program is legal, within limits */
1305 return GL_TRUE;
1306 }
1307
1308 void
1309 i915_update_program(GLcontext *ctx)
1310 {
1311 struct intel_context *intel = intel_context(ctx);
1312 struct i915_context *i915 = i915_context(&intel->ctx);
1313 struct i915_fragment_program *fp =
1314 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1315
1316 if (i915->current_program != fp) {
1317 if (i915->current_program) {
1318 i915->current_program->on_hardware = 0;
1319 i915->current_program->params_uptodate = 0;
1320 }
1321
1322 i915->current_program = fp;
1323 }
1324
1325 if (!fp->translated)
1326 translate_program(fp);
1327
1328 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1329 }
1330
1331 void
1332 i915ValidateFragmentProgram(struct i915_context *i915)
1333 {
1334 GLcontext *ctx = &i915->intel.ctx;
1335 struct intel_context *intel = intel_context(ctx);
1336 TNLcontext *tnl = TNL_CONTEXT(ctx);
1337 struct vertex_buffer *VB = &tnl->vb;
1338
1339 struct i915_fragment_program *p =
1340 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1341
1342 const GLuint inputsRead = p->FragProg.Base.InputsRead;
1343 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1344 GLuint s2 = S2_TEXCOORD_NONE;
1345 int i, offset = 0;
1346
1347 /* Important:
1348 */
1349 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1350
1351 if (!p->translated)
1352 translate_program(p);
1353
1354 intel->vertex_attr_count = 0;
1355 intel->wpos_offset = 0;
1356 intel->wpos_size = 0;
1357 intel->coloroffset = 0;
1358 intel->specoffset = 0;
1359
1360 if (inputsRead & FRAG_BITS_TEX_ANY) {
1361 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1362 }
1363 else {
1364 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1365 }
1366
1367 if (inputsRead & FRAG_BIT_COL0) {
1368 intel->coloroffset = offset / 4;
1369 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1370 }
1371
1372 if (inputsRead & FRAG_BIT_COL1) {
1373 intel->specoffset = offset / 4;
1374 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1375 }
1376
1377 if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE) {
1378 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1379 }
1380
1381 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1382 if (inputsRead & FRAG_BIT_TEX(i)) {
1383 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1384
1385 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1386 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1387
1388 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1389 }
1390 else if (inputsRead & FRAG_BIT_VAR(i)) {
1391 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1392
1393 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1394 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1395
1396 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1397 }
1398 else if (i == p->wpos_tex) {
1399
1400 /* If WPOS is required, duplicate the XYZ position data in an
1401 * unused texture coordinate:
1402 */
1403 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1404 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1405
1406 intel->wpos_offset = offset;
1407 intel->wpos_size = 3 * sizeof(GLuint);
1408
1409 EMIT_PAD(intel->wpos_size);
1410 }
1411 }
1412
1413 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1414 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1415 int k;
1416
1417 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1418
1419 /* Must do this *after* statechange, so as not to affect
1420 * buffered vertices reliant on the old state:
1421 */
1422 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1423 intel->vertex_attrs,
1424 intel->vertex_attr_count,
1425 intel->ViewportMatrix.m, 0);
1426
1427 intel->vertex_size >>= 2;
1428
1429 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1430 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1431
1432 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1433 assert(k);
1434 }
1435
1436 if (!p->params_uptodate)
1437 track_params(p);
1438
1439 if (!p->on_hardware)
1440 i915_upload_program(i915, p);
1441 }
1442
1443 void
1444 i915InitFragProgFuncs(struct dd_function_table *functions)
1445 {
1446 functions->BindProgram = i915BindProgram;
1447 functions->NewProgram = i915NewProgram;
1448 functions->DeleteProgram = i915DeleteProgram;
1449 functions->IsProgramNative = i915IsProgramNative;
1450 functions->ProgramStringNotify = i915ProgramStringNotify;
1451 }