projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge remote branch 'origin/7.8'
[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 "shader/prog_instruction.h"
33 #include "shader/prog_parameter.h"
34 #include "shader/program.h"
35 #include "shader/programopt.h"
36 #include "shader/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 /* Various paramters and env values. All emitted to
147 * hardware as program constants.
148 */
149 case PROGRAM_LOCAL_PARAM:
150 src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
151 break;
152
153 case PROGRAM_ENV_PARAM:
154 src =
155 i915_emit_param4fv(p,
156 p->ctx->FragmentProgram.Parameters[source->
157 Index]);
158 break;
159
160 case PROGRAM_CONSTANT:
161 case PROGRAM_STATE_VAR:
162 case PROGRAM_NAMED_PARAM:
163 case PROGRAM_UNIFORM:
164 src =
165 i915_emit_param4fv(p,
166 program->Base.Parameters->ParameterValues[source->
167 Index]);
168 break;
169
170 default:
171 i915_program_error(p, "Bad source->File: %d", source->File);
172 return 0;
173 }
174
175 src = swizzle(src,
176 GET_SWZ(source->Swizzle, 0),
177 GET_SWZ(source->Swizzle, 1),
178 GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
179
180 if (source->Negate)
181 src = negate(src,
182 GET_BIT(source->Negate, 0),
183 GET_BIT(source->Negate, 1),
184 GET_BIT(source->Negate, 2),
185 GET_BIT(source->Negate, 3));
186
187 return src;
188 }
189
190
191 static GLuint
192 get_result_vector(struct i915_fragment_program *p,
193 const struct prog_instruction *inst)
194 {
195 switch (inst->DstReg.File) {
196 case PROGRAM_OUTPUT:
197 switch (inst->DstReg.Index) {
198 case FRAG_RESULT_COLOR:
199 return UREG(REG_TYPE_OC, 0);
200 case FRAG_RESULT_DEPTH:
201 p->depth_written = 1;
202 return UREG(REG_TYPE_OD, 0);
203 default:
204 i915_program_error(p, "Bad inst->DstReg.Index: %d",
205 inst->DstReg.Index);
206 return 0;
207 }
208 case PROGRAM_TEMPORARY:
209 return UREG(REG_TYPE_R, inst->DstReg.Index);
210 default:
211 i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
212 return 0;
213 }
214 }
215
216 static GLuint
217 get_result_flags(const struct prog_instruction *inst)
218 {
219 GLuint flags = 0;
220
221 if (inst->SaturateMode == SATURATE_ZERO_ONE)
222 flags |= A0_DEST_SATURATE;
223 if (inst->DstReg.WriteMask & WRITEMASK_X)
224 flags |= A0_DEST_CHANNEL_X;
225 if (inst->DstReg.WriteMask & WRITEMASK_Y)
226 flags |= A0_DEST_CHANNEL_Y;
227 if (inst->DstReg.WriteMask & WRITEMASK_Z)
228 flags |= A0_DEST_CHANNEL_Z;
229 if (inst->DstReg.WriteMask & WRITEMASK_W)
230 flags |= A0_DEST_CHANNEL_W;
231
232 return flags;
233 }
234
235 static GLuint
236 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
237 {
238 switch (bit) {
239 case TEXTURE_1D_INDEX:
240 return D0_SAMPLE_TYPE_2D;
241 case TEXTURE_2D_INDEX:
242 return D0_SAMPLE_TYPE_2D;
243 case TEXTURE_RECT_INDEX:
244 return D0_SAMPLE_TYPE_2D;
245 case TEXTURE_3D_INDEX:
246 return D0_SAMPLE_TYPE_VOLUME;
247 case TEXTURE_CUBE_INDEX:
248 return D0_SAMPLE_TYPE_CUBE;
249 default:
250 i915_program_error(p, "TexSrcBit: %d", bit);
251 return 0;
252 }
253 }
254
255 #define EMIT_TEX( OP ) \
256 do { \
257 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
258 const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current; \
259 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
260 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
261 unit, dim); \
262 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
263 /* Texel lookup */ \
264 \
265 i915_emit_texld( p, get_live_regs(p, inst), \
266 get_result_vector( p, inst ), \
267 get_result_flags( inst ), \
268 sampler, \
269 coord, \
270 OP); \
271 } while (0)
272
273 #define EMIT_ARITH( OP, N ) \
274 do { \
275 i915_emit_arith( p, \
276 OP, \
277 get_result_vector( p, inst ), \
278 get_result_flags( inst ), 0, \
279 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
280 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
281 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
282 } while (0)
283
284 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
285 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
286 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
287
288 /*
289 * TODO: consider moving this into core
290 */
291 static void calc_live_regs( struct i915_fragment_program *p )
292 {
293 const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
294 GLuint regsUsed = 0xffff0000;
295 GLint i;
296
297 for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
298 struct prog_instruction *inst = &program->Base.Instructions[i];
299 int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
300 int a;
301
302 /* Register is written to: unmark as live for this and preceeding ops */
303 if (inst->DstReg.File == PROGRAM_TEMPORARY)
304 regsUsed &= ~(1 << inst->DstReg.Index);
305
306 for (a = 0; a < opArgs; a++) {
307 /* Register is read from: mark as live for this and preceeding ops */
308 if (inst->SrcReg[a].File == PROGRAM_TEMPORARY)
309 regsUsed |= 1 << inst->SrcReg[a].Index;
310 }
311
312 p->usedRegs[i] = regsUsed;
313 }
314 }
315
316 static GLuint get_live_regs( struct i915_fragment_program *p,
317 const struct prog_instruction *inst )
318 {
319 const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current;
320 GLuint nr = inst - program->Base.Instructions;
321
322 return p->usedRegs[nr];
323 }
324
325
326 /* Possible concerns:
327 *
328 * SIN, COS -- could use another taylor step?
329 * LIT -- results seem a little different to sw mesa
330 * LOG -- different to mesa on negative numbers, but this is conformant.
331 *
332 * Parse failures -- Mesa doesn't currently give a good indication
333 * internally whether a particular program string parsed or not. This
334 * can lead to confusion -- hopefully we cope with it ok now.
335 *
336 */
337 static void
338 upload_program(struct i915_fragment_program *p)
339 {
340 const struct gl_fragment_program *program =
341 p->ctx->FragmentProgram._Current;
342 const struct prog_instruction *inst = program->Base.Instructions;
343
344 if (INTEL_DEBUG & DEBUG_WM)
345 _mesa_print_program(&program->Base);
346
347 /* Is this a parse-failed program? Ensure a valid program is
348 * loaded, as the flagging of an error isn't sufficient to stop
349 * this being uploaded to hardware.
350 */
351 if (inst[0].Opcode == OPCODE_END) {
352 GLuint tmp = i915_get_utemp(p);
353 i915_emit_arith(p,
354 A0_MOV,
355 UREG(REG_TYPE_OC, 0),
356 A0_DEST_CHANNEL_ALL, 0,
357 swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
358 return;
359 }
360
361 if (program->Base.NumInstructions > I915_MAX_INSN) {
362 i915_program_error( p, "Exceeded max instructions" );
363 return;
364 }
365
366 /* Not always needed:
367 */
368 calc_live_regs(p);
369
370 while (1) {
371 GLuint src0, src1, src2, flags;
372 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
373
374 switch (inst->Opcode) {
375 case OPCODE_ABS:
376 src0 = src_vector(p, &inst->SrcReg[0], program);
377 i915_emit_arith(p,
378 A0_MAX,
379 get_result_vector(p, inst),
380 get_result_flags(inst), 0,
381 src0, negate(src0, 1, 1, 1, 1), 0);
382 break;
383
384 case OPCODE_ADD:
385 EMIT_2ARG_ARITH(A0_ADD);
386 break;
387
388 case OPCODE_CMP:
389 src0 = src_vector(p, &inst->SrcReg[0], program);
390 src1 = src_vector(p, &inst->SrcReg[1], program);
391 src2 = src_vector(p, &inst->SrcReg[2], program);
392 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
393 break;
394
395 case OPCODE_COS:
396 src0 = src_vector(p, &inst->SrcReg[0], program);
397 tmp = i915_get_utemp(p);
398 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
399 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
400
401 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
402 i915_emit_arith(p,
403 A0_MAD,
404 tmp, A0_DEST_CHANNEL_X, 0,
405 src0,
406 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
407 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
408
409 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
410
411 i915_emit_arith(p,
412 A0_MAD,
413 tmp, A0_DEST_CHANNEL_X, 0,
414 tmp,
415 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
416 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
417
418 /* Compute COS with the same calculation used for SIN, but a
419 * different source range has been mapped to [-1,1] this time.
420 */
421
422 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
423 i915_emit_arith(p,
424 A0_MAX,
425 tmp, A0_DEST_CHANNEL_Y, 0,
426 swizzle(tmp, ZERO, X, ZERO, ZERO),
427 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
428 0);
429
430 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
431 i915_emit_arith(p,
432 A0_MUL,
433 tmp, A0_DEST_CHANNEL_Y, 0,
434 swizzle(tmp, ZERO, X, ZERO, ZERO),
435 tmp,
436 0);
437
438 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
439 i915_emit_arith(p,
440 A0_DP3,
441 tmp, A0_DEST_CHANNEL_X, 0,
442 tmp,
443 swizzle(consts1, X, Y, ZERO, ZERO),
444 0);
445
446 /* tmp.x now contains a first approximation (y). Now, weight it
447 * against tmp.y**2 to get closer.
448 */
449 i915_emit_arith(p,
450 A0_MAX,
451 tmp, A0_DEST_CHANNEL_Y, 0,
452 swizzle(tmp, ZERO, X, ZERO, ZERO),
453 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
454 0);
455
456 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
457 i915_emit_arith(p,
458 A0_MAD,
459 tmp, A0_DEST_CHANNEL_Y, 0,
460 swizzle(tmp, ZERO, X, ZERO, ZERO),
461 swizzle(tmp, ZERO, Y, ZERO, ZERO),
462 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
463
464 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
465 i915_emit_arith(p,
466 A0_MAD,
467 get_result_vector(p, inst),
468 get_result_flags(inst), 0,
469 swizzle(consts1, W, W, W, W),
470 swizzle(tmp, Y, Y, Y, Y),
471 swizzle(tmp, X, X, X, X));
472 break;
473
474 case OPCODE_DP3:
475 EMIT_2ARG_ARITH(A0_DP3);
476 break;
477
478 case OPCODE_DP4:
479 EMIT_2ARG_ARITH(A0_DP4);
480 break;
481
482 case OPCODE_DPH:
483 src0 = src_vector(p, &inst->SrcReg[0], program);
484 src1 = src_vector(p, &inst->SrcReg[1], program);
485
486 i915_emit_arith(p,
487 A0_DP4,
488 get_result_vector(p, inst),
489 get_result_flags(inst), 0,
490 swizzle(src0, X, Y, Z, ONE), src1, 0);
491 break;
492
493 case OPCODE_DST:
494 src0 = src_vector(p, &inst->SrcReg[0], program);
495 src1 = src_vector(p, &inst->SrcReg[1], program);
496
497 /* result[0] = 1 * 1;
498 * result[1] = a[1] * b[1];
499 * result[2] = a[2] * 1;
500 * result[3] = 1 * b[3];
501 */
502 i915_emit_arith(p,
503 A0_MUL,
504 get_result_vector(p, inst),
505 get_result_flags(inst), 0,
506 swizzle(src0, ONE, Y, Z, ONE),
507 swizzle(src1, ONE, Y, ONE, W), 0);
508 break;
509
510 case OPCODE_EX2:
511 src0 = src_vector(p, &inst->SrcReg[0], program);
512
513 i915_emit_arith(p,
514 A0_EXP,
515 get_result_vector(p, inst),
516 get_result_flags(inst), 0,
517 swizzle(src0, X, X, X, X), 0, 0);
518 break;
519
520 case OPCODE_FLR:
521 EMIT_1ARG_ARITH(A0_FLR);
522 break;
523
524 case OPCODE_TRUNC:
525 EMIT_1ARG_ARITH(A0_TRC);
526 break;
527
528 case OPCODE_FRC:
529 EMIT_1ARG_ARITH(A0_FRC);
530 break;
531
532 case OPCODE_KIL:
533 src0 = src_vector(p, &inst->SrcReg[0], program);
534 tmp = i915_get_utemp(p);
535
536 i915_emit_texld(p, get_live_regs(p, inst),
537 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
538 0, src0, T0_TEXKILL);
539 break;
540
541 case OPCODE_KIL_NV:
542 if (inst->DstReg.CondMask == COND_TR) {
543 tmp = i915_get_utemp(p);
544
545 i915_emit_texld(p, get_live_regs(p, inst),
546 tmp, A0_DEST_CHANNEL_ALL,
547 0, /* use a dummy dest reg */
548 swizzle(tmp, ONE, ONE, ONE, ONE), /* always */
549 T0_TEXKILL);
550 } else {
551 p->error = 1;
552 i915_program_error(p, "Unsupported KIL_NV condition code: %d",
553 inst->DstReg.CondMask);
554 }
555 break;
556
557 case OPCODE_LG2:
558 src0 = src_vector(p, &inst->SrcReg[0], program);
559
560 i915_emit_arith(p,
561 A0_LOG,
562 get_result_vector(p, inst),
563 get_result_flags(inst), 0,
564 swizzle(src0, X, X, X, X), 0, 0);
565 break;
566
567 case OPCODE_LIT:
568 src0 = src_vector(p, &inst->SrcReg[0], program);
569 tmp = i915_get_utemp(p);
570
571 /* tmp = max( a.xyzw, a.00zw )
572 * XXX: Clamp tmp.w to -128..128
573 * tmp.y = log(tmp.y)
574 * tmp.y = tmp.w * tmp.y
575 * tmp.y = exp(tmp.y)
576 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
577 */
578 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
579 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
580
581 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
582 swizzle(tmp, Y, Y, Y, Y), 0, 0);
583
584 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
585 swizzle(tmp, ZERO, Y, ZERO, ZERO),
586 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
587
588 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
589 swizzle(tmp, Y, Y, Y, Y), 0, 0);
590
591 i915_emit_arith(p, A0_CMP,
592 get_result_vector(p, inst),
593 get_result_flags(inst), 0,
594 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
595 swizzle(tmp, ONE, X, ZERO, ONE),
596 swizzle(tmp, ONE, X, Y, ONE));
597
598 break;
599
600 case OPCODE_LRP:
601 src0 = src_vector(p, &inst->SrcReg[0], program);
602 src1 = src_vector(p, &inst->SrcReg[1], program);
603 src2 = src_vector(p, &inst->SrcReg[2], program);
604 flags = get_result_flags(inst);
605 tmp = i915_get_utemp(p);
606
607 /* b*a + c*(1-a)
608 *
609 * b*a + c - ca
610 *
611 * tmp = b*a + c,
612 * result = (-c)*a + tmp
613 */
614 i915_emit_arith(p, A0_MAD, tmp,
615 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
616
617 i915_emit_arith(p, A0_MAD,
618 get_result_vector(p, inst),
619 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
620 break;
621
622 case OPCODE_MAD:
623 EMIT_3ARG_ARITH(A0_MAD);
624 break;
625
626 case OPCODE_MAX:
627 EMIT_2ARG_ARITH(A0_MAX);
628 break;
629
630 case OPCODE_MIN:
631 src0 = src_vector(p, &inst->SrcReg[0], program);
632 src1 = src_vector(p, &inst->SrcReg[1], program);
633 tmp = i915_get_utemp(p);
634 flags = get_result_flags(inst);
635
636 i915_emit_arith(p,
637 A0_MAX,
638 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
639 negate(src0, 1, 1, 1, 1),
640 negate(src1, 1, 1, 1, 1), 0);
641
642 i915_emit_arith(p,
643 A0_MOV,
644 get_result_vector(p, inst),
645 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
646 break;
647
648 case OPCODE_MOV:
649 EMIT_1ARG_ARITH(A0_MOV);
650 break;
651
652 case OPCODE_MUL:
653 EMIT_2ARG_ARITH(A0_MUL);
654 break;
655
656 case OPCODE_NOISE1:
657 case OPCODE_NOISE2:
658 case OPCODE_NOISE3:
659 case OPCODE_NOISE4:
660 /* Don't implement noise because we just don't have the instructions
661 * to spare. We aren't the first vendor to do so.
662 */
663 i915_program_error(p, "Stubbed-out noise functions");
664 i915_emit_arith(p,
665 A0_MOV,
666 get_result_vector(p, inst),
667 get_result_flags(inst), 0,
668 swizzle(tmp, ZERO, ZERO, ZERO, ZERO), 0, 0);
669 break;
670
671 case OPCODE_POW:
672 src0 = src_vector(p, &inst->SrcReg[0], program);
673 src1 = src_vector(p, &inst->SrcReg[1], program);
674 tmp = i915_get_utemp(p);
675 flags = get_result_flags(inst);
676
677 /* XXX: masking on intermediate values, here and elsewhere.
678 */
679 i915_emit_arith(p,
680 A0_LOG,
681 tmp, A0_DEST_CHANNEL_X, 0,
682 swizzle(src0, X, X, X, X), 0, 0);
683
684 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
685
686
687 i915_emit_arith(p,
688 A0_EXP,
689 get_result_vector(p, inst),
690 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
691
692 break;
693
694 case OPCODE_RCP:
695 src0 = src_vector(p, &inst->SrcReg[0], program);
696
697 i915_emit_arith(p,
698 A0_RCP,
699 get_result_vector(p, inst),
700 get_result_flags(inst), 0,
701 swizzle(src0, X, X, X, X), 0, 0);
702 break;
703
704 case OPCODE_RSQ:
705
706 src0 = src_vector(p, &inst->SrcReg[0], program);
707
708 i915_emit_arith(p,
709 A0_RSQ,
710 get_result_vector(p, inst),
711 get_result_flags(inst), 0,
712 swizzle(src0, X, X, X, X), 0, 0);
713 break;
714
715 case OPCODE_SCS:
716 src0 = src_vector(p, &inst->SrcReg[0], program);
717 tmp = i915_get_utemp(p);
718
719 /*
720 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
721 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
722 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
723 * scs.x = DP4 t1, sin_constants
724 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
725 * scs.y = DP4 t1, cos_constants
726 */
727 i915_emit_arith(p,
728 A0_MUL,
729 tmp, A0_DEST_CHANNEL_XY, 0,
730 swizzle(src0, X, X, ONE, ONE),
731 swizzle(src0, X, ONE, ONE, ONE), 0);
732
733 i915_emit_arith(p,
734 A0_MUL,
735 tmp, A0_DEST_CHANNEL_ALL, 0,
736 swizzle(tmp, X, Y, X, Y),
737 swizzle(tmp, X, X, ONE, ONE), 0);
738
739 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
740 GLuint tmp1;
741
742 if (inst->DstReg.WriteMask & WRITEMASK_X)
743 tmp1 = i915_get_utemp(p);
744 else
745 tmp1 = tmp;
746
747 i915_emit_arith(p,
748 A0_MUL,
749 tmp1, A0_DEST_CHANNEL_ALL, 0,
750 swizzle(tmp, X, Y, Y, W),
751 swizzle(tmp, X, Z, ONE, ONE), 0);
752
753 i915_emit_arith(p,
754 A0_DP4,
755 get_result_vector(p, inst),
756 A0_DEST_CHANNEL_Y, 0,
757 swizzle(tmp1, W, Z, Y, X),
758 i915_emit_const4fv(p, sin_constants), 0);
759 }
760
761 if (inst->DstReg.WriteMask & WRITEMASK_X) {
762 i915_emit_arith(p,
763 A0_MUL,
764 tmp, A0_DEST_CHANNEL_XYZ, 0,
765 swizzle(tmp, X, X, Z, ONE),
766 swizzle(tmp, Z, ONE, ONE, ONE), 0);
767
768 i915_emit_arith(p,
769 A0_DP4,
770 get_result_vector(p, inst),
771 A0_DEST_CHANNEL_X, 0,
772 swizzle(tmp, ONE, Z, Y, X),
773 i915_emit_const4fv(p, cos_constants), 0);
774 }
775 break;
776
777 case OPCODE_SEQ:
778 tmp = i915_get_utemp(p);
779 flags = get_result_flags(inst);
780 dst = get_result_vector(p, inst);
781
782 /* dst = src1 >= src2 */
783 i915_emit_arith(p,
784 A0_SGE,
785 dst,
786 flags, 0,
787 src_vector(p, &inst->SrcReg[0], program),
788 src_vector(p, &inst->SrcReg[1], program),
789 0);
790 /* tmp = src1 <= src2 */
791 i915_emit_arith(p,
792 A0_SGE,
793 tmp,
794 flags, 0,
795 negate(src_vector(p, &inst->SrcReg[0], program),
796 1, 1, 1, 1),
797 negate(src_vector(p, &inst->SrcReg[1], program),
798 1, 1, 1, 1),
799 0);
800 /* dst = tmp && dst */
801 i915_emit_arith(p,
802 A0_MUL,
803 dst,
804 flags, 0,
805 dst,
806 tmp,
807 0);
808 break;
809
810 case OPCODE_SIN:
811 src0 = src_vector(p, &inst->SrcReg[0], program);
812 tmp = i915_get_utemp(p);
813 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
814 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
815
816 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
817 i915_emit_arith(p,
818 A0_MAD,
819 tmp, A0_DEST_CHANNEL_X, 0,
820 src0,
821 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
822 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
823
824 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
825
826 i915_emit_arith(p,
827 A0_MAD,
828 tmp, A0_DEST_CHANNEL_X, 0,
829 tmp,
830 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
831 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
832
833 /* Compute sin using a quadratic and quartic. It gives continuity
834 * that repeating the Taylor series lacks every 2*pi, and has
835 * reduced error.
836 *
837 * The idea was described at:
838 * http://www.devmaster.net/forums/showthread.php?t=5784
839 */
840
841 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
842 i915_emit_arith(p,
843 A0_MAX,
844 tmp, A0_DEST_CHANNEL_Y, 0,
845 swizzle(tmp, ZERO, X, ZERO, ZERO),
846 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
847 0);
848
849 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
850 i915_emit_arith(p,
851 A0_MUL,
852 tmp, A0_DEST_CHANNEL_Y, 0,
853 swizzle(tmp, ZERO, X, ZERO, ZERO),
854 tmp,
855 0);
856
857 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
858 i915_emit_arith(p,
859 A0_DP3,
860 tmp, A0_DEST_CHANNEL_X, 0,
861 tmp,
862 swizzle(consts1, X, Y, ZERO, ZERO),
863 0);
864
865 /* tmp.x now contains a first approximation (y). Now, weight it
866 * against tmp.y**2 to get closer.
867 */
868 i915_emit_arith(p,
869 A0_MAX,
870 tmp, A0_DEST_CHANNEL_Y, 0,
871 swizzle(tmp, ZERO, X, ZERO, ZERO),
872 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
873 0);
874
875 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
876 i915_emit_arith(p,
877 A0_MAD,
878 tmp, A0_DEST_CHANNEL_Y, 0,
879 swizzle(tmp, ZERO, X, ZERO, ZERO),
880 swizzle(tmp, ZERO, Y, ZERO, ZERO),
881 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
882
883 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
884 i915_emit_arith(p,
885 A0_MAD,
886 get_result_vector(p, inst),
887 get_result_flags(inst), 0,
888 swizzle(consts1, W, W, W, W),
889 swizzle(tmp, Y, Y, Y, Y),
890 swizzle(tmp, X, X, X, X));
891
892 break;
893
894 case OPCODE_SGE:
895 EMIT_2ARG_ARITH(A0_SGE);
896 break;
897
898 case OPCODE_SGT:
899 i915_emit_arith(p,
900 A0_SLT,
901 get_result_vector( p, inst ),
902 get_result_flags( inst ), 0,
903 negate(src_vector( p, &inst->SrcReg[0], program),
904 1, 1, 1, 1),
905 negate(src_vector( p, &inst->SrcReg[1], program),
906 1, 1, 1, 1),
907 0);
908 break;
909
910 case OPCODE_SLE:
911 i915_emit_arith(p,
912 A0_SGE,
913 get_result_vector( p, inst ),
914 get_result_flags( inst ), 0,
915 negate(src_vector( p, &inst->SrcReg[0], program),
916 1, 1, 1, 1),
917 negate(src_vector( p, &inst->SrcReg[1], program),
918 1, 1, 1, 1),
919 0);
920 break;
921
922 case OPCODE_SLT:
923 EMIT_2ARG_ARITH(A0_SLT);
924 break;
925
926 case OPCODE_SNE:
927 tmp = i915_get_utemp(p);
928 flags = get_result_flags(inst);
929 dst = get_result_vector(p, inst);
930
931 /* dst = src1 < src2 */
932 i915_emit_arith(p,
933 A0_SLT,
934 dst,
935 flags, 0,
936 src_vector(p, &inst->SrcReg[0], program),
937 src_vector(p, &inst->SrcReg[1], program),
938 0);
939 /* tmp = src1 > src2 */
940 i915_emit_arith(p,
941 A0_SLT,
942 tmp,
943 flags, 0,
944 negate(src_vector(p, &inst->SrcReg[0], program),
945 1, 1, 1, 1),
946 negate(src_vector(p, &inst->SrcReg[1], program),
947 1, 1, 1, 1),
948 0);
949 /* dst = tmp || dst */
950 i915_emit_arith(p,
951 A0_ADD,
952 dst,
953 flags | A0_DEST_SATURATE, 0,
954 dst,
955 tmp,
956 0);
957 break;
958
959 case OPCODE_SUB:
960 src0 = src_vector(p, &inst->SrcReg[0], program);
961 src1 = src_vector(p, &inst->SrcReg[1], program);
962
963 i915_emit_arith(p,
964 A0_ADD,
965 get_result_vector(p, inst),
966 get_result_flags(inst), 0,
967 src0, negate(src1, 1, 1, 1, 1), 0);
968 break;
969
970 case OPCODE_SWZ:
971 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
972 break;
973
974 case OPCODE_TEX:
975 EMIT_TEX(T0_TEXLD);
976 break;
977
978 case OPCODE_TXB:
979 EMIT_TEX(T0_TEXLDB);
980 break;
981
982 case OPCODE_TXP:
983 EMIT_TEX(T0_TEXLDP);
984 break;
985
986 case OPCODE_XPD:
987 /* Cross product:
988 * result.x = src0.y * src1.z - src0.z * src1.y;
989 * result.y = src0.z * src1.x - src0.x * src1.z;
990 * result.z = src0.x * src1.y - src0.y * src1.x;
991 * result.w = undef;
992 */
993 src0 = src_vector(p, &inst->SrcReg[0], program);
994 src1 = src_vector(p, &inst->SrcReg[1], program);
995 tmp = i915_get_utemp(p);
996
997 i915_emit_arith(p,
998 A0_MUL,
999 tmp, A0_DEST_CHANNEL_ALL, 0,
1000 swizzle(src0, Z, X, Y, ONE),
1001 swizzle(src1, Y, Z, X, ONE), 0);
1002
1003 i915_emit_arith(p,
1004 A0_MAD,
1005 get_result_vector(p, inst),
1006 get_result_flags(inst), 0,
1007 swizzle(src0, Y, Z, X, ONE),
1008 swizzle(src1, Z, X, Y, ONE),
1009 negate(tmp, 1, 1, 1, 0));
1010 break;
1011
1012 case OPCODE_END:
1013 return;
1014
1015 case OPCODE_BGNLOOP:
1016 case OPCODE_BGNSUB:
1017 case OPCODE_BRA:
1018 case OPCODE_BRK:
1019 case OPCODE_CAL:
1020 case OPCODE_CONT:
1021 case OPCODE_DDX:
1022 case OPCODE_DDY:
1023 case OPCODE_ELSE:
1024 case OPCODE_ENDIF:
1025 case OPCODE_ENDLOOP:
1026 case OPCODE_ENDSUB:
1027 case OPCODE_IF:
1028 case OPCODE_RET:
1029 p->error = 1;
1030 i915_program_error(p, "Unsupported opcode: %s",
1031 _mesa_opcode_string(inst->Opcode));
1032 return;
1033
1034 case OPCODE_EXP:
1035 case OPCODE_LOG:
1036 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1037 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1038 * Instead, it translates to EX2 or LG2.
1039 */
1040 case OPCODE_TXD:
1041 case OPCODE_TXL:
1042 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1043 * only NV_vp/fp appears to emit them.
1044 */
1045 default:
1046 i915_program_error(p, "bad opcode: %s",
1047 _mesa_opcode_string(inst->Opcode));
1048 return;
1049 }
1050
1051 inst++;
1052 i915_release_utemps(p);
1053 }
1054 }
1055
1056 /* Rather than trying to intercept and jiggle depth writes during
1057 * emit, just move the value into its correct position at the end of
1058 * the program:
1059 */
1060 static void
1061 fixup_depth_write(struct i915_fragment_program *p)
1062 {
1063 if (p->depth_written) {
1064 GLuint depth = UREG(REG_TYPE_OD, 0);
1065
1066 i915_emit_arith(p,
1067 A0_MOV,
1068 depth, A0_DEST_CHANNEL_W, 0,
1069 swizzle(depth, X, Y, Z, Z), 0, 0);
1070 }
1071 }
1072
1073
1074 static void
1075 check_wpos(struct i915_fragment_program *p)
1076 {
1077 GLuint inputs = p->FragProg.Base.InputsRead;
1078 GLint i;
1079
1080 p->wpos_tex = -1;
1081
1082 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1083 if (inputs & (FRAG_BIT_TEX(i) | FRAG_BIT_VAR(i)))
1084 continue;
1085 else if (inputs & FRAG_BIT_WPOS) {
1086 p->wpos_tex = i;
1087 inputs &= ~FRAG_BIT_WPOS;
1088 }
1089 }
1090
1091 if (inputs & FRAG_BIT_WPOS) {
1092 i915_program_error(p, "No free texcoord for wpos value");
1093 }
1094 }
1095
1096
1097 static void
1098 translate_program(struct i915_fragment_program *p)
1099 {
1100 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1101
1102 i915_init_program(i915, p);
1103 check_wpos(p);
1104 upload_program(p);
1105 fixup_depth_write(p);
1106 i915_fini_program(p);
1107
1108 p->translated = 1;
1109 }
1110
1111
1112 static void
1113 track_params(struct i915_fragment_program *p)
1114 {
1115 GLint i;
1116
1117 if (p->nr_params)
1118 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1119
1120 for (i = 0; i < p->nr_params; i++) {
1121 GLint reg = p->param[i].reg;
1122 COPY_4V(p->constant[reg], p->param[i].values);
1123 }
1124
1125 p->params_uptodate = 1;
1126 p->on_hardware = 0; /* overkill */
1127 }
1128
1129
1130 static void
1131 i915BindProgram(GLcontext * ctx, GLenum target, struct gl_program *prog)
1132 {
1133 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1134 struct i915_context *i915 = I915_CONTEXT(ctx);
1135 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1136
1137 if (i915->current_program == p)
1138 return;
1139
1140 if (i915->current_program) {
1141 i915->current_program->on_hardware = 0;
1142 i915->current_program->params_uptodate = 0;
1143 }
1144
1145 i915->current_program = p;
1146
1147 assert(p->on_hardware == 0);
1148 assert(p->params_uptodate == 0);
1149
1150 }
1151 }
1152
1153 static struct gl_program *
1154 i915NewProgram(GLcontext * ctx, GLenum target, GLuint id)
1155 {
1156 switch (target) {
1157 case GL_VERTEX_PROGRAM_ARB:
1158 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1159 target, id);
1160
1161 case GL_FRAGMENT_PROGRAM_ARB:{
1162 struct i915_fragment_program *prog =
1163 CALLOC_STRUCT(i915_fragment_program);
1164 if (prog) {
1165 i915_init_program(I915_CONTEXT(ctx), prog);
1166
1167 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1168 target, id);
1169 }
1170 else
1171 return NULL;
1172 }
1173
1174 default:
1175 /* Just fallback:
1176 */
1177 return _mesa_new_program(ctx, target, id);
1178 }
1179 }
1180
1181 static void
1182 i915DeleteProgram(GLcontext * ctx, struct gl_program *prog)
1183 {
1184 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1185 struct i915_context *i915 = I915_CONTEXT(ctx);
1186 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1187
1188 if (i915->current_program == p)
1189 i915->current_program = 0;
1190 }
1191
1192 _mesa_delete_program(ctx, prog);
1193 }
1194
1195
1196 static GLboolean
1197 i915IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
1198 {
1199 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1200 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1201
1202 if (!p->translated)
1203 translate_program(p);
1204
1205 return !p->error;
1206 }
1207 else
1208 return GL_TRUE;
1209 }
1210
1211 static GLboolean
1212 i915ProgramStringNotify(GLcontext * ctx,
1213 GLenum target, struct gl_program *prog)
1214 {
1215 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1216 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1217 p->translated = 0;
1218
1219 /* Hack: make sure fog is correctly enabled according to this
1220 * fragment program's fog options.
1221 */
1222 if (p->FragProg.FogOption) {
1223 /* add extra instructions to do fog, then turn off FogOption field */
1224 _mesa_append_fog_code(ctx, &p->FragProg);
1225 p->FragProg.FogOption = GL_NONE;
1226 }
1227 }
1228
1229 (void) _tnl_program_string(ctx, target, prog);
1230
1231 /* XXX check if program is legal, within limits */
1232 return GL_TRUE;
1233 }
1234
1235 void
1236 i915_update_program(GLcontext *ctx)
1237 {
1238 struct intel_context *intel = intel_context(ctx);
1239 struct i915_context *i915 = i915_context(&intel->ctx);
1240 struct i915_fragment_program *fp =
1241 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1242
1243 if (i915->current_program != fp) {
1244 if (i915->current_program) {
1245 i915->current_program->on_hardware = 0;
1246 i915->current_program->params_uptodate = 0;
1247 }
1248
1249 i915->current_program = fp;
1250 }
1251
1252 if (!fp->translated)
1253 translate_program(fp);
1254
1255 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1256 }
1257
1258 void
1259 i915ValidateFragmentProgram(struct i915_context *i915)
1260 {
1261 GLcontext *ctx = &i915->intel.ctx;
1262 struct intel_context *intel = intel_context(ctx);
1263 TNLcontext *tnl = TNL_CONTEXT(ctx);
1264 struct vertex_buffer *VB = &tnl->vb;
1265
1266 struct i915_fragment_program *p =
1267 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1268
1269 const GLuint inputsRead = p->FragProg.Base.InputsRead;
1270 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1271 GLuint s2 = S2_TEXCOORD_NONE;
1272 int i, offset = 0;
1273
1274 /* Important:
1275 */
1276 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1277
1278 if (!p->translated)
1279 translate_program(p);
1280
1281 intel->vertex_attr_count = 0;
1282 intel->wpos_offset = 0;
1283 intel->wpos_size = 0;
1284 intel->coloroffset = 0;
1285 intel->specoffset = 0;
1286
1287 if (inputsRead & FRAG_BITS_TEX_ANY) {
1288 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1289 }
1290 else {
1291 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1292 }
1293
1294 if (inputsRead & FRAG_BIT_COL0) {
1295 intel->coloroffset = offset / 4;
1296 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1297 }
1298
1299 if (inputsRead & FRAG_BIT_COL1) {
1300 intel->specoffset = offset / 4;
1301 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1302 }
1303
1304 if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE) {
1305 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1306 }
1307
1308 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1309 if (inputsRead & FRAG_BIT_TEX(i)) {
1310 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1311
1312 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1313 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1314
1315 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1316 }
1317 else if (inputsRead & FRAG_BIT_VAR(i)) {
1318 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1319
1320 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1321 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1322
1323 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1324 }
1325 else if (i == p->wpos_tex) {
1326
1327 /* If WPOS is required, duplicate the XYZ position data in an
1328 * unused texture coordinate:
1329 */
1330 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1331 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1332
1333 intel->wpos_offset = offset;
1334 intel->wpos_size = 3 * sizeof(GLuint);
1335
1336 EMIT_PAD(intel->wpos_size);
1337 }
1338 }
1339
1340 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1341 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1342 int k;
1343
1344 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1345
1346 /* Must do this *after* statechange, so as not to affect
1347 * buffered vertices reliant on the old state:
1348 */
1349 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1350 intel->vertex_attrs,
1351 intel->vertex_attr_count,
1352 intel->ViewportMatrix.m, 0);
1353
1354 intel->vertex_size >>= 2;
1355
1356 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1357 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1358
1359 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1360 assert(k);
1361 }
1362
1363 if (!p->params_uptodate)
1364 track_params(p);
1365
1366 if (!p->on_hardware)
1367 i915_upload_program(i915, p);
1368 }
1369
1370 void
1371 i915InitFragProgFuncs(struct dd_function_table *functions)
1372 {
1373 functions->BindProgram = i915BindProgram;
1374 functions->NewProgram = i915NewProgram;
1375 functions->DeleteProgram = i915DeleteProgram;
1376 functions->IsProgramNative = i915IsProgramNative;
1377 functions->ProgramStringNotify = i915ProgramStringNotify;
1378 }