projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
mesa: rename src/mesa/shader/ to src/mesa/program/
[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 /* 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 (%d out of %d)",
363 program->Base.NumInstructions, I915_MAX_INSN);
364 return;
365 }
366
367 /* Not always needed:
368 */
369 calc_live_regs(p);
370
371 while (1) {
372 GLuint src0, src1, src2, flags;
373 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
374
375 switch (inst->Opcode) {
376 case OPCODE_ABS:
377 src0 = src_vector(p, &inst->SrcReg[0], program);
378 i915_emit_arith(p,
379 A0_MAX,
380 get_result_vector(p, inst),
381 get_result_flags(inst), 0,
382 src0, negate(src0, 1, 1, 1, 1), 0);
383 break;
384
385 case OPCODE_ADD:
386 EMIT_2ARG_ARITH(A0_ADD);
387 break;
388
389 case OPCODE_CMP:
390 src0 = src_vector(p, &inst->SrcReg[0], program);
391 src1 = src_vector(p, &inst->SrcReg[1], program);
392 src2 = src_vector(p, &inst->SrcReg[2], program);
393 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
394 break;
395
396 case OPCODE_COS:
397 src0 = src_vector(p, &inst->SrcReg[0], program);
398 tmp = i915_get_utemp(p);
399 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
400 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
401
402 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
403 i915_emit_arith(p,
404 A0_MAD,
405 tmp, A0_DEST_CHANNEL_X, 0,
406 src0,
407 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
408 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
409
410 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
411
412 i915_emit_arith(p,
413 A0_MAD,
414 tmp, A0_DEST_CHANNEL_X, 0,
415 tmp,
416 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
417 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
418
419 /* Compute COS with the same calculation used for SIN, but a
420 * different source range has been mapped to [-1,1] this time.
421 */
422
423 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
424 i915_emit_arith(p,
425 A0_MAX,
426 tmp, A0_DEST_CHANNEL_Y, 0,
427 swizzle(tmp, ZERO, X, ZERO, ZERO),
428 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
429 0);
430
431 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
432 i915_emit_arith(p,
433 A0_MUL,
434 tmp, A0_DEST_CHANNEL_Y, 0,
435 swizzle(tmp, ZERO, X, ZERO, ZERO),
436 tmp,
437 0);
438
439 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
440 i915_emit_arith(p,
441 A0_DP3,
442 tmp, A0_DEST_CHANNEL_X, 0,
443 tmp,
444 swizzle(consts1, X, Y, ZERO, ZERO),
445 0);
446
447 /* tmp.x now contains a first approximation (y). Now, weight it
448 * against tmp.y**2 to get closer.
449 */
450 i915_emit_arith(p,
451 A0_MAX,
452 tmp, A0_DEST_CHANNEL_Y, 0,
453 swizzle(tmp, ZERO, X, ZERO, ZERO),
454 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
455 0);
456
457 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
458 i915_emit_arith(p,
459 A0_MAD,
460 tmp, A0_DEST_CHANNEL_Y, 0,
461 swizzle(tmp, ZERO, X, ZERO, ZERO),
462 swizzle(tmp, ZERO, Y, ZERO, ZERO),
463 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
464
465 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
466 i915_emit_arith(p,
467 A0_MAD,
468 get_result_vector(p, inst),
469 get_result_flags(inst), 0,
470 swizzle(consts1, W, W, W, W),
471 swizzle(tmp, Y, Y, Y, Y),
472 swizzle(tmp, X, X, X, X));
473 break;
474
475 case OPCODE_DP3:
476 EMIT_2ARG_ARITH(A0_DP3);
477 break;
478
479 case OPCODE_DP4:
480 EMIT_2ARG_ARITH(A0_DP4);
481 break;
482
483 case OPCODE_DPH:
484 src0 = src_vector(p, &inst->SrcReg[0], program);
485 src1 = src_vector(p, &inst->SrcReg[1], program);
486
487 i915_emit_arith(p,
488 A0_DP4,
489 get_result_vector(p, inst),
490 get_result_flags(inst), 0,
491 swizzle(src0, X, Y, Z, ONE), src1, 0);
492 break;
493
494 case OPCODE_DST:
495 src0 = src_vector(p, &inst->SrcReg[0], program);
496 src1 = src_vector(p, &inst->SrcReg[1], program);
497
498 /* result[0] = 1 * 1;
499 * result[1] = a[1] * b[1];
500 * result[2] = a[2] * 1;
501 * result[3] = 1 * b[3];
502 */
503 i915_emit_arith(p,
504 A0_MUL,
505 get_result_vector(p, inst),
506 get_result_flags(inst), 0,
507 swizzle(src0, ONE, Y, Z, ONE),
508 swizzle(src1, ONE, Y, ONE, W), 0);
509 break;
510
511 case OPCODE_EX2:
512 src0 = src_vector(p, &inst->SrcReg[0], program);
513
514 i915_emit_arith(p,
515 A0_EXP,
516 get_result_vector(p, inst),
517 get_result_flags(inst), 0,
518 swizzle(src0, X, X, X, X), 0, 0);
519 break;
520
521 case OPCODE_FLR:
522 EMIT_1ARG_ARITH(A0_FLR);
523 break;
524
525 case OPCODE_TRUNC:
526 EMIT_1ARG_ARITH(A0_TRC);
527 break;
528
529 case OPCODE_FRC:
530 EMIT_1ARG_ARITH(A0_FRC);
531 break;
532
533 case OPCODE_KIL:
534 src0 = src_vector(p, &inst->SrcReg[0], program);
535 tmp = i915_get_utemp(p);
536
537 i915_emit_texld(p, get_live_regs(p, inst),
538 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
539 0, src0, T0_TEXKILL);
540 break;
541
542 case OPCODE_KIL_NV:
543 if (inst->DstReg.CondMask == COND_TR) {
544 tmp = i915_get_utemp(p);
545
546 i915_emit_texld(p, get_live_regs(p, inst),
547 tmp, A0_DEST_CHANNEL_ALL,
548 0, /* use a dummy dest reg */
549 swizzle(tmp, ONE, ONE, ONE, ONE), /* always */
550 T0_TEXKILL);
551 } else {
552 p->error = 1;
553 i915_program_error(p, "Unsupported KIL_NV condition code: %d",
554 inst->DstReg.CondMask);
555 }
556 break;
557
558 case OPCODE_LG2:
559 src0 = src_vector(p, &inst->SrcReg[0], program);
560
561 i915_emit_arith(p,
562 A0_LOG,
563 get_result_vector(p, inst),
564 get_result_flags(inst), 0,
565 swizzle(src0, X, X, X, X), 0, 0);
566 break;
567
568 case OPCODE_LIT:
569 src0 = src_vector(p, &inst->SrcReg[0], program);
570 tmp = i915_get_utemp(p);
571
572 /* tmp = max( a.xyzw, a.00zw )
573 * XXX: Clamp tmp.w to -128..128
574 * tmp.y = log(tmp.y)
575 * tmp.y = tmp.w * tmp.y
576 * tmp.y = exp(tmp.y)
577 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
578 */
579 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
580 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
581
582 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
583 swizzle(tmp, Y, Y, Y, Y), 0, 0);
584
585 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
586 swizzle(tmp, ZERO, Y, ZERO, ZERO),
587 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
588
589 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
590 swizzle(tmp, Y, Y, Y, Y), 0, 0);
591
592 i915_emit_arith(p, A0_CMP,
593 get_result_vector(p, inst),
594 get_result_flags(inst), 0,
595 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
596 swizzle(tmp, ONE, X, ZERO, ONE),
597 swizzle(tmp, ONE, X, Y, ONE));
598
599 break;
600
601 case OPCODE_LRP:
602 src0 = src_vector(p, &inst->SrcReg[0], program);
603 src1 = src_vector(p, &inst->SrcReg[1], program);
604 src2 = src_vector(p, &inst->SrcReg[2], program);
605 flags = get_result_flags(inst);
606 tmp = i915_get_utemp(p);
607
608 /* b*a + c*(1-a)
609 *
610 * b*a + c - ca
611 *
612 * tmp = b*a + c,
613 * result = (-c)*a + tmp
614 */
615 i915_emit_arith(p, A0_MAD, tmp,
616 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
617
618 i915_emit_arith(p, A0_MAD,
619 get_result_vector(p, inst),
620 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
621 break;
622
623 case OPCODE_MAD:
624 EMIT_3ARG_ARITH(A0_MAD);
625 break;
626
627 case OPCODE_MAX:
628 EMIT_2ARG_ARITH(A0_MAX);
629 break;
630
631 case OPCODE_MIN:
632 src0 = src_vector(p, &inst->SrcReg[0], program);
633 src1 = src_vector(p, &inst->SrcReg[1], program);
634 tmp = i915_get_utemp(p);
635 flags = get_result_flags(inst);
636
637 i915_emit_arith(p,
638 A0_MAX,
639 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
640 negate(src0, 1, 1, 1, 1),
641 negate(src1, 1, 1, 1, 1), 0);
642
643 i915_emit_arith(p,
644 A0_MOV,
645 get_result_vector(p, inst),
646 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
647 break;
648
649 case OPCODE_MOV:
650 EMIT_1ARG_ARITH(A0_MOV);
651 break;
652
653 case OPCODE_MUL:
654 EMIT_2ARG_ARITH(A0_MUL);
655 break;
656
657 case OPCODE_NOISE1:
658 case OPCODE_NOISE2:
659 case OPCODE_NOISE3:
660 case OPCODE_NOISE4:
661 /* Don't implement noise because we just don't have the instructions
662 * to spare. We aren't the first vendor to do so.
663 */
664 i915_program_error(p, "Stubbed-out noise functions");
665 i915_emit_arith(p,
666 A0_MOV,
667 get_result_vector(p, inst),
668 get_result_flags(inst), 0,
669 swizzle(tmp, ZERO, ZERO, ZERO, ZERO), 0, 0);
670 break;
671
672 case OPCODE_POW:
673 src0 = src_vector(p, &inst->SrcReg[0], program);
674 src1 = src_vector(p, &inst->SrcReg[1], program);
675 tmp = i915_get_utemp(p);
676 flags = get_result_flags(inst);
677
678 /* XXX: masking on intermediate values, here and elsewhere.
679 */
680 i915_emit_arith(p,
681 A0_LOG,
682 tmp, A0_DEST_CHANNEL_X, 0,
683 swizzle(src0, X, X, X, X), 0, 0);
684
685 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
686
687
688 i915_emit_arith(p,
689 A0_EXP,
690 get_result_vector(p, inst),
691 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
692
693 break;
694
695 case OPCODE_RCP:
696 src0 = src_vector(p, &inst->SrcReg[0], program);
697
698 i915_emit_arith(p,
699 A0_RCP,
700 get_result_vector(p, inst),
701 get_result_flags(inst), 0,
702 swizzle(src0, X, X, X, X), 0, 0);
703 break;
704
705 case OPCODE_RSQ:
706
707 src0 = src_vector(p, &inst->SrcReg[0], program);
708
709 i915_emit_arith(p,
710 A0_RSQ,
711 get_result_vector(p, inst),
712 get_result_flags(inst), 0,
713 swizzle(src0, X, X, X, X), 0, 0);
714 break;
715
716 case OPCODE_SCS:
717 src0 = src_vector(p, &inst->SrcReg[0], program);
718 tmp = i915_get_utemp(p);
719
720 /*
721 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
722 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
723 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
724 * scs.x = DP4 t1, sin_constants
725 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
726 * scs.y = DP4 t1, cos_constants
727 */
728 i915_emit_arith(p,
729 A0_MUL,
730 tmp, A0_DEST_CHANNEL_XY, 0,
731 swizzle(src0, X, X, ONE, ONE),
732 swizzle(src0, X, ONE, ONE, ONE), 0);
733
734 i915_emit_arith(p,
735 A0_MUL,
736 tmp, A0_DEST_CHANNEL_ALL, 0,
737 swizzle(tmp, X, Y, X, Y),
738 swizzle(tmp, X, X, ONE, ONE), 0);
739
740 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
741 GLuint tmp1;
742
743 if (inst->DstReg.WriteMask & WRITEMASK_X)
744 tmp1 = i915_get_utemp(p);
745 else
746 tmp1 = tmp;
747
748 i915_emit_arith(p,
749 A0_MUL,
750 tmp1, A0_DEST_CHANNEL_ALL, 0,
751 swizzle(tmp, X, Y, Y, W),
752 swizzle(tmp, X, Z, ONE, ONE), 0);
753
754 i915_emit_arith(p,
755 A0_DP4,
756 get_result_vector(p, inst),
757 A0_DEST_CHANNEL_Y, 0,
758 swizzle(tmp1, W, Z, Y, X),
759 i915_emit_const4fv(p, sin_constants), 0);
760 }
761
762 if (inst->DstReg.WriteMask & WRITEMASK_X) {
763 i915_emit_arith(p,
764 A0_MUL,
765 tmp, A0_DEST_CHANNEL_XYZ, 0,
766 swizzle(tmp, X, X, Z, ONE),
767 swizzle(tmp, Z, ONE, ONE, ONE), 0);
768
769 i915_emit_arith(p,
770 A0_DP4,
771 get_result_vector(p, inst),
772 A0_DEST_CHANNEL_X, 0,
773 swizzle(tmp, ONE, Z, Y, X),
774 i915_emit_const4fv(p, cos_constants), 0);
775 }
776 break;
777
778 case OPCODE_SEQ:
779 tmp = i915_get_utemp(p);
780 flags = get_result_flags(inst);
781 dst = get_result_vector(p, inst);
782
783 /* dst = src1 >= src2 */
784 i915_emit_arith(p,
785 A0_SGE,
786 dst,
787 flags, 0,
788 src_vector(p, &inst->SrcReg[0], program),
789 src_vector(p, &inst->SrcReg[1], program),
790 0);
791 /* tmp = src1 <= src2 */
792 i915_emit_arith(p,
793 A0_SGE,
794 tmp,
795 flags, 0,
796 negate(src_vector(p, &inst->SrcReg[0], program),
797 1, 1, 1, 1),
798 negate(src_vector(p, &inst->SrcReg[1], program),
799 1, 1, 1, 1),
800 0);
801 /* dst = tmp && dst */
802 i915_emit_arith(p,
803 A0_MUL,
804 dst,
805 flags, 0,
806 dst,
807 tmp,
808 0);
809 break;
810
811 case OPCODE_SIN:
812 src0 = src_vector(p, &inst->SrcReg[0], program);
813 tmp = i915_get_utemp(p);
814 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
815 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
816
817 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
818 i915_emit_arith(p,
819 A0_MAD,
820 tmp, A0_DEST_CHANNEL_X, 0,
821 src0,
822 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
823 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
824
825 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
826
827 i915_emit_arith(p,
828 A0_MAD,
829 tmp, A0_DEST_CHANNEL_X, 0,
830 tmp,
831 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
832 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
833
834 /* Compute sin using a quadratic and quartic. It gives continuity
835 * that repeating the Taylor series lacks every 2*pi, and has
836 * reduced error.
837 *
838 * The idea was described at:
839 * http://www.devmaster.net/forums/showthread.php?t=5784
840 */
841
842 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
843 i915_emit_arith(p,
844 A0_MAX,
845 tmp, A0_DEST_CHANNEL_Y, 0,
846 swizzle(tmp, ZERO, X, ZERO, ZERO),
847 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
848 0);
849
850 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
851 i915_emit_arith(p,
852 A0_MUL,
853 tmp, A0_DEST_CHANNEL_Y, 0,
854 swizzle(tmp, ZERO, X, ZERO, ZERO),
855 tmp,
856 0);
857
858 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
859 i915_emit_arith(p,
860 A0_DP3,
861 tmp, A0_DEST_CHANNEL_X, 0,
862 tmp,
863 swizzle(consts1, X, Y, ZERO, ZERO),
864 0);
865
866 /* tmp.x now contains a first approximation (y). Now, weight it
867 * against tmp.y**2 to get closer.
868 */
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.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
877 i915_emit_arith(p,
878 A0_MAD,
879 tmp, A0_DEST_CHANNEL_Y, 0,
880 swizzle(tmp, ZERO, X, ZERO, ZERO),
881 swizzle(tmp, ZERO, Y, ZERO, ZERO),
882 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
883
884 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
885 i915_emit_arith(p,
886 A0_MAD,
887 get_result_vector(p, inst),
888 get_result_flags(inst), 0,
889 swizzle(consts1, W, W, W, W),
890 swizzle(tmp, Y, Y, Y, Y),
891 swizzle(tmp, X, X, X, X));
892
893 break;
894
895 case OPCODE_SGE:
896 EMIT_2ARG_ARITH(A0_SGE);
897 break;
898
899 case OPCODE_SGT:
900 i915_emit_arith(p,
901 A0_SLT,
902 get_result_vector( p, inst ),
903 get_result_flags( inst ), 0,
904 negate(src_vector( p, &inst->SrcReg[0], program),
905 1, 1, 1, 1),
906 negate(src_vector( p, &inst->SrcReg[1], program),
907 1, 1, 1, 1),
908 0);
909 break;
910
911 case OPCODE_SLE:
912 i915_emit_arith(p,
913 A0_SGE,
914 get_result_vector( p, inst ),
915 get_result_flags( inst ), 0,
916 negate(src_vector( p, &inst->SrcReg[0], program),
917 1, 1, 1, 1),
918 negate(src_vector( p, &inst->SrcReg[1], program),
919 1, 1, 1, 1),
920 0);
921 break;
922
923 case OPCODE_SLT:
924 EMIT_2ARG_ARITH(A0_SLT);
925 break;
926
927 case OPCODE_SNE:
928 tmp = i915_get_utemp(p);
929 flags = get_result_flags(inst);
930 dst = get_result_vector(p, inst);
931
932 /* dst = src1 < src2 */
933 i915_emit_arith(p,
934 A0_SLT,
935 dst,
936 flags, 0,
937 src_vector(p, &inst->SrcReg[0], program),
938 src_vector(p, &inst->SrcReg[1], program),
939 0);
940 /* tmp = src1 > src2 */
941 i915_emit_arith(p,
942 A0_SLT,
943 tmp,
944 flags, 0,
945 negate(src_vector(p, &inst->SrcReg[0], program),
946 1, 1, 1, 1),
947 negate(src_vector(p, &inst->SrcReg[1], program),
948 1, 1, 1, 1),
949 0);
950 /* dst = tmp || dst */
951 i915_emit_arith(p,
952 A0_ADD,
953 dst,
954 flags | A0_DEST_SATURATE, 0,
955 dst,
956 tmp,
957 0);
958 break;
959
960 case OPCODE_SUB:
961 src0 = src_vector(p, &inst->SrcReg[0], program);
962 src1 = src_vector(p, &inst->SrcReg[1], program);
963
964 i915_emit_arith(p,
965 A0_ADD,
966 get_result_vector(p, inst),
967 get_result_flags(inst), 0,
968 src0, negate(src1, 1, 1, 1, 1), 0);
969 break;
970
971 case OPCODE_SWZ:
972 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
973 break;
974
975 case OPCODE_TEX:
976 EMIT_TEX(T0_TEXLD);
977 break;
978
979 case OPCODE_TXB:
980 EMIT_TEX(T0_TEXLDB);
981 break;
982
983 case OPCODE_TXP:
984 EMIT_TEX(T0_TEXLDP);
985 break;
986
987 case OPCODE_XPD:
988 /* Cross product:
989 * result.x = src0.y * src1.z - src0.z * src1.y;
990 * result.y = src0.z * src1.x - src0.x * src1.z;
991 * result.z = src0.x * src1.y - src0.y * src1.x;
992 * result.w = undef;
993 */
994 src0 = src_vector(p, &inst->SrcReg[0], program);
995 src1 = src_vector(p, &inst->SrcReg[1], program);
996 tmp = i915_get_utemp(p);
997
998 i915_emit_arith(p,
999 A0_MUL,
1000 tmp, A0_DEST_CHANNEL_ALL, 0,
1001 swizzle(src0, Z, X, Y, ONE),
1002 swizzle(src1, Y, Z, X, ONE), 0);
1003
1004 i915_emit_arith(p,
1005 A0_MAD,
1006 get_result_vector(p, inst),
1007 get_result_flags(inst), 0,
1008 swizzle(src0, Y, Z, X, ONE),
1009 swizzle(src1, Z, X, Y, ONE),
1010 negate(tmp, 1, 1, 1, 0));
1011 break;
1012
1013 case OPCODE_END:
1014 return;
1015
1016 case OPCODE_BGNLOOP:
1017 case OPCODE_BGNSUB:
1018 case OPCODE_BRA:
1019 case OPCODE_BRK:
1020 case OPCODE_CAL:
1021 case OPCODE_CONT:
1022 case OPCODE_DDX:
1023 case OPCODE_DDY:
1024 case OPCODE_ELSE:
1025 case OPCODE_ENDIF:
1026 case OPCODE_ENDLOOP:
1027 case OPCODE_ENDSUB:
1028 case OPCODE_IF:
1029 case OPCODE_RET:
1030 p->error = 1;
1031 i915_program_error(p, "Unsupported opcode: %s",
1032 _mesa_opcode_string(inst->Opcode));
1033 return;
1034
1035 case OPCODE_EXP:
1036 case OPCODE_LOG:
1037 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1038 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1039 * Instead, it translates to EX2 or LG2.
1040 */
1041 case OPCODE_TXD:
1042 case OPCODE_TXL:
1043 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1044 * only NV_vp/fp appears to emit them.
1045 */
1046 default:
1047 i915_program_error(p, "bad opcode: %s",
1048 _mesa_opcode_string(inst->Opcode));
1049 return;
1050 }
1051
1052 inst++;
1053 i915_release_utemps(p);
1054 }
1055 }
1056
1057 /* Rather than trying to intercept and jiggle depth writes during
1058 * emit, just move the value into its correct position at the end of
1059 * the program:
1060 */
1061 static void
1062 fixup_depth_write(struct i915_fragment_program *p)
1063 {
1064 if (p->depth_written) {
1065 GLuint depth = UREG(REG_TYPE_OD, 0);
1066
1067 i915_emit_arith(p,
1068 A0_MOV,
1069 depth, A0_DEST_CHANNEL_W, 0,
1070 swizzle(depth, X, Y, Z, Z), 0, 0);
1071 }
1072 }
1073
1074
1075 static void
1076 check_wpos(struct i915_fragment_program *p)
1077 {
1078 GLuint inputs = p->FragProg.Base.InputsRead;
1079 GLint i;
1080
1081 p->wpos_tex = -1;
1082
1083 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1084 if (inputs & (FRAG_BIT_TEX(i) | FRAG_BIT_VAR(i)))
1085 continue;
1086 else if (inputs & FRAG_BIT_WPOS) {
1087 p->wpos_tex = i;
1088 inputs &= ~FRAG_BIT_WPOS;
1089 }
1090 }
1091
1092 if (inputs & FRAG_BIT_WPOS) {
1093 i915_program_error(p, "No free texcoord for wpos value");
1094 }
1095 }
1096
1097
1098 static void
1099 translate_program(struct i915_fragment_program *p)
1100 {
1101 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1102
1103 if (INTEL_DEBUG & DEBUG_WM) {
1104 printf("fp:\n");
1105 _mesa_print_program(&p->ctx->FragmentProgram._Current->Base);
1106 printf("\n");
1107 }
1108
1109 i915_init_program(i915, p);
1110 check_wpos(p);
1111 upload_program(p);
1112 fixup_depth_write(p);
1113 i915_fini_program(p);
1114
1115 if (INTEL_DEBUG & DEBUG_WM) {
1116 printf("i915:\n");
1117 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1118 }
1119
1120 p->translated = 1;
1121 }
1122
1123
1124 static void
1125 track_params(struct i915_fragment_program *p)
1126 {
1127 GLint i;
1128
1129 if (p->nr_params)
1130 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1131
1132 for (i = 0; i < p->nr_params; i++) {
1133 GLint reg = p->param[i].reg;
1134 COPY_4V(p->constant[reg], p->param[i].values);
1135 }
1136
1137 p->params_uptodate = 1;
1138 p->on_hardware = 0; /* overkill */
1139 }
1140
1141
1142 static void
1143 i915BindProgram(GLcontext * ctx, GLenum target, struct gl_program *prog)
1144 {
1145 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1146 struct i915_context *i915 = I915_CONTEXT(ctx);
1147 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1148
1149 if (i915->current_program == p)
1150 return;
1151
1152 if (i915->current_program) {
1153 i915->current_program->on_hardware = 0;
1154 i915->current_program->params_uptodate = 0;
1155 }
1156
1157 i915->current_program = p;
1158
1159 assert(p->on_hardware == 0);
1160 assert(p->params_uptodate == 0);
1161
1162 }
1163 }
1164
1165 static struct gl_program *
1166 i915NewProgram(GLcontext * ctx, GLenum target, GLuint id)
1167 {
1168 switch (target) {
1169 case GL_VERTEX_PROGRAM_ARB:
1170 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1171 target, id);
1172
1173 case GL_FRAGMENT_PROGRAM_ARB:{
1174 struct i915_fragment_program *prog =
1175 CALLOC_STRUCT(i915_fragment_program);
1176 if (prog) {
1177 i915_init_program(I915_CONTEXT(ctx), prog);
1178
1179 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1180 target, id);
1181 }
1182 else
1183 return NULL;
1184 }
1185
1186 default:
1187 /* Just fallback:
1188 */
1189 return _mesa_new_program(ctx, target, id);
1190 }
1191 }
1192
1193 static void
1194 i915DeleteProgram(GLcontext * ctx, struct gl_program *prog)
1195 {
1196 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1197 struct i915_context *i915 = I915_CONTEXT(ctx);
1198 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1199
1200 if (i915->current_program == p)
1201 i915->current_program = 0;
1202 }
1203
1204 _mesa_delete_program(ctx, prog);
1205 }
1206
1207
1208 static GLboolean
1209 i915IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
1210 {
1211 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1212 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1213
1214 if (!p->translated)
1215 translate_program(p);
1216
1217 return !p->error;
1218 }
1219 else
1220 return GL_TRUE;
1221 }
1222
1223 static GLboolean
1224 i915ProgramStringNotify(GLcontext * ctx,
1225 GLenum target, struct gl_program *prog)
1226 {
1227 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1228 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1229 p->translated = 0;
1230
1231 /* Hack: make sure fog is correctly enabled according to this
1232 * fragment program's fog options.
1233 */
1234 if (p->FragProg.FogOption) {
1235 /* add extra instructions to do fog, then turn off FogOption field */
1236 _mesa_append_fog_code(ctx, &p->FragProg);
1237 p->FragProg.FogOption = GL_NONE;
1238 }
1239 }
1240
1241 (void) _tnl_program_string(ctx, target, prog);
1242
1243 /* XXX check if program is legal, within limits */
1244 return GL_TRUE;
1245 }
1246
1247 void
1248 i915_update_program(GLcontext *ctx)
1249 {
1250 struct intel_context *intel = intel_context(ctx);
1251 struct i915_context *i915 = i915_context(&intel->ctx);
1252 struct i915_fragment_program *fp =
1253 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1254
1255 if (i915->current_program != fp) {
1256 if (i915->current_program) {
1257 i915->current_program->on_hardware = 0;
1258 i915->current_program->params_uptodate = 0;
1259 }
1260
1261 i915->current_program = fp;
1262 }
1263
1264 if (!fp->translated)
1265 translate_program(fp);
1266
1267 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1268 }
1269
1270 void
1271 i915ValidateFragmentProgram(struct i915_context *i915)
1272 {
1273 GLcontext *ctx = &i915->intel.ctx;
1274 struct intel_context *intel = intel_context(ctx);
1275 TNLcontext *tnl = TNL_CONTEXT(ctx);
1276 struct vertex_buffer *VB = &tnl->vb;
1277
1278 struct i915_fragment_program *p =
1279 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1280
1281 const GLuint inputsRead = p->FragProg.Base.InputsRead;
1282 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1283 GLuint s2 = S2_TEXCOORD_NONE;
1284 int i, offset = 0;
1285
1286 /* Important:
1287 */
1288 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1289
1290 if (!p->translated)
1291 translate_program(p);
1292
1293 intel->vertex_attr_count = 0;
1294 intel->wpos_offset = 0;
1295 intel->wpos_size = 0;
1296 intel->coloroffset = 0;
1297 intel->specoffset = 0;
1298
1299 if (inputsRead & FRAG_BITS_TEX_ANY) {
1300 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1301 }
1302 else {
1303 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1304 }
1305
1306 if (inputsRead & FRAG_BIT_COL0) {
1307 intel->coloroffset = offset / 4;
1308 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1309 }
1310
1311 if (inputsRead & FRAG_BIT_COL1) {
1312 intel->specoffset = offset / 4;
1313 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1314 }
1315
1316 if ((inputsRead & FRAG_BIT_FOGC) || i915->vertex_fog != I915_FOG_NONE) {
1317 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1318 }
1319
1320 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1321 if (inputsRead & FRAG_BIT_TEX(i)) {
1322 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1323
1324 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1325 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1326
1327 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1328 }
1329 else if (inputsRead & FRAG_BIT_VAR(i)) {
1330 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1331
1332 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1333 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1334
1335 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1336 }
1337 else if (i == p->wpos_tex) {
1338
1339 /* If WPOS is required, duplicate the XYZ position data in an
1340 * unused texture coordinate:
1341 */
1342 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1343 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(3));
1344
1345 intel->wpos_offset = offset;
1346 intel->wpos_size = 3 * sizeof(GLuint);
1347
1348 EMIT_PAD(intel->wpos_size);
1349 }
1350 }
1351
1352 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1353 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1354 int k;
1355
1356 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1357
1358 /* Must do this *after* statechange, so as not to affect
1359 * buffered vertices reliant on the old state:
1360 */
1361 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1362 intel->vertex_attrs,
1363 intel->vertex_attr_count,
1364 intel->ViewportMatrix.m, 0);
1365
1366 intel->vertex_size >>= 2;
1367
1368 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1369 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1370
1371 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1372 assert(k);
1373 }
1374
1375 if (!p->params_uptodate)
1376 track_params(p);
1377
1378 if (!p->on_hardware)
1379 i915_upload_program(i915, p);
1380 }
1381
1382 void
1383 i915InitFragProgFuncs(struct dd_function_table *functions)
1384 {
1385 functions->BindProgram = i915BindProgram;
1386 functions->NewProgram = i915NewProgram;
1387 functions->DeleteProgram = i915DeleteProgram;
1388 functions->IsProgramNative = i915IsProgramNative;
1389 functions->ProgramStringNotify = i915ProgramStringNotify;
1390 }