projects / mesa.git / blob
? search:


d42da5a6b61b2e0bafb4da9090a17afd2fc774ae
[mesa.git] / src / mesa / drivers / dri / i915 / i915_fragprog.c
1 /**************************************************************************
2 *
3 * Copyright 2003 VMware, Inc.
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 VMWARE 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 VARYING_SLOT_POS:
101 src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
102 break;
103 case VARYING_SLOT_COL0:
104 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
105 break;
106 case VARYING_SLOT_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 VARYING_SLOT_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 VARYING_SLOT_TEX0:
115 case VARYING_SLOT_TEX1:
116 case VARYING_SLOT_TEX2:
117 case VARYING_SLOT_TEX3:
118 case VARYING_SLOT_TEX4:
119 case VARYING_SLOT_TEX5:
120 case VARYING_SLOT_TEX6:
121 case VARYING_SLOT_TEX7:
122 src = i915_emit_decl(p, REG_TYPE_T,
123 T_TEX0 + (source->Index - VARYING_SLOT_TEX0),
124 D0_CHANNEL_ALL);
125 break;
126
127 case VARYING_SLOT_VAR0:
128 case VARYING_SLOT_VAR0 + 1:
129 case VARYING_SLOT_VAR0 + 2:
130 case VARYING_SLOT_VAR0 + 3:
131 case VARYING_SLOT_VAR0 + 4:
132 case VARYING_SLOT_VAR0 + 5:
133 case VARYING_SLOT_VAR0 + 6:
134 case VARYING_SLOT_VAR0 + 7:
135 src = i915_emit_decl(p, REG_TYPE_T,
136 T_TEX0 + (source->Index - VARYING_SLOT_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 case FRAG_RESULT_DATA0:
150 src = UREG(REG_TYPE_OC, 0);
151 break;
152 case FRAG_RESULT_DEPTH:
153 src = UREG(REG_TYPE_OD, 0);
154 break;
155 default:
156 i915_program_error(p, "Bad source->Index: %d", source->Index);
157 return 0;
158 }
159 break;
160
161 /* Various paramters and env values. All emitted to
162 * hardware as program constants.
163 */
164 case PROGRAM_CONSTANT:
165 case PROGRAM_STATE_VAR:
166 case PROGRAM_UNIFORM:
167 src = i915_emit_param4fv(p,
168 &program->Base.Parameters->ParameterValues[source->Index][0].f);
169 break;
170
171 default:
172 i915_program_error(p, "Bad source->File: %d", source->File);
173 return 0;
174 }
175
176 src = swizzle(src,
177 GET_SWZ(source->Swizzle, 0),
178 GET_SWZ(source->Swizzle, 1),
179 GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
180
181 if (source->Negate)
182 src = negate(src,
183 GET_BIT(source->Negate, 0),
184 GET_BIT(source->Negate, 1),
185 GET_BIT(source->Negate, 2),
186 GET_BIT(source->Negate, 3));
187
188 return src;
189 }
190
191
192 static GLuint
193 get_result_vector(struct i915_fragment_program *p,
194 const struct prog_instruction *inst)
195 {
196 switch (inst->DstReg.File) {
197 case PROGRAM_OUTPUT:
198 switch (inst->DstReg.Index) {
199 case FRAG_RESULT_COLOR:
200 case FRAG_RESULT_DATA0:
201 return UREG(REG_TYPE_OC, 0);
202 case FRAG_RESULT_DEPTH:
203 p->depth_written = 1;
204 return UREG(REG_TYPE_OD, 0);
205 default:
206 i915_program_error(p, "Bad inst->DstReg.Index: %d",
207 inst->DstReg.Index);
208 return 0;
209 }
210 case PROGRAM_TEMPORARY:
211 return UREG(REG_TYPE_R, inst->DstReg.Index);
212 default:
213 i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
214 return 0;
215 }
216 }
217
218 static GLuint
219 get_result_flags(const struct prog_instruction *inst)
220 {
221 GLuint flags = 0;
222
223 if (inst->SaturateMode == SATURATE_ZERO_ONE)
224 flags |= A0_DEST_SATURATE;
225 if (inst->DstReg.WriteMask & WRITEMASK_X)
226 flags |= A0_DEST_CHANNEL_X;
227 if (inst->DstReg.WriteMask & WRITEMASK_Y)
228 flags |= A0_DEST_CHANNEL_Y;
229 if (inst->DstReg.WriteMask & WRITEMASK_Z)
230 flags |= A0_DEST_CHANNEL_Z;
231 if (inst->DstReg.WriteMask & WRITEMASK_W)
232 flags |= A0_DEST_CHANNEL_W;
233
234 return flags;
235 }
236
237 static GLuint
238 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
239 {
240 switch (bit) {
241 case TEXTURE_1D_INDEX:
242 return D0_SAMPLE_TYPE_2D;
243 case TEXTURE_2D_INDEX:
244 return D0_SAMPLE_TYPE_2D;
245 case TEXTURE_RECT_INDEX:
246 return D0_SAMPLE_TYPE_2D;
247 case TEXTURE_3D_INDEX:
248 return D0_SAMPLE_TYPE_VOLUME;
249 case TEXTURE_CUBE_INDEX:
250 return D0_SAMPLE_TYPE_CUBE;
251 default:
252 i915_program_error(p, "TexSrcBit: %d", bit);
253 return 0;
254 }
255 }
256
257 #define EMIT_TEX( OP ) \
258 do { \
259 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
260 const struct gl_fragment_program *program = &p->FragProg; \
261 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
262 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
263 unit, dim); \
264 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
265 /* Texel lookup */ \
266 \
267 i915_emit_texld( p, get_live_regs(p, inst), \
268 get_result_vector( p, inst ), \
269 get_result_flags( inst ), \
270 sampler, \
271 coord, \
272 OP); \
273 } while (0)
274
275 #define EMIT_ARITH( OP, N ) \
276 do { \
277 i915_emit_arith( p, \
278 OP, \
279 get_result_vector( p, inst ), \
280 get_result_flags( inst ), 0, \
281 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
282 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
283 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
284 } while (0)
285
286 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
287 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
288 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
289
290 /*
291 * TODO: consider moving this into core
292 */
293 static bool calc_live_regs( struct i915_fragment_program *p )
294 {
295 const struct gl_fragment_program *program = &p->FragProg;
296 GLuint regsUsed = ~((1 << I915_MAX_TEMPORARY) - 1);
297 uint8_t live_components[I915_MAX_TEMPORARY] = { 0, };
298 GLint i;
299
300 for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
301 struct prog_instruction *inst = &program->Base.Instructions[i];
302 int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
303 int a;
304
305 /* Register is written to: unmark as live for this and preceeding ops */
306 if (inst->DstReg.File == PROGRAM_TEMPORARY) {
307 if (inst->DstReg.Index >= I915_MAX_TEMPORARY)
308 return false;
309
310 live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
311 if (live_components[inst->DstReg.Index] == 0)
312 regsUsed &= ~(1 << inst->DstReg.Index);
313 }
314
315 for (a = 0; a < opArgs; a++) {
316 /* Register is read from: mark as live for this and preceeding ops */
317 if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
318 unsigned c;
319
320 if (inst->SrcReg[a].Index >= I915_MAX_TEMPORARY)
321 return false;
322
323 regsUsed |= 1 << inst->SrcReg[a].Index;
324
325 for (c = 0; c < 4; c++) {
326 const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
327
328 if (field <= SWIZZLE_W)
329 live_components[inst->SrcReg[a].Index] |= (1U << field);
330 }
331 }
332 }
333
334 p->usedRegs[i] = regsUsed;
335 }
336
337 return true;
338 }
339
340 static GLuint get_live_regs( struct i915_fragment_program *p,
341 const struct prog_instruction *inst )
342 {
343 const struct gl_fragment_program *program = &p->FragProg;
344 GLuint nr = inst - program->Base.Instructions;
345
346 return p->usedRegs[nr];
347 }
348
349
350 /* Possible concerns:
351 *
352 * SIN, COS -- could use another taylor step?
353 * LIT -- results seem a little different to sw mesa
354 * LOG -- different to mesa on negative numbers, but this is conformant.
355 *
356 * Parse failures -- Mesa doesn't currently give a good indication
357 * internally whether a particular program string parsed or not. This
358 * can lead to confusion -- hopefully we cope with it ok now.
359 *
360 */
361 static void
362 upload_program(struct i915_fragment_program *p)
363 {
364 const struct gl_fragment_program *program = &p->FragProg;
365 const struct prog_instruction *inst = program->Base.Instructions;
366
367 if (INTEL_DEBUG & DEBUG_WM)
368 _mesa_print_program(&program->Base);
369
370 /* Is this a parse-failed program? Ensure a valid program is
371 * loaded, as the flagging of an error isn't sufficient to stop
372 * this being uploaded to hardware.
373 */
374 if (inst[0].Opcode == OPCODE_END) {
375 GLuint tmp = i915_get_utemp(p);
376 i915_emit_arith(p,
377 A0_MOV,
378 UREG(REG_TYPE_OC, 0),
379 A0_DEST_CHANNEL_ALL, 0,
380 swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
381 return;
382 }
383
384 if (program->Base.NumInstructions > I915_MAX_INSN) {
385 i915_program_error(p, "Exceeded max instructions (%d out of %d)",
386 program->Base.NumInstructions, I915_MAX_INSN);
387 return;
388 }
389
390 /* Not always needed:
391 */
392 if (!calc_live_regs(p)) {
393 i915_program_error(p, "Could not allocate registers");
394 return;
395 }
396
397 while (1) {
398 GLuint src0, src1, src2, flags;
399 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
400
401 switch (inst->Opcode) {
402 case OPCODE_ABS:
403 src0 = src_vector(p, &inst->SrcReg[0], program);
404 i915_emit_arith(p,
405 A0_MAX,
406 get_result_vector(p, inst),
407 get_result_flags(inst), 0,
408 src0, negate(src0, 1, 1, 1, 1), 0);
409 break;
410
411 case OPCODE_ADD:
412 EMIT_2ARG_ARITH(A0_ADD);
413 break;
414
415 case OPCODE_CMP:
416 src0 = src_vector(p, &inst->SrcReg[0], program);
417 src1 = src_vector(p, &inst->SrcReg[1], program);
418 src2 = src_vector(p, &inst->SrcReg[2], program);
419 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
420 break;
421
422 case OPCODE_COS:
423 src0 = src_vector(p, &inst->SrcReg[0], program);
424 tmp = i915_get_utemp(p);
425 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
426 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
427
428 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
429 i915_emit_arith(p,
430 A0_MAD,
431 tmp, A0_DEST_CHANNEL_X, 0,
432 src0,
433 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
434 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
435
436 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
437
438 i915_emit_arith(p,
439 A0_MAD,
440 tmp, A0_DEST_CHANNEL_X, 0,
441 tmp,
442 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
443 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
444
445 /* Compute COS with the same calculation used for SIN, but a
446 * different source range has been mapped to [-1,1] this time.
447 */
448
449 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
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.y * tmp.x; {x, x * abs(x), 0, 0} */
458 i915_emit_arith(p,
459 A0_MUL,
460 tmp, A0_DEST_CHANNEL_Y, 0,
461 swizzle(tmp, ZERO, X, ZERO, ZERO),
462 tmp,
463 0);
464
465 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
466 i915_emit_arith(p,
467 A0_DP3,
468 tmp, A0_DEST_CHANNEL_X, 0,
469 tmp,
470 swizzle(consts1, X, Y, ZERO, ZERO),
471 0);
472
473 /* tmp.x now contains a first approximation (y). Now, weight it
474 * against tmp.y**2 to get closer.
475 */
476 i915_emit_arith(p,
477 A0_MAX,
478 tmp, A0_DEST_CHANNEL_Y, 0,
479 swizzle(tmp, ZERO, X, ZERO, ZERO),
480 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
481 0);
482
483 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
484 i915_emit_arith(p,
485 A0_MAD,
486 tmp, A0_DEST_CHANNEL_Y, 0,
487 swizzle(tmp, ZERO, X, ZERO, ZERO),
488 swizzle(tmp, ZERO, Y, ZERO, ZERO),
489 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
490
491 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
492 i915_emit_arith(p,
493 A0_MAD,
494 get_result_vector(p, inst),
495 get_result_flags(inst), 0,
496 swizzle(consts1, W, W, W, W),
497 swizzle(tmp, Y, Y, Y, Y),
498 swizzle(tmp, X, X, X, X));
499 break;
500
501 case OPCODE_DP2:
502 src0 = src_vector(p, &inst->SrcReg[0], program);
503 src1 = src_vector(p, &inst->SrcReg[1], program);
504 i915_emit_arith(p,
505 A0_DP3,
506 get_result_vector(p, inst),
507 get_result_flags(inst), 0,
508 swizzle(src0, X, Y, ZERO, ZERO),
509 swizzle(src1, X, Y, ZERO, ZERO),
510 0);
511 break;
512
513 case OPCODE_DP3:
514 EMIT_2ARG_ARITH(A0_DP3);
515 break;
516
517 case OPCODE_DP4:
518 EMIT_2ARG_ARITH(A0_DP4);
519 break;
520
521 case OPCODE_DPH:
522 src0 = src_vector(p, &inst->SrcReg[0], program);
523 src1 = src_vector(p, &inst->SrcReg[1], program);
524
525 i915_emit_arith(p,
526 A0_DP4,
527 get_result_vector(p, inst),
528 get_result_flags(inst), 0,
529 swizzle(src0, X, Y, Z, ONE), src1, 0);
530 break;
531
532 case OPCODE_DST:
533 src0 = src_vector(p, &inst->SrcReg[0], program);
534 src1 = src_vector(p, &inst->SrcReg[1], program);
535
536 /* result[0] = 1 * 1;
537 * result[1] = a[1] * b[1];
538 * result[2] = a[2] * 1;
539 * result[3] = 1 * b[3];
540 */
541 i915_emit_arith(p,
542 A0_MUL,
543 get_result_vector(p, inst),
544 get_result_flags(inst), 0,
545 swizzle(src0, ONE, Y, Z, ONE),
546 swizzle(src1, ONE, Y, ONE, W), 0);
547 break;
548
549 case OPCODE_EX2:
550 src0 = src_vector(p, &inst->SrcReg[0], program);
551
552 i915_emit_arith(p,
553 A0_EXP,
554 get_result_vector(p, inst),
555 get_result_flags(inst), 0,
556 swizzle(src0, X, X, X, X), 0, 0);
557 break;
558
559 case OPCODE_FLR:
560 EMIT_1ARG_ARITH(A0_FLR);
561 break;
562
563 case OPCODE_TRUNC:
564 EMIT_1ARG_ARITH(A0_TRC);
565 break;
566
567 case OPCODE_FRC:
568 EMIT_1ARG_ARITH(A0_FRC);
569 break;
570
571 case OPCODE_KIL:
572 src0 = src_vector(p, &inst->SrcReg[0], program);
573 tmp = i915_get_utemp(p);
574
575 i915_emit_texld(p, get_live_regs(p, inst),
576 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
577 0, src0, T0_TEXKILL);
578 break;
579
580 case OPCODE_KIL_NV:
581 if (inst->DstReg.CondMask == COND_TR) {
582 tmp = i915_get_utemp(p);
583
584 /* The KIL instruction discards the fragment if any component of
585 * the source is < 0. Emit an immediate operand of {-1}.xywz.
586 */
587 i915_emit_texld(p, get_live_regs(p, inst),
588 tmp, A0_DEST_CHANNEL_ALL,
589 0, /* use a dummy dest reg */
590 negate(swizzle(tmp, ONE, ONE, ONE, ONE),
591 1, 1, 1, 1),
592 T0_TEXKILL);
593 } else {
594 p->error = 1;
595 i915_program_error(p, "Unsupported KIL_NV condition code: %d",
596 inst->DstReg.CondMask);
597 }
598 break;
599
600 case OPCODE_LG2:
601 src0 = src_vector(p, &inst->SrcReg[0], program);
602
603 i915_emit_arith(p,
604 A0_LOG,
605 get_result_vector(p, inst),
606 get_result_flags(inst), 0,
607 swizzle(src0, X, X, X, X), 0, 0);
608 break;
609
610 case OPCODE_LIT:
611 src0 = src_vector(p, &inst->SrcReg[0], program);
612 tmp = i915_get_utemp(p);
613
614 /* tmp = max( a.xyzw, a.00zw )
615 * XXX: Clamp tmp.w to -128..128
616 * tmp.y = log(tmp.y)
617 * tmp.y = tmp.w * tmp.y
618 * tmp.y = exp(tmp.y)
619 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
620 */
621 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
622 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
623
624 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
625 swizzle(tmp, Y, Y, Y, Y), 0, 0);
626
627 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
628 swizzle(tmp, ZERO, Y, ZERO, ZERO),
629 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
630
631 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
632 swizzle(tmp, Y, Y, Y, Y), 0, 0);
633
634 i915_emit_arith(p, A0_CMP,
635 get_result_vector(p, inst),
636 get_result_flags(inst), 0,
637 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
638 swizzle(tmp, ONE, X, ZERO, ONE),
639 swizzle(tmp, ONE, X, Y, ONE));
640
641 break;
642
643 case OPCODE_LRP:
644 src0 = src_vector(p, &inst->SrcReg[0], program);
645 src1 = src_vector(p, &inst->SrcReg[1], program);
646 src2 = src_vector(p, &inst->SrcReg[2], program);
647 flags = get_result_flags(inst);
648 tmp = i915_get_utemp(p);
649
650 /* b*a + c*(1-a)
651 *
652 * b*a + c - ca
653 *
654 * tmp = b*a + c,
655 * result = (-c)*a + tmp
656 */
657 i915_emit_arith(p, A0_MAD, tmp,
658 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
659
660 i915_emit_arith(p, A0_MAD,
661 get_result_vector(p, inst),
662 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
663 break;
664
665 case OPCODE_MAD:
666 EMIT_3ARG_ARITH(A0_MAD);
667 break;
668
669 case OPCODE_MAX:
670 EMIT_2ARG_ARITH(A0_MAX);
671 break;
672
673 case OPCODE_MIN:
674 EMIT_2ARG_ARITH(A0_MIN);
675 break;
676
677 case OPCODE_MOV:
678 EMIT_1ARG_ARITH(A0_MOV);
679 break;
680
681 case OPCODE_MUL:
682 EMIT_2ARG_ARITH(A0_MUL);
683 break;
684
685 case OPCODE_POW:
686 src0 = src_vector(p, &inst->SrcReg[0], program);
687 src1 = src_vector(p, &inst->SrcReg[1], program);
688 tmp = i915_get_utemp(p);
689 flags = get_result_flags(inst);
690
691 /* XXX: masking on intermediate values, here and elsewhere.
692 */
693 i915_emit_arith(p,
694 A0_LOG,
695 tmp, A0_DEST_CHANNEL_X, 0,
696 swizzle(src0, X, X, X, X), 0, 0);
697
698 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
699
700
701 i915_emit_arith(p,
702 A0_EXP,
703 get_result_vector(p, inst),
704 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
705
706 break;
707
708 case OPCODE_RCP:
709 src0 = src_vector(p, &inst->SrcReg[0], program);
710
711 i915_emit_arith(p,
712 A0_RCP,
713 get_result_vector(p, inst),
714 get_result_flags(inst), 0,
715 swizzle(src0, X, X, X, X), 0, 0);
716 break;
717
718 case OPCODE_RSQ:
719
720 src0 = src_vector(p, &inst->SrcReg[0], program);
721
722 i915_emit_arith(p,
723 A0_RSQ,
724 get_result_vector(p, inst),
725 get_result_flags(inst), 0,
726 swizzle(src0, X, X, X, X), 0, 0);
727 break;
728
729 case OPCODE_SCS:
730 src0 = src_vector(p, &inst->SrcReg[0], program);
731 tmp = i915_get_utemp(p);
732
733 /*
734 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
735 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
736 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
737 * scs.x = DP4 t1, sin_constants
738 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
739 * scs.y = DP4 t1, cos_constants
740 */
741 i915_emit_arith(p,
742 A0_MUL,
743 tmp, A0_DEST_CHANNEL_XY, 0,
744 swizzle(src0, X, X, ONE, ONE),
745 swizzle(src0, X, ONE, ONE, ONE), 0);
746
747 i915_emit_arith(p,
748 A0_MUL,
749 tmp, A0_DEST_CHANNEL_ALL, 0,
750 swizzle(tmp, X, Y, X, Y),
751 swizzle(tmp, X, X, ONE, ONE), 0);
752
753 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
754 GLuint tmp1;
755
756 if (inst->DstReg.WriteMask & WRITEMASK_X)
757 tmp1 = i915_get_utemp(p);
758 else
759 tmp1 = tmp;
760
761 i915_emit_arith(p,
762 A0_MUL,
763 tmp1, A0_DEST_CHANNEL_ALL, 0,
764 swizzle(tmp, X, Y, Y, W),
765 swizzle(tmp, X, Z, ONE, ONE), 0);
766
767 i915_emit_arith(p,
768 A0_DP4,
769 get_result_vector(p, inst),
770 A0_DEST_CHANNEL_Y, 0,
771 swizzle(tmp1, W, Z, Y, X),
772 i915_emit_const4fv(p, sin_constants), 0);
773 }
774
775 if (inst->DstReg.WriteMask & WRITEMASK_X) {
776 i915_emit_arith(p,
777 A0_MUL,
778 tmp, A0_DEST_CHANNEL_XYZ, 0,
779 swizzle(tmp, X, X, Z, ONE),
780 swizzle(tmp, Z, ONE, ONE, ONE), 0);
781
782 i915_emit_arith(p,
783 A0_DP4,
784 get_result_vector(p, inst),
785 A0_DEST_CHANNEL_X, 0,
786 swizzle(tmp, ONE, Z, Y, X),
787 i915_emit_const4fv(p, cos_constants), 0);
788 }
789 break;
790
791 case OPCODE_SEQ:
792 tmp = i915_get_utemp(p);
793 flags = get_result_flags(inst);
794 dst = get_result_vector(p, inst);
795
796 /* If both operands are uniforms or constants, we get 5 instructions
797 * like:
798 *
799 * U[1] = MOV CONST[1]
800 * U[0].xyz = SGE CONST[0].xxxx, U[1]
801 * U[1] = MOV CONST[1].-x-y-z-w
802 * R[0].xyz = SGE CONST[0].-x-x-x-x, U[1]
803 * R[0].xyz = MUL R[0], U[0]
804 *
805 * This code is stupid. Instead of having the individual calls to
806 * i915_emit_arith generate the moves to utemps, do it in the caller.
807 * This results in code like:
808 *
809 * U[1] = MOV CONST[1]
810 * U[0].xyz = SGE CONST[0].xxxx, U[1]
811 * R[0].xyz = SGE CONST[0].-x-x-x-x, U[1].-x-y-z-w
812 * R[0].xyz = MUL R[0], U[0]
813 */
814 src0 = src_vector(p, &inst->SrcReg[0], program);
815 src1 = src_vector(p, &inst->SrcReg[1], program);
816
817 if (GET_UREG_TYPE(src0) == REG_TYPE_CONST
818 && GET_UREG_TYPE(src1) == REG_TYPE_CONST) {
819 unsigned tmp = i915_get_utemp(p);
820
821 i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
822 src1, 0, 0);
823
824 src1 = tmp;
825 }
826
827 /* tmp = src1 >= src2 */
828 i915_emit_arith(p,
829 A0_SGE,
830 tmp,
831 flags, 0,
832 src0,
833 src1,
834 0);
835 /* dst = src1 <= src2 */
836 i915_emit_arith(p,
837 A0_SGE,
838 dst,
839 flags, 0,
840 negate(src0, 1, 1, 1, 1),
841 negate(src1, 1, 1, 1, 1),
842 0);
843 /* dst = tmp && dst */
844 i915_emit_arith(p,
845 A0_MUL,
846 dst,
847 flags, 0,
848 dst,
849 tmp,
850 0);
851 break;
852
853 case OPCODE_SIN:
854 src0 = src_vector(p, &inst->SrcReg[0], program);
855 tmp = i915_get_utemp(p);
856 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
857 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
858
859 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
860 i915_emit_arith(p,
861 A0_MAD,
862 tmp, A0_DEST_CHANNEL_X, 0,
863 src0,
864 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
865 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
866
867 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
868
869 i915_emit_arith(p,
870 A0_MAD,
871 tmp, A0_DEST_CHANNEL_X, 0,
872 tmp,
873 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
874 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
875
876 /* Compute sin using a quadratic and quartic. It gives continuity
877 * that repeating the Taylor series lacks every 2*pi, and has
878 * reduced error.
879 *
880 * The idea was described at:
881 * http://www.devmaster.net/forums/showthread.php?t=5784
882 */
883
884 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
885 i915_emit_arith(p,
886 A0_MAX,
887 tmp, A0_DEST_CHANNEL_Y, 0,
888 swizzle(tmp, ZERO, X, ZERO, ZERO),
889 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
890 0);
891
892 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
893 i915_emit_arith(p,
894 A0_MUL,
895 tmp, A0_DEST_CHANNEL_Y, 0,
896 swizzle(tmp, ZERO, X, ZERO, ZERO),
897 tmp,
898 0);
899
900 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
901 i915_emit_arith(p,
902 A0_DP3,
903 tmp, A0_DEST_CHANNEL_X, 0,
904 tmp,
905 swizzle(consts1, X, Y, ZERO, ZERO),
906 0);
907
908 /* tmp.x now contains a first approximation (y). Now, weight it
909 * against tmp.y**2 to get closer.
910 */
911 i915_emit_arith(p,
912 A0_MAX,
913 tmp, A0_DEST_CHANNEL_Y, 0,
914 swizzle(tmp, ZERO, X, ZERO, ZERO),
915 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
916 0);
917
918 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
919 i915_emit_arith(p,
920 A0_MAD,
921 tmp, A0_DEST_CHANNEL_Y, 0,
922 swizzle(tmp, ZERO, X, ZERO, ZERO),
923 swizzle(tmp, ZERO, Y, ZERO, ZERO),
924 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
925
926 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
927 i915_emit_arith(p,
928 A0_MAD,
929 get_result_vector(p, inst),
930 get_result_flags(inst), 0,
931 swizzle(consts1, W, W, W, W),
932 swizzle(tmp, Y, Y, Y, Y),
933 swizzle(tmp, X, X, X, X));
934
935 break;
936
937 case OPCODE_SGE:
938 EMIT_2ARG_ARITH(A0_SGE);
939 break;
940
941 case OPCODE_SGT:
942 i915_emit_arith(p,
943 A0_SLT,
944 get_result_vector( p, inst ),
945 get_result_flags( inst ), 0,
946 negate(src_vector( p, &inst->SrcReg[0], program),
947 1, 1, 1, 1),
948 negate(src_vector( p, &inst->SrcReg[1], program),
949 1, 1, 1, 1),
950 0);
951 break;
952
953 case OPCODE_SLE:
954 i915_emit_arith(p,
955 A0_SGE,
956 get_result_vector( p, inst ),
957 get_result_flags( inst ), 0,
958 negate(src_vector( p, &inst->SrcReg[0], program),
959 1, 1, 1, 1),
960 negate(src_vector( p, &inst->SrcReg[1], program),
961 1, 1, 1, 1),
962 0);
963 break;
964
965 case OPCODE_SLT:
966 EMIT_2ARG_ARITH(A0_SLT);
967 break;
968
969 case OPCODE_SNE:
970 tmp = i915_get_utemp(p);
971 flags = get_result_flags(inst);
972 dst = get_result_vector(p, inst);
973
974 /* If both operands are uniforms or constants, we get 5 instructions
975 * like:
976 *
977 * U[1] = MOV CONST[1]
978 * U[0].xyz = SLT CONST[0].xxxx, U[1]
979 * U[1] = MOV CONST[1].-x-y-z-w
980 * R[0].xyz = SLT CONST[0].-x-x-x-x, U[1]
981 * R[0].xyz = MUL R[0], U[0]
982 *
983 * This code is stupid. Instead of having the individual calls to
984 * i915_emit_arith generate the moves to utemps, do it in the caller.
985 * This results in code like:
986 *
987 * U[1] = MOV CONST[1]
988 * U[0].xyz = SLT CONST[0].xxxx, U[1]
989 * R[0].xyz = SLT CONST[0].-x-x-x-x, U[1].-x-y-z-w
990 * R[0].xyz = MUL R[0], U[0]
991 */
992 src0 = src_vector(p, &inst->SrcReg[0], program);
993 src1 = src_vector(p, &inst->SrcReg[1], program);
994
995 if (GET_UREG_TYPE(src0) == REG_TYPE_CONST
996 && GET_UREG_TYPE(src1) == REG_TYPE_CONST) {
997 unsigned tmp = i915_get_utemp(p);
998
999 i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
1000 src1, 0, 0);
1001
1002 src1 = tmp;
1003 }
1004
1005 /* tmp = src1 < src2 */
1006 i915_emit_arith(p,
1007 A0_SLT,
1008 tmp,
1009 flags, 0,
1010 src0,
1011 src1,
1012 0);
1013 /* dst = src1 > src2 */
1014 i915_emit_arith(p,
1015 A0_SLT,
1016 dst,
1017 flags, 0,
1018 negate(src0, 1, 1, 1, 1),
1019 negate(src1, 1, 1, 1, 1),
1020 0);
1021 /* dst = tmp || dst */
1022 i915_emit_arith(p,
1023 A0_ADD,
1024 dst,
1025 flags | A0_DEST_SATURATE, 0,
1026 dst,
1027 tmp,
1028 0);
1029 break;
1030
1031 case OPCODE_SSG:
1032 dst = get_result_vector(p, inst);
1033 flags = get_result_flags(inst);
1034 src0 = src_vector(p, &inst->SrcReg[0], program);
1035 tmp = i915_get_utemp(p);
1036
1037 /* tmp = (src < 0.0) */
1038 i915_emit_arith(p,
1039 A0_SLT,
1040 tmp,
1041 flags, 0,
1042 src0,
1043 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1044 0);
1045
1046 /* dst = (0.0 < src) */
1047 i915_emit_arith(p,
1048 A0_SLT,
1049 dst,
1050 flags, 0,
1051 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1052 src0,
1053 0);
1054
1055 /* dst = (src > 0.0) - (src < 0.0) */
1056 i915_emit_arith(p,
1057 A0_ADD,
1058 dst,
1059 flags, 0,
1060 dst,
1061 negate(tmp, 1, 1, 1, 1),
1062 0);
1063
1064 break;
1065
1066 case OPCODE_SUB:
1067 src0 = src_vector(p, &inst->SrcReg[0], program);
1068 src1 = src_vector(p, &inst->SrcReg[1], program);
1069
1070 i915_emit_arith(p,
1071 A0_ADD,
1072 get_result_vector(p, inst),
1073 get_result_flags(inst), 0,
1074 src0, negate(src1, 1, 1, 1, 1), 0);
1075 break;
1076
1077 case OPCODE_SWZ:
1078 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
1079 break;
1080
1081 case OPCODE_TEX:
1082 EMIT_TEX(T0_TEXLD);
1083 break;
1084
1085 case OPCODE_TXB:
1086 EMIT_TEX(T0_TEXLDB);
1087 break;
1088
1089 case OPCODE_TXP:
1090 EMIT_TEX(T0_TEXLDP);
1091 break;
1092
1093 case OPCODE_XPD:
1094 /* Cross product:
1095 * result.x = src0.y * src1.z - src0.z * src1.y;
1096 * result.y = src0.z * src1.x - src0.x * src1.z;
1097 * result.z = src0.x * src1.y - src0.y * src1.x;
1098 * result.w = undef;
1099 */
1100 src0 = src_vector(p, &inst->SrcReg[0], program);
1101 src1 = src_vector(p, &inst->SrcReg[1], program);
1102 tmp = i915_get_utemp(p);
1103
1104 i915_emit_arith(p,
1105 A0_MUL,
1106 tmp, A0_DEST_CHANNEL_ALL, 0,
1107 swizzle(src0, Z, X, Y, ONE),
1108 swizzle(src1, Y, Z, X, ONE), 0);
1109
1110 i915_emit_arith(p,
1111 A0_MAD,
1112 get_result_vector(p, inst),
1113 get_result_flags(inst), 0,
1114 swizzle(src0, Y, Z, X, ONE),
1115 swizzle(src1, Z, X, Y, ONE),
1116 negate(tmp, 1, 1, 1, 0));
1117 break;
1118
1119 case OPCODE_END:
1120 return;
1121
1122 case OPCODE_BGNLOOP:
1123 case OPCODE_BGNSUB:
1124 case OPCODE_BRK:
1125 case OPCODE_CAL:
1126 case OPCODE_CONT:
1127 case OPCODE_DDX:
1128 case OPCODE_DDY:
1129 case OPCODE_ELSE:
1130 case OPCODE_ENDIF:
1131 case OPCODE_ENDLOOP:
1132 case OPCODE_ENDSUB:
1133 case OPCODE_IF:
1134 case OPCODE_RET:
1135 p->error = 1;
1136 i915_program_error(p, "Unsupported opcode: %s",
1137 _mesa_opcode_string(inst->Opcode));
1138 return;
1139
1140 case OPCODE_EXP:
1141 case OPCODE_LOG:
1142 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1143 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1144 * Instead, it translates to EX2 or LG2.
1145 */
1146 case OPCODE_TXD:
1147 case OPCODE_TXL:
1148 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1149 * only NV_vp/fp appears to emit them.
1150 */
1151 default:
1152 i915_program_error(p, "bad opcode: %s",
1153 _mesa_opcode_string(inst->Opcode));
1154 return;
1155 }
1156
1157 inst++;
1158 i915_release_utemps(p);
1159 }
1160 }
1161
1162 /* Rather than trying to intercept and jiggle depth writes during
1163 * emit, just move the value into its correct position at the end of
1164 * the program:
1165 */
1166 static void
1167 fixup_depth_write(struct i915_fragment_program *p)
1168 {
1169 if (p->depth_written) {
1170 GLuint depth = UREG(REG_TYPE_OD, 0);
1171
1172 i915_emit_arith(p,
1173 A0_MOV,
1174 depth, A0_DEST_CHANNEL_W, 0,
1175 swizzle(depth, X, Y, Z, Z), 0, 0);
1176 }
1177 }
1178
1179
1180 static void
1181 check_wpos(struct i915_fragment_program *p)
1182 {
1183 GLbitfield64 inputs = p->FragProg.Base.InputsRead;
1184 GLint i;
1185
1186 p->wpos_tex = -1;
1187
1188 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1189 if (inputs & (VARYING_BIT_TEX(i) | VARYING_BIT_VAR(i)))
1190 continue;
1191 else if (inputs & VARYING_BIT_POS) {
1192 p->wpos_tex = i;
1193 inputs &= ~VARYING_BIT_POS;
1194 }
1195 }
1196
1197 if (inputs & VARYING_BIT_POS) {
1198 i915_program_error(p, "No free texcoord for wpos value");
1199 }
1200 }
1201
1202
1203 static void
1204 translate_program(struct i915_fragment_program *p)
1205 {
1206 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1207
1208 if (INTEL_DEBUG & DEBUG_WM) {
1209 printf("fp:\n");
1210 _mesa_print_program(&p->FragProg.Base);
1211 printf("\n");
1212 }
1213
1214 i915_init_program(i915, p);
1215 check_wpos(p);
1216 upload_program(p);
1217 fixup_depth_write(p);
1218 i915_fini_program(p);
1219
1220 p->translated = 1;
1221 }
1222
1223
1224 static void
1225 track_params(struct i915_fragment_program *p)
1226 {
1227 GLint i;
1228
1229 if (p->nr_params)
1230 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1231
1232 for (i = 0; i < p->nr_params; i++) {
1233 GLint reg = p->param[i].reg;
1234 COPY_4V(p->constant[reg], p->param[i].values);
1235 }
1236
1237 p->params_uptodate = 1;
1238 p->on_hardware = 0; /* overkill */
1239 }
1240
1241
1242 static void
1243 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1244 {
1245 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1246 struct i915_context *i915 = I915_CONTEXT(ctx);
1247 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1248
1249 if (i915->current_program == p)
1250 return;
1251
1252 if (i915->current_program) {
1253 i915->current_program->on_hardware = 0;
1254 i915->current_program->params_uptodate = 0;
1255 }
1256
1257 i915->current_program = p;
1258
1259 assert(p->on_hardware == 0);
1260 assert(p->params_uptodate == 0);
1261
1262 }
1263 }
1264
1265 static struct gl_program *
1266 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1267 {
1268 switch (target) {
1269 case GL_VERTEX_PROGRAM_ARB:
1270 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1271 target, id);
1272
1273 case GL_FRAGMENT_PROGRAM_ARB:{
1274 struct i915_fragment_program *prog =
1275 CALLOC_STRUCT(i915_fragment_program);
1276 if (prog) {
1277 i915_init_program(I915_CONTEXT(ctx), prog);
1278
1279 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1280 target, id);
1281 }
1282 else
1283 return NULL;
1284 }
1285
1286 default:
1287 /* Just fallback:
1288 */
1289 return _mesa_new_program(ctx, target, id);
1290 }
1291 }
1292
1293 static void
1294 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1295 {
1296 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1297 struct i915_context *i915 = I915_CONTEXT(ctx);
1298 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1299
1300 if (i915->current_program == p)
1301 i915->current_program = 0;
1302 }
1303
1304 _mesa_delete_program(ctx, prog);
1305 }
1306
1307
1308 static GLboolean
1309 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1310 {
1311 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1312 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1313
1314 if (!p->translated)
1315 translate_program(p);
1316
1317 return !p->error;
1318 }
1319 else
1320 return true;
1321 }
1322
1323 static GLboolean
1324 i915ProgramStringNotify(struct gl_context * ctx,
1325 GLenum target, struct gl_program *prog)
1326 {
1327 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1328 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1329 p->translated = 0;
1330 }
1331
1332 (void) _tnl_program_string(ctx, target, prog);
1333
1334 /* XXX check if program is legal, within limits */
1335 return true;
1336 }
1337
1338 static void
1339 i915SamplerUniformChange(struct gl_context *ctx,
1340 GLenum target, struct gl_program *prog)
1341 {
1342 i915ProgramStringNotify(ctx, target, prog);
1343 }
1344
1345 void
1346 i915_update_program(struct gl_context *ctx)
1347 {
1348 struct intel_context *intel = intel_context(ctx);
1349 struct i915_context *i915 = i915_context(&intel->ctx);
1350 struct i915_fragment_program *fp =
1351 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1352
1353 if (i915->current_program != fp) {
1354 if (i915->current_program) {
1355 i915->current_program->on_hardware = 0;
1356 i915->current_program->params_uptodate = 0;
1357 }
1358
1359 i915->current_program = fp;
1360 }
1361
1362 if (!fp->translated)
1363 translate_program(fp);
1364
1365 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1366 }
1367
1368 void
1369 i915ValidateFragmentProgram(struct i915_context *i915)
1370 {
1371 struct gl_context *ctx = &i915->intel.ctx;
1372 struct intel_context *intel = intel_context(ctx);
1373 TNLcontext *tnl = TNL_CONTEXT(ctx);
1374 struct vertex_buffer *VB = &tnl->vb;
1375
1376 struct i915_fragment_program *p =
1377 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1378
1379 const GLbitfield64 inputsRead = p->FragProg.Base.InputsRead;
1380 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1381 GLuint s2 = S2_TEXCOORD_NONE;
1382 int i, offset = 0;
1383
1384 /* Important:
1385 */
1386 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1387
1388 if (!p->translated)
1389 translate_program(p);
1390
1391 intel->vertex_attr_count = 0;
1392 intel->wpos_offset = 0;
1393 intel->coloroffset = 0;
1394 intel->specoffset = 0;
1395
1396 if (inputsRead & VARYING_BITS_TEX_ANY || p->wpos_tex != -1) {
1397 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1398 }
1399 else {
1400 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1401 }
1402
1403 /* Handle gl_PointSize builtin var here */
1404 if (ctx->Point._Attenuated || ctx->VertexProgram.PointSizeEnabled)
1405 EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, S4_VFMT_POINT_WIDTH, 4);
1406
1407 if (inputsRead & VARYING_BIT_COL0) {
1408 intel->coloroffset = offset / 4;
1409 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1410 }
1411
1412 if (inputsRead & VARYING_BIT_COL1) {
1413 intel->specoffset = offset / 4;
1414 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1415 }
1416
1417 if ((inputsRead & VARYING_BIT_FOGC)) {
1418 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1419 }
1420
1421 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1422 if (inputsRead & VARYING_BIT_TEX(i)) {
1423 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1424
1425 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1426 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1427
1428 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1429 }
1430 else if (inputsRead & VARYING_BIT_VAR(i)) {
1431 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1432
1433 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1434 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1435
1436 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1437 }
1438 else if (i == p->wpos_tex) {
1439 int wpos_size = 4 * sizeof(float);
1440 /* If WPOS is required, duplicate the XYZ position data in an
1441 * unused texture coordinate:
1442 */
1443 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1444 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(wpos_size));
1445
1446 intel->wpos_offset = offset;
1447 EMIT_PAD(wpos_size);
1448 }
1449 }
1450
1451 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1452 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1453 int k;
1454
1455 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1456
1457 /* Must do this *after* statechange, so as not to affect
1458 * buffered vertices reliant on the old state:
1459 */
1460 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1461 intel->vertex_attrs,
1462 intel->vertex_attr_count,
1463 intel->ViewportMatrix.m, 0);
1464
1465 assert(intel->prim.current_offset == intel->prim.start_offset);
1466 intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1467 intel->prim.current_offset = intel->prim.start_offset;
1468
1469 intel->vertex_size >>= 2;
1470
1471 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1472 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1473
1474 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1475 assert(k);
1476 }
1477
1478 if (!p->params_uptodate)
1479 track_params(p);
1480
1481 if (!p->on_hardware)
1482 i915_upload_program(i915, p);
1483
1484 if (INTEL_DEBUG & DEBUG_WM) {
1485 printf("i915:\n");
1486 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1487 }
1488 }
1489
1490 void
1491 i915InitFragProgFuncs(struct dd_function_table *functions)
1492 {
1493 functions->BindProgram = i915BindProgram;
1494 functions->NewProgram = i915NewProgram;
1495 functions->DeleteProgram = i915DeleteProgram;
1496 functions->IsProgramNative = i915IsProgramNative;
1497 functions->ProgramStringNotify = i915ProgramStringNotify;
1498 functions->SamplerUniformChange = i915SamplerUniformChange;
1499 }