projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
i965: Silence 'comparison is always true' warning
[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 src0 = src_vector(p, &inst->SrcReg[0], program);
675 src1 = src_vector(p, &inst->SrcReg[1], program);
676 tmp = i915_get_utemp(p);
677 flags = get_result_flags(inst);
678
679 i915_emit_arith(p,
680 A0_MAX,
681 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
682 negate(src0, 1, 1, 1, 1),
683 negate(src1, 1, 1, 1, 1), 0);
684
685 i915_emit_arith(p,
686 A0_MOV,
687 get_result_vector(p, inst),
688 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
689 break;
690
691 case OPCODE_MOV:
692 EMIT_1ARG_ARITH(A0_MOV);
693 break;
694
695 case OPCODE_MUL:
696 EMIT_2ARG_ARITH(A0_MUL);
697 break;
698
699 case OPCODE_POW:
700 src0 = src_vector(p, &inst->SrcReg[0], program);
701 src1 = src_vector(p, &inst->SrcReg[1], program);
702 tmp = i915_get_utemp(p);
703 flags = get_result_flags(inst);
704
705 /* XXX: masking on intermediate values, here and elsewhere.
706 */
707 i915_emit_arith(p,
708 A0_LOG,
709 tmp, A0_DEST_CHANNEL_X, 0,
710 swizzle(src0, X, X, X, X), 0, 0);
711
712 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
713
714
715 i915_emit_arith(p,
716 A0_EXP,
717 get_result_vector(p, inst),
718 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
719
720 break;
721
722 case OPCODE_RCP:
723 src0 = src_vector(p, &inst->SrcReg[0], program);
724
725 i915_emit_arith(p,
726 A0_RCP,
727 get_result_vector(p, inst),
728 get_result_flags(inst), 0,
729 swizzle(src0, X, X, X, X), 0, 0);
730 break;
731
732 case OPCODE_RSQ:
733
734 src0 = src_vector(p, &inst->SrcReg[0], program);
735
736 i915_emit_arith(p,
737 A0_RSQ,
738 get_result_vector(p, inst),
739 get_result_flags(inst), 0,
740 swizzle(src0, X, X, X, X), 0, 0);
741 break;
742
743 case OPCODE_SCS:
744 src0 = src_vector(p, &inst->SrcReg[0], program);
745 tmp = i915_get_utemp(p);
746
747 /*
748 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
749 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
750 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
751 * scs.x = DP4 t1, sin_constants
752 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
753 * scs.y = DP4 t1, cos_constants
754 */
755 i915_emit_arith(p,
756 A0_MUL,
757 tmp, A0_DEST_CHANNEL_XY, 0,
758 swizzle(src0, X, X, ONE, ONE),
759 swizzle(src0, X, ONE, ONE, ONE), 0);
760
761 i915_emit_arith(p,
762 A0_MUL,
763 tmp, A0_DEST_CHANNEL_ALL, 0,
764 swizzle(tmp, X, Y, X, Y),
765 swizzle(tmp, X, X, ONE, ONE), 0);
766
767 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
768 GLuint tmp1;
769
770 if (inst->DstReg.WriteMask & WRITEMASK_X)
771 tmp1 = i915_get_utemp(p);
772 else
773 tmp1 = tmp;
774
775 i915_emit_arith(p,
776 A0_MUL,
777 tmp1, A0_DEST_CHANNEL_ALL, 0,
778 swizzle(tmp, X, Y, Y, W),
779 swizzle(tmp, X, Z, ONE, ONE), 0);
780
781 i915_emit_arith(p,
782 A0_DP4,
783 get_result_vector(p, inst),
784 A0_DEST_CHANNEL_Y, 0,
785 swizzle(tmp1, W, Z, Y, X),
786 i915_emit_const4fv(p, sin_constants), 0);
787 }
788
789 if (inst->DstReg.WriteMask & WRITEMASK_X) {
790 i915_emit_arith(p,
791 A0_MUL,
792 tmp, A0_DEST_CHANNEL_XYZ, 0,
793 swizzle(tmp, X, X, Z, ONE),
794 swizzle(tmp, Z, ONE, ONE, ONE), 0);
795
796 i915_emit_arith(p,
797 A0_DP4,
798 get_result_vector(p, inst),
799 A0_DEST_CHANNEL_X, 0,
800 swizzle(tmp, ONE, Z, Y, X),
801 i915_emit_const4fv(p, cos_constants), 0);
802 }
803 break;
804
805 case OPCODE_SEQ:
806 tmp = i915_get_utemp(p);
807 flags = get_result_flags(inst);
808 dst = get_result_vector(p, inst);
809
810 /* If both operands are uniforms or constants, we get 5 instructions
811 * like:
812 *
813 * U[1] = MOV CONST[1]
814 * U[0].xyz = SGE CONST[0].xxxx, U[1]
815 * U[1] = MOV CONST[1].-x-y-z-w
816 * R[0].xyz = SGE CONST[0].-x-x-x-x, U[1]
817 * R[0].xyz = MUL R[0], U[0]
818 *
819 * This code is stupid. Instead of having the individual calls to
820 * i915_emit_arith generate the moves to utemps, do it in the caller.
821 * This results in code like:
822 *
823 * U[1] = MOV CONST[1]
824 * U[0].xyz = SGE CONST[0].xxxx, U[1]
825 * R[0].xyz = SGE CONST[0].-x-x-x-x, U[1].-x-y-z-w
826 * R[0].xyz = MUL R[0], U[0]
827 */
828 src0 = src_vector(p, &inst->SrcReg[0], program);
829 src1 = src_vector(p, &inst->SrcReg[1], program);
830
831 if (GET_UREG_TYPE(src0) == REG_TYPE_CONST
832 && GET_UREG_TYPE(src1) == REG_TYPE_CONST) {
833 unsigned tmp = i915_get_utemp(p);
834
835 i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
836 src1, 0, 0);
837
838 src1 = tmp;
839 }
840
841 /* tmp = src1 >= src2 */
842 i915_emit_arith(p,
843 A0_SGE,
844 tmp,
845 flags, 0,
846 src0,
847 src1,
848 0);
849 /* dst = src1 <= src2 */
850 i915_emit_arith(p,
851 A0_SGE,
852 dst,
853 flags, 0,
854 negate(src0, 1, 1, 1, 1),
855 negate(src1, 1, 1, 1, 1),
856 0);
857 /* dst = tmp && dst */
858 i915_emit_arith(p,
859 A0_MUL,
860 dst,
861 flags, 0,
862 dst,
863 tmp,
864 0);
865 break;
866
867 case OPCODE_SIN:
868 src0 = src_vector(p, &inst->SrcReg[0], program);
869 tmp = i915_get_utemp(p);
870 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
871 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
872
873 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
874 i915_emit_arith(p,
875 A0_MAD,
876 tmp, A0_DEST_CHANNEL_X, 0,
877 src0,
878 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
879 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
880
881 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
882
883 i915_emit_arith(p,
884 A0_MAD,
885 tmp, A0_DEST_CHANNEL_X, 0,
886 tmp,
887 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
888 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
889
890 /* Compute sin using a quadratic and quartic. It gives continuity
891 * that repeating the Taylor series lacks every 2*pi, and has
892 * reduced error.
893 *
894 * The idea was described at:
895 * http://www.devmaster.net/forums/showthread.php?t=5784
896 */
897
898 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
899 i915_emit_arith(p,
900 A0_MAX,
901 tmp, A0_DEST_CHANNEL_Y, 0,
902 swizzle(tmp, ZERO, X, ZERO, ZERO),
903 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
904 0);
905
906 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
907 i915_emit_arith(p,
908 A0_MUL,
909 tmp, A0_DEST_CHANNEL_Y, 0,
910 swizzle(tmp, ZERO, X, ZERO, ZERO),
911 tmp,
912 0);
913
914 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
915 i915_emit_arith(p,
916 A0_DP3,
917 tmp, A0_DEST_CHANNEL_X, 0,
918 tmp,
919 swizzle(consts1, X, Y, ZERO, ZERO),
920 0);
921
922 /* tmp.x now contains a first approximation (y). Now, weight it
923 * against tmp.y**2 to get closer.
924 */
925 i915_emit_arith(p,
926 A0_MAX,
927 tmp, A0_DEST_CHANNEL_Y, 0,
928 swizzle(tmp, ZERO, X, ZERO, ZERO),
929 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
930 0);
931
932 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
933 i915_emit_arith(p,
934 A0_MAD,
935 tmp, A0_DEST_CHANNEL_Y, 0,
936 swizzle(tmp, ZERO, X, ZERO, ZERO),
937 swizzle(tmp, ZERO, Y, ZERO, ZERO),
938 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
939
940 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
941 i915_emit_arith(p,
942 A0_MAD,
943 get_result_vector(p, inst),
944 get_result_flags(inst), 0,
945 swizzle(consts1, W, W, W, W),
946 swizzle(tmp, Y, Y, Y, Y),
947 swizzle(tmp, X, X, X, X));
948
949 break;
950
951 case OPCODE_SGE:
952 EMIT_2ARG_ARITH(A0_SGE);
953 break;
954
955 case OPCODE_SGT:
956 i915_emit_arith(p,
957 A0_SLT,
958 get_result_vector( p, inst ),
959 get_result_flags( inst ), 0,
960 negate(src_vector( p, &inst->SrcReg[0], program),
961 1, 1, 1, 1),
962 negate(src_vector( p, &inst->SrcReg[1], program),
963 1, 1, 1, 1),
964 0);
965 break;
966
967 case OPCODE_SLE:
968 i915_emit_arith(p,
969 A0_SGE,
970 get_result_vector( p, inst ),
971 get_result_flags( inst ), 0,
972 negate(src_vector( p, &inst->SrcReg[0], program),
973 1, 1, 1, 1),
974 negate(src_vector( p, &inst->SrcReg[1], program),
975 1, 1, 1, 1),
976 0);
977 break;
978
979 case OPCODE_SLT:
980 EMIT_2ARG_ARITH(A0_SLT);
981 break;
982
983 case OPCODE_SNE:
984 tmp = i915_get_utemp(p);
985 flags = get_result_flags(inst);
986 dst = get_result_vector(p, inst);
987
988 /* If both operands are uniforms or constants, we get 5 instructions
989 * like:
990 *
991 * U[1] = MOV CONST[1]
992 * U[0].xyz = SLT CONST[0].xxxx, U[1]
993 * U[1] = MOV CONST[1].-x-y-z-w
994 * R[0].xyz = SLT CONST[0].-x-x-x-x, U[1]
995 * R[0].xyz = MUL R[0], U[0]
996 *
997 * This code is stupid. Instead of having the individual calls to
998 * i915_emit_arith generate the moves to utemps, do it in the caller.
999 * This results in code like:
1000 *
1001 * U[1] = MOV CONST[1]
1002 * U[0].xyz = SLT CONST[0].xxxx, U[1]
1003 * R[0].xyz = SLT CONST[0].-x-x-x-x, U[1].-x-y-z-w
1004 * R[0].xyz = MUL R[0], U[0]
1005 */
1006 src0 = src_vector(p, &inst->SrcReg[0], program);
1007 src1 = src_vector(p, &inst->SrcReg[1], program);
1008
1009 if (GET_UREG_TYPE(src0) == REG_TYPE_CONST
1010 && GET_UREG_TYPE(src1) == REG_TYPE_CONST) {
1011 unsigned tmp = i915_get_utemp(p);
1012
1013 i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
1014 src1, 0, 0);
1015
1016 src1 = tmp;
1017 }
1018
1019 /* tmp = src1 < src2 */
1020 i915_emit_arith(p,
1021 A0_SLT,
1022 tmp,
1023 flags, 0,
1024 src0,
1025 src1,
1026 0);
1027 /* dst = src1 > src2 */
1028 i915_emit_arith(p,
1029 A0_SLT,
1030 dst,
1031 flags, 0,
1032 negate(src0, 1, 1, 1, 1),
1033 negate(src1, 1, 1, 1, 1),
1034 0);
1035 /* dst = tmp || dst */
1036 i915_emit_arith(p,
1037 A0_ADD,
1038 dst,
1039 flags | A0_DEST_SATURATE, 0,
1040 dst,
1041 tmp,
1042 0);
1043 break;
1044
1045 case OPCODE_SSG:
1046 dst = get_result_vector(p, inst);
1047 flags = get_result_flags(inst);
1048 src0 = src_vector(p, &inst->SrcReg[0], program);
1049 tmp = i915_get_utemp(p);
1050
1051 /* tmp = (src < 0.0) */
1052 i915_emit_arith(p,
1053 A0_SLT,
1054 tmp,
1055 flags, 0,
1056 src0,
1057 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1058 0);
1059
1060 /* dst = (0.0 < src) */
1061 i915_emit_arith(p,
1062 A0_SLT,
1063 dst,
1064 flags, 0,
1065 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1066 src0,
1067 0);
1068
1069 /* dst = (src > 0.0) - (src < 0.0) */
1070 i915_emit_arith(p,
1071 A0_ADD,
1072 dst,
1073 flags, 0,
1074 dst,
1075 negate(tmp, 1, 1, 1, 1),
1076 0);
1077
1078 break;
1079
1080 case OPCODE_SUB:
1081 src0 = src_vector(p, &inst->SrcReg[0], program);
1082 src1 = src_vector(p, &inst->SrcReg[1], program);
1083
1084 i915_emit_arith(p,
1085 A0_ADD,
1086 get_result_vector(p, inst),
1087 get_result_flags(inst), 0,
1088 src0, negate(src1, 1, 1, 1, 1), 0);
1089 break;
1090
1091 case OPCODE_SWZ:
1092 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
1093 break;
1094
1095 case OPCODE_TEX:
1096 EMIT_TEX(T0_TEXLD);
1097 break;
1098
1099 case OPCODE_TXB:
1100 EMIT_TEX(T0_TEXLDB);
1101 break;
1102
1103 case OPCODE_TXP:
1104 EMIT_TEX(T0_TEXLDP);
1105 break;
1106
1107 case OPCODE_XPD:
1108 /* Cross product:
1109 * result.x = src0.y * src1.z - src0.z * src1.y;
1110 * result.y = src0.z * src1.x - src0.x * src1.z;
1111 * result.z = src0.x * src1.y - src0.y * src1.x;
1112 * result.w = undef;
1113 */
1114 src0 = src_vector(p, &inst->SrcReg[0], program);
1115 src1 = src_vector(p, &inst->SrcReg[1], program);
1116 tmp = i915_get_utemp(p);
1117
1118 i915_emit_arith(p,
1119 A0_MUL,
1120 tmp, A0_DEST_CHANNEL_ALL, 0,
1121 swizzle(src0, Z, X, Y, ONE),
1122 swizzle(src1, Y, Z, X, ONE), 0);
1123
1124 i915_emit_arith(p,
1125 A0_MAD,
1126 get_result_vector(p, inst),
1127 get_result_flags(inst), 0,
1128 swizzle(src0, Y, Z, X, ONE),
1129 swizzle(src1, Z, X, Y, ONE),
1130 negate(tmp, 1, 1, 1, 0));
1131 break;
1132
1133 case OPCODE_END:
1134 return;
1135
1136 case OPCODE_BGNLOOP:
1137 case OPCODE_BGNSUB:
1138 case OPCODE_BRK:
1139 case OPCODE_CAL:
1140 case OPCODE_CONT:
1141 case OPCODE_DDX:
1142 case OPCODE_DDY:
1143 case OPCODE_ELSE:
1144 case OPCODE_ENDIF:
1145 case OPCODE_ENDLOOP:
1146 case OPCODE_ENDSUB:
1147 case OPCODE_IF:
1148 case OPCODE_RET:
1149 p->error = 1;
1150 i915_program_error(p, "Unsupported opcode: %s",
1151 _mesa_opcode_string(inst->Opcode));
1152 return;
1153
1154 case OPCODE_EXP:
1155 case OPCODE_LOG:
1156 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1157 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1158 * Instead, it translates to EX2 or LG2.
1159 */
1160 case OPCODE_TXD:
1161 case OPCODE_TXL:
1162 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1163 * only NV_vp/fp appears to emit them.
1164 */
1165 default:
1166 i915_program_error(p, "bad opcode: %s",
1167 _mesa_opcode_string(inst->Opcode));
1168 return;
1169 }
1170
1171 inst++;
1172 i915_release_utemps(p);
1173 }
1174 }
1175
1176 /* Rather than trying to intercept and jiggle depth writes during
1177 * emit, just move the value into its correct position at the end of
1178 * the program:
1179 */
1180 static void
1181 fixup_depth_write(struct i915_fragment_program *p)
1182 {
1183 if (p->depth_written) {
1184 GLuint depth = UREG(REG_TYPE_OD, 0);
1185
1186 i915_emit_arith(p,
1187 A0_MOV,
1188 depth, A0_DEST_CHANNEL_W, 0,
1189 swizzle(depth, X, Y, Z, Z), 0, 0);
1190 }
1191 }
1192
1193
1194 static void
1195 check_wpos(struct i915_fragment_program *p)
1196 {
1197 GLbitfield64 inputs = p->FragProg.Base.InputsRead;
1198 GLint i;
1199
1200 p->wpos_tex = -1;
1201
1202 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1203 if (inputs & (VARYING_BIT_TEX(i) | VARYING_BIT_VAR(i)))
1204 continue;
1205 else if (inputs & VARYING_BIT_POS) {
1206 p->wpos_tex = i;
1207 inputs &= ~VARYING_BIT_POS;
1208 }
1209 }
1210
1211 if (inputs & VARYING_BIT_POS) {
1212 i915_program_error(p, "No free texcoord for wpos value");
1213 }
1214 }
1215
1216
1217 static void
1218 translate_program(struct i915_fragment_program *p)
1219 {
1220 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1221
1222 if (INTEL_DEBUG & DEBUG_WM) {
1223 printf("fp:\n");
1224 _mesa_print_program(&p->FragProg.Base);
1225 printf("\n");
1226 }
1227
1228 i915_init_program(i915, p);
1229 check_wpos(p);
1230 upload_program(p);
1231 fixup_depth_write(p);
1232 i915_fini_program(p);
1233
1234 p->translated = 1;
1235 }
1236
1237
1238 static void
1239 track_params(struct i915_fragment_program *p)
1240 {
1241 GLint i;
1242
1243 if (p->nr_params)
1244 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1245
1246 for (i = 0; i < p->nr_params; i++) {
1247 GLint reg = p->param[i].reg;
1248 COPY_4V(p->constant[reg], p->param[i].values);
1249 }
1250
1251 p->params_uptodate = 1;
1252 p->on_hardware = 0; /* overkill */
1253 }
1254
1255
1256 static void
1257 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1258 {
1259 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1260 struct i915_context *i915 = I915_CONTEXT(ctx);
1261 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1262
1263 if (i915->current_program == p)
1264 return;
1265
1266 if (i915->current_program) {
1267 i915->current_program->on_hardware = 0;
1268 i915->current_program->params_uptodate = 0;
1269 }
1270
1271 i915->current_program = p;
1272
1273 assert(p->on_hardware == 0);
1274 assert(p->params_uptodate == 0);
1275
1276 }
1277 }
1278
1279 static struct gl_program *
1280 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1281 {
1282 switch (target) {
1283 case GL_VERTEX_PROGRAM_ARB:
1284 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1285 target, id);
1286
1287 case GL_FRAGMENT_PROGRAM_ARB:{
1288 struct i915_fragment_program *prog =
1289 CALLOC_STRUCT(i915_fragment_program);
1290 if (prog) {
1291 i915_init_program(I915_CONTEXT(ctx), prog);
1292
1293 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1294 target, id);
1295 }
1296 else
1297 return NULL;
1298 }
1299
1300 default:
1301 /* Just fallback:
1302 */
1303 return _mesa_new_program(ctx, target, id);
1304 }
1305 }
1306
1307 static void
1308 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1309 {
1310 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1311 struct i915_context *i915 = I915_CONTEXT(ctx);
1312 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1313
1314 if (i915->current_program == p)
1315 i915->current_program = 0;
1316 }
1317
1318 _mesa_delete_program(ctx, prog);
1319 }
1320
1321
1322 static GLboolean
1323 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1324 {
1325 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1326 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1327
1328 if (!p->translated)
1329 translate_program(p);
1330
1331 return !p->error;
1332 }
1333 else
1334 return true;
1335 }
1336
1337 static GLboolean
1338 i915ProgramStringNotify(struct gl_context * ctx,
1339 GLenum target, struct gl_program *prog)
1340 {
1341 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1342 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1343 p->translated = 0;
1344 }
1345
1346 (void) _tnl_program_string(ctx, target, prog);
1347
1348 /* XXX check if program is legal, within limits */
1349 return true;
1350 }
1351
1352 static void
1353 i915SamplerUniformChange(struct gl_context *ctx,
1354 GLenum target, struct gl_program *prog)
1355 {
1356 i915ProgramStringNotify(ctx, target, prog);
1357 }
1358
1359 void
1360 i915_update_program(struct gl_context *ctx)
1361 {
1362 struct intel_context *intel = intel_context(ctx);
1363 struct i915_context *i915 = i915_context(&intel->ctx);
1364 struct i915_fragment_program *fp =
1365 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1366
1367 if (i915->current_program != fp) {
1368 if (i915->current_program) {
1369 i915->current_program->on_hardware = 0;
1370 i915->current_program->params_uptodate = 0;
1371 }
1372
1373 i915->current_program = fp;
1374 }
1375
1376 if (!fp->translated)
1377 translate_program(fp);
1378
1379 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1380 }
1381
1382 void
1383 i915ValidateFragmentProgram(struct i915_context *i915)
1384 {
1385 struct gl_context *ctx = &i915->intel.ctx;
1386 struct intel_context *intel = intel_context(ctx);
1387 TNLcontext *tnl = TNL_CONTEXT(ctx);
1388 struct vertex_buffer *VB = &tnl->vb;
1389
1390 struct i915_fragment_program *p =
1391 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1392
1393 const GLbitfield64 inputsRead = p->FragProg.Base.InputsRead;
1394 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1395 GLuint s2 = S2_TEXCOORD_NONE;
1396 int i, offset = 0;
1397
1398 /* Important:
1399 */
1400 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1401
1402 if (!p->translated)
1403 translate_program(p);
1404
1405 intel->vertex_attr_count = 0;
1406 intel->wpos_offset = 0;
1407 intel->coloroffset = 0;
1408 intel->specoffset = 0;
1409
1410 if (inputsRead & VARYING_BITS_TEX_ANY || p->wpos_tex != -1) {
1411 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1412 }
1413 else {
1414 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1415 }
1416
1417 /* Handle gl_PointSize builtin var here */
1418 if (ctx->Point._Attenuated || ctx->VertexProgram.PointSizeEnabled)
1419 EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, S4_VFMT_POINT_WIDTH, 4);
1420
1421 if (inputsRead & VARYING_BIT_COL0) {
1422 intel->coloroffset = offset / 4;
1423 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1424 }
1425
1426 if (inputsRead & VARYING_BIT_COL1) {
1427 intel->specoffset = offset / 4;
1428 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1429 }
1430
1431 if ((inputsRead & VARYING_BIT_FOGC)) {
1432 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1433 }
1434
1435 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1436 if (inputsRead & VARYING_BIT_TEX(i)) {
1437 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1438
1439 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1440 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1441
1442 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1443 }
1444 else if (inputsRead & VARYING_BIT_VAR(i)) {
1445 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1446
1447 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1448 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1449
1450 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1451 }
1452 else if (i == p->wpos_tex) {
1453 int wpos_size = 4 * sizeof(float);
1454 /* If WPOS is required, duplicate the XYZ position data in an
1455 * unused texture coordinate:
1456 */
1457 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1458 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(wpos_size));
1459
1460 intel->wpos_offset = offset;
1461 EMIT_PAD(wpos_size);
1462 }
1463 }
1464
1465 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1466 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1467 int k;
1468
1469 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1470
1471 /* Must do this *after* statechange, so as not to affect
1472 * buffered vertices reliant on the old state:
1473 */
1474 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1475 intel->vertex_attrs,
1476 intel->vertex_attr_count,
1477 intel->ViewportMatrix.m, 0);
1478
1479 assert(intel->prim.current_offset == intel->prim.start_offset);
1480 intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1481 intel->prim.current_offset = intel->prim.start_offset;
1482
1483 intel->vertex_size >>= 2;
1484
1485 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1486 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1487
1488 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1489 assert(k);
1490 }
1491
1492 if (!p->params_uptodate)
1493 track_params(p);
1494
1495 if (!p->on_hardware)
1496 i915_upload_program(i915, p);
1497
1498 if (INTEL_DEBUG & DEBUG_WM) {
1499 printf("i915:\n");
1500 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1501 }
1502 }
1503
1504 void
1505 i915InitFragProgFuncs(struct dd_function_table *functions)
1506 {
1507 functions->BindProgram = i915BindProgram;
1508 functions->NewProgram = i915NewProgram;
1509 functions->DeleteProgram = i915DeleteProgram;
1510 functions->IsProgramNative = i915IsProgramNative;
1511 functions->ProgramStringNotify = i915ProgramStringNotify;
1512 functions->SamplerUniformChange = i915SamplerUniformChange;
1513 }