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
src_vector(struct i915_fragment_program * p,const struct prog_src_register * source,const struct gl_fragment_program * program)80 src_vector(struct i915_fragment_program *p,
81 const struct prog_src_register *source,
82 const struct gl_fragment_program *program)
83 {
84 GLuint src;
85
86 switch (source->File) {
87
88 /* Registers:
89 */
90 case PROGRAM_TEMPORARY:
91 if (source->Index >= I915_MAX_TEMPORARY) {
92 i915_program_error(p, "Exceeded max temporary reg: %d/%d",
93 source->Index, I915_MAX_TEMPORARY);
94 return 0;
95 }
96 src = UREG(REG_TYPE_R, source->Index);
97 break;
98 case PROGRAM_INPUT:
99 switch (source->Index) {
100 case FRAG_ATTRIB_WPOS:
101 src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
102 break;
103 case FRAG_ATTRIB_COL0:
104 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
105 break;
106 case FRAG_ATTRIB_COL1:
107 src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
108 src = swizzle(src, X, Y, Z, ONE);
109 break;
110 case FRAG_ATTRIB_FOGC:
111 src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
112 src = swizzle(src, W, ZERO, ZERO, ONE);
113 break;
114 case FRAG_ATTRIB_TEX0:
115 case FRAG_ATTRIB_TEX1:
116 case FRAG_ATTRIB_TEX2:
117 case FRAG_ATTRIB_TEX3:
118 case FRAG_ATTRIB_TEX4:
119 case FRAG_ATTRIB_TEX5:
120 case FRAG_ATTRIB_TEX6:
121 case FRAG_ATTRIB_TEX7:
122 src = i915_emit_decl(p, REG_TYPE_T,
123 T_TEX0 + (source->Index - FRAG_ATTRIB_TEX0),
124 D0_CHANNEL_ALL);
125 break;
126
127 case FRAG_ATTRIB_VAR0:
128 case FRAG_ATTRIB_VAR0 + 1:
129 case FRAG_ATTRIB_VAR0 + 2:
130 case FRAG_ATTRIB_VAR0 + 3:
131 case FRAG_ATTRIB_VAR0 + 4:
132 case FRAG_ATTRIB_VAR0 + 5:
133 case FRAG_ATTRIB_VAR0 + 6:
134 case FRAG_ATTRIB_VAR0 + 7:
135 src = i915_emit_decl(p, REG_TYPE_T,
136 T_TEX0 + (source->Index - FRAG_ATTRIB_VAR0),
137 D0_CHANNEL_ALL);
138 break;
139
140 default:
141 i915_program_error(p, "Bad source->Index: %d", source->Index);
142 return 0;
143 }
144 break;
145
146 case PROGRAM_OUTPUT:
147 switch (source->Index) {
148 case FRAG_RESULT_COLOR:
149 src = UREG(REG_TYPE_OC, 0);
150 break;
151 case FRAG_RESULT_DEPTH:
152 src = UREG(REG_TYPE_OD, 0);
153 break;
154 default:
155 i915_program_error(p, "Bad source->Index: %d", source->Index);
156 return 0;
157 }
158 break;
159
160 /* Various paramters and env values. All emitted to
161 * hardware as program constants.
162 */
163 case PROGRAM_LOCAL_PARAM:
164 src = i915_emit_param4fv(p, program->Base.LocalParams[source->Index]);
165 break;
166
167 case PROGRAM_ENV_PARAM:
168 src =
169 i915_emit_param4fv(p,
170 p->ctx->FragmentProgram.Parameters[source->
171 Index]);
172 break;
173
174 case PROGRAM_CONSTANT:
175 case PROGRAM_STATE_VAR:
176 case PROGRAM_NAMED_PARAM:
177 case PROGRAM_UNIFORM:
178 src = i915_emit_param4fv(p,
179 &program->Base.Parameters->ParameterValues[source->Index][0].f);
180 break;
181
182 default:
183 i915_program_error(p, "Bad source->File: %d", source->File);
184 return 0;
185 }
186
187 src = swizzle(src,
188 GET_SWZ(source->Swizzle, 0),
189 GET_SWZ(source->Swizzle, 1),
190 GET_SWZ(source->Swizzle, 2), GET_SWZ(source->Swizzle, 3));
191
192 if (source->Negate)
193 src = negate(src,
194 GET_BIT(source->Negate, 0),
195 GET_BIT(source->Negate, 1),
196 GET_BIT(source->Negate, 2),
197 GET_BIT(source->Negate, 3));
198
199 return src;
200 }
201
202
203 static GLuint
get_result_vector(struct i915_fragment_program * p,const struct prog_instruction * inst)204 get_result_vector(struct i915_fragment_program *p,
205 const struct prog_instruction *inst)
206 {
207 switch (inst->DstReg.File) {
208 case PROGRAM_OUTPUT:
209 switch (inst->DstReg.Index) {
210 case FRAG_RESULT_COLOR:
211 case FRAG_RESULT_DATA0:
212 return UREG(REG_TYPE_OC, 0);
213 case FRAG_RESULT_DEPTH:
214 p->depth_written = 1;
215 return UREG(REG_TYPE_OD, 0);
216 default:
217 i915_program_error(p, "Bad inst->DstReg.Index: %d",
218 inst->DstReg.Index);
219 return 0;
220 }
221 case PROGRAM_TEMPORARY:
222 return UREG(REG_TYPE_R, inst->DstReg.Index);
223 default:
224 i915_program_error(p, "Bad inst->DstReg.File: %d", inst->DstReg.File);
225 return 0;
226 }
227 }
228
229 static GLuint
get_result_flags(const struct prog_instruction * inst)230 get_result_flags(const struct prog_instruction *inst)
231 {
232 GLuint flags = 0;
233
234 if (inst->SaturateMode == SATURATE_ZERO_ONE)
235 flags |= A0_DEST_SATURATE;
236 if (inst->DstReg.WriteMask & WRITEMASK_X)
237 flags |= A0_DEST_CHANNEL_X;
238 if (inst->DstReg.WriteMask & WRITEMASK_Y)
239 flags |= A0_DEST_CHANNEL_Y;
240 if (inst->DstReg.WriteMask & WRITEMASK_Z)
241 flags |= A0_DEST_CHANNEL_Z;
242 if (inst->DstReg.WriteMask & WRITEMASK_W)
243 flags |= A0_DEST_CHANNEL_W;
244
245 return flags;
246 }
247
248 static GLuint
translate_tex_src_target(struct i915_fragment_program * p,GLubyte bit)249 translate_tex_src_target(struct i915_fragment_program *p, GLubyte bit)
250 {
251 switch (bit) {
252 case TEXTURE_1D_INDEX:
253 return D0_SAMPLE_TYPE_2D;
254 case TEXTURE_2D_INDEX:
255 return D0_SAMPLE_TYPE_2D;
256 case TEXTURE_RECT_INDEX:
257 return D0_SAMPLE_TYPE_2D;
258 case TEXTURE_3D_INDEX:
259 return D0_SAMPLE_TYPE_VOLUME;
260 case TEXTURE_CUBE_INDEX:
261 return D0_SAMPLE_TYPE_CUBE;
262 default:
263 i915_program_error(p, "TexSrcBit: %d", bit);
264 return 0;
265 }
266 }
267
268 #define EMIT_TEX( OP ) \
269 do { \
270 GLuint dim = translate_tex_src_target( p, inst->TexSrcTarget ); \
271 const struct gl_fragment_program *program = &p->FragProg; \
272 GLuint unit = program->Base.SamplerUnits[inst->TexSrcUnit]; \
273 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, \
274 unit, dim); \
275 GLuint coord = src_vector( p, &inst->SrcReg[0], program); \
276 /* Texel lookup */ \
277 \
278 i915_emit_texld( p, get_live_regs(p, inst), \
279 get_result_vector( p, inst ), \
280 get_result_flags( inst ), \
281 sampler, \
282 coord, \
283 OP); \
284 } while (0)
285
286 #define EMIT_ARITH( OP, N ) \
287 do { \
288 i915_emit_arith( p, \
289 OP, \
290 get_result_vector( p, inst ), \
291 get_result_flags( inst ), 0, \
292 (N<1)?0:src_vector( p, &inst->SrcReg[0], program), \
293 (N<2)?0:src_vector( p, &inst->SrcReg[1], program), \
294 (N<3)?0:src_vector( p, &inst->SrcReg[2], program)); \
295 } while (0)
296
297 #define EMIT_1ARG_ARITH( OP ) EMIT_ARITH( OP, 1 )
298 #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 )
299 #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 )
300
301 /*
302 * TODO: consider moving this into core
303 */
calc_live_regs(struct i915_fragment_program * p)304 static bool calc_live_regs( struct i915_fragment_program *p )
305 {
306 const struct gl_fragment_program *program = &p->FragProg;
307 GLuint regsUsed = ~((1 << I915_MAX_TEMPORARY) - 1);
308 uint8_t live_components[I915_MAX_TEMPORARY] = { 0, };
309 GLint i;
310
311 for (i = program->Base.NumInstructions - 1; i >= 0; i--) {
312 struct prog_instruction *inst = &program->Base.Instructions[i];
313 int opArgs = _mesa_num_inst_src_regs(inst->Opcode);
314 int a;
315
316 /* Register is written to: unmark as live for this and preceeding ops */
317 if (inst->DstReg.File == PROGRAM_TEMPORARY) {
318 if (inst->DstReg.Index >= I915_MAX_TEMPORARY)
319 return false;
320
321 live_components[inst->DstReg.Index] &= ~inst->DstReg.WriteMask;
322 if (live_components[inst->DstReg.Index] == 0)
323 regsUsed &= ~(1 << inst->DstReg.Index);
324 }
325
326 for (a = 0; a < opArgs; a++) {
327 /* Register is read from: mark as live for this and preceeding ops */
328 if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) {
329 unsigned c;
330
331 if (inst->SrcReg[a].Index >= I915_MAX_TEMPORARY)
332 return false;
333
334 regsUsed |= 1 << inst->SrcReg[a].Index;
335
336 for (c = 0; c < 4; c++) {
337 const unsigned field = GET_SWZ(inst->SrcReg[a].Swizzle, c);
338
339 if (field <= SWIZZLE_W)
340 live_components[inst->SrcReg[a].Index] |= (1U << field);
341 }
342 }
343 }
344
345 p->usedRegs[i] = regsUsed;
346 }
347
348 return true;
349 }
350
get_live_regs(struct i915_fragment_program * p,const struct prog_instruction * inst)351 static GLuint get_live_regs( struct i915_fragment_program *p,
352 const struct prog_instruction *inst )
353 {
354 const struct gl_fragment_program *program = &p->FragProg;
355 GLuint nr = inst - program->Base.Instructions;
356
357 return p->usedRegs[nr];
358 }
359
360
361 /* Possible concerns:
362 *
363 * SIN, COS -- could use another taylor step?
364 * LIT -- results seem a little different to sw mesa
365 * LOG -- different to mesa on negative numbers, but this is conformant.
366 *
367 * Parse failures -- Mesa doesn't currently give a good indication
368 * internally whether a particular program string parsed or not. This
369 * can lead to confusion -- hopefully we cope with it ok now.
370 *
371 */
372 static void
upload_program(struct i915_fragment_program * p)373 upload_program(struct i915_fragment_program *p)
374 {
375 const struct gl_fragment_program *program = &p->FragProg;
376 const struct prog_instruction *inst = program->Base.Instructions;
377
378 if (INTEL_DEBUG & DEBUG_WM)
379 _mesa_print_program(&program->Base);
380
381 /* Is this a parse-failed program? Ensure a valid program is
382 * loaded, as the flagging of an error isn't sufficient to stop
383 * this being uploaded to hardware.
384 */
385 if (inst[0].Opcode == OPCODE_END) {
386 GLuint tmp = i915_get_utemp(p);
387 i915_emit_arith(p,
388 A0_MOV,
389 UREG(REG_TYPE_OC, 0),
390 A0_DEST_CHANNEL_ALL, 0,
391 swizzle(tmp, ONE, ZERO, ONE, ONE), 0, 0);
392 return;
393 }
394
395 if (program->Base.NumInstructions > I915_MAX_INSN) {
396 i915_program_error(p, "Exceeded max instructions (%d out of %d)",
397 program->Base.NumInstructions, I915_MAX_INSN);
398 return;
399 }
400
401 /* Not always needed:
402 */
403 if (!calc_live_regs(p)) {
404 i915_program_error(p, "Could not allocate registers");
405 return;
406 }
407
408 while (1) {
409 GLuint src0, src1, src2, flags;
410 GLuint tmp = 0, dst, consts0 = 0, consts1 = 0;
411
412 switch (inst->Opcode) {
413 case OPCODE_ABS:
414 src0 = src_vector(p, &inst->SrcReg[0], program);
415 i915_emit_arith(p,
416 A0_MAX,
417 get_result_vector(p, inst),
418 get_result_flags(inst), 0,
419 src0, negate(src0, 1, 1, 1, 1), 0);
420 break;
421
422 case OPCODE_ADD:
423 EMIT_2ARG_ARITH(A0_ADD);
424 break;
425
426 case OPCODE_CMP:
427 src0 = src_vector(p, &inst->SrcReg[0], program);
428 src1 = src_vector(p, &inst->SrcReg[1], program);
429 src2 = src_vector(p, &inst->SrcReg[2], program);
430 i915_emit_arith(p, A0_CMP, get_result_vector(p, inst), get_result_flags(inst), 0, src0, src2, src1); /* NOTE: order of src2, src1 */
431 break;
432
433 case OPCODE_COS:
434 src0 = src_vector(p, &inst->SrcReg[0], program);
435 tmp = i915_get_utemp(p);
436 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
437 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
438
439 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
440 i915_emit_arith(p,
441 A0_MAD,
442 tmp, A0_DEST_CHANNEL_X, 0,
443 src0,
444 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
445 swizzle(consts0, W, ZERO, ZERO, ZERO)); /* .75 */
446
447 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
448
449 i915_emit_arith(p,
450 A0_MAD,
451 tmp, A0_DEST_CHANNEL_X, 0,
452 tmp,
453 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
454 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
455
456 /* Compute COS with the same calculation used for SIN, but a
457 * different source range has been mapped to [-1,1] this time.
458 */
459
460 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
461 i915_emit_arith(p,
462 A0_MAX,
463 tmp, A0_DEST_CHANNEL_Y, 0,
464 swizzle(tmp, ZERO, X, ZERO, ZERO),
465 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
466 0);
467
468 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
469 i915_emit_arith(p,
470 A0_MUL,
471 tmp, A0_DEST_CHANNEL_Y, 0,
472 swizzle(tmp, ZERO, X, ZERO, ZERO),
473 tmp,
474 0);
475
476 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
477 i915_emit_arith(p,
478 A0_DP3,
479 tmp, A0_DEST_CHANNEL_X, 0,
480 tmp,
481 swizzle(consts1, X, Y, ZERO, ZERO),
482 0);
483
484 /* tmp.x now contains a first approximation (y). Now, weight it
485 * against tmp.y**2 to get closer.
486 */
487 i915_emit_arith(p,
488 A0_MAX,
489 tmp, A0_DEST_CHANNEL_Y, 0,
490 swizzle(tmp, ZERO, X, ZERO, ZERO),
491 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
492 0);
493
494 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
495 i915_emit_arith(p,
496 A0_MAD,
497 tmp, A0_DEST_CHANNEL_Y, 0,
498 swizzle(tmp, ZERO, X, ZERO, ZERO),
499 swizzle(tmp, ZERO, Y, ZERO, ZERO),
500 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
501
502 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
503 i915_emit_arith(p,
504 A0_MAD,
505 get_result_vector(p, inst),
506 get_result_flags(inst), 0,
507 swizzle(consts1, W, W, W, W),
508 swizzle(tmp, Y, Y, Y, Y),
509 swizzle(tmp, X, X, X, X));
510 break;
511
512 case OPCODE_DP2:
513 src0 = src_vector(p, &inst->SrcReg[0], program);
514 src1 = src_vector(p, &inst->SrcReg[1], program);
515 i915_emit_arith(p,
516 A0_DP3,
517 get_result_vector(p, inst),
518 get_result_flags(inst), 0,
519 swizzle(src0, X, Y, ZERO, ZERO),
520 swizzle(src1, X, Y, ZERO, ZERO),
521 0);
522 break;
523
524 case OPCODE_DP3:
525 EMIT_2ARG_ARITH(A0_DP3);
526 break;
527
528 case OPCODE_DP4:
529 EMIT_2ARG_ARITH(A0_DP4);
530 break;
531
532 case OPCODE_DPH:
533 src0 = src_vector(p, &inst->SrcReg[0], program);
534 src1 = src_vector(p, &inst->SrcReg[1], program);
535
536 i915_emit_arith(p,
537 A0_DP4,
538 get_result_vector(p, inst),
539 get_result_flags(inst), 0,
540 swizzle(src0, X, Y, Z, ONE), src1, 0);
541 break;
542
543 case OPCODE_DST:
544 src0 = src_vector(p, &inst->SrcReg[0], program);
545 src1 = src_vector(p, &inst->SrcReg[1], program);
546
547 /* result[0] = 1 * 1;
548 * result[1] = a[1] * b[1];
549 * result[2] = a[2] * 1;
550 * result[3] = 1 * b[3];
551 */
552 i915_emit_arith(p,
553 A0_MUL,
554 get_result_vector(p, inst),
555 get_result_flags(inst), 0,
556 swizzle(src0, ONE, Y, Z, ONE),
557 swizzle(src1, ONE, Y, ONE, W), 0);
558 break;
559
560 case OPCODE_EX2:
561 src0 = src_vector(p, &inst->SrcReg[0], program);
562
563 i915_emit_arith(p,
564 A0_EXP,
565 get_result_vector(p, inst),
566 get_result_flags(inst), 0,
567 swizzle(src0, X, X, X, X), 0, 0);
568 break;
569
570 case OPCODE_FLR:
571 EMIT_1ARG_ARITH(A0_FLR);
572 break;
573
574 case OPCODE_TRUNC:
575 EMIT_1ARG_ARITH(A0_TRC);
576 break;
577
578 case OPCODE_FRC:
579 EMIT_1ARG_ARITH(A0_FRC);
580 break;
581
582 case OPCODE_KIL:
583 src0 = src_vector(p, &inst->SrcReg[0], program);
584 tmp = i915_get_utemp(p);
585
586 i915_emit_texld(p, get_live_regs(p, inst),
587 tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */
588 0, src0, T0_TEXKILL);
589 break;
590
591 case OPCODE_KIL_NV:
592 if (inst->DstReg.CondMask == COND_TR) {
593 tmp = i915_get_utemp(p);
594
595 /* The KIL instruction discards the fragment if any component of
596 * the source is < 0. Emit an immediate operand of {-1}.xywz.
597 */
598 i915_emit_texld(p, get_live_regs(p, inst),
599 tmp, A0_DEST_CHANNEL_ALL,
600 0, /* use a dummy dest reg */
601 negate(swizzle(tmp, ONE, ONE, ONE, ONE),
602 1, 1, 1, 1),
603 T0_TEXKILL);
604 } else {
605 p->error = 1;
606 i915_program_error(p, "Unsupported KIL_NV condition code: %d",
607 inst->DstReg.CondMask);
608 }
609 break;
610
611 case OPCODE_LG2:
612 src0 = src_vector(p, &inst->SrcReg[0], program);
613
614 i915_emit_arith(p,
615 A0_LOG,
616 get_result_vector(p, inst),
617 get_result_flags(inst), 0,
618 swizzle(src0, X, X, X, X), 0, 0);
619 break;
620
621 case OPCODE_LIT:
622 src0 = src_vector(p, &inst->SrcReg[0], program);
623 tmp = i915_get_utemp(p);
624
625 /* tmp = max( a.xyzw, a.00zw )
626 * XXX: Clamp tmp.w to -128..128
627 * tmp.y = log(tmp.y)
628 * tmp.y = tmp.w * tmp.y
629 * tmp.y = exp(tmp.y)
630 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
631 */
632 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
633 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
634
635 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
636 swizzle(tmp, Y, Y, Y, Y), 0, 0);
637
638 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
639 swizzle(tmp, ZERO, Y, ZERO, ZERO),
640 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
641
642 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
643 swizzle(tmp, Y, Y, Y, Y), 0, 0);
644
645 i915_emit_arith(p, A0_CMP,
646 get_result_vector(p, inst),
647 get_result_flags(inst), 0,
648 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
649 swizzle(tmp, ONE, X, ZERO, ONE),
650 swizzle(tmp, ONE, X, Y, ONE));
651
652 break;
653
654 case OPCODE_LRP:
655 src0 = src_vector(p, &inst->SrcReg[0], program);
656 src1 = src_vector(p, &inst->SrcReg[1], program);
657 src2 = src_vector(p, &inst->SrcReg[2], program);
658 flags = get_result_flags(inst);
659 tmp = i915_get_utemp(p);
660
661 /* b*a + c*(1-a)
662 *
663 * b*a + c - ca
664 *
665 * tmp = b*a + c,
666 * result = (-c)*a + tmp
667 */
668 i915_emit_arith(p, A0_MAD, tmp,
669 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
670
671 i915_emit_arith(p, A0_MAD,
672 get_result_vector(p, inst),
673 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
674 break;
675
676 case OPCODE_MAD:
677 EMIT_3ARG_ARITH(A0_MAD);
678 break;
679
680 case OPCODE_MAX:
681 EMIT_2ARG_ARITH(A0_MAX);
682 break;
683
684 case OPCODE_MIN:
685 src0 = src_vector(p, &inst->SrcReg[0], program);
686 src1 = src_vector(p, &inst->SrcReg[1], program);
687 tmp = i915_get_utemp(p);
688 flags = get_result_flags(inst);
689
690 i915_emit_arith(p,
691 A0_MAX,
692 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
693 negate(src0, 1, 1, 1, 1),
694 negate(src1, 1, 1, 1, 1), 0);
695
696 i915_emit_arith(p,
697 A0_MOV,
698 get_result_vector(p, inst),
699 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
700 break;
701
702 case OPCODE_MOV:
703 EMIT_1ARG_ARITH(A0_MOV);
704 break;
705
706 case OPCODE_MUL:
707 EMIT_2ARG_ARITH(A0_MUL);
708 break;
709
710 case OPCODE_POW:
711 src0 = src_vector(p, &inst->SrcReg[0], program);
712 src1 = src_vector(p, &inst->SrcReg[1], program);
713 tmp = i915_get_utemp(p);
714 flags = get_result_flags(inst);
715
716 /* XXX: masking on intermediate values, here and elsewhere.
717 */
718 i915_emit_arith(p,
719 A0_LOG,
720 tmp, A0_DEST_CHANNEL_X, 0,
721 swizzle(src0, X, X, X, X), 0, 0);
722
723 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
724
725
726 i915_emit_arith(p,
727 A0_EXP,
728 get_result_vector(p, inst),
729 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
730
731 break;
732
733 case OPCODE_RCP:
734 src0 = src_vector(p, &inst->SrcReg[0], program);
735
736 i915_emit_arith(p,
737 A0_RCP,
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_RSQ:
744
745 src0 = src_vector(p, &inst->SrcReg[0], program);
746
747 i915_emit_arith(p,
748 A0_RSQ,
749 get_result_vector(p, inst),
750 get_result_flags(inst), 0,
751 swizzle(src0, X, X, X, X), 0, 0);
752 break;
753
754 case OPCODE_SCS:
755 src0 = src_vector(p, &inst->SrcReg[0], program);
756 tmp = i915_get_utemp(p);
757
758 /*
759 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
760 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
761 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
762 * scs.x = DP4 t1, sin_constants
763 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
764 * scs.y = DP4 t1, cos_constants
765 */
766 i915_emit_arith(p,
767 A0_MUL,
768 tmp, A0_DEST_CHANNEL_XY, 0,
769 swizzle(src0, X, X, ONE, ONE),
770 swizzle(src0, X, ONE, ONE, ONE), 0);
771
772 i915_emit_arith(p,
773 A0_MUL,
774 tmp, A0_DEST_CHANNEL_ALL, 0,
775 swizzle(tmp, X, Y, X, Y),
776 swizzle(tmp, X, X, ONE, ONE), 0);
777
778 if (inst->DstReg.WriteMask & WRITEMASK_Y) {
779 GLuint tmp1;
780
781 if (inst->DstReg.WriteMask & WRITEMASK_X)
782 tmp1 = i915_get_utemp(p);
783 else
784 tmp1 = tmp;
785
786 i915_emit_arith(p,
787 A0_MUL,
788 tmp1, A0_DEST_CHANNEL_ALL, 0,
789 swizzle(tmp, X, Y, Y, W),
790 swizzle(tmp, X, Z, ONE, ONE), 0);
791
792 i915_emit_arith(p,
793 A0_DP4,
794 get_result_vector(p, inst),
795 A0_DEST_CHANNEL_Y, 0,
796 swizzle(tmp1, W, Z, Y, X),
797 i915_emit_const4fv(p, sin_constants), 0);
798 }
799
800 if (inst->DstReg.WriteMask & WRITEMASK_X) {
801 i915_emit_arith(p,
802 A0_MUL,
803 tmp, A0_DEST_CHANNEL_XYZ, 0,
804 swizzle(tmp, X, X, Z, ONE),
805 swizzle(tmp, Z, ONE, ONE, ONE), 0);
806
807 i915_emit_arith(p,
808 A0_DP4,
809 get_result_vector(p, inst),
810 A0_DEST_CHANNEL_X, 0,
811 swizzle(tmp, ONE, Z, Y, X),
812 i915_emit_const4fv(p, cos_constants), 0);
813 }
814 break;
815
816 case OPCODE_SEQ:
817 tmp = i915_get_utemp(p);
818 flags = get_result_flags(inst);
819 dst = get_result_vector(p, inst);
820
821 /* tmp = src1 >= src2 */
822 i915_emit_arith(p,
823 A0_SGE,
824 tmp,
825 flags, 0,
826 src_vector(p, &inst->SrcReg[0], program),
827 src_vector(p, &inst->SrcReg[1], program),
828 0);
829 /* dst = src1 <= src2 */
830 i915_emit_arith(p,
831 A0_SGE,
832 dst,
833 flags, 0,
834 negate(src_vector(p, &inst->SrcReg[0], program),
835 1, 1, 1, 1),
836 negate(src_vector(p, &inst->SrcReg[1], program),
837 1, 1, 1, 1),
838 0);
839 /* dst = tmp && dst */
840 i915_emit_arith(p,
841 A0_MUL,
842 dst,
843 flags, 0,
844 dst,
845 tmp,
846 0);
847 break;
848
849 case OPCODE_SIN:
850 src0 = src_vector(p, &inst->SrcReg[0], program);
851 tmp = i915_get_utemp(p);
852 consts0 = i915_emit_const4fv(p, sin_quad_constants[0]);
853 consts1 = i915_emit_const4fv(p, sin_quad_constants[1]);
854
855 /* Reduce range from repeating about [-pi,pi] to [-1,1] */
856 i915_emit_arith(p,
857 A0_MAD,
858 tmp, A0_DEST_CHANNEL_X, 0,
859 src0,
860 swizzle(consts1, Z, ZERO, ZERO, ZERO), /* 1/(2pi) */
861 swizzle(consts0, Z, ZERO, ZERO, ZERO)); /* .5 */
862
863 i915_emit_arith(p, A0_FRC, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
864
865 i915_emit_arith(p,
866 A0_MAD,
867 tmp, A0_DEST_CHANNEL_X, 0,
868 tmp,
869 swizzle(consts0, X, ZERO, ZERO, ZERO), /* 2 */
870 swizzle(consts0, Y, ZERO, ZERO, ZERO)); /* -1 */
871
872 /* Compute sin using a quadratic and quartic. It gives continuity
873 * that repeating the Taylor series lacks every 2*pi, and has
874 * reduced error.
875 *
876 * The idea was described at:
877 * http://www.devmaster.net/forums/showthread.php?t=5784
878 */
879
880 /* tmp.y = abs(tmp.x); {x, abs(x), 0, 0} */
881 i915_emit_arith(p,
882 A0_MAX,
883 tmp, A0_DEST_CHANNEL_Y, 0,
884 swizzle(tmp, ZERO, X, ZERO, ZERO),
885 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
886 0);
887
888 /* tmp.y = tmp.y * tmp.x; {x, x * abs(x), 0, 0} */
889 i915_emit_arith(p,
890 A0_MUL,
891 tmp, A0_DEST_CHANNEL_Y, 0,
892 swizzle(tmp, ZERO, X, ZERO, ZERO),
893 tmp,
894 0);
895
896 /* tmp.x = tmp.xy DP sin_quad_constants[2].xy */
897 i915_emit_arith(p,
898 A0_DP3,
899 tmp, A0_DEST_CHANNEL_X, 0,
900 tmp,
901 swizzle(consts1, X, Y, ZERO, ZERO),
902 0);
903
904 /* tmp.x now contains a first approximation (y). Now, weight it
905 * against tmp.y**2 to get closer.
906 */
907 i915_emit_arith(p,
908 A0_MAX,
909 tmp, A0_DEST_CHANNEL_Y, 0,
910 swizzle(tmp, ZERO, X, ZERO, ZERO),
911 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0),
912 0);
913
914 /* tmp.y = tmp.x * tmp.y - tmp.x; {y, y * abs(y) - y, 0, 0} */
915 i915_emit_arith(p,
916 A0_MAD,
917 tmp, A0_DEST_CHANNEL_Y, 0,
918 swizzle(tmp, ZERO, X, ZERO, ZERO),
919 swizzle(tmp, ZERO, Y, ZERO, ZERO),
920 negate(swizzle(tmp, ZERO, X, ZERO, ZERO), 0, 1, 0, 0));
921
922 /* result = .2225 * tmp.y + tmp.x =.2225(y * abs(y) - y) + y= */
923 i915_emit_arith(p,
924 A0_MAD,
925 get_result_vector(p, inst),
926 get_result_flags(inst), 0,
927 swizzle(consts1, W, W, W, W),
928 swizzle(tmp, Y, Y, Y, Y),
929 swizzle(tmp, X, X, X, X));
930
931 break;
932
933 case OPCODE_SGE:
934 EMIT_2ARG_ARITH(A0_SGE);
935 break;
936
937 case OPCODE_SGT:
938 i915_emit_arith(p,
939 A0_SLT,
940 get_result_vector( p, inst ),
941 get_result_flags( inst ), 0,
942 negate(src_vector( p, &inst->SrcReg[0], program),
943 1, 1, 1, 1),
944 negate(src_vector( p, &inst->SrcReg[1], program),
945 1, 1, 1, 1),
946 0);
947 break;
948
949 case OPCODE_SLE:
950 i915_emit_arith(p,
951 A0_SGE,
952 get_result_vector( p, inst ),
953 get_result_flags( inst ), 0,
954 negate(src_vector( p, &inst->SrcReg[0], program),
955 1, 1, 1, 1),
956 negate(src_vector( p, &inst->SrcReg[1], program),
957 1, 1, 1, 1),
958 0);
959 break;
960
961 case OPCODE_SLT:
962 EMIT_2ARG_ARITH(A0_SLT);
963 break;
964
965 case OPCODE_SNE:
966 tmp = i915_get_utemp(p);
967 flags = get_result_flags(inst);
968 dst = get_result_vector(p, inst);
969
970 /* tmp = src1 < src2 */
971 i915_emit_arith(p,
972 A0_SLT,
973 tmp,
974 flags, 0,
975 src_vector(p, &inst->SrcReg[0], program),
976 src_vector(p, &inst->SrcReg[1], program),
977 0);
978 /* dst = src1 > src2 */
979 i915_emit_arith(p,
980 A0_SLT,
981 dst,
982 flags, 0,
983 negate(src_vector(p, &inst->SrcReg[0], program),
984 1, 1, 1, 1),
985 negate(src_vector(p, &inst->SrcReg[1], program),
986 1, 1, 1, 1),
987 0);
988 /* dst = tmp || dst */
989 i915_emit_arith(p,
990 A0_ADD,
991 dst,
992 flags | A0_DEST_SATURATE, 0,
993 dst,
994 tmp,
995 0);
996 break;
997
998 case OPCODE_SSG:
999 dst = get_result_vector(p, inst);
1000 flags = get_result_flags(inst);
1001 src0 = src_vector(p, &inst->SrcReg[0], program);
1002 tmp = i915_get_utemp(p);
1003
1004 /* tmp = (src < 0.0) */
1005 i915_emit_arith(p,
1006 A0_SLT,
1007 tmp,
1008 flags, 0,
1009 src0,
1010 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1011 0);
1012
1013 /* dst = (0.0 < src) */
1014 i915_emit_arith(p,
1015 A0_SLT,
1016 dst,
1017 flags, 0,
1018 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1019 src0,
1020 0);
1021
1022 /* dst = (src > 0.0) - (src < 0.0) */
1023 i915_emit_arith(p,
1024 A0_ADD,
1025 dst,
1026 flags, 0,
1027 dst,
1028 negate(tmp, 1, 1, 1, 1),
1029 0);
1030
1031 break;
1032
1033 case OPCODE_SUB:
1034 src0 = src_vector(p, &inst->SrcReg[0], program);
1035 src1 = src_vector(p, &inst->SrcReg[1], program);
1036
1037 i915_emit_arith(p,
1038 A0_ADD,
1039 get_result_vector(p, inst),
1040 get_result_flags(inst), 0,
1041 src0, negate(src1, 1, 1, 1, 1), 0);
1042 break;
1043
1044 case OPCODE_SWZ:
1045 EMIT_1ARG_ARITH(A0_MOV); /* extended swizzle handled natively */
1046 break;
1047
1048 case OPCODE_TEX:
1049 EMIT_TEX(T0_TEXLD);
1050 break;
1051
1052 case OPCODE_TXB:
1053 EMIT_TEX(T0_TEXLDB);
1054 break;
1055
1056 case OPCODE_TXP:
1057 EMIT_TEX(T0_TEXLDP);
1058 break;
1059
1060 case OPCODE_XPD:
1061 /* Cross product:
1062 * result.x = src0.y * src1.z - src0.z * src1.y;
1063 * result.y = src0.z * src1.x - src0.x * src1.z;
1064 * result.z = src0.x * src1.y - src0.y * src1.x;
1065 * result.w = undef;
1066 */
1067 src0 = src_vector(p, &inst->SrcReg[0], program);
1068 src1 = src_vector(p, &inst->SrcReg[1], program);
1069 tmp = i915_get_utemp(p);
1070
1071 i915_emit_arith(p,
1072 A0_MUL,
1073 tmp, A0_DEST_CHANNEL_ALL, 0,
1074 swizzle(src0, Z, X, Y, ONE),
1075 swizzle(src1, Y, Z, X, ONE), 0);
1076
1077 i915_emit_arith(p,
1078 A0_MAD,
1079 get_result_vector(p, inst),
1080 get_result_flags(inst), 0,
1081 swizzle(src0, Y, Z, X, ONE),
1082 swizzle(src1, Z, X, Y, ONE),
1083 negate(tmp, 1, 1, 1, 0));
1084 break;
1085
1086 case OPCODE_END:
1087 return;
1088
1089 case OPCODE_BGNLOOP:
1090 case OPCODE_BGNSUB:
1091 case OPCODE_BRA:
1092 case OPCODE_BRK:
1093 case OPCODE_CAL:
1094 case OPCODE_CONT:
1095 case OPCODE_DDX:
1096 case OPCODE_DDY:
1097 case OPCODE_ELSE:
1098 case OPCODE_ENDIF:
1099 case OPCODE_ENDLOOP:
1100 case OPCODE_ENDSUB:
1101 case OPCODE_IF:
1102 case OPCODE_RET:
1103 p->error = 1;
1104 i915_program_error(p, "Unsupported opcode: %s",
1105 _mesa_opcode_string(inst->Opcode));
1106 return;
1107
1108 case OPCODE_EXP:
1109 case OPCODE_LOG:
1110 /* These opcodes are claimed as GLSL, NV_vp, and ARB_vp in
1111 * prog_instruction.h, but apparently GLSL doesn't ever emit them.
1112 * Instead, it translates to EX2 or LG2.
1113 */
1114 case OPCODE_TXD:
1115 case OPCODE_TXL:
1116 /* These opcodes are claimed by GLSL in prog_instruction.h, but
1117 * only NV_vp/fp appears to emit them.
1118 */
1119 default:
1120 i915_program_error(p, "bad opcode: %s",
1121 _mesa_opcode_string(inst->Opcode));
1122 return;
1123 }
1124
1125 inst++;
1126 i915_release_utemps(p);
1127 }
1128 }
1129
1130 /* Rather than trying to intercept and jiggle depth writes during
1131 * emit, just move the value into its correct position at the end of
1132 * the program:
1133 */
1134 static void
fixup_depth_write(struct i915_fragment_program * p)1135 fixup_depth_write(struct i915_fragment_program *p)
1136 {
1137 if (p->depth_written) {
1138 GLuint depth = UREG(REG_TYPE_OD, 0);
1139
1140 i915_emit_arith(p,
1141 A0_MOV,
1142 depth, A0_DEST_CHANNEL_W, 0,
1143 swizzle(depth, X, Y, Z, Z), 0, 0);
1144 }
1145 }
1146
1147
1148 static void
check_wpos(struct i915_fragment_program * p)1149 check_wpos(struct i915_fragment_program *p)
1150 {
1151 GLbitfield64 inputs = p->FragProg.Base.InputsRead;
1152 GLint i;
1153
1154 p->wpos_tex = -1;
1155
1156 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1157 if (inputs & (FRAG_BIT_TEX(i) | FRAG_BIT_VAR(i)))
1158 continue;
1159 else if (inputs & FRAG_BIT_WPOS) {
1160 p->wpos_tex = i;
1161 inputs &= ~FRAG_BIT_WPOS;
1162 }
1163 }
1164
1165 if (inputs & FRAG_BIT_WPOS) {
1166 i915_program_error(p, "No free texcoord for wpos value");
1167 }
1168 }
1169
1170
1171 static void
translate_program(struct i915_fragment_program * p)1172 translate_program(struct i915_fragment_program *p)
1173 {
1174 struct i915_context *i915 = I915_CONTEXT(p->ctx);
1175
1176 if (INTEL_DEBUG & DEBUG_WM) {
1177 printf("fp:\n");
1178 _mesa_print_program(&p->FragProg.Base);
1179 printf("\n");
1180 }
1181
1182 i915_init_program(i915, p);
1183 check_wpos(p);
1184 upload_program(p);
1185 fixup_depth_write(p);
1186 i915_fini_program(p);
1187
1188 p->translated = 1;
1189 }
1190
1191
1192 static void
track_params(struct i915_fragment_program * p)1193 track_params(struct i915_fragment_program *p)
1194 {
1195 GLint i;
1196
1197 if (p->nr_params)
1198 _mesa_load_state_parameters(p->ctx, p->FragProg.Base.Parameters);
1199
1200 for (i = 0; i < p->nr_params; i++) {
1201 GLint reg = p->param[i].reg;
1202 COPY_4V(p->constant[reg], p->param[i].values);
1203 }
1204
1205 p->params_uptodate = 1;
1206 p->on_hardware = 0; /* overkill */
1207 }
1208
1209
1210 static void
i915BindProgram(struct gl_context * ctx,GLenum target,struct gl_program * prog)1211 i915BindProgram(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1212 {
1213 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1214 struct i915_context *i915 = I915_CONTEXT(ctx);
1215 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1216
1217 if (i915->current_program == p)
1218 return;
1219
1220 if (i915->current_program) {
1221 i915->current_program->on_hardware = 0;
1222 i915->current_program->params_uptodate = 0;
1223 }
1224
1225 i915->current_program = p;
1226
1227 assert(p->on_hardware == 0);
1228 assert(p->params_uptodate == 0);
1229
1230 }
1231 }
1232
1233 static struct gl_program *
i915NewProgram(struct gl_context * ctx,GLenum target,GLuint id)1234 i915NewProgram(struct gl_context * ctx, GLenum target, GLuint id)
1235 {
1236 switch (target) {
1237 case GL_VERTEX_PROGRAM_ARB:
1238 return _mesa_init_vertex_program(ctx, CALLOC_STRUCT(gl_vertex_program),
1239 target, id);
1240
1241 case GL_FRAGMENT_PROGRAM_ARB:{
1242 struct i915_fragment_program *prog =
1243 CALLOC_STRUCT(i915_fragment_program);
1244 if (prog) {
1245 i915_init_program(I915_CONTEXT(ctx), prog);
1246
1247 return _mesa_init_fragment_program(ctx, &prog->FragProg,
1248 target, id);
1249 }
1250 else
1251 return NULL;
1252 }
1253
1254 default:
1255 /* Just fallback:
1256 */
1257 return _mesa_new_program(ctx, target, id);
1258 }
1259 }
1260
1261 static void
i915DeleteProgram(struct gl_context * ctx,struct gl_program * prog)1262 i915DeleteProgram(struct gl_context * ctx, struct gl_program *prog)
1263 {
1264 if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
1265 struct i915_context *i915 = I915_CONTEXT(ctx);
1266 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1267
1268 if (i915->current_program == p)
1269 i915->current_program = 0;
1270 }
1271
1272 _mesa_delete_program(ctx, prog);
1273 }
1274
1275
1276 static GLboolean
i915IsProgramNative(struct gl_context * ctx,GLenum target,struct gl_program * prog)1277 i915IsProgramNative(struct gl_context * ctx, GLenum target, struct gl_program *prog)
1278 {
1279 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1280 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1281
1282 if (!p->translated)
1283 translate_program(p);
1284
1285 return !p->error;
1286 }
1287 else
1288 return true;
1289 }
1290
1291 static GLboolean
i915ProgramStringNotify(struct gl_context * ctx,GLenum target,struct gl_program * prog)1292 i915ProgramStringNotify(struct gl_context * ctx,
1293 GLenum target, struct gl_program *prog)
1294 {
1295 if (target == GL_FRAGMENT_PROGRAM_ARB) {
1296 struct i915_fragment_program *p = (struct i915_fragment_program *) prog;
1297 p->translated = 0;
1298 }
1299
1300 (void) _tnl_program_string(ctx, target, prog);
1301
1302 /* XXX check if program is legal, within limits */
1303 return true;
1304 }
1305
1306 static void
i915SamplerUniformChange(struct gl_context * ctx,GLenum target,struct gl_program * prog)1307 i915SamplerUniformChange(struct gl_context *ctx,
1308 GLenum target, struct gl_program *prog)
1309 {
1310 i915ProgramStringNotify(ctx, target, prog);
1311 }
1312
1313 void
i915_update_program(struct gl_context * ctx)1314 i915_update_program(struct gl_context *ctx)
1315 {
1316 struct intel_context *intel = intel_context(ctx);
1317 struct i915_context *i915 = i915_context(&intel->ctx);
1318 struct i915_fragment_program *fp =
1319 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1320
1321 if (i915->current_program != fp) {
1322 if (i915->current_program) {
1323 i915->current_program->on_hardware = 0;
1324 i915->current_program->params_uptodate = 0;
1325 }
1326
1327 i915->current_program = fp;
1328 }
1329
1330 if (!fp->translated)
1331 translate_program(fp);
1332
1333 FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, fp->error);
1334 }
1335
1336 void
i915ValidateFragmentProgram(struct i915_context * i915)1337 i915ValidateFragmentProgram(struct i915_context *i915)
1338 {
1339 struct gl_context *ctx = &i915->intel.ctx;
1340 struct intel_context *intel = intel_context(ctx);
1341 TNLcontext *tnl = TNL_CONTEXT(ctx);
1342 struct vertex_buffer *VB = &tnl->vb;
1343
1344 struct i915_fragment_program *p =
1345 (struct i915_fragment_program *) ctx->FragmentProgram._Current;
1346
1347 const GLbitfield64 inputsRead = p->FragProg.Base.InputsRead;
1348 GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
1349 GLuint s2 = S2_TEXCOORD_NONE;
1350 int i, offset = 0;
1351
1352 /* Important:
1353 */
1354 VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
1355
1356 if (!p->translated)
1357 translate_program(p);
1358
1359 intel->vertex_attr_count = 0;
1360 intel->wpos_offset = 0;
1361 intel->coloroffset = 0;
1362 intel->specoffset = 0;
1363
1364 if (inputsRead & FRAG_BITS_TEX_ANY || p->wpos_tex != -1) {
1365 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16);
1366 }
1367 else {
1368 EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12);
1369 }
1370
1371 /* Handle gl_PointSize builtin var here */
1372 if (ctx->Point._Attenuated || ctx->VertexProgram.PointSizeEnabled)
1373 EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, S4_VFMT_POINT_WIDTH, 4);
1374
1375 if (inputsRead & FRAG_BIT_COL0) {
1376 intel->coloroffset = offset / 4;
1377 EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4);
1378 }
1379
1380 if (inputsRead & FRAG_BIT_COL1) {
1381 intel->specoffset = offset / 4;
1382 EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_4UB_4F_BGRA, S4_VFMT_SPEC_FOG, 4);
1383 }
1384
1385 if ((inputsRead & FRAG_BIT_FOGC)) {
1386 EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1F, S4_VFMT_FOG_PARAM, 4);
1387 }
1388
1389 for (i = 0; i < p->ctx->Const.MaxTextureCoordUnits; i++) {
1390 if (inputsRead & FRAG_BIT_TEX(i)) {
1391 int sz = VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]->size;
1392
1393 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1394 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1395
1396 EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, EMIT_SZ(sz), 0, sz * 4);
1397 }
1398 else if (inputsRead & FRAG_BIT_VAR(i)) {
1399 int sz = VB->AttribPtr[_TNL_ATTRIB_GENERIC0 + i]->size;
1400
1401 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1402 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
1403
1404 EMIT_ATTR(_TNL_ATTRIB_GENERIC0 + i, EMIT_SZ(sz), 0, sz * 4);
1405 }
1406 else if (i == p->wpos_tex) {
1407 int wpos_size = 4 * sizeof(float);
1408 /* If WPOS is required, duplicate the XYZ position data in an
1409 * unused texture coordinate:
1410 */
1411 s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
1412 s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(wpos_size));
1413
1414 intel->wpos_offset = offset;
1415 EMIT_PAD(wpos_size);
1416 }
1417 }
1418
1419 if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
1420 s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
1421 int k;
1422
1423 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
1424
1425 /* Must do this *after* statechange, so as not to affect
1426 * buffered vertices reliant on the old state:
1427 */
1428 intel->vertex_size = _tnl_install_attrs(&intel->ctx,
1429 intel->vertex_attrs,
1430 intel->vertex_attr_count,
1431 intel->ViewportMatrix.m, 0);
1432
1433 assert(intel->prim.current_offset == intel->prim.start_offset);
1434 intel->prim.start_offset = (intel->prim.current_offset + intel->vertex_size-1) / intel->vertex_size * intel->vertex_size;
1435 intel->prim.current_offset = intel->prim.start_offset;
1436
1437 intel->vertex_size >>= 2;
1438
1439 i915->state.Ctx[I915_CTXREG_LIS2] = s2;
1440 i915->state.Ctx[I915_CTXREG_LIS4] = s4;
1441
1442 k = intel->vtbl.check_vertex_size(intel, intel->vertex_size);
1443 assert(k);
1444 }
1445
1446 if (!p->params_uptodate)
1447 track_params(p);
1448
1449 if (!p->on_hardware)
1450 i915_upload_program(i915, p);
1451
1452 if (INTEL_DEBUG & DEBUG_WM) {
1453 printf("i915:\n");
1454 i915_disassemble_program(i915->state.Program, i915->state.ProgramSize);
1455 }
1456 }
1457
1458 void
i915InitFragProgFuncs(struct dd_function_table * functions)1459 i915InitFragProgFuncs(struct dd_function_table *functions)
1460 {
1461 functions->BindProgram = i915BindProgram;
1462 functions->NewProgram = i915NewProgram;
1463 functions->DeleteProgram = i915DeleteProgram;
1464 functions->IsProgramNative = i915IsProgramNative;
1465 functions->ProgramStringNotify = i915ProgramStringNotify;
1466 functions->SamplerUniformChange = i915SamplerUniformChange;
1467 }
1468