projects / mesa.git / blob
? search:


738d331a56c5553be28c9e73070d820d736a7c2b
[mesa.git] / src / gallium / drivers / i915 / i915_fpc_translate.c
1 /**************************************************************************
2 *
3 * Copyright 2007 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
29 #include <stdarg.h>
30
31 #include "i915_reg.h"
32 #include "i915_context.h"
33 #include "i915_fpc.h"
34
35 #include "pipe/p_shader_tokens.h"
36 #include "util/u_math.h"
37 #include "util/u_memory.h"
38 #include "util/u_string.h"
39 #include "tgsi/tgsi_parse.h"
40 #include "tgsi/tgsi_dump.h"
41
42 #include "draw/draw_vertex.h"
43
44
45 /**
46 * Simple pass-through fragment shader to use when we don't have
47 * a real shader (or it fails to compile for some reason).
48 */
49 static unsigned passthrough[] =
50 {
51 _3DSTATE_PIXEL_SHADER_PROGRAM | ((2*3)-1),
52
53 /* declare input color:
54 */
55 (D0_DCL |
56 (REG_TYPE_T << D0_TYPE_SHIFT) |
57 (T_DIFFUSE << D0_NR_SHIFT) |
58 D0_CHANNEL_ALL),
59 0,
60 0,
61
62 /* move to output color:
63 */
64 (A0_MOV |
65 (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) |
66 A0_DEST_CHANNEL_ALL |
67 (REG_TYPE_T << A0_SRC0_TYPE_SHIFT) |
68 (T_DIFFUSE << A0_SRC0_NR_SHIFT)),
69 0x01230000, /* .xyzw */
70 0
71 };
72
73
74 /* 1, -1/3!, 1/5!, -1/7! */
75 static const float sin_constants[4] = { 1.0,
76 -1.0f / (3 * 2 * 1),
77 1.0f / (5 * 4 * 3 * 2 * 1),
78 -1.0f / (7 * 6 * 5 * 4 * 3 * 2 * 1)
79 };
80
81 /* 1, -1/2!, 1/4!, -1/6! */
82 static const float cos_constants[4] = { 1.0,
83 -1.0f / (2 * 1),
84 1.0f / (4 * 3 * 2 * 1),
85 -1.0f / (6 * 5 * 4 * 3 * 2 * 1)
86 };
87
88
89
90 /**
91 * component-wise negation of ureg
92 */
93 static INLINE int
94 negate(int reg, int x, int y, int z, int w)
95 {
96 /* Another neat thing about the UREG representation */
97 return reg ^ (((x & 1) << UREG_CHANNEL_X_NEGATE_SHIFT) |
98 ((y & 1) << UREG_CHANNEL_Y_NEGATE_SHIFT) |
99 ((z & 1) << UREG_CHANNEL_Z_NEGATE_SHIFT) |
100 ((w & 1) << UREG_CHANNEL_W_NEGATE_SHIFT));
101 }
102
103
104 /**
105 * In the event of a translation failure, we'll generate a simple color
106 * pass-through program.
107 */
108 static void
109 i915_use_passthrough_shader(struct i915_fragment_shader *fs)
110 {
111 fs->program = (uint *) MALLOC(sizeof(passthrough));
112 if (fs->program) {
113 memcpy(fs->program, passthrough, sizeof(passthrough));
114 fs->program_len = Elements(passthrough);
115 }
116 fs->num_constants = 0;
117 }
118
119
120 void
121 i915_program_error(struct i915_fp_compile *p, const char *msg, ...)
122 {
123 va_list args;
124 char buffer[1024];
125
126 debug_printf("i915_program_error: ");
127 va_start( args, msg );
128 util_vsnprintf( buffer, sizeof(buffer), msg, args );
129 va_end( args );
130 debug_printf("%s", buffer);
131 debug_printf("\n");
132
133 p->error = 1;
134 }
135
136 static uint get_mapping(struct i915_fragment_shader* fs, int unit)
137 {
138 int i;
139 for (i = 0; i < I915_TEX_UNITS; i++)
140 {
141 if (fs->generic_mapping[i] == -1) {
142 fs->generic_mapping[i] = unit;
143 return i;
144 }
145 if (fs->generic_mapping[i] == unit)
146 return i;
147 }
148 debug_printf("Exceeded max generics\n");
149 return 0;
150 }
151
152 /**
153 * Construct a ureg for the given source register. Will emit
154 * constants, apply swizzling and negation as needed.
155 */
156 static uint
157 src_vector(struct i915_fp_compile *p,
158 const struct tgsi_full_src_register *source,
159 struct i915_fragment_shader* fs)
160 {
161 uint index = source->Register.Index;
162 uint src = 0, sem_name, sem_ind;
163
164 switch (source->Register.File) {
165 case TGSI_FILE_TEMPORARY:
166 if (source->Register.Index >= I915_MAX_TEMPORARY) {
167 i915_program_error(p, "Exceeded max temporary reg");
168 return 0;
169 }
170 src = UREG(REG_TYPE_R, index);
171 break;
172 case TGSI_FILE_INPUT:
173 /* XXX: Packing COL1, FOGC into a single attribute works for
174 * texenv programs, but will fail for real fragment programs
175 * that use these attributes and expect them to be a full 4
176 * components wide. Could use a texcoord to pass these
177 * attributes if necessary, but that won't work in the general
178 * case.
179 *
180 * We also use a texture coordinate to pass wpos when possible.
181 */
182
183 sem_name = p->shader->info.input_semantic_name[index];
184 sem_ind = p->shader->info.input_semantic_index[index];
185
186 switch (sem_name) {
187 case TGSI_SEMANTIC_POSITION:
188 debug_printf("SKIP SEM POS\n");
189 /*
190 assert(p->wpos_tex != -1);
191 src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
192 */
193 break;
194 case TGSI_SEMANTIC_COLOR:
195 if (sem_ind == 0) {
196 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
197 }
198 else {
199 /* secondary color */
200 assert(sem_ind == 1);
201 src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
202 src = swizzle(src, X, Y, Z, ONE);
203 }
204 break;
205 case TGSI_SEMANTIC_FOG:
206 src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
207 src = swizzle(src, W, W, W, W);
208 break;
209 case TGSI_SEMANTIC_GENERIC:
210 {
211 int real_tex_unit = get_mapping(fs, sem_ind);
212 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_ALL);
213 break;
214 }
215 default:
216 i915_program_error(p, "Bad source->Index");
217 return 0;
218 }
219 break;
220
221 case TGSI_FILE_IMMEDIATE:
222 assert(index < p->num_immediates);
223 index = p->immediates_map[index];
224 /* fall-through */
225 case TGSI_FILE_CONSTANT:
226 src = UREG(REG_TYPE_CONST, index);
227 break;
228
229 default:
230 i915_program_error(p, "Bad source->File");
231 return 0;
232 }
233
234 src = swizzle(src,
235 source->Register.SwizzleX,
236 source->Register.SwizzleY,
237 source->Register.SwizzleZ,
238 source->Register.SwizzleW);
239
240
241 /* There's both negate-all-components and per-component negation.
242 * Try to handle both here.
243 */
244 {
245 int n = source->Register.Negate;
246 src = negate(src, n, n, n, n);
247 }
248
249 /* no abs() */
250 #if 0
251 /* XXX assertions disabled to allow arbfplight.c to run */
252 /* XXX enable these assertions, or fix things */
253 assert(!source->Register.Absolute);
254 #endif
255 return src;
256 }
257
258
259 /**
260 * Construct a ureg for a destination register.
261 */
262 static uint
263 get_result_vector(struct i915_fp_compile *p,
264 const struct tgsi_full_dst_register *dest)
265 {
266 switch (dest->Register.File) {
267 case TGSI_FILE_OUTPUT:
268 {
269 uint sem_name = p->shader->info.output_semantic_name[dest->Register.Index];
270 switch (sem_name) {
271 case TGSI_SEMANTIC_POSITION:
272 return UREG(REG_TYPE_OD, 0);
273 case TGSI_SEMANTIC_COLOR:
274 return UREG(REG_TYPE_OC, 0);
275 default:
276 i915_program_error(p, "Bad inst->DstReg.Index/semantics");
277 return 0;
278 }
279 }
280 case TGSI_FILE_TEMPORARY:
281 return UREG(REG_TYPE_R, dest->Register.Index);
282 default:
283 i915_program_error(p, "Bad inst->DstReg.File");
284 return 0;
285 }
286 }
287
288
289 /**
290 * Compute flags for saturation and writemask.
291 */
292 static uint
293 get_result_flags(const struct tgsi_full_instruction *inst)
294 {
295 const uint writeMask
296 = inst->Dst[0].Register.WriteMask;
297 uint flags = 0x0;
298
299 if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE)
300 flags |= A0_DEST_SATURATE;
301
302 if (writeMask & TGSI_WRITEMASK_X)
303 flags |= A0_DEST_CHANNEL_X;
304 if (writeMask & TGSI_WRITEMASK_Y)
305 flags |= A0_DEST_CHANNEL_Y;
306 if (writeMask & TGSI_WRITEMASK_Z)
307 flags |= A0_DEST_CHANNEL_Z;
308 if (writeMask & TGSI_WRITEMASK_W)
309 flags |= A0_DEST_CHANNEL_W;
310
311 return flags;
312 }
313
314
315 /**
316 * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
317 */
318 static uint
319 translate_tex_src_target(struct i915_fp_compile *p, uint tex)
320 {
321 switch (tex) {
322 case TGSI_TEXTURE_SHADOW1D:
323 /* fall-through */
324 case TGSI_TEXTURE_1D:
325 return D0_SAMPLE_TYPE_2D;
326
327 case TGSI_TEXTURE_SHADOW2D:
328 /* fall-through */
329 case TGSI_TEXTURE_2D:
330 return D0_SAMPLE_TYPE_2D;
331
332 case TGSI_TEXTURE_SHADOWRECT:
333 /* fall-through */
334 case TGSI_TEXTURE_RECT:
335 return D0_SAMPLE_TYPE_2D;
336
337 case TGSI_TEXTURE_3D:
338 return D0_SAMPLE_TYPE_VOLUME;
339
340 case TGSI_TEXTURE_CUBE:
341 return D0_SAMPLE_TYPE_CUBE;
342
343 default:
344 i915_program_error(p, "TexSrc type");
345 return 0;
346 }
347 }
348
349
350 /**
351 * Generate texel lookup instruction.
352 */
353 static void
354 emit_tex(struct i915_fp_compile *p,
355 const struct tgsi_full_instruction *inst,
356 uint opcode,
357 struct i915_fragment_shader* fs)
358 {
359 uint texture = inst->Texture.Texture;
360 uint unit = inst->Src[1].Register.Index;
361 uint tex = translate_tex_src_target( p, texture );
362 uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex);
363 uint coord = src_vector( p, &inst->Src[0], fs);
364
365 i915_emit_texld( p,
366 get_result_vector( p, &inst->Dst[0] ),
367 get_result_flags( inst ),
368 sampler,
369 coord,
370 opcode);
371 }
372
373
374 /**
375 * Generate a simple arithmetic instruction
376 * \param opcode the i915 opcode
377 * \param numArgs the number of input/src arguments
378 */
379 static void
380 emit_simple_arith(struct i915_fp_compile *p,
381 const struct tgsi_full_instruction *inst,
382 uint opcode, uint numArgs,
383 struct i915_fragment_shader* fs)
384 {
385 uint arg1, arg2, arg3;
386
387 assert(numArgs <= 3);
388
389 arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->Src[0], fs );
390 arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->Src[1], fs );
391 arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->Src[2], fs );
392
393 i915_emit_arith( p,
394 opcode,
395 get_result_vector( p, &inst->Dst[0]),
396 get_result_flags( inst ), 0,
397 arg1,
398 arg2,
399 arg3 );
400 }
401
402
403 /** As above, but swap the first two src regs */
404 static void
405 emit_simple_arith_swap2(struct i915_fp_compile *p,
406 const struct tgsi_full_instruction *inst,
407 uint opcode, uint numArgs,
408 struct i915_fragment_shader* fs)
409 {
410 struct tgsi_full_instruction inst2;
411
412 assert(numArgs == 2);
413
414 /* transpose first two registers */
415 inst2 = *inst;
416 inst2.Src[0] = inst->Src[1];
417 inst2.Src[1] = inst->Src[0];
418
419 emit_simple_arith(p, &inst2, opcode, numArgs, fs);
420 }
421
422
423 #ifndef M_PI
424 #define M_PI 3.14159265358979323846
425 #endif
426
427 /*
428 * Translate TGSI instruction to i915 instruction.
429 *
430 * Possible concerns:
431 *
432 * SIN, COS -- could use another taylor step?
433 * LIT -- results seem a little different to sw mesa
434 * LOG -- different to mesa on negative numbers, but this is conformant.
435 */
436 static void
437 i915_translate_instruction(struct i915_fp_compile *p,
438 const struct tgsi_full_instruction *inst,
439 struct i915_fragment_shader *fs)
440 {
441 uint writemask;
442 uint src0, src1, src2, flags;
443 uint tmp = 0;
444
445 switch (inst->Instruction.Opcode) {
446 case TGSI_OPCODE_ABS:
447 src0 = src_vector(p, &inst->Src[0], fs);
448 i915_emit_arith(p,
449 A0_MAX,
450 get_result_vector(p, &inst->Dst[0]),
451 get_result_flags(inst), 0,
452 src0, negate(src0, 1, 1, 1, 1), 0);
453 break;
454
455 case TGSI_OPCODE_ADD:
456 emit_simple_arith(p, inst, A0_ADD, 2, fs);
457 break;
458
459 case TGSI_OPCODE_CMP:
460 src0 = src_vector(p, &inst->Src[0], fs);
461 src1 = src_vector(p, &inst->Src[1], fs);
462 src2 = src_vector(p, &inst->Src[2], fs);
463 i915_emit_arith(p, A0_CMP,
464 get_result_vector(p, &inst->Dst[0]),
465 get_result_flags(inst),
466 0, src0, src2, src1); /* NOTE: order of src2, src1 */
467 break;
468
469 case TGSI_OPCODE_COS:
470 src0 = src_vector(p, &inst->Src[0], fs);
471 tmp = i915_get_utemp(p);
472
473 i915_emit_arith(p,
474 A0_MUL,
475 tmp, A0_DEST_CHANNEL_X, 0,
476 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
477
478 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
479
480 /* By choosing different taylor constants, could get rid of this mul:
481 */
482 i915_emit_arith(p,
483 A0_MUL,
484 tmp, A0_DEST_CHANNEL_X, 0,
485 tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0);
486
487 /*
488 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
489 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
490 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
491 * result = DP4 t0, cos_constants
492 */
493 i915_emit_arith(p,
494 A0_MUL,
495 tmp, A0_DEST_CHANNEL_XY, 0,
496 swizzle(tmp, X, X, ONE, ONE),
497 swizzle(tmp, X, ONE, ONE, ONE), 0);
498
499 i915_emit_arith(p,
500 A0_MUL,
501 tmp, A0_DEST_CHANNEL_XYZ, 0,
502 swizzle(tmp, X, Y, X, ONE),
503 swizzle(tmp, X, X, ONE, ONE), 0);
504
505 i915_emit_arith(p,
506 A0_MUL,
507 tmp, A0_DEST_CHANNEL_XYZ, 0,
508 swizzle(tmp, X, X, Z, ONE),
509 swizzle(tmp, Z, ONE, ONE, ONE), 0);
510
511 i915_emit_arith(p,
512 A0_DP4,
513 get_result_vector(p, &inst->Dst[0]),
514 get_result_flags(inst), 0,
515 swizzle(tmp, ONE, Z, Y, X),
516 i915_emit_const4fv(p, cos_constants), 0);
517 break;
518
519 case TGSI_OPCODE_DP2:
520 src0 = src_vector(p, &inst->Src[0], fs);
521 src1 = src_vector(p, &inst->Src[1], fs);
522
523 i915_emit_arith(p,
524 A0_DP3,
525 get_result_vector(p, &inst->Dst[0]),
526 get_result_flags(inst), 0,
527 swizzle(src0, X, Y, ZERO, ZERO), src1, 0);
528 break;
529
530 case TGSI_OPCODE_DP3:
531 emit_simple_arith(p, inst, A0_DP3, 2, fs);
532 break;
533
534 case TGSI_OPCODE_DP4:
535 emit_simple_arith(p, inst, A0_DP4, 2, fs);
536 break;
537
538 case TGSI_OPCODE_DPH:
539 src0 = src_vector(p, &inst->Src[0], fs);
540 src1 = src_vector(p, &inst->Src[1], fs);
541
542 i915_emit_arith(p,
543 A0_DP4,
544 get_result_vector(p, &inst->Dst[0]),
545 get_result_flags(inst), 0,
546 swizzle(src0, X, Y, Z, ONE), src1, 0);
547 break;
548
549 case TGSI_OPCODE_DST:
550 src0 = src_vector(p, &inst->Src[0], fs);
551 src1 = src_vector(p, &inst->Src[1], fs);
552
553 /* result[0] = 1 * 1;
554 * result[1] = a[1] * b[1];
555 * result[2] = a[2] * 1;
556 * result[3] = 1 * b[3];
557 */
558 i915_emit_arith(p,
559 A0_MUL,
560 get_result_vector(p, &inst->Dst[0]),
561 get_result_flags(inst), 0,
562 swizzle(src0, ONE, Y, Z, ONE),
563 swizzle(src1, ONE, Y, ONE, W), 0);
564 break;
565
566 case TGSI_OPCODE_END:
567 /* no-op */
568 break;
569
570 case TGSI_OPCODE_EX2:
571 src0 = src_vector(p, &inst->Src[0], fs);
572
573 i915_emit_arith(p,
574 A0_EXP,
575 get_result_vector(p, &inst->Dst[0]),
576 get_result_flags(inst), 0,
577 swizzle(src0, X, X, X, X), 0, 0);
578 break;
579
580 case TGSI_OPCODE_FLR:
581 emit_simple_arith(p, inst, A0_FLR, 1, fs);
582 break;
583
584 case TGSI_OPCODE_FRC:
585 emit_simple_arith(p, inst, A0_FRC, 1, fs);
586 break;
587
588 case TGSI_OPCODE_KIL:
589 /* kill if src[0].x < 0 || src[0].y < 0 ... */
590 src0 = src_vector(p, &inst->Src[0], fs);
591 tmp = i915_get_utemp(p);
592
593 i915_emit_texld(p,
594 tmp, /* dest reg: a dummy reg */
595 A0_DEST_CHANNEL_ALL, /* dest writemask */
596 0, /* sampler */
597 src0, /* coord*/
598 T0_TEXKILL); /* opcode */
599 break;
600
601 case TGSI_OPCODE_KILP:
602 assert(0); /* not tested yet */
603 break;
604
605 case TGSI_OPCODE_LG2:
606 src0 = src_vector(p, &inst->Src[0], fs);
607
608 i915_emit_arith(p,
609 A0_LOG,
610 get_result_vector(p, &inst->Dst[0]),
611 get_result_flags(inst), 0,
612 swizzle(src0, X, X, X, X), 0, 0);
613 break;
614
615 case TGSI_OPCODE_LIT:
616 src0 = src_vector(p, &inst->Src[0], fs);
617 tmp = i915_get_utemp(p);
618
619 /* tmp = max( a.xyzw, a.00zw )
620 * XXX: Clamp tmp.w to -128..128
621 * tmp.y = log(tmp.y)
622 * tmp.y = tmp.w * tmp.y
623 * tmp.y = exp(tmp.y)
624 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
625 */
626 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
627 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
628
629 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
630 swizzle(tmp, Y, Y, Y, Y), 0, 0);
631
632 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
633 swizzle(tmp, ZERO, Y, ZERO, ZERO),
634 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
635
636 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
637 swizzle(tmp, Y, Y, Y, Y), 0, 0);
638
639 i915_emit_arith(p, A0_CMP,
640 get_result_vector(p, &inst->Dst[0]),
641 get_result_flags(inst), 0,
642 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
643 swizzle(tmp, ONE, X, ZERO, ONE),
644 swizzle(tmp, ONE, X, Y, ONE));
645
646 break;
647
648 case TGSI_OPCODE_LRP:
649 src0 = src_vector(p, &inst->Src[0], fs);
650 src1 = src_vector(p, &inst->Src[1], fs);
651 src2 = src_vector(p, &inst->Src[2], fs);
652 flags = get_result_flags(inst);
653 tmp = i915_get_utemp(p);
654
655 /* b*a + c*(1-a)
656 *
657 * b*a + c - ca
658 *
659 * tmp = b*a + c,
660 * result = (-c)*a + tmp
661 */
662 i915_emit_arith(p, A0_MAD, tmp,
663 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
664
665 i915_emit_arith(p, A0_MAD,
666 get_result_vector(p, &inst->Dst[0]),
667 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
668 break;
669
670 case TGSI_OPCODE_MAD:
671 emit_simple_arith(p, inst, A0_MAD, 3, fs);
672 break;
673
674 case TGSI_OPCODE_MAX:
675 emit_simple_arith(p, inst, A0_MAX, 2, fs);
676 break;
677
678 case TGSI_OPCODE_MIN:
679 src0 = src_vector(p, &inst->Src[0], fs);
680 src1 = src_vector(p, &inst->Src[1], fs);
681 tmp = i915_get_utemp(p);
682 flags = get_result_flags(inst);
683
684 i915_emit_arith(p,
685 A0_MAX,
686 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
687 negate(src0, 1, 1, 1, 1),
688 negate(src1, 1, 1, 1, 1), 0);
689
690 i915_emit_arith(p,
691 A0_MOV,
692 get_result_vector(p, &inst->Dst[0]),
693 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
694 break;
695
696 case TGSI_OPCODE_MOV:
697 emit_simple_arith(p, inst, A0_MOV, 1, fs);
698 break;
699
700 case TGSI_OPCODE_MUL:
701 emit_simple_arith(p, inst, A0_MUL, 2, fs);
702 break;
703
704 case TGSI_OPCODE_POW:
705 src0 = src_vector(p, &inst->Src[0], fs);
706 src1 = src_vector(p, &inst->Src[1], fs);
707 tmp = i915_get_utemp(p);
708 flags = get_result_flags(inst);
709
710 /* XXX: masking on intermediate values, here and elsewhere.
711 */
712 i915_emit_arith(p,
713 A0_LOG,
714 tmp, A0_DEST_CHANNEL_X, 0,
715 swizzle(src0, X, X, X, X), 0, 0);
716
717 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
718
719 i915_emit_arith(p,
720 A0_EXP,
721 get_result_vector(p, &inst->Dst[0]),
722 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
723 break;
724
725 case TGSI_OPCODE_RET:
726 /* XXX: no-op? */
727 break;
728
729 case TGSI_OPCODE_RCP:
730 src0 = src_vector(p, &inst->Src[0], fs);
731
732 i915_emit_arith(p,
733 A0_RCP,
734 get_result_vector(p, &inst->Dst[0]),
735 get_result_flags(inst), 0,
736 swizzle(src0, X, X, X, X), 0, 0);
737 break;
738
739 case TGSI_OPCODE_RSQ:
740 src0 = src_vector(p, &inst->Src[0], fs);
741
742 i915_emit_arith(p,
743 A0_RSQ,
744 get_result_vector(p, &inst->Dst[0]),
745 get_result_flags(inst), 0,
746 swizzle(src0, X, X, X, X), 0, 0);
747 break;
748
749 case TGSI_OPCODE_SCS:
750 src0 = src_vector(p, &inst->Src[0], fs);
751 tmp = i915_get_utemp(p);
752
753 /*
754 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
755 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
756 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
757 * scs.x = DP4 t1, sin_constants
758 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
759 * scs.y = DP4 t1, cos_constants
760 */
761 i915_emit_arith(p,
762 A0_MUL,
763 tmp, A0_DEST_CHANNEL_XY, 0,
764 swizzle(src0, X, X, ONE, ONE),
765 swizzle(src0, X, ONE, ONE, ONE), 0);
766
767 i915_emit_arith(p,
768 A0_MUL,
769 tmp, A0_DEST_CHANNEL_ALL, 0,
770 swizzle(tmp, X, Y, X, Y),
771 swizzle(tmp, X, X, ONE, ONE), 0);
772
773 writemask = inst->Dst[0].Register.WriteMask;
774
775 if (writemask & TGSI_WRITEMASK_Y) {
776 uint tmp1;
777
778 if (writemask & TGSI_WRITEMASK_X)
779 tmp1 = i915_get_utemp(p);
780 else
781 tmp1 = tmp;
782
783 i915_emit_arith(p,
784 A0_MUL,
785 tmp1, A0_DEST_CHANNEL_ALL, 0,
786 swizzle(tmp, X, Y, Y, W),
787 swizzle(tmp, X, Z, ONE, ONE), 0);
788
789 i915_emit_arith(p,
790 A0_DP4,
791 get_result_vector(p, &inst->Dst[0]),
792 A0_DEST_CHANNEL_Y, 0,
793 swizzle(tmp1, W, Z, Y, X),
794 i915_emit_const4fv(p, sin_constants), 0);
795 }
796
797 if (writemask & TGSI_WRITEMASK_X) {
798 i915_emit_arith(p,
799 A0_MUL,
800 tmp, A0_DEST_CHANNEL_XYZ, 0,
801 swizzle(tmp, X, X, Z, ONE),
802 swizzle(tmp, Z, ONE, ONE, ONE), 0);
803
804 i915_emit_arith(p,
805 A0_DP4,
806 get_result_vector(p, &inst->Dst[0]),
807 A0_DEST_CHANNEL_X, 0,
808 swizzle(tmp, ONE, Z, Y, X),
809 i915_emit_const4fv(p, cos_constants), 0);
810 }
811 break;
812
813 case TGSI_OPCODE_SEQ:
814 /* if we're both >= and <= then we're == */
815 src0 = src_vector(p, &inst->Src[0], fs);
816 src1 = src_vector(p, &inst->Src[1], fs);
817 tmp = i915_get_utemp(p);
818
819 i915_emit_arith(p,
820 A0_SGE,
821 tmp, A0_DEST_CHANNEL_ALL, 0,
822 src0,
823 src1, 0);
824
825 i915_emit_arith(p,
826 A0_SGE,
827 get_result_vector(p, &inst->Dst[0]),
828 A0_DEST_CHANNEL_ALL, 0,
829 src1,
830 src0, 0);
831
832 i915_emit_arith(p,
833 A0_MUL,
834 get_result_vector(p, &inst->Dst[0]),
835 A0_DEST_CHANNEL_ALL, 0,
836 get_result_vector(p, &inst->Dst[0]),
837 tmp, 0);
838
839 break;
840
841 case TGSI_OPCODE_SGE:
842 emit_simple_arith(p, inst, A0_SGE, 2, fs);
843 break;
844
845 case TGSI_OPCODE_SIN:
846 src0 = src_vector(p, &inst->Src[0], fs);
847 tmp = i915_get_utemp(p);
848
849 i915_emit_arith(p,
850 A0_MUL,
851 tmp, A0_DEST_CHANNEL_X, 0,
852 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
853
854 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
855
856 /* By choosing different taylor constants, could get rid of this mul:
857 */
858 i915_emit_arith(p,
859 A0_MUL,
860 tmp, A0_DEST_CHANNEL_X, 0,
861 tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0);
862
863 /*
864 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
865 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
866 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
867 * result = DP4 t1.wzyx, sin_constants
868 */
869 i915_emit_arith(p,
870 A0_MUL,
871 tmp, A0_DEST_CHANNEL_XY, 0,
872 swizzle(tmp, X, X, ONE, ONE),
873 swizzle(tmp, X, ONE, ONE, ONE), 0);
874
875 i915_emit_arith(p,
876 A0_MUL,
877 tmp, A0_DEST_CHANNEL_ALL, 0,
878 swizzle(tmp, X, Y, X, Y),
879 swizzle(tmp, X, X, ONE, ONE), 0);
880
881 i915_emit_arith(p,
882 A0_MUL,
883 tmp, A0_DEST_CHANNEL_ALL, 0,
884 swizzle(tmp, X, Y, Y, W),
885 swizzle(tmp, X, Z, ONE, ONE), 0);
886
887 i915_emit_arith(p,
888 A0_DP4,
889 get_result_vector(p, &inst->Dst[0]),
890 get_result_flags(inst), 0,
891 swizzle(tmp, W, Z, Y, X),
892 i915_emit_const4fv(p, sin_constants), 0);
893 break;
894
895 case TGSI_OPCODE_SLE:
896 /* like SGE, but swap reg0, reg1 */
897 emit_simple_arith_swap2(p, inst, A0_SGE, 2, fs);
898 break;
899
900 case TGSI_OPCODE_SLT:
901 emit_simple_arith(p, inst, A0_SLT, 2, fs);
902 break;
903
904 case TGSI_OPCODE_SGT:
905 /* like SLT, but swap reg0, reg1 */
906 emit_simple_arith_swap2(p, inst, A0_SLT, 2, fs);
907 break;
908
909 case TGSI_OPCODE_SNE:
910 /* if we're < or > then we're != */
911 src0 = src_vector(p, &inst->Src[0], fs);
912 src1 = src_vector(p, &inst->Src[1], fs);
913 tmp = i915_get_utemp(p);
914
915 i915_emit_arith(p,
916 A0_SLT,
917 tmp,
918 A0_DEST_CHANNEL_ALL, 0,
919 src0,
920 src1, 0);
921
922 i915_emit_arith(p,
923 A0_SLT,
924 get_result_vector(p, &inst->Dst[0]),
925 A0_DEST_CHANNEL_ALL, 0,
926 src1,
927 src0, 0);
928
929 i915_emit_arith(p,
930 A0_ADD,
931 get_result_vector(p, &inst->Dst[0]),
932 A0_DEST_CHANNEL_ALL, 0,
933 get_result_vector(p, &inst->Dst[0]),
934 tmp, 0);
935 break;
936
937 case TGSI_OPCODE_SSG:
938 /* compute (src>0) - (src<0) */
939 src0 = src_vector(p, &inst->Src[0], fs);
940 tmp = i915_get_utemp(p);
941
942 i915_emit_arith(p,
943 A0_SLT,
944 tmp,
945 A0_DEST_CHANNEL_ALL, 0,
946 src0,
947 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0);
948
949 i915_emit_arith(p,
950 A0_SLT,
951 get_result_vector(p, &inst->Dst[0]),
952 A0_DEST_CHANNEL_ALL, 0,
953 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
954 src0, 0);
955
956 i915_emit_arith(p,
957 A0_ADD,
958 get_result_vector(p, &inst->Dst[0]),
959 A0_DEST_CHANNEL_ALL, 0,
960 get_result_vector(p, &inst->Dst[0]),
961 negate(tmp, 1, 1, 1, 1), 0);
962 break;
963
964 case TGSI_OPCODE_SUB:
965 src0 = src_vector(p, &inst->Src[0], fs);
966 src1 = src_vector(p, &inst->Src[1], fs);
967
968 i915_emit_arith(p,
969 A0_ADD,
970 get_result_vector(p, &inst->Dst[0]),
971 get_result_flags(inst), 0,
972 src0, negate(src1, 1, 1, 1, 1), 0);
973 break;
974
975 case TGSI_OPCODE_TEX:
976 emit_tex(p, inst, T0_TEXLD, fs);
977 break;
978
979 case TGSI_OPCODE_TRUNC:
980 emit_simple_arith(p, inst, A0_TRC, 1, fs);
981 break;
982
983 case TGSI_OPCODE_TXB:
984 emit_tex(p, inst, T0_TEXLDB, fs);
985 break;
986
987 case TGSI_OPCODE_TXP:
988 emit_tex(p, inst, T0_TEXLDP, fs);
989 break;
990
991 case TGSI_OPCODE_XPD:
992 /* Cross product:
993 * result.x = src0.y * src1.z - src0.z * src1.y;
994 * result.y = src0.z * src1.x - src0.x * src1.z;
995 * result.z = src0.x * src1.y - src0.y * src1.x;
996 * result.w = undef;
997 */
998 src0 = src_vector(p, &inst->Src[0], fs);
999 src1 = src_vector(p, &inst->Src[1], fs);
1000 tmp = i915_get_utemp(p);
1001
1002 i915_emit_arith(p,
1003 A0_MUL,
1004 tmp, A0_DEST_CHANNEL_ALL, 0,
1005 swizzle(src0, Z, X, Y, ONE),
1006 swizzle(src1, Y, Z, X, ONE), 0);
1007
1008 i915_emit_arith(p,
1009 A0_MAD,
1010 get_result_vector(p, &inst->Dst[0]),
1011 get_result_flags(inst), 0,
1012 swizzle(src0, Y, Z, X, ONE),
1013 swizzle(src1, Z, X, Y, ONE),
1014 negate(tmp, 1, 1, 1, 0));
1015 break;
1016
1017 default:
1018 i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
1019 p->error = 1;
1020 return;
1021 }
1022
1023 i915_release_utemps(p);
1024 }
1025
1026
1027 /**
1028 * Translate TGSI fragment shader into i915 hardware instructions.
1029 * \param p the translation state
1030 * \param tokens the TGSI token array
1031 */
1032 static void
1033 i915_translate_instructions(struct i915_fp_compile *p,
1034 const struct tgsi_token *tokens,
1035 struct i915_fragment_shader *fs)
1036 {
1037 struct i915_fragment_shader *ifs = p->shader;
1038 struct tgsi_parse_context parse;
1039
1040 tgsi_parse_init( &parse, tokens );
1041
1042 while( !tgsi_parse_end_of_tokens( &parse ) ) {
1043
1044 tgsi_parse_token( &parse );
1045
1046 switch( parse.FullToken.Token.Type ) {
1047 case TGSI_TOKEN_TYPE_PROPERTY:
1048 /*
1049 * We only support one cbuf, but we still need to ignore the property
1050 * correctly so we don't hit the assert at the end of the switch case.
1051 */
1052 assert(parse.FullToken.FullProperty.Property.PropertyName ==
1053 TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS);
1054 break;
1055 case TGSI_TOKEN_TYPE_DECLARATION:
1056 if (parse.FullToken.FullDeclaration.Declaration.File
1057 == TGSI_FILE_CONSTANT) {
1058 uint i;
1059 for (i = parse.FullToken.FullDeclaration.Range.First;
1060 i <= parse.FullToken.FullDeclaration.Range.Last;
1061 i++) {
1062 assert(ifs->constant_flags[i] == 0x0);
1063 ifs->constant_flags[i] = I915_CONSTFLAG_USER;
1064 ifs->num_constants = MAX2(ifs->num_constants, i + 1);
1065 }
1066 }
1067 else if (parse.FullToken.FullDeclaration.Declaration.File
1068 == TGSI_FILE_TEMPORARY) {
1069 uint i;
1070 for (i = parse.FullToken.FullDeclaration.Range.First;
1071 i <= parse.FullToken.FullDeclaration.Range.Last;
1072 i++) {
1073 assert(i < I915_MAX_TEMPORARY);
1074 /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */
1075 p->temp_flag |= (1 << i); /* mark temp as used */
1076 }
1077 }
1078 break;
1079
1080 case TGSI_TOKEN_TYPE_IMMEDIATE:
1081 {
1082 const struct tgsi_full_immediate *imm
1083 = &parse.FullToken.FullImmediate;
1084 const uint pos = p->num_immediates++;
1085 uint j;
1086 assert( imm->Immediate.NrTokens <= 4 + 1 );
1087 for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
1088 p->immediates[pos][j] = imm->u[j].Float;
1089 }
1090 }
1091 break;
1092
1093 case TGSI_TOKEN_TYPE_INSTRUCTION:
1094 if (p->first_instruction) {
1095 /* resolve location of immediates */
1096 uint i, j;
1097 for (i = 0; i < p->num_immediates; i++) {
1098 /* find constant slot for this immediate */
1099 for (j = 0; j < I915_MAX_CONSTANT; j++) {
1100 if (ifs->constant_flags[j] == 0x0) {
1101 memcpy(ifs->constants[j],
1102 p->immediates[i],
1103 4 * sizeof(float));
1104 /*printf("immediate %d maps to const %d\n", i, j);*/
1105 ifs->constant_flags[j] = 0xf; /* all four comps used */
1106 p->immediates_map[i] = j;
1107 ifs->num_constants = MAX2(ifs->num_constants, j + 1);
1108 break;
1109 }
1110 }
1111 }
1112
1113 p->first_instruction = FALSE;
1114 }
1115
1116 i915_translate_instruction(p, &parse.FullToken.FullInstruction, fs);
1117 break;
1118
1119 default:
1120 assert( 0 );
1121 }
1122
1123 } /* while */
1124
1125 tgsi_parse_free (&parse);
1126 }
1127
1128
1129 static struct i915_fp_compile *
1130 i915_init_compile(struct i915_context *i915,
1131 struct i915_fragment_shader *ifs)
1132 {
1133 struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
1134 int i;
1135
1136 p->shader = ifs;
1137
1138 /* Put new constants at end of const buffer, growing downward.
1139 * The problem is we don't know how many user-defined constants might
1140 * be specified with pipe->set_constant_buffer().
1141 * Should pre-scan the user's program to determine the highest-numbered
1142 * constant referenced.
1143 */
1144 ifs->num_constants = 0;
1145 memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags));
1146
1147 for (i = 0; i < I915_TEX_UNITS; i++)
1148 ifs->generic_mapping[i] = -1;
1149
1150 p->first_instruction = TRUE;
1151
1152 p->nr_tex_indirect = 1; /* correct? */
1153 p->nr_tex_insn = 0;
1154 p->nr_alu_insn = 0;
1155 p->nr_decl_insn = 0;
1156
1157 p->csr = p->program;
1158 p->decl = p->declarations;
1159 p->decl_s = 0;
1160 p->decl_t = 0;
1161 p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
1162 p->utemp_flag = ~0x7;
1163
1164 p->wpos_tex = -1;
1165
1166 /* initialize the first program word */
1167 *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
1168
1169 return p;
1170 }
1171
1172
1173 /* Copy compile results to the fragment program struct and destroy the
1174 * compilation context.
1175 */
1176 static void
1177 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
1178 {
1179 struct i915_fragment_shader *ifs = p->shader;
1180 unsigned long program_size = (unsigned long) (p->csr - p->program);
1181 unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
1182
1183 if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
1184 i915_program_error(p, "Exceeded max nr indirect texture lookups");
1185
1186 if (p->nr_tex_insn > I915_MAX_TEX_INSN)
1187 i915_program_error(p, "Exceeded max TEX instructions");
1188
1189 if (p->nr_alu_insn > I915_MAX_ALU_INSN)
1190 i915_program_error(p, "Exceeded max ALU instructions");
1191
1192 if (p->nr_decl_insn > I915_MAX_DECL_INSN)
1193 i915_program_error(p, "Exceeded max DECL instructions");
1194
1195 if (p->error) {
1196 p->NumNativeInstructions = 0;
1197 p->NumNativeAluInstructions = 0;
1198 p->NumNativeTexInstructions = 0;
1199 p->NumNativeTexIndirections = 0;
1200
1201 i915_use_passthrough_shader(ifs);
1202 }
1203 else {
1204 p->NumNativeInstructions
1205 = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
1206 p->NumNativeAluInstructions = p->nr_alu_insn;
1207 p->NumNativeTexInstructions = p->nr_tex_insn;
1208 p->NumNativeTexIndirections = p->nr_tex_indirect;
1209
1210 /* patch in the program length */
1211 p->declarations[0] |= program_size + decl_size - 2;
1212
1213 /* Copy compilation results to fragment program struct:
1214 */
1215 assert(!ifs->program);
1216 ifs->program
1217 = (uint *) MALLOC((program_size + decl_size) * sizeof(uint));
1218 if (ifs->program) {
1219 ifs->program_len = program_size + decl_size;
1220
1221 memcpy(ifs->program,
1222 p->declarations,
1223 decl_size * sizeof(uint));
1224
1225 memcpy(ifs->program + decl_size,
1226 p->program,
1227 program_size * sizeof(uint));
1228 }
1229 }
1230
1231 /* Release the compilation struct:
1232 */
1233 FREE(p);
1234 }
1235
1236
1237 /**
1238 * Find an unused texture coordinate slot to use for fragment WPOS.
1239 * Update p->fp->wpos_tex with the result (-1 if no used texcoord slot is found).
1240 */
1241 static void
1242 i915_find_wpos_space(struct i915_fp_compile *p)
1243 {
1244 #if 0
1245 const uint inputs
1246 = p->shader->inputs_read | (1 << TGSI_ATTRIB_POS); /*XXX hack*/
1247 uint i;
1248
1249 p->wpos_tex = -1;
1250
1251 if (inputs & (1 << TGSI_ATTRIB_POS)) {
1252 for (i = 0; i < I915_TEX_UNITS; i++) {
1253 if ((inputs & (1 << (TGSI_ATTRIB_TEX0 + i))) == 0) {
1254 p->wpos_tex = i;
1255 return;
1256 }
1257 }
1258
1259 i915_program_error(p, "No free texcoord for wpos value");
1260 }
1261 #else
1262 if (p->shader->info.input_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1263 /* frag shader using the fragment position input */
1264 #if 0
1265 assert(0);
1266 #endif
1267 }
1268 #endif
1269 }
1270
1271
1272
1273
1274 /**
1275 * Rather than trying to intercept and jiggle depth writes during
1276 * emit, just move the value into its correct position at the end of
1277 * the program:
1278 */
1279 static void
1280 i915_fixup_depth_write(struct i915_fp_compile *p)
1281 {
1282 /* XXX assuming pos/depth is always in output[0] */
1283 if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1284 const uint depth = UREG(REG_TYPE_OD, 0);
1285
1286 i915_emit_arith(p,
1287 A0_MOV, /* opcode */
1288 depth, /* dest reg */
1289 A0_DEST_CHANNEL_W, /* write mask */
1290 0, /* saturate? */
1291 swizzle(depth, X, Y, Z, Z), /* src0 */
1292 0, 0 /* src1, src2 */);
1293 }
1294 }
1295
1296
1297 void
1298 i915_translate_fragment_program( struct i915_context *i915,
1299 struct i915_fragment_shader *fs)
1300 {
1301 struct i915_fp_compile *p;
1302 const struct tgsi_token *tokens = fs->state.tokens;
1303
1304 #if 0
1305 tgsi_dump(tokens, 0);
1306 #endif
1307
1308 /* hw doesn't seem to like empty frag programs, even when the depth write
1309 * fixup gets emitted below - may that one is fishy, too? */
1310 if (fs->info.num_instructions == 1) {
1311 i915_use_passthrough_shader(fs);
1312
1313 return;
1314 }
1315
1316 p = i915_init_compile(i915, fs);
1317 i915_find_wpos_space(p);
1318
1319 i915_translate_instructions(p, tokens, fs);
1320 i915_fixup_depth_write(p);
1321
1322 i915_fini_compile(i915, p);
1323 }