projects / mesa.git / blob
? search:


f39b020537701bf5e83013ef2871d3e57ff2549c
[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 {
189 /* for fragcoord */
190 int real_tex_unit = get_mapping(fs, I915_SEMANTIC_POS);
191 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_ALL);
192 break;
193 }
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 case TGSI_SEMANTIC_FACE:
216 {
217 /* for back/front faces */
218 /* XXX also emit something from 0,1 to -1,1 */
219 int real_tex_unit = get_mapping(fs, I915_SEMANTIC_FACE);
220 printf("semantic face fpc at %d\n",real_tex_unit);
221 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_X);
222 break;
223 }
224 default:
225 i915_program_error(p, "Bad source->Index");
226 return 0;
227 }
228 break;
229
230 case TGSI_FILE_IMMEDIATE:
231 assert(index < p->num_immediates);
232 index = p->immediates_map[index];
233 /* fall-through */
234 case TGSI_FILE_CONSTANT:
235 src = UREG(REG_TYPE_CONST, index);
236 break;
237
238 default:
239 i915_program_error(p, "Bad source->File");
240 return 0;
241 }
242
243 src = swizzle(src,
244 source->Register.SwizzleX,
245 source->Register.SwizzleY,
246 source->Register.SwizzleZ,
247 source->Register.SwizzleW);
248
249
250 /* There's both negate-all-components and per-component negation.
251 * Try to handle both here.
252 */
253 {
254 int n = source->Register.Negate;
255 src = negate(src, n, n, n, n);
256 }
257
258 /* no abs() */
259 #if 0
260 /* XXX assertions disabled to allow arbfplight.c to run */
261 /* XXX enable these assertions, or fix things */
262 assert(!source->Register.Absolute);
263 #endif
264 return src;
265 }
266
267
268 /**
269 * Construct a ureg for a destination register.
270 */
271 static uint
272 get_result_vector(struct i915_fp_compile *p,
273 const struct tgsi_full_dst_register *dest)
274 {
275 switch (dest->Register.File) {
276 case TGSI_FILE_OUTPUT:
277 {
278 uint sem_name = p->shader->info.output_semantic_name[dest->Register.Index];
279 switch (sem_name) {
280 case TGSI_SEMANTIC_POSITION:
281 return UREG(REG_TYPE_OD, 0);
282 case TGSI_SEMANTIC_COLOR:
283 return UREG(REG_TYPE_OC, 0);
284 default:
285 i915_program_error(p, "Bad inst->DstReg.Index/semantics");
286 return 0;
287 }
288 }
289 case TGSI_FILE_TEMPORARY:
290 return UREG(REG_TYPE_R, dest->Register.Index);
291 default:
292 i915_program_error(p, "Bad inst->DstReg.File");
293 return 0;
294 }
295 }
296
297
298 /**
299 * Compute flags for saturation and writemask.
300 */
301 static uint
302 get_result_flags(const struct tgsi_full_instruction *inst)
303 {
304 const uint writeMask
305 = inst->Dst[0].Register.WriteMask;
306 uint flags = 0x0;
307
308 if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE)
309 flags |= A0_DEST_SATURATE;
310
311 if (writeMask & TGSI_WRITEMASK_X)
312 flags |= A0_DEST_CHANNEL_X;
313 if (writeMask & TGSI_WRITEMASK_Y)
314 flags |= A0_DEST_CHANNEL_Y;
315 if (writeMask & TGSI_WRITEMASK_Z)
316 flags |= A0_DEST_CHANNEL_Z;
317 if (writeMask & TGSI_WRITEMASK_W)
318 flags |= A0_DEST_CHANNEL_W;
319
320 return flags;
321 }
322
323
324 /**
325 * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
326 */
327 static uint
328 translate_tex_src_target(struct i915_fp_compile *p, uint tex)
329 {
330 switch (tex) {
331 case TGSI_TEXTURE_SHADOW1D:
332 /* fall-through */
333 case TGSI_TEXTURE_1D:
334 return D0_SAMPLE_TYPE_2D;
335
336 case TGSI_TEXTURE_SHADOW2D:
337 /* fall-through */
338 case TGSI_TEXTURE_2D:
339 return D0_SAMPLE_TYPE_2D;
340
341 case TGSI_TEXTURE_SHADOWRECT:
342 /* fall-through */
343 case TGSI_TEXTURE_RECT:
344 return D0_SAMPLE_TYPE_2D;
345
346 case TGSI_TEXTURE_3D:
347 return D0_SAMPLE_TYPE_VOLUME;
348
349 case TGSI_TEXTURE_CUBE:
350 return D0_SAMPLE_TYPE_CUBE;
351
352 default:
353 i915_program_error(p, "TexSrc type");
354 return 0;
355 }
356 }
357
358
359 /**
360 * Generate texel lookup instruction.
361 */
362 static void
363 emit_tex(struct i915_fp_compile *p,
364 const struct tgsi_full_instruction *inst,
365 uint opcode,
366 struct i915_fragment_shader* fs)
367 {
368 uint texture = inst->Texture.Texture;
369 uint unit = inst->Src[1].Register.Index;
370 uint tex = translate_tex_src_target( p, texture );
371 uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex);
372 uint coord = src_vector( p, &inst->Src[0], fs);
373
374 i915_emit_texld( p,
375 get_result_vector( p, &inst->Dst[0] ),
376 get_result_flags( inst ),
377 sampler,
378 coord,
379 opcode);
380 }
381
382
383 /**
384 * Generate a simple arithmetic instruction
385 * \param opcode the i915 opcode
386 * \param numArgs the number of input/src arguments
387 */
388 static void
389 emit_simple_arith(struct i915_fp_compile *p,
390 const struct tgsi_full_instruction *inst,
391 uint opcode, uint numArgs,
392 struct i915_fragment_shader* fs)
393 {
394 uint arg1, arg2, arg3;
395
396 assert(numArgs <= 3);
397
398 arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->Src[0], fs );
399 arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->Src[1], fs );
400 arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->Src[2], fs );
401
402 i915_emit_arith( p,
403 opcode,
404 get_result_vector( p, &inst->Dst[0]),
405 get_result_flags( inst ), 0,
406 arg1,
407 arg2,
408 arg3 );
409 }
410
411
412 /** As above, but swap the first two src regs */
413 static void
414 emit_simple_arith_swap2(struct i915_fp_compile *p,
415 const struct tgsi_full_instruction *inst,
416 uint opcode, uint numArgs,
417 struct i915_fragment_shader* fs)
418 {
419 struct tgsi_full_instruction inst2;
420
421 assert(numArgs == 2);
422
423 /* transpose first two registers */
424 inst2 = *inst;
425 inst2.Src[0] = inst->Src[1];
426 inst2.Src[1] = inst->Src[0];
427
428 emit_simple_arith(p, &inst2, opcode, numArgs, fs);
429 }
430
431
432 #ifndef M_PI
433 #define M_PI 3.14159265358979323846
434 #endif
435
436 /*
437 * Translate TGSI instruction to i915 instruction.
438 *
439 * Possible concerns:
440 *
441 * SIN, COS -- could use another taylor step?
442 * LIT -- results seem a little different to sw mesa
443 * LOG -- different to mesa on negative numbers, but this is conformant.
444 */
445 static void
446 i915_translate_instruction(struct i915_fp_compile *p,
447 const struct tgsi_full_instruction *inst,
448 struct i915_fragment_shader *fs)
449 {
450 uint writemask;
451 uint src0, src1, src2, flags;
452 uint tmp = 0;
453
454 switch (inst->Instruction.Opcode) {
455 case TGSI_OPCODE_ABS:
456 src0 = src_vector(p, &inst->Src[0], fs);
457 i915_emit_arith(p,
458 A0_MAX,
459 get_result_vector(p, &inst->Dst[0]),
460 get_result_flags(inst), 0,
461 src0, negate(src0, 1, 1, 1, 1), 0);
462 break;
463
464 case TGSI_OPCODE_ADD:
465 emit_simple_arith(p, inst, A0_ADD, 2, fs);
466 break;
467
468 case TGSI_OPCODE_CMP:
469 src0 = src_vector(p, &inst->Src[0], fs);
470 src1 = src_vector(p, &inst->Src[1], fs);
471 src2 = src_vector(p, &inst->Src[2], fs);
472 i915_emit_arith(p, A0_CMP,
473 get_result_vector(p, &inst->Dst[0]),
474 get_result_flags(inst),
475 0, src0, src2, src1); /* NOTE: order of src2, src1 */
476 break;
477
478 case TGSI_OPCODE_COS:
479 src0 = src_vector(p, &inst->Src[0], fs);
480 tmp = i915_get_utemp(p);
481
482 i915_emit_arith(p,
483 A0_MUL,
484 tmp, A0_DEST_CHANNEL_X, 0,
485 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
486
487 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
488
489 /* By choosing different taylor constants, could get rid of this mul:
490 */
491 i915_emit_arith(p,
492 A0_MUL,
493 tmp, A0_DEST_CHANNEL_X, 0,
494 tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0);
495
496 /*
497 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
498 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
499 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
500 * result = DP4 t0, cos_constants
501 */
502 i915_emit_arith(p,
503 A0_MUL,
504 tmp, A0_DEST_CHANNEL_XY, 0,
505 swizzle(tmp, X, X, ONE, ONE),
506 swizzle(tmp, X, ONE, ONE, ONE), 0);
507
508 i915_emit_arith(p,
509 A0_MUL,
510 tmp, A0_DEST_CHANNEL_XYZ, 0,
511 swizzle(tmp, X, Y, X, ONE),
512 swizzle(tmp, X, X, ONE, ONE), 0);
513
514 i915_emit_arith(p,
515 A0_MUL,
516 tmp, A0_DEST_CHANNEL_XYZ, 0,
517 swizzle(tmp, X, X, Z, ONE),
518 swizzle(tmp, Z, ONE, ONE, ONE), 0);
519
520 i915_emit_arith(p,
521 A0_DP4,
522 get_result_vector(p, &inst->Dst[0]),
523 get_result_flags(inst), 0,
524 swizzle(tmp, ONE, Z, Y, X),
525 i915_emit_const4fv(p, cos_constants), 0);
526 break;
527
528 case TGSI_OPCODE_DDX:
529 case TGSI_OPCODE_DDY:
530 /* XXX We just output 0 here */
531 debug_printf("Punting DDX/DDX\n");
532 src0 = get_result_vector(p, &inst->Dst[0]);
533 i915_emit_arith(p,
534 A0_MOV,
535 get_result_vector(p, &inst->Dst[0]),
536 get_result_flags(inst), 0,
537 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0, 0);
538 break;
539
540 case TGSI_OPCODE_DP2:
541 src0 = src_vector(p, &inst->Src[0], fs);
542 src1 = src_vector(p, &inst->Src[1], fs);
543
544 i915_emit_arith(p,
545 A0_DP3,
546 get_result_vector(p, &inst->Dst[0]),
547 get_result_flags(inst), 0,
548 swizzle(src0, X, Y, ZERO, ZERO), src1, 0);
549 break;
550
551 case TGSI_OPCODE_DP3:
552 emit_simple_arith(p, inst, A0_DP3, 2, fs);
553 break;
554
555 case TGSI_OPCODE_DP4:
556 emit_simple_arith(p, inst, A0_DP4, 2, fs);
557 break;
558
559 case TGSI_OPCODE_DPH:
560 src0 = src_vector(p, &inst->Src[0], fs);
561 src1 = src_vector(p, &inst->Src[1], fs);
562
563 i915_emit_arith(p,
564 A0_DP4,
565 get_result_vector(p, &inst->Dst[0]),
566 get_result_flags(inst), 0,
567 swizzle(src0, X, Y, Z, ONE), src1, 0);
568 break;
569
570 case TGSI_OPCODE_DST:
571 src0 = src_vector(p, &inst->Src[0], fs);
572 src1 = src_vector(p, &inst->Src[1], fs);
573
574 /* result[0] = 1 * 1;
575 * result[1] = a[1] * b[1];
576 * result[2] = a[2] * 1;
577 * result[3] = 1 * b[3];
578 */
579 i915_emit_arith(p,
580 A0_MUL,
581 get_result_vector(p, &inst->Dst[0]),
582 get_result_flags(inst), 0,
583 swizzle(src0, ONE, Y, Z, ONE),
584 swizzle(src1, ONE, Y, ONE, W), 0);
585 break;
586
587 case TGSI_OPCODE_END:
588 /* no-op */
589 break;
590
591 case TGSI_OPCODE_EX2:
592 src0 = src_vector(p, &inst->Src[0], fs);
593
594 i915_emit_arith(p,
595 A0_EXP,
596 get_result_vector(p, &inst->Dst[0]),
597 get_result_flags(inst), 0,
598 swizzle(src0, X, X, X, X), 0, 0);
599 break;
600
601 case TGSI_OPCODE_FLR:
602 emit_simple_arith(p, inst, A0_FLR, 1, fs);
603 break;
604
605 case TGSI_OPCODE_FRC:
606 emit_simple_arith(p, inst, A0_FRC, 1, fs);
607 break;
608
609 case TGSI_OPCODE_KIL:
610 /* kill if src[0].x < 0 || src[0].y < 0 ... */
611 src0 = src_vector(p, &inst->Src[0], fs);
612 tmp = i915_get_utemp(p);
613
614 i915_emit_texld(p,
615 tmp, /* dest reg: a dummy reg */
616 A0_DEST_CHANNEL_ALL, /* dest writemask */
617 0, /* sampler */
618 src0, /* coord*/
619 T0_TEXKILL); /* opcode */
620 break;
621
622 case TGSI_OPCODE_KILP:
623 assert(0); /* not tested yet */
624 break;
625
626 case TGSI_OPCODE_LG2:
627 src0 = src_vector(p, &inst->Src[0], fs);
628
629 i915_emit_arith(p,
630 A0_LOG,
631 get_result_vector(p, &inst->Dst[0]),
632 get_result_flags(inst), 0,
633 swizzle(src0, X, X, X, X), 0, 0);
634 break;
635
636 case TGSI_OPCODE_LIT:
637 src0 = src_vector(p, &inst->Src[0], fs);
638 tmp = i915_get_utemp(p);
639
640 /* tmp = max( a.xyzw, a.00zw )
641 * XXX: Clamp tmp.w to -128..128
642 * tmp.y = log(tmp.y)
643 * tmp.y = tmp.w * tmp.y
644 * tmp.y = exp(tmp.y)
645 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
646 */
647 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
648 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
649
650 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
651 swizzle(tmp, Y, Y, Y, Y), 0, 0);
652
653 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
654 swizzle(tmp, ZERO, Y, ZERO, ZERO),
655 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
656
657 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
658 swizzle(tmp, Y, Y, Y, Y), 0, 0);
659
660 i915_emit_arith(p, A0_CMP,
661 get_result_vector(p, &inst->Dst[0]),
662 get_result_flags(inst), 0,
663 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
664 swizzle(tmp, ONE, X, ZERO, ONE),
665 swizzle(tmp, ONE, X, Y, ONE));
666
667 break;
668
669 case TGSI_OPCODE_LRP:
670 src0 = src_vector(p, &inst->Src[0], fs);
671 src1 = src_vector(p, &inst->Src[1], fs);
672 src2 = src_vector(p, &inst->Src[2], fs);
673 flags = get_result_flags(inst);
674 tmp = i915_get_utemp(p);
675
676 /* b*a + c*(1-a)
677 *
678 * b*a + c - ca
679 *
680 * tmp = b*a + c,
681 * result = (-c)*a + tmp
682 */
683 i915_emit_arith(p, A0_MAD, tmp,
684 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
685
686 i915_emit_arith(p, A0_MAD,
687 get_result_vector(p, &inst->Dst[0]),
688 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
689 break;
690
691 case TGSI_OPCODE_MAD:
692 emit_simple_arith(p, inst, A0_MAD, 3, fs);
693 break;
694
695 case TGSI_OPCODE_MAX:
696 emit_simple_arith(p, inst, A0_MAX, 2, fs);
697 break;
698
699 case TGSI_OPCODE_MIN:
700 src0 = src_vector(p, &inst->Src[0], fs);
701 src1 = src_vector(p, &inst->Src[1], fs);
702 tmp = i915_get_utemp(p);
703 flags = get_result_flags(inst);
704
705 i915_emit_arith(p,
706 A0_MAX,
707 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
708 negate(src0, 1, 1, 1, 1),
709 negate(src1, 1, 1, 1, 1), 0);
710
711 i915_emit_arith(p,
712 A0_MOV,
713 get_result_vector(p, &inst->Dst[0]),
714 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
715 break;
716
717 case TGSI_OPCODE_MOV:
718 emit_simple_arith(p, inst, A0_MOV, 1, fs);
719 break;
720
721 case TGSI_OPCODE_MUL:
722 emit_simple_arith(p, inst, A0_MUL, 2, fs);
723 break;
724
725 case TGSI_OPCODE_POW:
726 src0 = src_vector(p, &inst->Src[0], fs);
727 src1 = src_vector(p, &inst->Src[1], fs);
728 tmp = i915_get_utemp(p);
729 flags = get_result_flags(inst);
730
731 /* XXX: masking on intermediate values, here and elsewhere.
732 */
733 i915_emit_arith(p,
734 A0_LOG,
735 tmp, A0_DEST_CHANNEL_X, 0,
736 swizzle(src0, X, X, X, X), 0, 0);
737
738 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
739
740 i915_emit_arith(p,
741 A0_EXP,
742 get_result_vector(p, &inst->Dst[0]),
743 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
744 break;
745
746 case TGSI_OPCODE_RET:
747 /* XXX: no-op? */
748 break;
749
750 case TGSI_OPCODE_RCP:
751 src0 = src_vector(p, &inst->Src[0], fs);
752
753 i915_emit_arith(p,
754 A0_RCP,
755 get_result_vector(p, &inst->Dst[0]),
756 get_result_flags(inst), 0,
757 swizzle(src0, X, X, X, X), 0, 0);
758 break;
759
760 case TGSI_OPCODE_RSQ:
761 src0 = src_vector(p, &inst->Src[0], fs);
762
763 i915_emit_arith(p,
764 A0_RSQ,
765 get_result_vector(p, &inst->Dst[0]),
766 get_result_flags(inst), 0,
767 swizzle(src0, X, X, X, X), 0, 0);
768 break;
769
770 case TGSI_OPCODE_SCS:
771 src0 = src_vector(p, &inst->Src[0], fs);
772 tmp = i915_get_utemp(p);
773
774 /*
775 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
776 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
777 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
778 * scs.x = DP4 t1, sin_constants
779 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
780 * scs.y = DP4 t1, cos_constants
781 */
782 i915_emit_arith(p,
783 A0_MUL,
784 tmp, A0_DEST_CHANNEL_XY, 0,
785 swizzle(src0, X, X, ONE, ONE),
786 swizzle(src0, X, ONE, ONE, ONE), 0);
787
788 i915_emit_arith(p,
789 A0_MUL,
790 tmp, A0_DEST_CHANNEL_ALL, 0,
791 swizzle(tmp, X, Y, X, Y),
792 swizzle(tmp, X, X, ONE, ONE), 0);
793
794 writemask = inst->Dst[0].Register.WriteMask;
795
796 if (writemask & TGSI_WRITEMASK_Y) {
797 uint tmp1;
798
799 if (writemask & TGSI_WRITEMASK_X)
800 tmp1 = i915_get_utemp(p);
801 else
802 tmp1 = tmp;
803
804 i915_emit_arith(p,
805 A0_MUL,
806 tmp1, A0_DEST_CHANNEL_ALL, 0,
807 swizzle(tmp, X, Y, Y, W),
808 swizzle(tmp, X, Z, ONE, ONE), 0);
809
810 i915_emit_arith(p,
811 A0_DP4,
812 get_result_vector(p, &inst->Dst[0]),
813 A0_DEST_CHANNEL_Y, 0,
814 swizzle(tmp1, W, Z, Y, X),
815 i915_emit_const4fv(p, sin_constants), 0);
816 }
817
818 if (writemask & TGSI_WRITEMASK_X) {
819 i915_emit_arith(p,
820 A0_MUL,
821 tmp, A0_DEST_CHANNEL_XYZ, 0,
822 swizzle(tmp, X, X, Z, ONE),
823 swizzle(tmp, Z, ONE, ONE, ONE), 0);
824
825 i915_emit_arith(p,
826 A0_DP4,
827 get_result_vector(p, &inst->Dst[0]),
828 A0_DEST_CHANNEL_X, 0,
829 swizzle(tmp, ONE, Z, Y, X),
830 i915_emit_const4fv(p, cos_constants), 0);
831 }
832 break;
833
834 case TGSI_OPCODE_SEQ:
835 /* if we're both >= and <= then we're == */
836 src0 = src_vector(p, &inst->Src[0], fs);
837 src1 = src_vector(p, &inst->Src[1], fs);
838 tmp = i915_get_utemp(p);
839
840 i915_emit_arith(p,
841 A0_SGE,
842 tmp, A0_DEST_CHANNEL_ALL, 0,
843 src0,
844 src1, 0);
845
846 i915_emit_arith(p,
847 A0_SGE,
848 get_result_vector(p, &inst->Dst[0]),
849 A0_DEST_CHANNEL_ALL, 0,
850 src1,
851 src0, 0);
852
853 i915_emit_arith(p,
854 A0_MUL,
855 get_result_vector(p, &inst->Dst[0]),
856 A0_DEST_CHANNEL_ALL, 0,
857 get_result_vector(p, &inst->Dst[0]),
858 tmp, 0);
859
860 break;
861
862 case TGSI_OPCODE_SGE:
863 emit_simple_arith(p, inst, A0_SGE, 2, fs);
864 break;
865
866 case TGSI_OPCODE_SIN:
867 src0 = src_vector(p, &inst->Src[0], fs);
868 tmp = i915_get_utemp(p);
869
870 i915_emit_arith(p,
871 A0_MUL,
872 tmp, A0_DEST_CHANNEL_X, 0,
873 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
874
875 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
876
877 /* By choosing different taylor constants, could get rid of this mul:
878 */
879 i915_emit_arith(p,
880 A0_MUL,
881 tmp, A0_DEST_CHANNEL_X, 0,
882 tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0);
883
884 /*
885 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
886 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
887 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
888 * result = DP4 t1.wzyx, sin_constants
889 */
890 i915_emit_arith(p,
891 A0_MUL,
892 tmp, A0_DEST_CHANNEL_XY, 0,
893 swizzle(tmp, X, X, ONE, ONE),
894 swizzle(tmp, X, ONE, ONE, ONE), 0);
895
896 i915_emit_arith(p,
897 A0_MUL,
898 tmp, A0_DEST_CHANNEL_ALL, 0,
899 swizzle(tmp, X, Y, X, Y),
900 swizzle(tmp, X, X, ONE, ONE), 0);
901
902 i915_emit_arith(p,
903 A0_MUL,
904 tmp, A0_DEST_CHANNEL_ALL, 0,
905 swizzle(tmp, X, Y, Y, W),
906 swizzle(tmp, X, Z, ONE, ONE), 0);
907
908 i915_emit_arith(p,
909 A0_DP4,
910 get_result_vector(p, &inst->Dst[0]),
911 get_result_flags(inst), 0,
912 swizzle(tmp, W, Z, Y, X),
913 i915_emit_const4fv(p, sin_constants), 0);
914 break;
915
916 case TGSI_OPCODE_SLE:
917 /* like SGE, but swap reg0, reg1 */
918 emit_simple_arith_swap2(p, inst, A0_SGE, 2, fs);
919 break;
920
921 case TGSI_OPCODE_SLT:
922 emit_simple_arith(p, inst, A0_SLT, 2, fs);
923 break;
924
925 case TGSI_OPCODE_SGT:
926 /* like SLT, but swap reg0, reg1 */
927 emit_simple_arith_swap2(p, inst, A0_SLT, 2, fs);
928 break;
929
930 case TGSI_OPCODE_SNE:
931 /* if we're < or > then we're != */
932 src0 = src_vector(p, &inst->Src[0], fs);
933 src1 = src_vector(p, &inst->Src[1], fs);
934 tmp = i915_get_utemp(p);
935
936 i915_emit_arith(p,
937 A0_SLT,
938 tmp,
939 A0_DEST_CHANNEL_ALL, 0,
940 src0,
941 src1, 0);
942
943 i915_emit_arith(p,
944 A0_SLT,
945 get_result_vector(p, &inst->Dst[0]),
946 A0_DEST_CHANNEL_ALL, 0,
947 src1,
948 src0, 0);
949
950 i915_emit_arith(p,
951 A0_ADD,
952 get_result_vector(p, &inst->Dst[0]),
953 A0_DEST_CHANNEL_ALL, 0,
954 get_result_vector(p, &inst->Dst[0]),
955 tmp, 0);
956 break;
957
958 case TGSI_OPCODE_SSG:
959 /* compute (src>0) - (src<0) */
960 src0 = src_vector(p, &inst->Src[0], fs);
961 tmp = i915_get_utemp(p);
962
963 i915_emit_arith(p,
964 A0_SLT,
965 tmp,
966 A0_DEST_CHANNEL_ALL, 0,
967 src0,
968 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0);
969
970 i915_emit_arith(p,
971 A0_SLT,
972 get_result_vector(p, &inst->Dst[0]),
973 A0_DEST_CHANNEL_ALL, 0,
974 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
975 src0, 0);
976
977 i915_emit_arith(p,
978 A0_ADD,
979 get_result_vector(p, &inst->Dst[0]),
980 A0_DEST_CHANNEL_ALL, 0,
981 get_result_vector(p, &inst->Dst[0]),
982 negate(tmp, 1, 1, 1, 1), 0);
983 break;
984
985 case TGSI_OPCODE_SUB:
986 src0 = src_vector(p, &inst->Src[0], fs);
987 src1 = src_vector(p, &inst->Src[1], fs);
988
989 i915_emit_arith(p,
990 A0_ADD,
991 get_result_vector(p, &inst->Dst[0]),
992 get_result_flags(inst), 0,
993 src0, negate(src1, 1, 1, 1, 1), 0);
994 break;
995
996 case TGSI_OPCODE_TEX:
997 emit_tex(p, inst, T0_TEXLD, fs);
998 break;
999
1000 case TGSI_OPCODE_TRUNC:
1001 emit_simple_arith(p, inst, A0_TRC, 1, fs);
1002 break;
1003
1004 case TGSI_OPCODE_TXB:
1005 emit_tex(p, inst, T0_TEXLDB, fs);
1006 break;
1007
1008 case TGSI_OPCODE_TXP:
1009 emit_tex(p, inst, T0_TEXLDP, fs);
1010 break;
1011
1012 case TGSI_OPCODE_XPD:
1013 /* Cross product:
1014 * result.x = src0.y * src1.z - src0.z * src1.y;
1015 * result.y = src0.z * src1.x - src0.x * src1.z;
1016 * result.z = src0.x * src1.y - src0.y * src1.x;
1017 * result.w = undef;
1018 */
1019 src0 = src_vector(p, &inst->Src[0], fs);
1020 src1 = src_vector(p, &inst->Src[1], fs);
1021 tmp = i915_get_utemp(p);
1022
1023 i915_emit_arith(p,
1024 A0_MUL,
1025 tmp, A0_DEST_CHANNEL_ALL, 0,
1026 swizzle(src0, Z, X, Y, ONE),
1027 swizzle(src1, Y, Z, X, ONE), 0);
1028
1029 i915_emit_arith(p,
1030 A0_MAD,
1031 get_result_vector(p, &inst->Dst[0]),
1032 get_result_flags(inst), 0,
1033 swizzle(src0, Y, Z, X, ONE),
1034 swizzle(src1, Z, X, Y, ONE),
1035 negate(tmp, 1, 1, 1, 0));
1036 break;
1037
1038 default:
1039 i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
1040 p->error = 1;
1041 return;
1042 }
1043
1044 i915_release_utemps(p);
1045 }
1046
1047
1048 /**
1049 * Translate TGSI fragment shader into i915 hardware instructions.
1050 * \param p the translation state
1051 * \param tokens the TGSI token array
1052 */
1053 static void
1054 i915_translate_instructions(struct i915_fp_compile *p,
1055 const struct tgsi_token *tokens,
1056 struct i915_fragment_shader *fs)
1057 {
1058 struct i915_fragment_shader *ifs = p->shader;
1059 struct tgsi_parse_context parse;
1060
1061 tgsi_parse_init( &parse, tokens );
1062
1063 while( !tgsi_parse_end_of_tokens( &parse ) ) {
1064
1065 tgsi_parse_token( &parse );
1066
1067 switch( parse.FullToken.Token.Type ) {
1068 case TGSI_TOKEN_TYPE_PROPERTY:
1069 /*
1070 * We only support one cbuf, but we still need to ignore the property
1071 * correctly so we don't hit the assert at the end of the switch case.
1072 */
1073 assert(parse.FullToken.FullProperty.Property.PropertyName ==
1074 TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS);
1075 break;
1076 case TGSI_TOKEN_TYPE_DECLARATION:
1077 if (parse.FullToken.FullDeclaration.Declaration.File
1078 == TGSI_FILE_CONSTANT) {
1079 uint i;
1080 for (i = parse.FullToken.FullDeclaration.Range.First;
1081 i <= parse.FullToken.FullDeclaration.Range.Last;
1082 i++) {
1083 assert(ifs->constant_flags[i] == 0x0);
1084 ifs->constant_flags[i] = I915_CONSTFLAG_USER;
1085 ifs->num_constants = MAX2(ifs->num_constants, i + 1);
1086 }
1087 }
1088 else if (parse.FullToken.FullDeclaration.Declaration.File
1089 == TGSI_FILE_TEMPORARY) {
1090 uint i;
1091 for (i = parse.FullToken.FullDeclaration.Range.First;
1092 i <= parse.FullToken.FullDeclaration.Range.Last;
1093 i++) {
1094 assert(i < I915_MAX_TEMPORARY);
1095 /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */
1096 p->temp_flag |= (1 << i); /* mark temp as used */
1097 }
1098 }
1099 break;
1100
1101 case TGSI_TOKEN_TYPE_IMMEDIATE:
1102 {
1103 const struct tgsi_full_immediate *imm
1104 = &parse.FullToken.FullImmediate;
1105 const uint pos = p->num_immediates++;
1106 uint j;
1107 assert( imm->Immediate.NrTokens <= 4 + 1 );
1108 for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
1109 p->immediates[pos][j] = imm->u[j].Float;
1110 }
1111 }
1112 break;
1113
1114 case TGSI_TOKEN_TYPE_INSTRUCTION:
1115 if (p->first_instruction) {
1116 /* resolve location of immediates */
1117 uint i, j;
1118 for (i = 0; i < p->num_immediates; i++) {
1119 /* find constant slot for this immediate */
1120 for (j = 0; j < I915_MAX_CONSTANT; j++) {
1121 if (ifs->constant_flags[j] == 0x0) {
1122 memcpy(ifs->constants[j],
1123 p->immediates[i],
1124 4 * sizeof(float));
1125 /*printf("immediate %d maps to const %d\n", i, j);*/
1126 ifs->constant_flags[j] = 0xf; /* all four comps used */
1127 p->immediates_map[i] = j;
1128 ifs->num_constants = MAX2(ifs->num_constants, j + 1);
1129 break;
1130 }
1131 }
1132 }
1133
1134 p->first_instruction = FALSE;
1135 }
1136
1137 i915_translate_instruction(p, &parse.FullToken.FullInstruction, fs);
1138 break;
1139
1140 default:
1141 assert( 0 );
1142 }
1143
1144 } /* while */
1145
1146 tgsi_parse_free (&parse);
1147 }
1148
1149
1150 static struct i915_fp_compile *
1151 i915_init_compile(struct i915_context *i915,
1152 struct i915_fragment_shader *ifs)
1153 {
1154 struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
1155 int i;
1156
1157 p->shader = ifs;
1158
1159 /* Put new constants at end of const buffer, growing downward.
1160 * The problem is we don't know how many user-defined constants might
1161 * be specified with pipe->set_constant_buffer().
1162 * Should pre-scan the user's program to determine the highest-numbered
1163 * constant referenced.
1164 */
1165 ifs->num_constants = 0;
1166 memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags));
1167
1168 for (i = 0; i < I915_TEX_UNITS; i++)
1169 ifs->generic_mapping[i] = -1;
1170
1171 p->first_instruction = TRUE;
1172
1173 p->nr_tex_indirect = 1; /* correct? */
1174 p->nr_tex_insn = 0;
1175 p->nr_alu_insn = 0;
1176 p->nr_decl_insn = 0;
1177
1178 p->csr = p->program;
1179 p->decl = p->declarations;
1180 p->decl_s = 0;
1181 p->decl_t = 0;
1182 p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
1183 p->utemp_flag = ~0x7;
1184
1185 /* initialize the first program word */
1186 *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
1187
1188 return p;
1189 }
1190
1191
1192 /* Copy compile results to the fragment program struct and destroy the
1193 * compilation context.
1194 */
1195 static void
1196 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
1197 {
1198 struct i915_fragment_shader *ifs = p->shader;
1199 unsigned long program_size = (unsigned long) (p->csr - p->program);
1200 unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
1201
1202 if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
1203 i915_program_error(p, "Exceeded max nr indirect texture lookups");
1204
1205 if (p->nr_tex_insn > I915_MAX_TEX_INSN)
1206 i915_program_error(p, "Exceeded max TEX instructions");
1207
1208 if (p->nr_alu_insn > I915_MAX_ALU_INSN)
1209 i915_program_error(p, "Exceeded max ALU instructions");
1210
1211 if (p->nr_decl_insn > I915_MAX_DECL_INSN)
1212 i915_program_error(p, "Exceeded max DECL instructions");
1213
1214 if (p->error) {
1215 p->NumNativeInstructions = 0;
1216 p->NumNativeAluInstructions = 0;
1217 p->NumNativeTexInstructions = 0;
1218 p->NumNativeTexIndirections = 0;
1219
1220 i915_use_passthrough_shader(ifs);
1221 }
1222 else {
1223 p->NumNativeInstructions
1224 = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
1225 p->NumNativeAluInstructions = p->nr_alu_insn;
1226 p->NumNativeTexInstructions = p->nr_tex_insn;
1227 p->NumNativeTexIndirections = p->nr_tex_indirect;
1228
1229 /* patch in the program length */
1230 p->declarations[0] |= program_size + decl_size - 2;
1231
1232 /* Copy compilation results to fragment program struct:
1233 */
1234 assert(!ifs->program);
1235 ifs->program
1236 = (uint *) MALLOC((program_size + decl_size) * sizeof(uint));
1237 if (ifs->program) {
1238 ifs->program_len = program_size + decl_size;
1239
1240 memcpy(ifs->program,
1241 p->declarations,
1242 decl_size * sizeof(uint));
1243
1244 memcpy(ifs->program + decl_size,
1245 p->program,
1246 program_size * sizeof(uint));
1247 }
1248 }
1249
1250 /* Release the compilation struct:
1251 */
1252 FREE(p);
1253 }
1254
1255
1256
1257
1258
1259 /**
1260 * Rather than trying to intercept and jiggle depth writes during
1261 * emit, just move the value into its correct position at the end of
1262 * the program:
1263 */
1264 static void
1265 i915_fixup_depth_write(struct i915_fp_compile *p)
1266 {
1267 /* XXX assuming pos/depth is always in output[0] */
1268 if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1269 const uint depth = UREG(REG_TYPE_OD, 0);
1270
1271 i915_emit_arith(p,
1272 A0_MOV, /* opcode */
1273 depth, /* dest reg */
1274 A0_DEST_CHANNEL_W, /* write mask */
1275 0, /* saturate? */
1276 swizzle(depth, X, Y, Z, Z), /* src0 */
1277 0, 0 /* src1, src2 */);
1278 }
1279 }
1280
1281
1282 void
1283 i915_translate_fragment_program( struct i915_context *i915,
1284 struct i915_fragment_shader *fs)
1285 {
1286 struct i915_fp_compile *p;
1287 const struct tgsi_token *tokens = fs->state.tokens;
1288
1289 #if 0
1290 tgsi_dump(tokens, 0);
1291 #endif
1292
1293 /* hw doesn't seem to like empty frag programs, even when the depth write
1294 * fixup gets emitted below - may that one is fishy, too? */
1295 if (fs->info.num_instructions == 1) {
1296 i915_use_passthrough_shader(fs);
1297
1298 return;
1299 }
1300
1301 p = i915_init_compile(i915, fs);
1302
1303 i915_translate_instructions(p, tokens, fs);
1304 i915_fixup_depth_write(p);
1305
1306 i915_fini_compile(i915, p);
1307 }