projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge branch 'xa_branch'
[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 #ifndef M_PI
45 #define M_PI 3.14159265358979323846
46 #endif
47
48 /**
49 * Simple pass-through fragment shader to use when we don't have
50 * a real shader (or it fails to compile for some reason).
51 */
52 static unsigned passthrough[] =
53 {
54 _3DSTATE_PIXEL_SHADER_PROGRAM | ((2*3)-1),
55
56 /* declare input color:
57 */
58 (D0_DCL |
59 (REG_TYPE_T << D0_TYPE_SHIFT) |
60 (T_DIFFUSE << D0_NR_SHIFT) |
61 D0_CHANNEL_ALL),
62 0,
63 0,
64
65 /* move to output color:
66 */
67 (A0_MOV |
68 (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) |
69 A0_DEST_CHANNEL_ALL |
70 (REG_TYPE_T << A0_SRC0_TYPE_SHIFT) |
71 (T_DIFFUSE << A0_SRC0_NR_SHIFT)),
72 0x01230000, /* .xyzw */
73 0
74 };
75
76
77 /* 1, -1/3!, 1/5!, -1/7! */
78 static const float scs_sin_constants[4] = { 1.0,
79 -1.0f / (3 * 2 * 1),
80 1.0f / (5 * 4 * 3 * 2 * 1),
81 -1.0f / (7 * 6 * 5 * 4 * 3 * 2 * 1)
82 };
83
84 /* 1, -1/2!, 1/4!, -1/6! */
85 static const float scs_cos_constants[4] = { 1.0,
86 -1.0f / (2 * 1),
87 1.0f / (4 * 3 * 2 * 1),
88 -1.0f / (6 * 5 * 4 * 3 * 2 * 1)
89 };
90
91 /* 2*pi, -(2*pi)^3/3!, (2*pi)^5/5!, -(2*pi)^7/7! */
92 static const float sin_constants[4] = { 2.0 * M_PI,
93 -8.0f * M_PI * M_PI * M_PI / (3 * 2 * 1),
94 32.0f * M_PI * M_PI * M_PI * M_PI * M_PI / (5 * 4 * 3 * 2 * 1),
95 -128.0f * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI / (7 * 6 * 5 * 4 * 3 * 2 * 1)
96 };
97
98 /* 1, -(2*pi)^2/2!, (2*pi)^4/4!, -(2*pi)^6/6! */
99 static const float cos_constants[4] = { 1.0,
100 -4.0f * M_PI * M_PI / (2 * 1),
101 16.0f * M_PI * M_PI * M_PI * M_PI / (4 * 3 * 2 * 1),
102 -64.0f * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI / (6 * 5 * 4 * 3 * 2 * 1)
103 };
104
105
106
107 /**
108 * component-wise negation of ureg
109 */
110 static INLINE int
111 negate(int reg, int x, int y, int z, int w)
112 {
113 /* Another neat thing about the UREG representation */
114 return reg ^ (((x & 1) << UREG_CHANNEL_X_NEGATE_SHIFT) |
115 ((y & 1) << UREG_CHANNEL_Y_NEGATE_SHIFT) |
116 ((z & 1) << UREG_CHANNEL_Z_NEGATE_SHIFT) |
117 ((w & 1) << UREG_CHANNEL_W_NEGATE_SHIFT));
118 }
119
120
121 /**
122 * In the event of a translation failure, we'll generate a simple color
123 * pass-through program.
124 */
125 static void
126 i915_use_passthrough_shader(struct i915_fragment_shader *fs)
127 {
128 fs->program = (uint *) MALLOC(sizeof(passthrough));
129 if (fs->program) {
130 memcpy(fs->program, passthrough, sizeof(passthrough));
131 fs->program_len = Elements(passthrough);
132 }
133 fs->num_constants = 0;
134 }
135
136
137 void
138 i915_program_error(struct i915_fp_compile *p, const char *msg, ...)
139 {
140 va_list args;
141 char buffer[1024];
142
143 debug_printf("i915_program_error: ");
144 va_start( args, msg );
145 util_vsnprintf( buffer, sizeof(buffer), msg, args );
146 va_end( args );
147 debug_printf("%s", buffer);
148 debug_printf("\n");
149
150 p->error = 1;
151 }
152
153 static uint get_mapping(struct i915_fragment_shader* fs, int unit)
154 {
155 int i;
156 for (i = 0; i < I915_TEX_UNITS; i++)
157 {
158 if (fs->generic_mapping[i] == -1) {
159 fs->generic_mapping[i] = unit;
160 return i;
161 }
162 if (fs->generic_mapping[i] == unit)
163 return i;
164 }
165 debug_printf("Exceeded max generics\n");
166 return 0;
167 }
168
169 /**
170 * Construct a ureg for the given source register. Will emit
171 * constants, apply swizzling and negation as needed.
172 */
173 static uint
174 src_vector(struct i915_fp_compile *p,
175 const struct tgsi_full_src_register *source,
176 struct i915_fragment_shader* fs)
177 {
178 uint index = source->Register.Index;
179 uint src = 0, sem_name, sem_ind;
180
181 switch (source->Register.File) {
182 case TGSI_FILE_TEMPORARY:
183 if (source->Register.Index >= I915_MAX_TEMPORARY) {
184 i915_program_error(p, "Exceeded max temporary reg");
185 return 0;
186 }
187 src = UREG(REG_TYPE_R, index);
188 break;
189 case TGSI_FILE_INPUT:
190 /* XXX: Packing COL1, FOGC into a single attribute works for
191 * texenv programs, but will fail for real fragment programs
192 * that use these attributes and expect them to be a full 4
193 * components wide. Could use a texcoord to pass these
194 * attributes if necessary, but that won't work in the general
195 * case.
196 *
197 * We also use a texture coordinate to pass wpos when possible.
198 */
199
200 sem_name = p->shader->info.input_semantic_name[index];
201 sem_ind = p->shader->info.input_semantic_index[index];
202
203 switch (sem_name) {
204 case TGSI_SEMANTIC_POSITION:
205 {
206 /* for fragcoord */
207 int real_tex_unit = get_mapping(fs, I915_SEMANTIC_POS);
208 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_ALL);
209 break;
210 }
211 case TGSI_SEMANTIC_COLOR:
212 if (sem_ind == 0) {
213 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
214 }
215 else {
216 /* secondary color */
217 assert(sem_ind == 1);
218 src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
219 src = swizzle(src, X, Y, Z, ONE);
220 }
221 break;
222 case TGSI_SEMANTIC_FOG:
223 src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
224 src = swizzle(src, W, W, W, W);
225 break;
226 case TGSI_SEMANTIC_GENERIC:
227 {
228 int real_tex_unit = get_mapping(fs, sem_ind);
229 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_ALL);
230 break;
231 }
232 case TGSI_SEMANTIC_FACE:
233 {
234 /* for back/front faces */
235 int real_tex_unit = get_mapping(fs, I915_SEMANTIC_FACE);
236 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_X);
237 break;
238 }
239 default:
240 i915_program_error(p, "Bad source->Index");
241 return 0;
242 }
243 break;
244
245 case TGSI_FILE_IMMEDIATE:
246 assert(index < p->num_immediates);
247 index = p->immediates_map[index];
248 /* fall-through */
249 case TGSI_FILE_CONSTANT:
250 src = UREG(REG_TYPE_CONST, index);
251 break;
252
253 default:
254 i915_program_error(p, "Bad source->File");
255 return 0;
256 }
257
258 src = swizzle(src,
259 source->Register.SwizzleX,
260 source->Register.SwizzleY,
261 source->Register.SwizzleZ,
262 source->Register.SwizzleW);
263
264 /* There's both negate-all-components and per-component negation.
265 * Try to handle both here.
266 */
267 {
268 int n = source->Register.Negate;
269 src = negate(src, n, n, n, n);
270 }
271
272 /* no abs() */
273 #if 0
274 /* XXX assertions disabled to allow arbfplight.c to run */
275 /* XXX enable these assertions, or fix things */
276 assert(!source->Register.Absolute);
277 #endif
278 if (source->Register.Absolute)
279 debug_printf("Unhandled absolute value\n");
280
281 return src;
282 }
283
284
285 /**
286 * Construct a ureg for a destination register.
287 */
288 static uint
289 get_result_vector(struct i915_fp_compile *p,
290 const struct tgsi_full_dst_register *dest)
291 {
292 switch (dest->Register.File) {
293 case TGSI_FILE_OUTPUT:
294 {
295 uint sem_name = p->shader->info.output_semantic_name[dest->Register.Index];
296 switch (sem_name) {
297 case TGSI_SEMANTIC_POSITION:
298 return UREG(REG_TYPE_OD, 0);
299 case TGSI_SEMANTIC_COLOR:
300 return UREG(REG_TYPE_OC, 0);
301 default:
302 i915_program_error(p, "Bad inst->DstReg.Index/semantics");
303 return 0;
304 }
305 }
306 case TGSI_FILE_TEMPORARY:
307 return UREG(REG_TYPE_R, dest->Register.Index);
308 default:
309 i915_program_error(p, "Bad inst->DstReg.File");
310 return 0;
311 }
312 }
313
314
315 /**
316 * Compute flags for saturation and writemask.
317 */
318 static uint
319 get_result_flags(const struct tgsi_full_instruction *inst)
320 {
321 const uint writeMask
322 = inst->Dst[0].Register.WriteMask;
323 uint flags = 0x0;
324
325 if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE)
326 flags |= A0_DEST_SATURATE;
327
328 if (writeMask & TGSI_WRITEMASK_X)
329 flags |= A0_DEST_CHANNEL_X;
330 if (writeMask & TGSI_WRITEMASK_Y)
331 flags |= A0_DEST_CHANNEL_Y;
332 if (writeMask & TGSI_WRITEMASK_Z)
333 flags |= A0_DEST_CHANNEL_Z;
334 if (writeMask & TGSI_WRITEMASK_W)
335 flags |= A0_DEST_CHANNEL_W;
336
337 return flags;
338 }
339
340
341 /**
342 * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
343 */
344 static uint
345 translate_tex_src_target(struct i915_fp_compile *p, uint tex)
346 {
347 switch (tex) {
348 case TGSI_TEXTURE_SHADOW1D:
349 /* fall-through */
350 case TGSI_TEXTURE_1D:
351 return D0_SAMPLE_TYPE_2D;
352
353 case TGSI_TEXTURE_SHADOW2D:
354 /* fall-through */
355 case TGSI_TEXTURE_2D:
356 return D0_SAMPLE_TYPE_2D;
357
358 case TGSI_TEXTURE_SHADOWRECT:
359 /* fall-through */
360 case TGSI_TEXTURE_RECT:
361 return D0_SAMPLE_TYPE_2D;
362
363 case TGSI_TEXTURE_3D:
364 return D0_SAMPLE_TYPE_VOLUME;
365
366 case TGSI_TEXTURE_CUBE:
367 return D0_SAMPLE_TYPE_CUBE;
368
369 default:
370 i915_program_error(p, "TexSrc type");
371 return 0;
372 }
373 }
374
375
376 /**
377 * Generate texel lookup instruction.
378 */
379 static void
380 emit_tex(struct i915_fp_compile *p,
381 const struct tgsi_full_instruction *inst,
382 uint opcode,
383 struct i915_fragment_shader* fs)
384 {
385 uint texture = inst->Texture.Texture;
386 uint unit = inst->Src[1].Register.Index;
387 uint tex = translate_tex_src_target( p, texture );
388 uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex);
389 uint coord = src_vector( p, &inst->Src[0], fs);
390
391 i915_emit_texld( p,
392 get_result_vector( p, &inst->Dst[0] ),
393 get_result_flags( inst ),
394 sampler,
395 coord,
396 opcode);
397 }
398
399
400 /**
401 * Generate a simple arithmetic instruction
402 * \param opcode the i915 opcode
403 * \param numArgs the number of input/src arguments
404 */
405 static void
406 emit_simple_arith(struct i915_fp_compile *p,
407 const struct tgsi_full_instruction *inst,
408 uint opcode, uint numArgs,
409 struct i915_fragment_shader* fs)
410 {
411 uint arg1, arg2, arg3;
412
413 assert(numArgs <= 3);
414
415 arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->Src[0], fs );
416 arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->Src[1], fs );
417 arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->Src[2], fs );
418
419 i915_emit_arith( p,
420 opcode,
421 get_result_vector( p, &inst->Dst[0]),
422 get_result_flags( inst ), 0,
423 arg1,
424 arg2,
425 arg3 );
426 }
427
428
429 /** As above, but swap the first two src regs */
430 static void
431 emit_simple_arith_swap2(struct i915_fp_compile *p,
432 const struct tgsi_full_instruction *inst,
433 uint opcode, uint numArgs,
434 struct i915_fragment_shader* fs)
435 {
436 struct tgsi_full_instruction inst2;
437
438 assert(numArgs == 2);
439
440 /* transpose first two registers */
441 inst2 = *inst;
442 inst2.Src[0] = inst->Src[1];
443 inst2.Src[1] = inst->Src[0];
444
445 emit_simple_arith(p, &inst2, opcode, numArgs, fs);
446 }
447
448 /*
449 * Translate TGSI instruction to i915 instruction.
450 *
451 * Possible concerns:
452 *
453 * SIN, COS -- could use another taylor step?
454 * LIT -- results seem a little different to sw mesa
455 * LOG -- different to mesa on negative numbers, but this is conformant.
456 */
457 static void
458 i915_translate_instruction(struct i915_fp_compile *p,
459 const struct tgsi_full_instruction *inst,
460 struct i915_fragment_shader *fs)
461 {
462 uint writemask;
463 uint src0, src1, src2, flags;
464 uint tmp = 0;
465
466 switch (inst->Instruction.Opcode) {
467 case TGSI_OPCODE_ABS:
468 src0 = src_vector(p, &inst->Src[0], fs);
469 i915_emit_arith(p,
470 A0_MAX,
471 get_result_vector(p, &inst->Dst[0]),
472 get_result_flags(inst), 0,
473 src0, negate(src0, 1, 1, 1, 1), 0);
474 break;
475
476 case TGSI_OPCODE_ADD:
477 emit_simple_arith(p, inst, A0_ADD, 2, fs);
478 break;
479
480 case TGSI_OPCODE_CMP:
481 src0 = src_vector(p, &inst->Src[0], fs);
482 src1 = src_vector(p, &inst->Src[1], fs);
483 src2 = src_vector(p, &inst->Src[2], fs);
484 i915_emit_arith(p, A0_CMP,
485 get_result_vector(p, &inst->Dst[0]),
486 get_result_flags(inst),
487 0, src0, src2, src1); /* NOTE: order of src2, src1 */
488 break;
489
490 case TGSI_OPCODE_COS:
491 src0 = src_vector(p, &inst->Src[0], fs);
492 tmp = i915_get_utemp(p);
493
494 i915_emit_arith(p,
495 A0_MUL,
496 tmp, A0_DEST_CHANNEL_X, 0,
497 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
498
499 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
500
501 /*
502 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
503 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
504 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
505 * result = DP4 t0, cos_constants
506 */
507 i915_emit_arith(p,
508 A0_MUL,
509 tmp, A0_DEST_CHANNEL_XY, 0,
510 swizzle(tmp, X, X, ONE, ONE),
511 swizzle(tmp, X, ONE, ONE, ONE), 0);
512
513 i915_emit_arith(p,
514 A0_MUL,
515 tmp, A0_DEST_CHANNEL_XYZ, 0,
516 swizzle(tmp, X, Y, X, ONE),
517 swizzle(tmp, X, X, ONE, ONE), 0);
518
519 i915_emit_arith(p,
520 A0_MUL,
521 tmp, A0_DEST_CHANNEL_XYZ, 0,
522 swizzle(tmp, X, X, Z, ONE),
523 swizzle(tmp, Z, ONE, ONE, ONE), 0);
524
525 i915_emit_arith(p,
526 A0_DP4,
527 get_result_vector(p, &inst->Dst[0]),
528 get_result_flags(inst), 0,
529 swizzle(tmp, ONE, Z, Y, X),
530 i915_emit_const4fv(p, cos_constants), 0);
531 break;
532
533 case TGSI_OPCODE_DDX:
534 case TGSI_OPCODE_DDY:
535 /* XXX We just output 0 here */
536 debug_printf("Punting DDX/DDX\n");
537 src0 = get_result_vector(p, &inst->Dst[0]);
538 i915_emit_arith(p,
539 A0_MOV,
540 get_result_vector(p, &inst->Dst[0]),
541 get_result_flags(inst), 0,
542 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0, 0);
543 break;
544
545 case TGSI_OPCODE_DP2:
546 src0 = src_vector(p, &inst->Src[0], fs);
547 src1 = src_vector(p, &inst->Src[1], fs);
548
549 i915_emit_arith(p,
550 A0_DP3,
551 get_result_vector(p, &inst->Dst[0]),
552 get_result_flags(inst), 0,
553 swizzle(src0, X, Y, ZERO, ZERO), src1, 0);
554 break;
555
556 case TGSI_OPCODE_DP3:
557 emit_simple_arith(p, inst, A0_DP3, 2, fs);
558 break;
559
560 case TGSI_OPCODE_DP4:
561 emit_simple_arith(p, inst, A0_DP4, 2, fs);
562 break;
563
564 case TGSI_OPCODE_DPH:
565 src0 = src_vector(p, &inst->Src[0], fs);
566 src1 = src_vector(p, &inst->Src[1], fs);
567
568 i915_emit_arith(p,
569 A0_DP4,
570 get_result_vector(p, &inst->Dst[0]),
571 get_result_flags(inst), 0,
572 swizzle(src0, X, Y, Z, ONE), src1, 0);
573 break;
574
575 case TGSI_OPCODE_DST:
576 src0 = src_vector(p, &inst->Src[0], fs);
577 src1 = src_vector(p, &inst->Src[1], fs);
578
579 /* result[0] = 1 * 1;
580 * result[1] = a[1] * b[1];
581 * result[2] = a[2] * 1;
582 * result[3] = 1 * b[3];
583 */
584 i915_emit_arith(p,
585 A0_MUL,
586 get_result_vector(p, &inst->Dst[0]),
587 get_result_flags(inst), 0,
588 swizzle(src0, ONE, Y, Z, ONE),
589 swizzle(src1, ONE, Y, ONE, W), 0);
590 break;
591
592 case TGSI_OPCODE_END:
593 /* no-op */
594 break;
595
596 case TGSI_OPCODE_EX2:
597 src0 = src_vector(p, &inst->Src[0], fs);
598
599 i915_emit_arith(p,
600 A0_EXP,
601 get_result_vector(p, &inst->Dst[0]),
602 get_result_flags(inst), 0,
603 swizzle(src0, X, X, X, X), 0, 0);
604 break;
605
606 case TGSI_OPCODE_FLR:
607 emit_simple_arith(p, inst, A0_FLR, 1, fs);
608 break;
609
610 case TGSI_OPCODE_FRC:
611 emit_simple_arith(p, inst, A0_FRC, 1, fs);
612 break;
613
614 case TGSI_OPCODE_KIL:
615 /* kill if src[0].x < 0 || src[0].y < 0 ... */
616 src0 = src_vector(p, &inst->Src[0], fs);
617 tmp = i915_get_utemp(p);
618
619 i915_emit_texld(p,
620 tmp, /* dest reg: a dummy reg */
621 A0_DEST_CHANNEL_ALL, /* dest writemask */
622 0, /* sampler */
623 src0, /* coord*/
624 T0_TEXKILL); /* opcode */
625 break;
626
627 case TGSI_OPCODE_KILP:
628 assert(0); /* not tested yet */
629 break;
630
631 case TGSI_OPCODE_LG2:
632 src0 = src_vector(p, &inst->Src[0], fs);
633
634 i915_emit_arith(p,
635 A0_LOG,
636 get_result_vector(p, &inst->Dst[0]),
637 get_result_flags(inst), 0,
638 swizzle(src0, X, X, X, X), 0, 0);
639 break;
640
641 case TGSI_OPCODE_LIT:
642 src0 = src_vector(p, &inst->Src[0], fs);
643 tmp = i915_get_utemp(p);
644
645 /* tmp = max( a.xyzw, a.00zw )
646 * XXX: Clamp tmp.w to -128..128
647 * tmp.y = log(tmp.y)
648 * tmp.y = tmp.w * tmp.y
649 * tmp.y = exp(tmp.y)
650 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
651 */
652 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
653 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
654
655 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
656 swizzle(tmp, Y, Y, Y, Y), 0, 0);
657
658 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
659 swizzle(tmp, ZERO, Y, ZERO, ZERO),
660 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
661
662 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
663 swizzle(tmp, Y, Y, Y, Y), 0, 0);
664
665 i915_emit_arith(p, A0_CMP,
666 get_result_vector(p, &inst->Dst[0]),
667 get_result_flags(inst), 0,
668 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
669 swizzle(tmp, ONE, X, ZERO, ONE),
670 swizzle(tmp, ONE, X, Y, ONE));
671
672 break;
673
674 case TGSI_OPCODE_LRP:
675 src0 = src_vector(p, &inst->Src[0], fs);
676 src1 = src_vector(p, &inst->Src[1], fs);
677 src2 = src_vector(p, &inst->Src[2], fs);
678 flags = get_result_flags(inst);
679 tmp = i915_get_utemp(p);
680
681 /* b*a + c*(1-a)
682 *
683 * b*a + c - ca
684 *
685 * tmp = b*a + c,
686 * result = (-c)*a + tmp
687 */
688 i915_emit_arith(p, A0_MAD, tmp,
689 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
690
691 i915_emit_arith(p, A0_MAD,
692 get_result_vector(p, &inst->Dst[0]),
693 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
694 break;
695
696 case TGSI_OPCODE_MAD:
697 emit_simple_arith(p, inst, A0_MAD, 3, fs);
698 break;
699
700 case TGSI_OPCODE_MAX:
701 emit_simple_arith(p, inst, A0_MAX, 2, fs);
702 break;
703
704 case TGSI_OPCODE_MIN:
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 i915_emit_arith(p,
711 A0_MAX,
712 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
713 negate(src0, 1, 1, 1, 1),
714 negate(src1, 1, 1, 1, 1), 0);
715
716 i915_emit_arith(p,
717 A0_MOV,
718 get_result_vector(p, &inst->Dst[0]),
719 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
720 break;
721
722 case TGSI_OPCODE_MOV:
723 emit_simple_arith(p, inst, A0_MOV, 1, fs);
724 break;
725
726 case TGSI_OPCODE_MUL:
727 emit_simple_arith(p, inst, A0_MUL, 2, fs);
728 break;
729
730 case TGSI_OPCODE_POW:
731 src0 = src_vector(p, &inst->Src[0], fs);
732 src1 = src_vector(p, &inst->Src[1], fs);
733 tmp = i915_get_utemp(p);
734 flags = get_result_flags(inst);
735
736 /* XXX: masking on intermediate values, here and elsewhere.
737 */
738 i915_emit_arith(p,
739 A0_LOG,
740 tmp, A0_DEST_CHANNEL_X, 0,
741 swizzle(src0, X, X, X, X), 0, 0);
742
743 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
744
745 i915_emit_arith(p,
746 A0_EXP,
747 get_result_vector(p, &inst->Dst[0]),
748 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
749 break;
750
751 case TGSI_OPCODE_RET:
752 /* XXX: no-op? */
753 break;
754
755 case TGSI_OPCODE_RCP:
756 src0 = src_vector(p, &inst->Src[0], fs);
757
758 i915_emit_arith(p,
759 A0_RCP,
760 get_result_vector(p, &inst->Dst[0]),
761 get_result_flags(inst), 0,
762 swizzle(src0, X, X, X, X), 0, 0);
763 break;
764
765 case TGSI_OPCODE_RSQ:
766 src0 = src_vector(p, &inst->Src[0], fs);
767
768 i915_emit_arith(p,
769 A0_RSQ,
770 get_result_vector(p, &inst->Dst[0]),
771 get_result_flags(inst), 0,
772 swizzle(src0, X, X, X, X), 0, 0);
773 break;
774
775 case TGSI_OPCODE_SCS:
776 src0 = src_vector(p, &inst->Src[0], fs);
777 tmp = i915_get_utemp(p);
778
779 /*
780 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
781 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
782 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
783 * scs.x = DP4 t1, scs_sin_constants
784 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
785 * scs.y = DP4 t1, scs_cos_constants
786 */
787 i915_emit_arith(p,
788 A0_MUL,
789 tmp, A0_DEST_CHANNEL_XY, 0,
790 swizzle(src0, X, X, ONE, ONE),
791 swizzle(src0, X, ONE, ONE, ONE), 0);
792
793 i915_emit_arith(p,
794 A0_MUL,
795 tmp, A0_DEST_CHANNEL_ALL, 0,
796 swizzle(tmp, X, Y, X, Y),
797 swizzle(tmp, X, X, ONE, ONE), 0);
798
799 writemask = inst->Dst[0].Register.WriteMask;
800
801 if (writemask & TGSI_WRITEMASK_Y) {
802 uint tmp1;
803
804 if (writemask & TGSI_WRITEMASK_X)
805 tmp1 = i915_get_utemp(p);
806 else
807 tmp1 = tmp;
808
809 i915_emit_arith(p,
810 A0_MUL,
811 tmp1, A0_DEST_CHANNEL_ALL, 0,
812 swizzle(tmp, X, Y, Y, W),
813 swizzle(tmp, X, Z, ONE, ONE), 0);
814
815 i915_emit_arith(p,
816 A0_DP4,
817 get_result_vector(p, &inst->Dst[0]),
818 A0_DEST_CHANNEL_Y, 0,
819 swizzle(tmp1, W, Z, Y, X),
820 i915_emit_const4fv(p, scs_sin_constants), 0);
821 }
822
823 if (writemask & TGSI_WRITEMASK_X) {
824 i915_emit_arith(p,
825 A0_MUL,
826 tmp, A0_DEST_CHANNEL_XYZ, 0,
827 swizzle(tmp, X, X, Z, ONE),
828 swizzle(tmp, Z, ONE, ONE, ONE), 0);
829
830 i915_emit_arith(p,
831 A0_DP4,
832 get_result_vector(p, &inst->Dst[0]),
833 A0_DEST_CHANNEL_X, 0,
834 swizzle(tmp, ONE, Z, Y, X),
835 i915_emit_const4fv(p, scs_cos_constants), 0);
836 }
837 break;
838
839 case TGSI_OPCODE_SEQ:
840 /* if we're both >= and <= then we're == */
841 src0 = src_vector(p, &inst->Src[0], fs);
842 src1 = src_vector(p, &inst->Src[1], fs);
843 tmp = i915_get_utemp(p);
844
845 i915_emit_arith(p,
846 A0_SGE,
847 tmp, A0_DEST_CHANNEL_ALL, 0,
848 src0,
849 src1, 0);
850
851 i915_emit_arith(p,
852 A0_SGE,
853 get_result_vector(p, &inst->Dst[0]),
854 A0_DEST_CHANNEL_ALL, 0,
855 src1,
856 src0, 0);
857
858 i915_emit_arith(p,
859 A0_MUL,
860 get_result_vector(p, &inst->Dst[0]),
861 A0_DEST_CHANNEL_ALL, 0,
862 get_result_vector(p, &inst->Dst[0]),
863 tmp, 0);
864
865 break;
866
867 case TGSI_OPCODE_SGE:
868 emit_simple_arith(p, inst, A0_SGE, 2, fs);
869 break;
870
871 case TGSI_OPCODE_SIN:
872 src0 = src_vector(p, &inst->Src[0], fs);
873 tmp = i915_get_utemp(p);
874
875 i915_emit_arith(p,
876 A0_MUL,
877 tmp, A0_DEST_CHANNEL_X, 0,
878 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
879
880 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
881
882 /*
883 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
884 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
885 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
886 * result = DP4 t1.wzyx, sin_constants
887 */
888 i915_emit_arith(p,
889 A0_MUL,
890 tmp, A0_DEST_CHANNEL_XY, 0,
891 swizzle(tmp, X, X, ONE, ONE),
892 swizzle(tmp, X, ONE, ONE, ONE), 0);
893
894 i915_emit_arith(p,
895 A0_MUL,
896 tmp, A0_DEST_CHANNEL_ALL, 0,
897 swizzle(tmp, X, Y, X, Y),
898 swizzle(tmp, X, X, ONE, ONE), 0);
899
900 i915_emit_arith(p,
901 A0_MUL,
902 tmp, A0_DEST_CHANNEL_ALL, 0,
903 swizzle(tmp, X, Y, Y, W),
904 swizzle(tmp, X, Z, ONE, ONE), 0);
905
906 i915_emit_arith(p,
907 A0_DP4,
908 get_result_vector(p, &inst->Dst[0]),
909 get_result_flags(inst), 0,
910 swizzle(tmp, W, Z, Y, X),
911 i915_emit_const4fv(p, sin_constants), 0);
912 break;
913
914 case TGSI_OPCODE_SLE:
915 /* like SGE, but swap reg0, reg1 */
916 emit_simple_arith_swap2(p, inst, A0_SGE, 2, fs);
917 break;
918
919 case TGSI_OPCODE_SLT:
920 emit_simple_arith(p, inst, A0_SLT, 2, fs);
921 break;
922
923 case TGSI_OPCODE_SGT:
924 /* like SLT, but swap reg0, reg1 */
925 emit_simple_arith_swap2(p, inst, A0_SLT, 2, fs);
926 break;
927
928 case TGSI_OPCODE_SNE:
929 /* if we're < or > then we're != */
930 src0 = src_vector(p, &inst->Src[0], fs);
931 src1 = src_vector(p, &inst->Src[1], fs);
932 tmp = i915_get_utemp(p);
933
934 i915_emit_arith(p,
935 A0_SLT,
936 tmp,
937 A0_DEST_CHANNEL_ALL, 0,
938 src0,
939 src1, 0);
940
941 i915_emit_arith(p,
942 A0_SLT,
943 get_result_vector(p, &inst->Dst[0]),
944 A0_DEST_CHANNEL_ALL, 0,
945 src1,
946 src0, 0);
947
948 i915_emit_arith(p,
949 A0_ADD,
950 get_result_vector(p, &inst->Dst[0]),
951 A0_DEST_CHANNEL_ALL, 0,
952 get_result_vector(p, &inst->Dst[0]),
953 tmp, 0);
954 break;
955
956 case TGSI_OPCODE_SSG:
957 /* compute (src>0) - (src<0) */
958 src0 = src_vector(p, &inst->Src[0], fs);
959 tmp = i915_get_utemp(p);
960
961 i915_emit_arith(p,
962 A0_SLT,
963 tmp,
964 A0_DEST_CHANNEL_ALL, 0,
965 src0,
966 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0);
967
968 i915_emit_arith(p,
969 A0_SLT,
970 get_result_vector(p, &inst->Dst[0]),
971 A0_DEST_CHANNEL_ALL, 0,
972 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
973 src0, 0);
974
975 i915_emit_arith(p,
976 A0_ADD,
977 get_result_vector(p, &inst->Dst[0]),
978 A0_DEST_CHANNEL_ALL, 0,
979 get_result_vector(p, &inst->Dst[0]),
980 negate(tmp, 1, 1, 1, 1), 0);
981 break;
982
983 case TGSI_OPCODE_SUB:
984 src0 = src_vector(p, &inst->Src[0], fs);
985 src1 = src_vector(p, &inst->Src[1], fs);
986
987 i915_emit_arith(p,
988 A0_ADD,
989 get_result_vector(p, &inst->Dst[0]),
990 get_result_flags(inst), 0,
991 src0, negate(src1, 1, 1, 1, 1), 0);
992 break;
993
994 case TGSI_OPCODE_TEX:
995 emit_tex(p, inst, T0_TEXLD, fs);
996 break;
997
998 case TGSI_OPCODE_TRUNC:
999 emit_simple_arith(p, inst, A0_TRC, 1, fs);
1000 break;
1001
1002 case TGSI_OPCODE_TXB:
1003 emit_tex(p, inst, T0_TEXLDB, fs);
1004 break;
1005
1006 case TGSI_OPCODE_TXP:
1007 emit_tex(p, inst, T0_TEXLDP, fs);
1008 break;
1009
1010 case TGSI_OPCODE_XPD:
1011 /* Cross product:
1012 * result.x = src0.y * src1.z - src0.z * src1.y;
1013 * result.y = src0.z * src1.x - src0.x * src1.z;
1014 * result.z = src0.x * src1.y - src0.y * src1.x;
1015 * result.w = undef;
1016 */
1017 src0 = src_vector(p, &inst->Src[0], fs);
1018 src1 = src_vector(p, &inst->Src[1], fs);
1019 tmp = i915_get_utemp(p);
1020
1021 i915_emit_arith(p,
1022 A0_MUL,
1023 tmp, A0_DEST_CHANNEL_ALL, 0,
1024 swizzle(src0, Z, X, Y, ONE),
1025 swizzle(src1, Y, Z, X, ONE), 0);
1026
1027 i915_emit_arith(p,
1028 A0_MAD,
1029 get_result_vector(p, &inst->Dst[0]),
1030 get_result_flags(inst), 0,
1031 swizzle(src0, Y, Z, X, ONE),
1032 swizzle(src1, Z, X, Y, ONE),
1033 negate(tmp, 1, 1, 1, 0));
1034 break;
1035
1036 default:
1037 i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
1038 p->error = 1;
1039 return;
1040 }
1041
1042 i915_release_utemps(p);
1043 }
1044
1045
1046 /**
1047 * Translate TGSI fragment shader into i915 hardware instructions.
1048 * \param p the translation state
1049 * \param tokens the TGSI token array
1050 */
1051 static void
1052 i915_translate_instructions(struct i915_fp_compile *p,
1053 const struct tgsi_token *tokens,
1054 struct i915_fragment_shader *fs)
1055 {
1056 struct i915_fragment_shader *ifs = p->shader;
1057 struct tgsi_parse_context parse;
1058
1059 tgsi_parse_init( &parse, tokens );
1060
1061 while( !tgsi_parse_end_of_tokens( &parse ) ) {
1062
1063 tgsi_parse_token( &parse );
1064
1065 switch( parse.FullToken.Token.Type ) {
1066 case TGSI_TOKEN_TYPE_PROPERTY:
1067 /*
1068 * We only support one cbuf, but we still need to ignore the property
1069 * correctly so we don't hit the assert at the end of the switch case.
1070 */
1071 assert(parse.FullToken.FullProperty.Property.PropertyName ==
1072 TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS);
1073 break;
1074 case TGSI_TOKEN_TYPE_DECLARATION:
1075 if (parse.FullToken.FullDeclaration.Declaration.File
1076 == TGSI_FILE_CONSTANT) {
1077 uint i;
1078 for (i = parse.FullToken.FullDeclaration.Range.First;
1079 i <= parse.FullToken.FullDeclaration.Range.Last;
1080 i++) {
1081 assert(ifs->constant_flags[i] == 0x0);
1082 ifs->constant_flags[i] = I915_CONSTFLAG_USER;
1083 ifs->num_constants = MAX2(ifs->num_constants, i + 1);
1084 }
1085 }
1086 else if (parse.FullToken.FullDeclaration.Declaration.File
1087 == TGSI_FILE_TEMPORARY) {
1088 uint i;
1089 for (i = parse.FullToken.FullDeclaration.Range.First;
1090 i <= parse.FullToken.FullDeclaration.Range.Last;
1091 i++) {
1092 if (i >= I915_MAX_TEMPORARY)
1093 debug_printf("Too many temps (%d)\n",i);
1094 else
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 memset(&p->register_phases, 0, sizeof(p->register_phases));
1169
1170 for (i = 0; i < I915_TEX_UNITS; i++)
1171 ifs->generic_mapping[i] = -1;
1172
1173 p->first_instruction = TRUE;
1174
1175 p->nr_tex_indirect = 1; /* correct? */
1176 p->nr_tex_insn = 0;
1177 p->nr_alu_insn = 0;
1178 p->nr_decl_insn = 0;
1179
1180 p->csr = p->program;
1181 p->decl = p->declarations;
1182 p->decl_s = 0;
1183 p->decl_t = 0;
1184 p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
1185 p->utemp_flag = ~0x7;
1186
1187 /* initialize the first program word */
1188 *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
1189
1190 return p;
1191 }
1192
1193
1194 /* Copy compile results to the fragment program struct and destroy the
1195 * compilation context.
1196 */
1197 static void
1198 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
1199 {
1200 struct i915_fragment_shader *ifs = p->shader;
1201 unsigned long program_size = (unsigned long) (p->csr - p->program);
1202 unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
1203
1204 if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
1205 debug_printf("Exceeded max nr indirect texture lookups\n");
1206
1207 if (p->nr_tex_insn > I915_MAX_TEX_INSN)
1208 i915_program_error(p, "Exceeded max TEX instructions");
1209
1210 if (p->nr_alu_insn > I915_MAX_ALU_INSN)
1211 i915_program_error(p, "Exceeded max ALU instructions");
1212
1213 if (p->nr_decl_insn > I915_MAX_DECL_INSN)
1214 i915_program_error(p, "Exceeded max DECL instructions");
1215
1216 if (p->error) {
1217 p->NumNativeInstructions = 0;
1218 p->NumNativeAluInstructions = 0;
1219 p->NumNativeTexInstructions = 0;
1220 p->NumNativeTexIndirections = 0;
1221
1222 i915_use_passthrough_shader(ifs);
1223 }
1224 else {
1225 p->NumNativeInstructions
1226 = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
1227 p->NumNativeAluInstructions = p->nr_alu_insn;
1228 p->NumNativeTexInstructions = p->nr_tex_insn;
1229 p->NumNativeTexIndirections = p->nr_tex_indirect;
1230
1231 /* patch in the program length */
1232 p->declarations[0] |= program_size + decl_size - 2;
1233
1234 /* Copy compilation results to fragment program struct:
1235 */
1236 assert(!ifs->program);
1237 ifs->program
1238 = (uint *) MALLOC((program_size + decl_size) * sizeof(uint));
1239 if (ifs->program) {
1240 ifs->program_len = program_size + decl_size;
1241
1242 memcpy(ifs->program,
1243 p->declarations,
1244 decl_size * sizeof(uint));
1245
1246 memcpy(ifs->program + decl_size,
1247 p->program,
1248 program_size * sizeof(uint));
1249 }
1250 }
1251
1252 /* Release the compilation struct:
1253 */
1254 FREE(p);
1255 }
1256
1257
1258
1259
1260
1261 /**
1262 * Rather than trying to intercept and jiggle depth writes during
1263 * emit, just move the value into its correct position at the end of
1264 * the program:
1265 */
1266 static void
1267 i915_fixup_depth_write(struct i915_fp_compile *p)
1268 {
1269 /* XXX assuming pos/depth is always in output[0] */
1270 if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1271 const uint depth = UREG(REG_TYPE_OD, 0);
1272
1273 i915_emit_arith(p,
1274 A0_MOV, /* opcode */
1275 depth, /* dest reg */
1276 A0_DEST_CHANNEL_W, /* write mask */
1277 0, /* saturate? */
1278 swizzle(depth, X, Y, Z, Z), /* src0 */
1279 0, 0 /* src1, src2 */);
1280 }
1281 }
1282
1283
1284 void
1285 i915_translate_fragment_program( struct i915_context *i915,
1286 struct i915_fragment_shader *fs)
1287 {
1288 struct i915_fp_compile *p;
1289 const struct tgsi_token *tokens = fs->state.tokens;
1290
1291 #if 0
1292 tgsi_dump(tokens, 0);
1293 #endif
1294
1295 /* hw doesn't seem to like empty frag programs, even when the depth write
1296 * fixup gets emitted below - may that one is fishy, too? */
1297 if (fs->info.num_instructions == 1) {
1298 i915_use_passthrough_shader(fs);
1299
1300 return;
1301 }
1302
1303 p = i915_init_compile(i915, fs);
1304
1305 i915_translate_instructions(p, tokens, fs);
1306 i915_fixup_depth_write(p);
1307
1308 i915_fini_compile(i915, p);
1309 }