projects / mesa.git / blob
? search:


a9601e82ca9cf2d28d3385688424434d26ebd0a2
[mesa.git] / src / gallium / drivers / i915 / i915_fpc_translate.c
1 /**************************************************************************
2 *
3 * Copyright 2007 VMware, Inc.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28
29 #include <stdarg.h>
30
31 #include "i915_reg.h"
32 #include "i915_context.h"
33 #include "i915_fpc.h"
34 #include "i915_debug_private.h"
35
36 #include "pipe/p_shader_tokens.h"
37 #include "util/u_math.h"
38 #include "util/u_memory.h"
39 #include "util/u_string.h"
40 #include "tgsi/tgsi_parse.h"
41 #include "tgsi/tgsi_dump.h"
42
43 #include "draw/draw_vertex.h"
44
45 #ifndef M_PI
46 #define M_PI 3.14159265358979323846
47 #endif
48
49 /**
50 * Simple pass-through fragment shader to use when we don't have
51 * a real shader (or it fails to compile for some reason).
52 */
53 static unsigned passthrough_decl[] =
54 {
55 _3DSTATE_PIXEL_SHADER_PROGRAM | ((2*3)-1),
56
57 /* declare input color:
58 */
59 (D0_DCL |
60 (REG_TYPE_T << D0_TYPE_SHIFT) |
61 (T_DIFFUSE << D0_NR_SHIFT) |
62 D0_CHANNEL_ALL),
63 0,
64 0,
65 };
66
67 static unsigned passthrough_program[] =
68 {
69 /* move to output color:
70 */
71 (A0_MOV |
72 (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) |
73 A0_DEST_CHANNEL_ALL |
74 (REG_TYPE_T << A0_SRC0_TYPE_SHIFT) |
75 (T_DIFFUSE << A0_SRC0_NR_SHIFT)),
76 0x01230000, /* .xyzw */
77 0
78 };
79
80
81 /* 1, -1/3!, 1/5!, -1/7! */
82 static const float scs_sin_constants[4] = { 1.0,
83 -1.0f / (3 * 2 * 1),
84 1.0f / (5 * 4 * 3 * 2 * 1),
85 -1.0f / (7 * 6 * 5 * 4 * 3 * 2 * 1)
86 };
87
88 /* 1, -1/2!, 1/4!, -1/6! */
89 static const float scs_cos_constants[4] = { 1.0,
90 -1.0f / (2 * 1),
91 1.0f / (4 * 3 * 2 * 1),
92 -1.0f / (6 * 5 * 4 * 3 * 2 * 1)
93 };
94
95 /* 2*pi, -(2*pi)^3/3!, (2*pi)^5/5!, -(2*pi)^7/7! */
96 static const float sin_constants[4] = { 2.0 * M_PI,
97 -8.0f * M_PI * M_PI * M_PI / (3 * 2 * 1),
98 32.0f * M_PI * M_PI * M_PI * M_PI * M_PI / (5 * 4 * 3 * 2 * 1),
99 -128.0f * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI / (7 * 6 * 5 * 4 * 3 * 2 * 1)
100 };
101
102 /* 1, -(2*pi)^2/2!, (2*pi)^4/4!, -(2*pi)^6/6! */
103 static const float cos_constants[4] = { 1.0,
104 -4.0f * M_PI * M_PI / (2 * 1),
105 16.0f * M_PI * M_PI * M_PI * M_PI / (4 * 3 * 2 * 1),
106 -64.0f * M_PI * M_PI * M_PI * M_PI * M_PI * M_PI / (6 * 5 * 4 * 3 * 2 * 1)
107 };
108
109
110
111 /**
112 * component-wise negation of ureg
113 */
114 static inline int
115 negate(int reg, int x, int y, int z, int w)
116 {
117 /* Another neat thing about the UREG representation */
118 return reg ^ (((x & 1) << UREG_CHANNEL_X_NEGATE_SHIFT) |
119 ((y & 1) << UREG_CHANNEL_Y_NEGATE_SHIFT) |
120 ((z & 1) << UREG_CHANNEL_Z_NEGATE_SHIFT) |
121 ((w & 1) << UREG_CHANNEL_W_NEGATE_SHIFT));
122 }
123
124
125 /**
126 * In the event of a translation failure, we'll generate a simple color
127 * pass-through program.
128 */
129 static void
130 i915_use_passthrough_shader(struct i915_fragment_shader *fs)
131 {
132 fs->program = (uint *) MALLOC(sizeof(passthrough_program));
133 fs->decl = (uint *) MALLOC(sizeof(passthrough_decl));
134 if (fs->program) {
135 memcpy(fs->program, passthrough_program, sizeof(passthrough_program));
136 memcpy(fs->decl, passthrough_decl, sizeof(passthrough_decl));
137 fs->program_len = ARRAY_SIZE(passthrough_program);
138 fs->decl_len = ARRAY_SIZE(passthrough_decl);
139 }
140 fs->num_constants = 0;
141 }
142
143
144 void
145 i915_program_error(struct i915_fp_compile *p, const char *msg, ...)
146 {
147 va_list args;
148 char buffer[1024];
149
150 debug_printf("i915_program_error: ");
151 va_start( args, msg );
152 util_vsnprintf( buffer, sizeof(buffer), msg, args );
153 va_end( args );
154 debug_printf("%s", buffer);
155 debug_printf("\n");
156
157 p->error = 1;
158 }
159
160 static uint get_mapping(struct i915_fragment_shader* fs, int unit)
161 {
162 int i;
163 for (i = 0; i < I915_TEX_UNITS; i++)
164 {
165 if (fs->generic_mapping[i] == -1) {
166 fs->generic_mapping[i] = unit;
167 return i;
168 }
169 if (fs->generic_mapping[i] == unit)
170 return i;
171 }
172 debug_printf("Exceeded max generics\n");
173 return 0;
174 }
175
176 /**
177 * Construct a ureg for the given source register. Will emit
178 * constants, apply swizzling and negation as needed.
179 */
180 static uint
181 src_vector(struct i915_fp_compile *p,
182 const struct i915_full_src_register *source,
183 struct i915_fragment_shader *fs)
184 {
185 uint index = source->Register.Index;
186 uint src = 0, sem_name, sem_ind;
187
188 switch (source->Register.File) {
189 case TGSI_FILE_TEMPORARY:
190 if (source->Register.Index >= I915_MAX_TEMPORARY) {
191 i915_program_error(p, "Exceeded max temporary reg");
192 return 0;
193 }
194 src = UREG(REG_TYPE_R, index);
195 break;
196 case TGSI_FILE_INPUT:
197 /* XXX: Packing COL1, FOGC into a single attribute works for
198 * texenv programs, but will fail for real fragment programs
199 * that use these attributes and expect them to be a full 4
200 * components wide. Could use a texcoord to pass these
201 * attributes if necessary, but that won't work in the general
202 * case.
203 *
204 * We also use a texture coordinate to pass wpos when possible.
205 */
206
207 sem_name = p->shader->info.input_semantic_name[index];
208 sem_ind = p->shader->info.input_semantic_index[index];
209
210 switch (sem_name) {
211 case TGSI_SEMANTIC_POSITION:
212 {
213 /* for fragcoord */
214 int real_tex_unit = get_mapping(fs, I915_SEMANTIC_POS);
215 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_ALL);
216 break;
217 }
218 case TGSI_SEMANTIC_COLOR:
219 if (sem_ind == 0) {
220 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
221 }
222 else {
223 /* secondary color */
224 assert(sem_ind == 1);
225 src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
226 src = swizzle(src, X, Y, Z, ONE);
227 }
228 break;
229 case TGSI_SEMANTIC_FOG:
230 src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
231 src = swizzle(src, W, W, W, W);
232 break;
233 case TGSI_SEMANTIC_GENERIC:
234 {
235 int real_tex_unit = get_mapping(fs, sem_ind);
236 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_ALL);
237 break;
238 }
239 case TGSI_SEMANTIC_FACE:
240 {
241 /* for back/front faces */
242 int real_tex_unit = get_mapping(fs, I915_SEMANTIC_FACE);
243 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + real_tex_unit, D0_CHANNEL_X);
244 break;
245 }
246 default:
247 i915_program_error(p, "Bad source->Index");
248 return 0;
249 }
250 break;
251
252 case TGSI_FILE_IMMEDIATE:
253 assert(index < p->num_immediates);
254 index = p->immediates_map[index];
255 /* fall-through */
256 case TGSI_FILE_CONSTANT:
257 src = UREG(REG_TYPE_CONST, index);
258 break;
259
260 default:
261 i915_program_error(p, "Bad source->File");
262 return 0;
263 }
264
265 src = swizzle(src,
266 source->Register.SwizzleX,
267 source->Register.SwizzleY,
268 source->Register.SwizzleZ,
269 source->Register.SwizzleW);
270
271 /* There's both negate-all-components and per-component negation.
272 * Try to handle both here.
273 */
274 {
275 int n = source->Register.Negate;
276 src = negate(src, n, n, n, n);
277 }
278
279 /* no abs() */
280 #if 0
281 /* XXX assertions disabled to allow arbfplight.c to run */
282 /* XXX enable these assertions, or fix things */
283 assert(!source->Register.Absolute);
284 #endif
285 if (source->Register.Absolute)
286 debug_printf("Unhandled absolute value\n");
287
288 return src;
289 }
290
291
292 /**
293 * Construct a ureg for a destination register.
294 */
295 static uint
296 get_result_vector(struct i915_fp_compile *p,
297 const struct i915_full_dst_register *dest)
298 {
299 switch (dest->Register.File) {
300 case TGSI_FILE_OUTPUT:
301 {
302 uint sem_name = p->shader->info.output_semantic_name[dest->Register.Index];
303 switch (sem_name) {
304 case TGSI_SEMANTIC_POSITION:
305 return UREG(REG_TYPE_OD, 0);
306 case TGSI_SEMANTIC_COLOR:
307 return UREG(REG_TYPE_OC, 0);
308 default:
309 i915_program_error(p, "Bad inst->DstReg.Index/semantics");
310 return 0;
311 }
312 }
313 case TGSI_FILE_TEMPORARY:
314 return UREG(REG_TYPE_R, dest->Register.Index);
315 default:
316 i915_program_error(p, "Bad inst->DstReg.File");
317 return 0;
318 }
319 }
320
321
322 /**
323 * Compute flags for saturation and writemask.
324 */
325 static uint
326 get_result_flags(const struct i915_full_instruction *inst)
327 {
328 const uint writeMask
329 = inst->Dst[0].Register.WriteMask;
330 uint flags = 0x0;
331
332 if (inst->Instruction.Saturate)
333 flags |= A0_DEST_SATURATE;
334
335 if (writeMask & TGSI_WRITEMASK_X)
336 flags |= A0_DEST_CHANNEL_X;
337 if (writeMask & TGSI_WRITEMASK_Y)
338 flags |= A0_DEST_CHANNEL_Y;
339 if (writeMask & TGSI_WRITEMASK_Z)
340 flags |= A0_DEST_CHANNEL_Z;
341 if (writeMask & TGSI_WRITEMASK_W)
342 flags |= A0_DEST_CHANNEL_W;
343
344 return flags;
345 }
346
347
348 /**
349 * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
350 */
351 static uint
352 translate_tex_src_target(struct i915_fp_compile *p, uint tex)
353 {
354 switch (tex) {
355 case TGSI_TEXTURE_SHADOW1D:
356 /* fall-through */
357 case TGSI_TEXTURE_1D:
358 return D0_SAMPLE_TYPE_2D;
359
360 case TGSI_TEXTURE_SHADOW2D:
361 /* fall-through */
362 case TGSI_TEXTURE_2D:
363 return D0_SAMPLE_TYPE_2D;
364
365 case TGSI_TEXTURE_SHADOWRECT:
366 /* fall-through */
367 case TGSI_TEXTURE_RECT:
368 return D0_SAMPLE_TYPE_2D;
369
370 case TGSI_TEXTURE_3D:
371 return D0_SAMPLE_TYPE_VOLUME;
372
373 case TGSI_TEXTURE_CUBE:
374 return D0_SAMPLE_TYPE_CUBE;
375
376 default:
377 i915_program_error(p, "TexSrc type");
378 return 0;
379 }
380 }
381
382 /**
383 * Return the number of coords needed to access a given TGSI_TEXTURE_*
384 */
385 uint
386 i915_num_coords(uint tex)
387 {
388 switch (tex) {
389 case TGSI_TEXTURE_SHADOW1D:
390 case TGSI_TEXTURE_1D:
391 return 1;
392
393 case TGSI_TEXTURE_SHADOW2D:
394 case TGSI_TEXTURE_2D:
395 case TGSI_TEXTURE_SHADOWRECT:
396 case TGSI_TEXTURE_RECT:
397 return 2;
398
399 case TGSI_TEXTURE_3D:
400 case TGSI_TEXTURE_CUBE:
401 return 3;
402
403 default:
404 debug_printf("Unknown texture target for num coords");
405 return 2;
406 }
407 }
408
409
410 /**
411 * Generate texel lookup instruction.
412 */
413 static void
414 emit_tex(struct i915_fp_compile *p,
415 const struct i915_full_instruction *inst,
416 uint opcode,
417 struct i915_fragment_shader* fs)
418 {
419 uint texture = inst->Texture.Texture;
420 uint unit = inst->Src[1].Register.Index;
421 uint tex = translate_tex_src_target( p, texture );
422 uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex);
423 uint coord = src_vector( p, &inst->Src[0], fs);
424
425 i915_emit_texld( p,
426 get_result_vector( p, &inst->Dst[0] ),
427 get_result_flags( inst ),
428 sampler,
429 coord,
430 opcode,
431 i915_num_coords(texture) );
432 }
433
434
435 /**
436 * Generate a simple arithmetic instruction
437 * \param opcode the i915 opcode
438 * \param numArgs the number of input/src arguments
439 */
440 static void
441 emit_simple_arith(struct i915_fp_compile *p,
442 const struct i915_full_instruction *inst,
443 uint opcode, uint numArgs,
444 struct i915_fragment_shader *fs)
445 {
446 uint arg1, arg2, arg3;
447
448 assert(numArgs <= 3);
449
450 arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->Src[0], fs );
451 arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->Src[1], fs );
452 arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->Src[2], fs );
453
454 i915_emit_arith( p,
455 opcode,
456 get_result_vector( p, &inst->Dst[0]),
457 get_result_flags( inst ), 0,
458 arg1,
459 arg2,
460 arg3 );
461 }
462
463
464 /** As above, but swap the first two src regs */
465 static void
466 emit_simple_arith_swap2(struct i915_fp_compile *p,
467 const struct i915_full_instruction *inst,
468 uint opcode, uint numArgs,
469 struct i915_fragment_shader *fs)
470 {
471 struct i915_full_instruction inst2;
472
473 assert(numArgs == 2);
474
475 /* transpose first two registers */
476 inst2 = *inst;
477 inst2.Src[0] = inst->Src[1];
478 inst2.Src[1] = inst->Src[0];
479
480 emit_simple_arith(p, &inst2, opcode, numArgs, fs);
481 }
482
483 /*
484 * Translate TGSI instruction to i915 instruction.
485 *
486 * Possible concerns:
487 *
488 * DDX, DDY -- return 0
489 * SIN, COS -- could use another taylor step?
490 * LIT -- results seem a little different to sw mesa
491 * LOG -- different to mesa on negative numbers, but this is conformant.
492 */
493 static void
494 i915_translate_instruction(struct i915_fp_compile *p,
495 const struct i915_full_instruction *inst,
496 struct i915_fragment_shader *fs)
497 {
498 uint writemask;
499 uint src0, src1, src2, flags;
500 uint tmp = 0;
501
502 switch (inst->Instruction.Opcode) {
503 case TGSI_OPCODE_ADD:
504 emit_simple_arith(p, inst, A0_ADD, 2, fs);
505 break;
506
507 case TGSI_OPCODE_CEIL:
508 src0 = src_vector(p, &inst->Src[0], fs);
509 tmp = i915_get_utemp(p);
510 flags = get_result_flags(inst);
511 i915_emit_arith(p,
512 A0_FLR,
513 tmp,
514 flags & A0_DEST_CHANNEL_ALL, 0,
515 negate(src0, 1, 1, 1, 1), 0, 0);
516 i915_emit_arith(p,
517 A0_MOV,
518 get_result_vector(p, &inst->Dst[0]),
519 flags, 0,
520 negate(tmp, 1, 1, 1, 1), 0, 0);
521 break;
522
523 case TGSI_OPCODE_CMP:
524 src0 = src_vector(p, &inst->Src[0], fs);
525 src1 = src_vector(p, &inst->Src[1], fs);
526 src2 = src_vector(p, &inst->Src[2], fs);
527 i915_emit_arith(p, A0_CMP,
528 get_result_vector(p, &inst->Dst[0]),
529 get_result_flags(inst),
530 0, src0, src2, src1); /* NOTE: order of src2, src1 */
531 break;
532
533 case TGSI_OPCODE_COS:
534 src0 = src_vector(p, &inst->Src[0], fs);
535 tmp = i915_get_utemp(p);
536
537 i915_emit_arith(p,
538 A0_MUL,
539 tmp, A0_DEST_CHANNEL_X, 0,
540 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
541
542 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
543
544 /*
545 * t0.xy = MUL x.xx11, x.x111 ; x^2, x, 1, 1
546 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
547 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
548 * result = DP4 t0, cos_constants
549 */
550 i915_emit_arith(p,
551 A0_MUL,
552 tmp, A0_DEST_CHANNEL_XY, 0,
553 swizzle(tmp, X, X, ONE, ONE),
554 swizzle(tmp, X, ONE, ONE, ONE), 0);
555
556 i915_emit_arith(p,
557 A0_MUL,
558 tmp, A0_DEST_CHANNEL_XYZ, 0,
559 swizzle(tmp, X, Y, X, ONE),
560 swizzle(tmp, X, X, ONE, ONE), 0);
561
562 i915_emit_arith(p,
563 A0_MUL,
564 tmp, A0_DEST_CHANNEL_XYZ, 0,
565 swizzle(tmp, X, X, Z, ONE),
566 swizzle(tmp, Z, ONE, ONE, ONE), 0);
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(tmp, ONE, Z, Y, X),
573 i915_emit_const4fv(p, cos_constants), 0);
574 break;
575
576 case TGSI_OPCODE_DDX:
577 case TGSI_OPCODE_DDY:
578 /* XXX We just output 0 here */
579 debug_printf("Punting DDX/DDY\n");
580 src0 = get_result_vector(p, &inst->Dst[0]);
581 i915_emit_arith(p,
582 A0_MOV,
583 get_result_vector(p, &inst->Dst[0]),
584 get_result_flags(inst), 0,
585 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0, 0);
586 break;
587
588 case TGSI_OPCODE_DP2:
589 src0 = src_vector(p, &inst->Src[0], fs);
590 src1 = src_vector(p, &inst->Src[1], fs);
591
592 i915_emit_arith(p,
593 A0_DP3,
594 get_result_vector(p, &inst->Dst[0]),
595 get_result_flags(inst), 0,
596 swizzle(src0, X, Y, ZERO, ZERO), src1, 0);
597 break;
598
599 case TGSI_OPCODE_DP3:
600 emit_simple_arith(p, inst, A0_DP3, 2, fs);
601 break;
602
603 case TGSI_OPCODE_DP4:
604 emit_simple_arith(p, inst, A0_DP4, 2, fs);
605 break;
606
607 case TGSI_OPCODE_DST:
608 src0 = src_vector(p, &inst->Src[0], fs);
609 src1 = src_vector(p, &inst->Src[1], fs);
610
611 /* result[0] = 1 * 1;
612 * result[1] = a[1] * b[1];
613 * result[2] = a[2] * 1;
614 * result[3] = 1 * b[3];
615 */
616 i915_emit_arith(p,
617 A0_MUL,
618 get_result_vector(p, &inst->Dst[0]),
619 get_result_flags(inst), 0,
620 swizzle(src0, ONE, Y, Z, ONE),
621 swizzle(src1, ONE, Y, ONE, W), 0);
622 break;
623
624 case TGSI_OPCODE_END:
625 /* no-op */
626 break;
627
628 case TGSI_OPCODE_EX2:
629 src0 = src_vector(p, &inst->Src[0], fs);
630
631 i915_emit_arith(p,
632 A0_EXP,
633 get_result_vector(p, &inst->Dst[0]),
634 get_result_flags(inst), 0,
635 swizzle(src0, X, X, X, X), 0, 0);
636 break;
637
638 case TGSI_OPCODE_FLR:
639 emit_simple_arith(p, inst, A0_FLR, 1, fs);
640 break;
641
642 case TGSI_OPCODE_FRC:
643 emit_simple_arith(p, inst, A0_FRC, 1, fs);
644 break;
645
646 case TGSI_OPCODE_KILL_IF:
647 /* kill if src[0].x < 0 || src[0].y < 0 ... */
648 src0 = src_vector(p, &inst->Src[0], fs);
649 tmp = i915_get_utemp(p);
650
651 i915_emit_texld(p,
652 tmp, /* dest reg: a dummy reg */
653 A0_DEST_CHANNEL_ALL, /* dest writemask */
654 0, /* sampler */
655 src0, /* coord*/
656 T0_TEXKILL, /* opcode */
657 1); /* num_coord */
658 break;
659
660 case TGSI_OPCODE_KILL:
661 /* unconditional kill */
662 tmp = i915_get_utemp(p);
663
664 i915_emit_texld(p,
665 tmp, /* dest reg: a dummy reg */
666 A0_DEST_CHANNEL_ALL, /* dest writemask */
667 0, /* sampler */
668 negate(swizzle(0, ONE, ONE, ONE, ONE), 1, 1, 1, 1), /* coord */
669 T0_TEXKILL, /* opcode */
670 1); /* num_coord */
671 break;
672
673 case TGSI_OPCODE_LG2:
674 src0 = src_vector(p, &inst->Src[0], fs);
675
676 i915_emit_arith(p,
677 A0_LOG,
678 get_result_vector(p, &inst->Dst[0]),
679 get_result_flags(inst), 0,
680 swizzle(src0, X, X, X, X), 0, 0);
681 break;
682
683 case TGSI_OPCODE_LIT:
684 src0 = src_vector(p, &inst->Src[0], fs);
685 tmp = i915_get_utemp(p);
686
687 /* tmp = max( a.xyzw, a.00zw )
688 * XXX: Clamp tmp.w to -128..128
689 * tmp.y = log(tmp.y)
690 * tmp.y = tmp.w * tmp.y
691 * tmp.y = exp(tmp.y)
692 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
693 */
694 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
695 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
696
697 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
698 swizzle(tmp, Y, Y, Y, Y), 0, 0);
699
700 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
701 swizzle(tmp, ZERO, Y, ZERO, ZERO),
702 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
703
704 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
705 swizzle(tmp, Y, Y, Y, Y), 0, 0);
706
707 i915_emit_arith(p, A0_CMP,
708 get_result_vector(p, &inst->Dst[0]),
709 get_result_flags(inst), 0,
710 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
711 swizzle(tmp, ONE, X, ZERO, ONE),
712 swizzle(tmp, ONE, X, Y, ONE));
713
714 break;
715
716 case TGSI_OPCODE_LRP:
717 src0 = src_vector(p, &inst->Src[0], fs);
718 src1 = src_vector(p, &inst->Src[1], fs);
719 src2 = src_vector(p, &inst->Src[2], fs);
720 flags = get_result_flags(inst);
721 tmp = i915_get_utemp(p);
722
723 /* b*a + c*(1-a)
724 *
725 * b*a + c - ca
726 *
727 * tmp = b*a + c,
728 * result = (-c)*a + tmp
729 */
730 i915_emit_arith(p, A0_MAD, tmp,
731 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
732
733 i915_emit_arith(p, A0_MAD,
734 get_result_vector(p, &inst->Dst[0]),
735 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
736 break;
737
738 case TGSI_OPCODE_MAD:
739 emit_simple_arith(p, inst, A0_MAD, 3, fs);
740 break;
741
742 case TGSI_OPCODE_MAX:
743 emit_simple_arith(p, inst, A0_MAX, 2, fs);
744 break;
745
746 case TGSI_OPCODE_MIN:
747 emit_simple_arith(p, inst, A0_MIN, 2, fs);
748 break;
749
750 case TGSI_OPCODE_MOV:
751 emit_simple_arith(p, inst, A0_MOV, 1, fs);
752 break;
753
754 case TGSI_OPCODE_MUL:
755 emit_simple_arith(p, inst, A0_MUL, 2, fs);
756 break;
757
758 case TGSI_OPCODE_NOP:
759 break;
760
761 case TGSI_OPCODE_POW:
762 src0 = src_vector(p, &inst->Src[0], fs);
763 src1 = src_vector(p, &inst->Src[1], fs);
764 tmp = i915_get_utemp(p);
765 flags = get_result_flags(inst);
766
767 /* XXX: masking on intermediate values, here and elsewhere.
768 */
769 i915_emit_arith(p,
770 A0_LOG,
771 tmp, A0_DEST_CHANNEL_X, 0,
772 swizzle(src0, X, X, X, X), 0, 0);
773
774 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
775
776 i915_emit_arith(p,
777 A0_EXP,
778 get_result_vector(p, &inst->Dst[0]),
779 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
780 break;
781
782 case TGSI_OPCODE_RET:
783 /* XXX: no-op? */
784 break;
785
786 case TGSI_OPCODE_RCP:
787 src0 = src_vector(p, &inst->Src[0], fs);
788
789 i915_emit_arith(p,
790 A0_RCP,
791 get_result_vector(p, &inst->Dst[0]),
792 get_result_flags(inst), 0,
793 swizzle(src0, X, X, X, X), 0, 0);
794 break;
795
796 case TGSI_OPCODE_RSQ:
797 src0 = src_vector(p, &inst->Src[0], fs);
798
799 i915_emit_arith(p,
800 A0_RSQ,
801 get_result_vector(p, &inst->Dst[0]),
802 get_result_flags(inst), 0,
803 swizzle(src0, X, X, X, X), 0, 0);
804 break;
805
806 case TGSI_OPCODE_SEQ:
807 /* if we're both >= and <= then we're == */
808 src0 = src_vector(p, &inst->Src[0], fs);
809 src1 = src_vector(p, &inst->Src[1], fs);
810 tmp = i915_get_utemp(p);
811
812 i915_emit_arith(p,
813 A0_SGE,
814 tmp, A0_DEST_CHANNEL_ALL, 0,
815 src0,
816 src1, 0);
817
818 i915_emit_arith(p,
819 A0_SGE,
820 get_result_vector(p, &inst->Dst[0]),
821 A0_DEST_CHANNEL_ALL, 0,
822 src1,
823 src0, 0);
824
825 i915_emit_arith(p,
826 A0_MUL,
827 get_result_vector(p, &inst->Dst[0]),
828 A0_DEST_CHANNEL_ALL, 0,
829 get_result_vector(p, &inst->Dst[0]),
830 tmp, 0);
831
832 break;
833
834 case TGSI_OPCODE_SGE:
835 emit_simple_arith(p, inst, A0_SGE, 2, fs);
836 break;
837
838 case TGSI_OPCODE_SIN:
839 src0 = src_vector(p, &inst->Src[0], fs);
840 tmp = i915_get_utemp(p);
841
842 i915_emit_arith(p,
843 A0_MUL,
844 tmp, A0_DEST_CHANNEL_X, 0,
845 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
846
847 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
848
849 /*
850 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
851 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
852 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
853 * result = DP4 t1.wzyx, sin_constants
854 */
855 i915_emit_arith(p,
856 A0_MUL,
857 tmp, A0_DEST_CHANNEL_XY, 0,
858 swizzle(tmp, X, X, ONE, ONE),
859 swizzle(tmp, X, ONE, ONE, ONE), 0);
860
861 i915_emit_arith(p,
862 A0_MUL,
863 tmp, A0_DEST_CHANNEL_ALL, 0,
864 swizzle(tmp, X, Y, X, Y),
865 swizzle(tmp, X, X, ONE, ONE), 0);
866
867 i915_emit_arith(p,
868 A0_MUL,
869 tmp, A0_DEST_CHANNEL_ALL, 0,
870 swizzle(tmp, X, Y, Y, W),
871 swizzle(tmp, X, Z, ONE, ONE), 0);
872
873 i915_emit_arith(p,
874 A0_DP4,
875 get_result_vector(p, &inst->Dst[0]),
876 get_result_flags(inst), 0,
877 swizzle(tmp, W, Z, Y, X),
878 i915_emit_const4fv(p, sin_constants), 0);
879 break;
880
881 case TGSI_OPCODE_SLE:
882 /* like SGE, but swap reg0, reg1 */
883 emit_simple_arith_swap2(p, inst, A0_SGE, 2, fs);
884 break;
885
886 case TGSI_OPCODE_SLT:
887 emit_simple_arith(p, inst, A0_SLT, 2, fs);
888 break;
889
890 case TGSI_OPCODE_SGT:
891 /* like SLT, but swap reg0, reg1 */
892 emit_simple_arith_swap2(p, inst, A0_SLT, 2, fs);
893 break;
894
895 case TGSI_OPCODE_SNE:
896 /* if we're < or > then we're != */
897 src0 = src_vector(p, &inst->Src[0], fs);
898 src1 = src_vector(p, &inst->Src[1], fs);
899 tmp = i915_get_utemp(p);
900
901 i915_emit_arith(p,
902 A0_SLT,
903 tmp,
904 A0_DEST_CHANNEL_ALL, 0,
905 src0,
906 src1, 0);
907
908 i915_emit_arith(p,
909 A0_SLT,
910 get_result_vector(p, &inst->Dst[0]),
911 A0_DEST_CHANNEL_ALL, 0,
912 src1,
913 src0, 0);
914
915 i915_emit_arith(p,
916 A0_ADD,
917 get_result_vector(p, &inst->Dst[0]),
918 A0_DEST_CHANNEL_ALL, 0,
919 get_result_vector(p, &inst->Dst[0]),
920 tmp, 0);
921 break;
922
923 case TGSI_OPCODE_SSG:
924 /* compute (src>0) - (src<0) */
925 src0 = src_vector(p, &inst->Src[0], fs);
926 tmp = i915_get_utemp(p);
927
928 i915_emit_arith(p,
929 A0_SLT,
930 tmp,
931 A0_DEST_CHANNEL_ALL, 0,
932 src0,
933 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0);
934
935 i915_emit_arith(p,
936 A0_SLT,
937 get_result_vector(p, &inst->Dst[0]),
938 A0_DEST_CHANNEL_ALL, 0,
939 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
940 src0, 0);
941
942 i915_emit_arith(p,
943 A0_ADD,
944 get_result_vector(p, &inst->Dst[0]),
945 A0_DEST_CHANNEL_ALL, 0,
946 get_result_vector(p, &inst->Dst[0]),
947 negate(tmp, 1, 1, 1, 1), 0);
948 break;
949
950 case TGSI_OPCODE_TEX:
951 emit_tex(p, inst, T0_TEXLD, fs);
952 break;
953
954 case TGSI_OPCODE_TRUNC:
955 emit_simple_arith(p, inst, A0_TRC, 1, fs);
956 break;
957
958 case TGSI_OPCODE_TXB:
959 emit_tex(p, inst, T0_TEXLDB, fs);
960 break;
961
962 case TGSI_OPCODE_TXP:
963 emit_tex(p, inst, T0_TEXLDP, fs);
964 break;
965
966 default:
967 i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
968 p->error = 1;
969 return;
970 }
971
972 i915_release_utemps(p);
973 }
974
975
976 static void i915_translate_token(struct i915_fp_compile *p,
977 const union i915_full_token *token,
978 struct i915_fragment_shader *fs)
979 {
980 struct i915_fragment_shader *ifs = p->shader;
981 switch( token->Token.Type ) {
982 case TGSI_TOKEN_TYPE_PROPERTY:
983 /*
984 * We only support one cbuf, but we still need to ignore the property
985 * correctly so we don't hit the assert at the end of the switch case.
986 */
987 assert(token->FullProperty.Property.PropertyName ==
988 TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS);
989 break;
990
991 case TGSI_TOKEN_TYPE_DECLARATION:
992 if (token->FullDeclaration.Declaration.File
993 == TGSI_FILE_CONSTANT) {
994 uint i;
995 for (i = token->FullDeclaration.Range.First;
996 i <= MIN2(token->FullDeclaration.Range.Last, I915_MAX_CONSTANT - 1);
997 i++) {
998 assert(ifs->constant_flags[i] == 0x0);
999 ifs->constant_flags[i] = I915_CONSTFLAG_USER;
1000 ifs->num_constants = MAX2(ifs->num_constants, i + 1);
1001 }
1002 }
1003 else if (token->FullDeclaration.Declaration.File
1004 == TGSI_FILE_TEMPORARY) {
1005 uint i;
1006 for (i = token->FullDeclaration.Range.First;
1007 i <= token->FullDeclaration.Range.Last;
1008 i++) {
1009 if (i >= I915_MAX_TEMPORARY)
1010 debug_printf("Too many temps (%d)\n",i);
1011 else
1012 /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */
1013 p->temp_flag |= (1 << i); /* mark temp as used */
1014 }
1015 }
1016 break;
1017
1018 case TGSI_TOKEN_TYPE_IMMEDIATE:
1019 {
1020 const struct tgsi_full_immediate *imm
1021 = &token->FullImmediate;
1022 const uint pos = p->num_immediates++;
1023 uint j;
1024 assert( imm->Immediate.NrTokens <= 4 + 1 );
1025 for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
1026 p->immediates[pos][j] = imm->u[j].Float;
1027 }
1028 }
1029 break;
1030
1031 case TGSI_TOKEN_TYPE_INSTRUCTION:
1032 if (p->first_instruction) {
1033 /* resolve location of immediates */
1034 uint i, j;
1035 for (i = 0; i < p->num_immediates; i++) {
1036 /* find constant slot for this immediate */
1037 for (j = 0; j < I915_MAX_CONSTANT; j++) {
1038 if (ifs->constant_flags[j] == 0x0) {
1039 memcpy(ifs->constants[j],
1040 p->immediates[i],
1041 4 * sizeof(float));
1042 /*printf("immediate %d maps to const %d\n", i, j);*/
1043 ifs->constant_flags[j] = 0xf; /* all four comps used */
1044 p->immediates_map[i] = j;
1045 ifs->num_constants = MAX2(ifs->num_constants, j + 1);
1046 break;
1047 }
1048 }
1049 }
1050
1051 p->first_instruction = FALSE;
1052 }
1053
1054 i915_translate_instruction(p, &token->FullInstruction, fs);
1055 break;
1056
1057 default:
1058 assert( 0 );
1059 }
1060
1061 }
1062
1063 /**
1064 * Translate TGSI fragment shader into i915 hardware instructions.
1065 * \param p the translation state
1066 * \param tokens the TGSI token array
1067 */
1068 static void
1069 i915_translate_instructions(struct i915_fp_compile *p,
1070 const struct i915_token_list *tokens,
1071 struct i915_fragment_shader *fs)
1072 {
1073 int i;
1074 for(i = 0; i<tokens->NumTokens; i++) {
1075 i915_translate_token(p, &tokens->Tokens[i], fs);
1076 }
1077 }
1078
1079
1080 static struct i915_fp_compile *
1081 i915_init_compile(struct i915_context *i915,
1082 struct i915_fragment_shader *ifs)
1083 {
1084 struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
1085 int i;
1086
1087 p->shader = ifs;
1088
1089 /* Put new constants at end of const buffer, growing downward.
1090 * The problem is we don't know how many user-defined constants might
1091 * be specified with pipe->set_constant_buffer().
1092 * Should pre-scan the user's program to determine the highest-numbered
1093 * constant referenced.
1094 */
1095 ifs->num_constants = 0;
1096 memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags));
1097
1098 memset(&p->register_phases, 0, sizeof(p->register_phases));
1099
1100 for (i = 0; i < I915_TEX_UNITS; i++)
1101 ifs->generic_mapping[i] = -1;
1102
1103 p->first_instruction = TRUE;
1104
1105 p->nr_tex_indirect = 1; /* correct? */
1106 p->nr_tex_insn = 0;
1107 p->nr_alu_insn = 0;
1108 p->nr_decl_insn = 0;
1109
1110 p->csr = p->program;
1111 p->decl = p->declarations;
1112 p->decl_s = 0;
1113 p->decl_t = 0;
1114 p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
1115 p->utemp_flag = ~0x7;
1116
1117 /* initialize the first program word */
1118 *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
1119
1120 return p;
1121 }
1122
1123
1124 /* Copy compile results to the fragment program struct and destroy the
1125 * compilation context.
1126 */
1127 static void
1128 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
1129 {
1130 struct i915_fragment_shader *ifs = p->shader;
1131 unsigned long program_size = (unsigned long) (p->csr - p->program);
1132 unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
1133
1134 if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
1135 debug_printf("Exceeded max nr indirect texture lookups\n");
1136
1137 if (p->nr_tex_insn > I915_MAX_TEX_INSN)
1138 i915_program_error(p, "Exceeded max TEX instructions");
1139
1140 if (p->nr_alu_insn > I915_MAX_ALU_INSN)
1141 i915_program_error(p, "Exceeded max ALU instructions");
1142
1143 if (p->nr_decl_insn > I915_MAX_DECL_INSN)
1144 i915_program_error(p, "Exceeded max DECL instructions");
1145
1146 if (p->error) {
1147 p->NumNativeInstructions = 0;
1148 p->NumNativeAluInstructions = 0;
1149 p->NumNativeTexInstructions = 0;
1150 p->NumNativeTexIndirections = 0;
1151
1152 i915_use_passthrough_shader(ifs);
1153 }
1154 else {
1155 p->NumNativeInstructions
1156 = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
1157 p->NumNativeAluInstructions = p->nr_alu_insn;
1158 p->NumNativeTexInstructions = p->nr_tex_insn;
1159 p->NumNativeTexIndirections = p->nr_tex_indirect;
1160
1161 /* patch in the program length */
1162 p->declarations[0] |= program_size + decl_size - 2;
1163
1164 /* Copy compilation results to fragment program struct:
1165 */
1166 assert(!ifs->decl);
1167 assert(!ifs->program);
1168
1169 ifs->decl
1170 = (uint *) MALLOC(decl_size * sizeof(uint));
1171 ifs->program
1172 = (uint *) MALLOC(program_size * sizeof(uint));
1173
1174 if (ifs->decl) {
1175 ifs->decl_len = decl_size;
1176
1177 memcpy(ifs->decl,
1178 p->declarations,
1179 decl_size * sizeof(uint));
1180 }
1181
1182 if (ifs->program) {
1183 ifs->program_len = program_size;
1184
1185 memcpy(ifs->program,
1186 p->program,
1187 program_size * sizeof(uint));
1188 }
1189 }
1190
1191 /* Release the compilation struct:
1192 */
1193 FREE(p);
1194 }
1195
1196
1197
1198
1199
1200 /**
1201 * Rather than trying to intercept and jiggle depth writes during
1202 * emit, just move the value into its correct position at the end of
1203 * the program:
1204 */
1205 static void
1206 i915_fixup_depth_write(struct i915_fp_compile *p)
1207 {
1208 /* XXX assuming pos/depth is always in output[0] */
1209 if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1210 const uint depth = UREG(REG_TYPE_OD, 0);
1211
1212 i915_emit_arith(p,
1213 A0_MOV, /* opcode */
1214 depth, /* dest reg */
1215 A0_DEST_CHANNEL_W, /* write mask */
1216 0, /* saturate? */
1217 swizzle(depth, X, Y, Z, Z), /* src0 */
1218 0, 0 /* src1, src2 */);
1219 }
1220 }
1221
1222
1223 void
1224 i915_translate_fragment_program( struct i915_context *i915,
1225 struct i915_fragment_shader *fs)
1226 {
1227 struct i915_fp_compile *p;
1228 const struct tgsi_token *tokens = fs->state.tokens;
1229 struct i915_token_list* i_tokens;
1230
1231 #if 0
1232 tgsi_dump(tokens, 0);
1233 #endif
1234
1235 /* hw doesn't seem to like empty frag programs, even when the depth write
1236 * fixup gets emitted below - may that one is fishy, too? */
1237 if (fs->info.num_instructions == 1) {
1238 i915_use_passthrough_shader(fs);
1239
1240 return;
1241 }
1242
1243 p = i915_init_compile(i915, fs);
1244
1245 i_tokens = i915_optimize(tokens);
1246 i915_translate_instructions(p, i_tokens, fs);
1247 i915_fixup_depth_write(p);
1248
1249 i915_fini_compile(i915, p);
1250 i915_optimize_free(i_tokens);
1251
1252 #if 0
1253 i915_disassemble_program(NULL, fs->program, fs->program_len);
1254 #endif
1255 }