projects / mesa.git / blob
? search:


4f5fdd84e66fdb126ad85690277af28d449dd7fe
[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 = Elements(passthrough_program);
138 fs->decl_len = Elements(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 == TGSI_SAT_ZERO_ONE)
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_ABS:
504 src0 = src_vector(p, &inst->Src[0], fs);
505 i915_emit_arith(p,
506 A0_MAX,
507 get_result_vector(p, &inst->Dst[0]),
508 get_result_flags(inst), 0,
509 src0, negate(src0, 1, 1, 1, 1), 0);
510 break;
511
512 case TGSI_OPCODE_ADD:
513 emit_simple_arith(p, inst, A0_ADD, 2, fs);
514 break;
515
516 case TGSI_OPCODE_CEIL:
517 src0 = src_vector(p, &inst->Src[0], fs);
518 tmp = i915_get_utemp(p);
519 flags = get_result_flags(inst);
520 i915_emit_arith(p,
521 A0_FLR,
522 tmp,
523 flags & A0_DEST_CHANNEL_ALL, 0,
524 negate(src0, 1, 1, 1, 1), 0, 0);
525 i915_emit_arith(p,
526 A0_MOV,
527 get_result_vector(p, &inst->Dst[0]),
528 flags, 0,
529 negate(tmp, 1, 1, 1, 1), 0, 0);
530 break;
531
532 case TGSI_OPCODE_CMP:
533 src0 = src_vector(p, &inst->Src[0], fs);
534 src1 = src_vector(p, &inst->Src[1], fs);
535 src2 = src_vector(p, &inst->Src[2], fs);
536 i915_emit_arith(p, A0_CMP,
537 get_result_vector(p, &inst->Dst[0]),
538 get_result_flags(inst),
539 0, src0, src2, src1); /* NOTE: order of src2, src1 */
540 break;
541
542 case TGSI_OPCODE_COS:
543 src0 = src_vector(p, &inst->Src[0], fs);
544 tmp = i915_get_utemp(p);
545
546 i915_emit_arith(p,
547 A0_MUL,
548 tmp, A0_DEST_CHANNEL_X, 0,
549 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
550
551 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
552
553 /*
554 * t0.xy = MUL x.xx11, x.x111 ; x^2, x, 1, 1
555 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
556 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
557 * result = DP4 t0, cos_constants
558 */
559 i915_emit_arith(p,
560 A0_MUL,
561 tmp, A0_DEST_CHANNEL_XY, 0,
562 swizzle(tmp, X, X, ONE, ONE),
563 swizzle(tmp, X, ONE, ONE, ONE), 0);
564
565 i915_emit_arith(p,
566 A0_MUL,
567 tmp, A0_DEST_CHANNEL_XYZ, 0,
568 swizzle(tmp, X, Y, X, ONE),
569 swizzle(tmp, X, X, ONE, ONE), 0);
570
571 i915_emit_arith(p,
572 A0_MUL,
573 tmp, A0_DEST_CHANNEL_XYZ, 0,
574 swizzle(tmp, X, X, Z, ONE),
575 swizzle(tmp, Z, ONE, ONE, ONE), 0);
576
577 i915_emit_arith(p,
578 A0_DP4,
579 get_result_vector(p, &inst->Dst[0]),
580 get_result_flags(inst), 0,
581 swizzle(tmp, ONE, Z, Y, X),
582 i915_emit_const4fv(p, cos_constants), 0);
583 break;
584
585 case TGSI_OPCODE_DDX:
586 case TGSI_OPCODE_DDY:
587 /* XXX We just output 0 here */
588 debug_printf("Punting DDX/DDX\n");
589 src0 = get_result_vector(p, &inst->Dst[0]);
590 i915_emit_arith(p,
591 A0_MOV,
592 get_result_vector(p, &inst->Dst[0]),
593 get_result_flags(inst), 0,
594 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0, 0);
595 break;
596
597 case TGSI_OPCODE_DP2:
598 src0 = src_vector(p, &inst->Src[0], fs);
599 src1 = src_vector(p, &inst->Src[1], fs);
600
601 i915_emit_arith(p,
602 A0_DP3,
603 get_result_vector(p, &inst->Dst[0]),
604 get_result_flags(inst), 0,
605 swizzle(src0, X, Y, ZERO, ZERO), src1, 0);
606 break;
607
608 case TGSI_OPCODE_DP3:
609 emit_simple_arith(p, inst, A0_DP3, 2, fs);
610 break;
611
612 case TGSI_OPCODE_DP4:
613 emit_simple_arith(p, inst, A0_DP4, 2, fs);
614 break;
615
616 case TGSI_OPCODE_DPH:
617 src0 = src_vector(p, &inst->Src[0], fs);
618 src1 = src_vector(p, &inst->Src[1], fs);
619
620 i915_emit_arith(p,
621 A0_DP4,
622 get_result_vector(p, &inst->Dst[0]),
623 get_result_flags(inst), 0,
624 swizzle(src0, X, Y, Z, ONE), src1, 0);
625 break;
626
627 case TGSI_OPCODE_DST:
628 src0 = src_vector(p, &inst->Src[0], fs);
629 src1 = src_vector(p, &inst->Src[1], fs);
630
631 /* result[0] = 1 * 1;
632 * result[1] = a[1] * b[1];
633 * result[2] = a[2] * 1;
634 * result[3] = 1 * b[3];
635 */
636 i915_emit_arith(p,
637 A0_MUL,
638 get_result_vector(p, &inst->Dst[0]),
639 get_result_flags(inst), 0,
640 swizzle(src0, ONE, Y, Z, ONE),
641 swizzle(src1, ONE, Y, ONE, W), 0);
642 break;
643
644 case TGSI_OPCODE_END:
645 /* no-op */
646 break;
647
648 case TGSI_OPCODE_EX2:
649 src0 = src_vector(p, &inst->Src[0], fs);
650
651 i915_emit_arith(p,
652 A0_EXP,
653 get_result_vector(p, &inst->Dst[0]),
654 get_result_flags(inst), 0,
655 swizzle(src0, X, X, X, X), 0, 0);
656 break;
657
658 case TGSI_OPCODE_FLR:
659 emit_simple_arith(p, inst, A0_FLR, 1, fs);
660 break;
661
662 case TGSI_OPCODE_FRC:
663 emit_simple_arith(p, inst, A0_FRC, 1, fs);
664 break;
665
666 case TGSI_OPCODE_KILL_IF:
667 /* kill if src[0].x < 0 || src[0].y < 0 ... */
668 src0 = src_vector(p, &inst->Src[0], fs);
669 tmp = i915_get_utemp(p);
670
671 i915_emit_texld(p,
672 tmp, /* dest reg: a dummy reg */
673 A0_DEST_CHANNEL_ALL, /* dest writemask */
674 0, /* sampler */
675 src0, /* coord*/
676 T0_TEXKILL, /* opcode */
677 1); /* num_coord */
678 break;
679
680 case TGSI_OPCODE_KILL:
681 /* unconditional kill */
682 tmp = i915_get_utemp(p);
683
684 i915_emit_texld(p,
685 tmp, /* dest reg: a dummy reg */
686 A0_DEST_CHANNEL_ALL, /* dest writemask */
687 0, /* sampler */
688 negate(swizzle(0, ONE, ONE, ONE, ONE), 1, 1, 1, 1), /* coord */
689 T0_TEXKILL, /* opcode */
690 1); /* num_coord */
691 break;
692
693 case TGSI_OPCODE_LG2:
694 src0 = src_vector(p, &inst->Src[0], fs);
695
696 i915_emit_arith(p,
697 A0_LOG,
698 get_result_vector(p, &inst->Dst[0]),
699 get_result_flags(inst), 0,
700 swizzle(src0, X, X, X, X), 0, 0);
701 break;
702
703 case TGSI_OPCODE_LIT:
704 src0 = src_vector(p, &inst->Src[0], fs);
705 tmp = i915_get_utemp(p);
706
707 /* tmp = max( a.xyzw, a.00zw )
708 * XXX: Clamp tmp.w to -128..128
709 * tmp.y = log(tmp.y)
710 * tmp.y = tmp.w * tmp.y
711 * tmp.y = exp(tmp.y)
712 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
713 */
714 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
715 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
716
717 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
718 swizzle(tmp, Y, Y, Y, Y), 0, 0);
719
720 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
721 swizzle(tmp, ZERO, Y, ZERO, ZERO),
722 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
723
724 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
725 swizzle(tmp, Y, Y, Y, Y), 0, 0);
726
727 i915_emit_arith(p, A0_CMP,
728 get_result_vector(p, &inst->Dst[0]),
729 get_result_flags(inst), 0,
730 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
731 swizzle(tmp, ONE, X, ZERO, ONE),
732 swizzle(tmp, ONE, X, Y, ONE));
733
734 break;
735
736 case TGSI_OPCODE_LRP:
737 src0 = src_vector(p, &inst->Src[0], fs);
738 src1 = src_vector(p, &inst->Src[1], fs);
739 src2 = src_vector(p, &inst->Src[2], fs);
740 flags = get_result_flags(inst);
741 tmp = i915_get_utemp(p);
742
743 /* b*a + c*(1-a)
744 *
745 * b*a + c - ca
746 *
747 * tmp = b*a + c,
748 * result = (-c)*a + tmp
749 */
750 i915_emit_arith(p, A0_MAD, tmp,
751 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
752
753 i915_emit_arith(p, A0_MAD,
754 get_result_vector(p, &inst->Dst[0]),
755 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
756 break;
757
758 case TGSI_OPCODE_MAD:
759 emit_simple_arith(p, inst, A0_MAD, 3, fs);
760 break;
761
762 case TGSI_OPCODE_MAX:
763 emit_simple_arith(p, inst, A0_MAX, 2, fs);
764 break;
765
766 case TGSI_OPCODE_MIN:
767 src0 = src_vector(p, &inst->Src[0], fs);
768 src1 = src_vector(p, &inst->Src[1], fs);
769 tmp = i915_get_utemp(p);
770 flags = get_result_flags(inst);
771
772 i915_emit_arith(p,
773 A0_MAX,
774 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
775 negate(src0, 1, 1, 1, 1),
776 negate(src1, 1, 1, 1, 1), 0);
777
778 i915_emit_arith(p,
779 A0_MOV,
780 get_result_vector(p, &inst->Dst[0]),
781 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
782 break;
783
784 case TGSI_OPCODE_MOV:
785 emit_simple_arith(p, inst, A0_MOV, 1, fs);
786 break;
787
788 case TGSI_OPCODE_MUL:
789 emit_simple_arith(p, inst, A0_MUL, 2, fs);
790 break;
791
792 case TGSI_OPCODE_NOP:
793 break;
794
795 case TGSI_OPCODE_POW:
796 src0 = src_vector(p, &inst->Src[0], fs);
797 src1 = src_vector(p, &inst->Src[1], fs);
798 tmp = i915_get_utemp(p);
799 flags = get_result_flags(inst);
800
801 /* XXX: masking on intermediate values, here and elsewhere.
802 */
803 i915_emit_arith(p,
804 A0_LOG,
805 tmp, A0_DEST_CHANNEL_X, 0,
806 swizzle(src0, X, X, X, X), 0, 0);
807
808 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
809
810 i915_emit_arith(p,
811 A0_EXP,
812 get_result_vector(p, &inst->Dst[0]),
813 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
814 break;
815
816 case TGSI_OPCODE_RET:
817 /* XXX: no-op? */
818 break;
819
820 case TGSI_OPCODE_RCP:
821 src0 = src_vector(p, &inst->Src[0], fs);
822
823 i915_emit_arith(p,
824 A0_RCP,
825 get_result_vector(p, &inst->Dst[0]),
826 get_result_flags(inst), 0,
827 swizzle(src0, X, X, X, X), 0, 0);
828 break;
829
830 case TGSI_OPCODE_RSQ:
831 src0 = src_vector(p, &inst->Src[0], fs);
832
833 i915_emit_arith(p,
834 A0_RSQ,
835 get_result_vector(p, &inst->Dst[0]),
836 get_result_flags(inst), 0,
837 swizzle(src0, X, X, X, X), 0, 0);
838 break;
839
840 case TGSI_OPCODE_SCS:
841 src0 = src_vector(p, &inst->Src[0], fs);
842 tmp = i915_get_utemp(p);
843
844 /*
845 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
846 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
847 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
848 * scs.x = DP4 t1, scs_sin_constants
849 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
850 * scs.y = DP4 t1, scs_cos_constants
851 */
852 i915_emit_arith(p,
853 A0_MUL,
854 tmp, A0_DEST_CHANNEL_XY, 0,
855 swizzle(src0, X, X, ONE, ONE),
856 swizzle(src0, X, ONE, ONE, ONE), 0);
857
858 i915_emit_arith(p,
859 A0_MUL,
860 tmp, A0_DEST_CHANNEL_ALL, 0,
861 swizzle(tmp, X, Y, X, Y),
862 swizzle(tmp, X, X, ONE, ONE), 0);
863
864 writemask = inst->Dst[0].Register.WriteMask;
865
866 if (writemask & TGSI_WRITEMASK_Y) {
867 uint tmp1;
868
869 if (writemask & TGSI_WRITEMASK_X)
870 tmp1 = i915_get_utemp(p);
871 else
872 tmp1 = tmp;
873
874 i915_emit_arith(p,
875 A0_MUL,
876 tmp1, A0_DEST_CHANNEL_ALL, 0,
877 swizzle(tmp, X, Y, Y, W),
878 swizzle(tmp, X, Z, ONE, ONE), 0);
879
880 i915_emit_arith(p,
881 A0_DP4,
882 get_result_vector(p, &inst->Dst[0]),
883 A0_DEST_CHANNEL_Y, 0,
884 swizzle(tmp1, W, Z, Y, X),
885 i915_emit_const4fv(p, scs_sin_constants), 0);
886 }
887
888 if (writemask & TGSI_WRITEMASK_X) {
889 i915_emit_arith(p,
890 A0_MUL,
891 tmp, A0_DEST_CHANNEL_XYZ, 0,
892 swizzle(tmp, X, X, Z, ONE),
893 swizzle(tmp, Z, ONE, ONE, ONE), 0);
894
895 i915_emit_arith(p,
896 A0_DP4,
897 get_result_vector(p, &inst->Dst[0]),
898 A0_DEST_CHANNEL_X, 0,
899 swizzle(tmp, ONE, Z, Y, X),
900 i915_emit_const4fv(p, scs_cos_constants), 0);
901 }
902 break;
903
904 case TGSI_OPCODE_SEQ:
905 /* if we're both >= and <= then we're == */
906 src0 = src_vector(p, &inst->Src[0], fs);
907 src1 = src_vector(p, &inst->Src[1], fs);
908 tmp = i915_get_utemp(p);
909
910 i915_emit_arith(p,
911 A0_SGE,
912 tmp, A0_DEST_CHANNEL_ALL, 0,
913 src0,
914 src1, 0);
915
916 i915_emit_arith(p,
917 A0_SGE,
918 get_result_vector(p, &inst->Dst[0]),
919 A0_DEST_CHANNEL_ALL, 0,
920 src1,
921 src0, 0);
922
923 i915_emit_arith(p,
924 A0_MUL,
925 get_result_vector(p, &inst->Dst[0]),
926 A0_DEST_CHANNEL_ALL, 0,
927 get_result_vector(p, &inst->Dst[0]),
928 tmp, 0);
929
930 break;
931
932 case TGSI_OPCODE_SGE:
933 emit_simple_arith(p, inst, A0_SGE, 2, fs);
934 break;
935
936 case TGSI_OPCODE_SIN:
937 src0 = src_vector(p, &inst->Src[0], fs);
938 tmp = i915_get_utemp(p);
939
940 i915_emit_arith(p,
941 A0_MUL,
942 tmp, A0_DEST_CHANNEL_X, 0,
943 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
944
945 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
946
947 /*
948 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
949 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
950 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
951 * result = DP4 t1.wzyx, sin_constants
952 */
953 i915_emit_arith(p,
954 A0_MUL,
955 tmp, A0_DEST_CHANNEL_XY, 0,
956 swizzle(tmp, X, X, ONE, ONE),
957 swizzle(tmp, X, ONE, ONE, ONE), 0);
958
959 i915_emit_arith(p,
960 A0_MUL,
961 tmp, A0_DEST_CHANNEL_ALL, 0,
962 swizzle(tmp, X, Y, X, Y),
963 swizzle(tmp, X, X, ONE, ONE), 0);
964
965 i915_emit_arith(p,
966 A0_MUL,
967 tmp, A0_DEST_CHANNEL_ALL, 0,
968 swizzle(tmp, X, Y, Y, W),
969 swizzle(tmp, X, Z, ONE, ONE), 0);
970
971 i915_emit_arith(p,
972 A0_DP4,
973 get_result_vector(p, &inst->Dst[0]),
974 get_result_flags(inst), 0,
975 swizzle(tmp, W, Z, Y, X),
976 i915_emit_const4fv(p, sin_constants), 0);
977 break;
978
979 case TGSI_OPCODE_SLE:
980 /* like SGE, but swap reg0, reg1 */
981 emit_simple_arith_swap2(p, inst, A0_SGE, 2, fs);
982 break;
983
984 case TGSI_OPCODE_SLT:
985 emit_simple_arith(p, inst, A0_SLT, 2, fs);
986 break;
987
988 case TGSI_OPCODE_SGT:
989 /* like SLT, but swap reg0, reg1 */
990 emit_simple_arith_swap2(p, inst, A0_SLT, 2, fs);
991 break;
992
993 case TGSI_OPCODE_SNE:
994 /* if we're < or > then we're != */
995 src0 = src_vector(p, &inst->Src[0], fs);
996 src1 = src_vector(p, &inst->Src[1], fs);
997 tmp = i915_get_utemp(p);
998
999 i915_emit_arith(p,
1000 A0_SLT,
1001 tmp,
1002 A0_DEST_CHANNEL_ALL, 0,
1003 src0,
1004 src1, 0);
1005
1006 i915_emit_arith(p,
1007 A0_SLT,
1008 get_result_vector(p, &inst->Dst[0]),
1009 A0_DEST_CHANNEL_ALL, 0,
1010 src1,
1011 src0, 0);
1012
1013 i915_emit_arith(p,
1014 A0_ADD,
1015 get_result_vector(p, &inst->Dst[0]),
1016 A0_DEST_CHANNEL_ALL, 0,
1017 get_result_vector(p, &inst->Dst[0]),
1018 tmp, 0);
1019 break;
1020
1021 case TGSI_OPCODE_SSG:
1022 /* compute (src>0) - (src<0) */
1023 src0 = src_vector(p, &inst->Src[0], fs);
1024 tmp = i915_get_utemp(p);
1025
1026 i915_emit_arith(p,
1027 A0_SLT,
1028 tmp,
1029 A0_DEST_CHANNEL_ALL, 0,
1030 src0,
1031 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0);
1032
1033 i915_emit_arith(p,
1034 A0_SLT,
1035 get_result_vector(p, &inst->Dst[0]),
1036 A0_DEST_CHANNEL_ALL, 0,
1037 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1038 src0, 0);
1039
1040 i915_emit_arith(p,
1041 A0_ADD,
1042 get_result_vector(p, &inst->Dst[0]),
1043 A0_DEST_CHANNEL_ALL, 0,
1044 get_result_vector(p, &inst->Dst[0]),
1045 negate(tmp, 1, 1, 1, 1), 0);
1046 break;
1047
1048 case TGSI_OPCODE_SUB:
1049 src0 = src_vector(p, &inst->Src[0], fs);
1050 src1 = src_vector(p, &inst->Src[1], fs);
1051
1052 i915_emit_arith(p,
1053 A0_ADD,
1054 get_result_vector(p, &inst->Dst[0]),
1055 get_result_flags(inst), 0,
1056 src0, negate(src1, 1, 1, 1, 1), 0);
1057 break;
1058
1059 case TGSI_OPCODE_TEX:
1060 emit_tex(p, inst, T0_TEXLD, fs);
1061 break;
1062
1063 case TGSI_OPCODE_TRUNC:
1064 emit_simple_arith(p, inst, A0_TRC, 1, fs);
1065 break;
1066
1067 case TGSI_OPCODE_TXB:
1068 emit_tex(p, inst, T0_TEXLDB, fs);
1069 break;
1070
1071 case TGSI_OPCODE_TXP:
1072 emit_tex(p, inst, T0_TEXLDP, fs);
1073 break;
1074
1075 case TGSI_OPCODE_XPD:
1076 /* Cross product:
1077 * result.x = src0.y * src1.z - src0.z * src1.y;
1078 * result.y = src0.z * src1.x - src0.x * src1.z;
1079 * result.z = src0.x * src1.y - src0.y * src1.x;
1080 * result.w = undef;
1081 */
1082 src0 = src_vector(p, &inst->Src[0], fs);
1083 src1 = src_vector(p, &inst->Src[1], fs);
1084 tmp = i915_get_utemp(p);
1085
1086 i915_emit_arith(p,
1087 A0_MUL,
1088 tmp, A0_DEST_CHANNEL_ALL, 0,
1089 swizzle(src0, Z, X, Y, ONE),
1090 swizzle(src1, Y, Z, X, ONE), 0);
1091
1092 i915_emit_arith(p,
1093 A0_MAD,
1094 get_result_vector(p, &inst->Dst[0]),
1095 get_result_flags(inst), 0,
1096 swizzle(src0, Y, Z, X, ONE),
1097 swizzle(src1, Z, X, Y, ONE),
1098 negate(tmp, 1, 1, 1, 0));
1099 break;
1100
1101 default:
1102 i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
1103 p->error = 1;
1104 return;
1105 }
1106
1107 i915_release_utemps(p);
1108 }
1109
1110
1111 static void i915_translate_token(struct i915_fp_compile *p,
1112 const union i915_full_token *token,
1113 struct i915_fragment_shader *fs)
1114 {
1115 struct i915_fragment_shader *ifs = p->shader;
1116 switch( token->Token.Type ) {
1117 case TGSI_TOKEN_TYPE_PROPERTY:
1118 /*
1119 * We only support one cbuf, but we still need to ignore the property
1120 * correctly so we don't hit the assert at the end of the switch case.
1121 */
1122 assert(token->FullProperty.Property.PropertyName ==
1123 TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS);
1124 break;
1125
1126 case TGSI_TOKEN_TYPE_DECLARATION:
1127 if (token->FullDeclaration.Declaration.File
1128 == TGSI_FILE_CONSTANT) {
1129 uint i;
1130 for (i = token->FullDeclaration.Range.First;
1131 i <= MIN2(token->FullDeclaration.Range.Last, I915_MAX_CONSTANT - 1);
1132 i++) {
1133 assert(ifs->constant_flags[i] == 0x0);
1134 ifs->constant_flags[i] = I915_CONSTFLAG_USER;
1135 ifs->num_constants = MAX2(ifs->num_constants, i + 1);
1136 }
1137 }
1138 else if (token->FullDeclaration.Declaration.File
1139 == TGSI_FILE_TEMPORARY) {
1140 uint i;
1141 for (i = token->FullDeclaration.Range.First;
1142 i <= token->FullDeclaration.Range.Last;
1143 i++) {
1144 if (i >= I915_MAX_TEMPORARY)
1145 debug_printf("Too many temps (%d)\n",i);
1146 else
1147 /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */
1148 p->temp_flag |= (1 << i); /* mark temp as used */
1149 }
1150 }
1151 break;
1152
1153 case TGSI_TOKEN_TYPE_IMMEDIATE:
1154 {
1155 const struct tgsi_full_immediate *imm
1156 = &token->FullImmediate;
1157 const uint pos = p->num_immediates++;
1158 uint j;
1159 assert( imm->Immediate.NrTokens <= 4 + 1 );
1160 for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
1161 p->immediates[pos][j] = imm->u[j].Float;
1162 }
1163 }
1164 break;
1165
1166 case TGSI_TOKEN_TYPE_INSTRUCTION:
1167 if (p->first_instruction) {
1168 /* resolve location of immediates */
1169 uint i, j;
1170 for (i = 0; i < p->num_immediates; i++) {
1171 /* find constant slot for this immediate */
1172 for (j = 0; j < I915_MAX_CONSTANT; j++) {
1173 if (ifs->constant_flags[j] == 0x0) {
1174 memcpy(ifs->constants[j],
1175 p->immediates[i],
1176 4 * sizeof(float));
1177 /*printf("immediate %d maps to const %d\n", i, j);*/
1178 ifs->constant_flags[j] = 0xf; /* all four comps used */
1179 p->immediates_map[i] = j;
1180 ifs->num_constants = MAX2(ifs->num_constants, j + 1);
1181 break;
1182 }
1183 }
1184 }
1185
1186 p->first_instruction = FALSE;
1187 }
1188
1189 i915_translate_instruction(p, &token->FullInstruction, fs);
1190 break;
1191
1192 default:
1193 assert( 0 );
1194 }
1195
1196 }
1197
1198 /**
1199 * Translate TGSI fragment shader into i915 hardware instructions.
1200 * \param p the translation state
1201 * \param tokens the TGSI token array
1202 */
1203 static void
1204 i915_translate_instructions(struct i915_fp_compile *p,
1205 const struct i915_token_list *tokens,
1206 struct i915_fragment_shader *fs)
1207 {
1208 int i;
1209 for(i = 0; i<tokens->NumTokens; i++) {
1210 i915_translate_token(p, &tokens->Tokens[i], fs);
1211 }
1212 }
1213
1214
1215 static struct i915_fp_compile *
1216 i915_init_compile(struct i915_context *i915,
1217 struct i915_fragment_shader *ifs)
1218 {
1219 struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
1220 int i;
1221
1222 p->shader = ifs;
1223
1224 /* Put new constants at end of const buffer, growing downward.
1225 * The problem is we don't know how many user-defined constants might
1226 * be specified with pipe->set_constant_buffer().
1227 * Should pre-scan the user's program to determine the highest-numbered
1228 * constant referenced.
1229 */
1230 ifs->num_constants = 0;
1231 memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags));
1232
1233 memset(&p->register_phases, 0, sizeof(p->register_phases));
1234
1235 for (i = 0; i < I915_TEX_UNITS; i++)
1236 ifs->generic_mapping[i] = -1;
1237
1238 p->first_instruction = TRUE;
1239
1240 p->nr_tex_indirect = 1; /* correct? */
1241 p->nr_tex_insn = 0;
1242 p->nr_alu_insn = 0;
1243 p->nr_decl_insn = 0;
1244
1245 p->csr = p->program;
1246 p->decl = p->declarations;
1247 p->decl_s = 0;
1248 p->decl_t = 0;
1249 p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
1250 p->utemp_flag = ~0x7;
1251
1252 /* initialize the first program word */
1253 *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
1254
1255 return p;
1256 }
1257
1258
1259 /* Copy compile results to the fragment program struct and destroy the
1260 * compilation context.
1261 */
1262 static void
1263 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
1264 {
1265 struct i915_fragment_shader *ifs = p->shader;
1266 unsigned long program_size = (unsigned long) (p->csr - p->program);
1267 unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
1268
1269 if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
1270 debug_printf("Exceeded max nr indirect texture lookups\n");
1271
1272 if (p->nr_tex_insn > I915_MAX_TEX_INSN)
1273 i915_program_error(p, "Exceeded max TEX instructions");
1274
1275 if (p->nr_alu_insn > I915_MAX_ALU_INSN)
1276 i915_program_error(p, "Exceeded max ALU instructions");
1277
1278 if (p->nr_decl_insn > I915_MAX_DECL_INSN)
1279 i915_program_error(p, "Exceeded max DECL instructions");
1280
1281 if (p->error) {
1282 p->NumNativeInstructions = 0;
1283 p->NumNativeAluInstructions = 0;
1284 p->NumNativeTexInstructions = 0;
1285 p->NumNativeTexIndirections = 0;
1286
1287 i915_use_passthrough_shader(ifs);
1288 }
1289 else {
1290 p->NumNativeInstructions
1291 = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
1292 p->NumNativeAluInstructions = p->nr_alu_insn;
1293 p->NumNativeTexInstructions = p->nr_tex_insn;
1294 p->NumNativeTexIndirections = p->nr_tex_indirect;
1295
1296 /* patch in the program length */
1297 p->declarations[0] |= program_size + decl_size - 2;
1298
1299 /* Copy compilation results to fragment program struct:
1300 */
1301 assert(!ifs->decl);
1302 assert(!ifs->program);
1303
1304 ifs->decl
1305 = (uint *) MALLOC(decl_size * sizeof(uint));
1306 ifs->program
1307 = (uint *) MALLOC(program_size * sizeof(uint));
1308
1309 if (ifs->decl) {
1310 ifs->decl_len = decl_size;
1311
1312 memcpy(ifs->decl,
1313 p->declarations,
1314 decl_size * sizeof(uint));
1315 }
1316
1317 if (ifs->program) {
1318 ifs->program_len = program_size;
1319
1320 memcpy(ifs->program,
1321 p->program,
1322 program_size * sizeof(uint));
1323 }
1324 }
1325
1326 /* Release the compilation struct:
1327 */
1328 FREE(p);
1329 }
1330
1331
1332
1333
1334
1335 /**
1336 * Rather than trying to intercept and jiggle depth writes during
1337 * emit, just move the value into its correct position at the end of
1338 * the program:
1339 */
1340 static void
1341 i915_fixup_depth_write(struct i915_fp_compile *p)
1342 {
1343 /* XXX assuming pos/depth is always in output[0] */
1344 if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1345 const uint depth = UREG(REG_TYPE_OD, 0);
1346
1347 i915_emit_arith(p,
1348 A0_MOV, /* opcode */
1349 depth, /* dest reg */
1350 A0_DEST_CHANNEL_W, /* write mask */
1351 0, /* saturate? */
1352 swizzle(depth, X, Y, Z, Z), /* src0 */
1353 0, 0 /* src1, src2 */);
1354 }
1355 }
1356
1357
1358 void
1359 i915_translate_fragment_program( struct i915_context *i915,
1360 struct i915_fragment_shader *fs)
1361 {
1362 struct i915_fp_compile *p;
1363 const struct tgsi_token *tokens = fs->state.tokens;
1364 struct i915_token_list* i_tokens;
1365
1366 #if 0
1367 tgsi_dump(tokens, 0);
1368 #endif
1369
1370 /* hw doesn't seem to like empty frag programs, even when the depth write
1371 * fixup gets emitted below - may that one is fishy, too? */
1372 if (fs->info.num_instructions == 1) {
1373 i915_use_passthrough_shader(fs);
1374
1375 return;
1376 }
1377
1378 p = i915_init_compile(i915, fs);
1379
1380 i_tokens = i915_optimize(tokens);
1381 i915_translate_instructions(p, i_tokens, fs);
1382 i915_fixup_depth_write(p);
1383
1384 i915_fini_compile(i915, p);
1385 i915_optimize_free(i_tokens);
1386
1387 #if 0
1388 i915_disassemble_program(NULL, fs->program, fs->program_len);
1389 #endif
1390 }