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
negate(int reg,int x,int y,int z,int w)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
i915_use_passthrough_shader(struct i915_fragment_shader * fs)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
i915_program_error(struct i915_fp_compile * p,const char * msg,...)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
get_mapping(struct i915_fragment_shader * fs,int unit)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
src_vector(struct i915_fp_compile * p,const struct i915_full_src_register * source,struct i915_fragment_shader * fs)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
get_result_vector(struct i915_fp_compile * p,const struct i915_full_dst_register * dest)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
get_result_flags(const struct i915_full_instruction * inst)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
translate_tex_src_target(struct i915_fp_compile * p,uint tex)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
i915_num_coords(uint tex)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
emit_tex(struct i915_fp_compile * p,const struct i915_full_instruction * inst,uint opcode,struct i915_fragment_shader * fs)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
emit_simple_arith(struct i915_fp_compile * p,const struct i915_full_instruction * inst,uint opcode,uint numArgs,struct i915_fragment_shader * fs)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
emit_simple_arith_swap2(struct i915_fp_compile * p,const struct i915_full_instruction * inst,uint opcode,uint numArgs,struct i915_fragment_shader * fs)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
i915_translate_instruction(struct i915_fp_compile * p,const struct i915_full_instruction * inst,struct i915_fragment_shader * fs)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_DPH:
608 src0 = src_vector(p, &inst->Src[0], fs);
609 src1 = src_vector(p, &inst->Src[1], fs);
610
611 i915_emit_arith(p,
612 A0_DP4,
613 get_result_vector(p, &inst->Dst[0]),
614 get_result_flags(inst), 0,
615 swizzle(src0, X, Y, Z, ONE), src1, 0);
616 break;
617
618 case TGSI_OPCODE_DST:
619 src0 = src_vector(p, &inst->Src[0], fs);
620 src1 = src_vector(p, &inst->Src[1], fs);
621
622 /* result[0] = 1 * 1;
623 * result[1] = a[1] * b[1];
624 * result[2] = a[2] * 1;
625 * result[3] = 1 * b[3];
626 */
627 i915_emit_arith(p,
628 A0_MUL,
629 get_result_vector(p, &inst->Dst[0]),
630 get_result_flags(inst), 0,
631 swizzle(src0, ONE, Y, Z, ONE),
632 swizzle(src1, ONE, Y, ONE, W), 0);
633 break;
634
635 case TGSI_OPCODE_END:
636 /* no-op */
637 break;
638
639 case TGSI_OPCODE_EX2:
640 src0 = src_vector(p, &inst->Src[0], fs);
641
642 i915_emit_arith(p,
643 A0_EXP,
644 get_result_vector(p, &inst->Dst[0]),
645 get_result_flags(inst), 0,
646 swizzle(src0, X, X, X, X), 0, 0);
647 break;
648
649 case TGSI_OPCODE_FLR:
650 emit_simple_arith(p, inst, A0_FLR, 1, fs);
651 break;
652
653 case TGSI_OPCODE_FRC:
654 emit_simple_arith(p, inst, A0_FRC, 1, fs);
655 break;
656
657 case TGSI_OPCODE_KILL_IF:
658 /* kill if src[0].x < 0 || src[0].y < 0 ... */
659 src0 = src_vector(p, &inst->Src[0], fs);
660 tmp = i915_get_utemp(p);
661
662 i915_emit_texld(p,
663 tmp, /* dest reg: a dummy reg */
664 A0_DEST_CHANNEL_ALL, /* dest writemask */
665 0, /* sampler */
666 src0, /* coord*/
667 T0_TEXKILL, /* opcode */
668 1); /* num_coord */
669 break;
670
671 case TGSI_OPCODE_KILL:
672 /* unconditional kill */
673 tmp = i915_get_utemp(p);
674
675 i915_emit_texld(p,
676 tmp, /* dest reg: a dummy reg */
677 A0_DEST_CHANNEL_ALL, /* dest writemask */
678 0, /* sampler */
679 negate(swizzle(0, ONE, ONE, ONE, ONE), 1, 1, 1, 1), /* coord */
680 T0_TEXKILL, /* opcode */
681 1); /* num_coord */
682 break;
683
684 case TGSI_OPCODE_LG2:
685 src0 = src_vector(p, &inst->Src[0], fs);
686
687 i915_emit_arith(p,
688 A0_LOG,
689 get_result_vector(p, &inst->Dst[0]),
690 get_result_flags(inst), 0,
691 swizzle(src0, X, X, X, X), 0, 0);
692 break;
693
694 case TGSI_OPCODE_LIT:
695 src0 = src_vector(p, &inst->Src[0], fs);
696 tmp = i915_get_utemp(p);
697
698 /* tmp = max( a.xyzw, a.00zw )
699 * XXX: Clamp tmp.w to -128..128
700 * tmp.y = log(tmp.y)
701 * tmp.y = tmp.w * tmp.y
702 * tmp.y = exp(tmp.y)
703 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
704 */
705 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
706 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
707
708 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
709 swizzle(tmp, Y, Y, Y, Y), 0, 0);
710
711 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
712 swizzle(tmp, ZERO, Y, ZERO, ZERO),
713 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
714
715 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
716 swizzle(tmp, Y, Y, Y, Y), 0, 0);
717
718 i915_emit_arith(p, A0_CMP,
719 get_result_vector(p, &inst->Dst[0]),
720 get_result_flags(inst), 0,
721 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
722 swizzle(tmp, ONE, X, ZERO, ONE),
723 swizzle(tmp, ONE, X, Y, ONE));
724
725 break;
726
727 case TGSI_OPCODE_LRP:
728 src0 = src_vector(p, &inst->Src[0], fs);
729 src1 = src_vector(p, &inst->Src[1], fs);
730 src2 = src_vector(p, &inst->Src[2], fs);
731 flags = get_result_flags(inst);
732 tmp = i915_get_utemp(p);
733
734 /* b*a + c*(1-a)
735 *
736 * b*a + c - ca
737 *
738 * tmp = b*a + c,
739 * result = (-c)*a + tmp
740 */
741 i915_emit_arith(p, A0_MAD, tmp,
742 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
743
744 i915_emit_arith(p, A0_MAD,
745 get_result_vector(p, &inst->Dst[0]),
746 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
747 break;
748
749 case TGSI_OPCODE_MAD:
750 emit_simple_arith(p, inst, A0_MAD, 3, fs);
751 break;
752
753 case TGSI_OPCODE_MAX:
754 emit_simple_arith(p, inst, A0_MAX, 2, fs);
755 break;
756
757 case TGSI_OPCODE_MIN:
758 emit_simple_arith(p, inst, A0_MIN, 2, fs);
759 break;
760
761 case TGSI_OPCODE_MOV:
762 emit_simple_arith(p, inst, A0_MOV, 1, fs);
763 break;
764
765 case TGSI_OPCODE_MUL:
766 emit_simple_arith(p, inst, A0_MUL, 2, fs);
767 break;
768
769 case TGSI_OPCODE_NOP:
770 break;
771
772 case TGSI_OPCODE_POW:
773 src0 = src_vector(p, &inst->Src[0], fs);
774 src1 = src_vector(p, &inst->Src[1], fs);
775 tmp = i915_get_utemp(p);
776 flags = get_result_flags(inst);
777
778 /* XXX: masking on intermediate values, here and elsewhere.
779 */
780 i915_emit_arith(p,
781 A0_LOG,
782 tmp, A0_DEST_CHANNEL_X, 0,
783 swizzle(src0, X, X, X, X), 0, 0);
784
785 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
786
787 i915_emit_arith(p,
788 A0_EXP,
789 get_result_vector(p, &inst->Dst[0]),
790 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
791 break;
792
793 case TGSI_OPCODE_RET:
794 /* XXX: no-op? */
795 break;
796
797 case TGSI_OPCODE_RCP:
798 src0 = src_vector(p, &inst->Src[0], fs);
799
800 i915_emit_arith(p,
801 A0_RCP,
802 get_result_vector(p, &inst->Dst[0]),
803 get_result_flags(inst), 0,
804 swizzle(src0, X, X, X, X), 0, 0);
805 break;
806
807 case TGSI_OPCODE_RSQ:
808 src0 = src_vector(p, &inst->Src[0], fs);
809
810 i915_emit_arith(p,
811 A0_RSQ,
812 get_result_vector(p, &inst->Dst[0]),
813 get_result_flags(inst), 0,
814 swizzle(src0, X, X, X, X), 0, 0);
815 break;
816
817 case TGSI_OPCODE_SCS:
818 src0 = src_vector(p, &inst->Src[0], fs);
819 tmp = i915_get_utemp(p);
820
821 /*
822 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
823 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
824 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
825 * scs.x = DP4 t1, scs_sin_constants
826 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
827 * scs.y = DP4 t1, scs_cos_constants
828 */
829 i915_emit_arith(p,
830 A0_MUL,
831 tmp, A0_DEST_CHANNEL_XY, 0,
832 swizzle(src0, X, X, ONE, ONE),
833 swizzle(src0, X, ONE, ONE, ONE), 0);
834
835 i915_emit_arith(p,
836 A0_MUL,
837 tmp, A0_DEST_CHANNEL_ALL, 0,
838 swizzle(tmp, X, Y, X, Y),
839 swizzle(tmp, X, X, ONE, ONE), 0);
840
841 writemask = inst->Dst[0].Register.WriteMask;
842
843 if (writemask & TGSI_WRITEMASK_Y) {
844 uint tmp1;
845
846 if (writemask & TGSI_WRITEMASK_X)
847 tmp1 = i915_get_utemp(p);
848 else
849 tmp1 = tmp;
850
851 i915_emit_arith(p,
852 A0_MUL,
853 tmp1, A0_DEST_CHANNEL_ALL, 0,
854 swizzle(tmp, X, Y, Y, W),
855 swizzle(tmp, X, Z, ONE, ONE), 0);
856
857 i915_emit_arith(p,
858 A0_DP4,
859 get_result_vector(p, &inst->Dst[0]),
860 A0_DEST_CHANNEL_Y, 0,
861 swizzle(tmp1, W, Z, Y, X),
862 i915_emit_const4fv(p, scs_sin_constants), 0);
863 }
864
865 if (writemask & TGSI_WRITEMASK_X) {
866 i915_emit_arith(p,
867 A0_MUL,
868 tmp, A0_DEST_CHANNEL_XYZ, 0,
869 swizzle(tmp, X, X, Z, ONE),
870 swizzle(tmp, Z, ONE, ONE, ONE), 0);
871
872 i915_emit_arith(p,
873 A0_DP4,
874 get_result_vector(p, &inst->Dst[0]),
875 A0_DEST_CHANNEL_X, 0,
876 swizzle(tmp, ONE, Z, Y, X),
877 i915_emit_const4fv(p, scs_cos_constants), 0);
878 }
879 break;
880
881 case TGSI_OPCODE_SEQ:
882 /* if we're both >= and <= then we're == */
883 src0 = src_vector(p, &inst->Src[0], fs);
884 src1 = src_vector(p, &inst->Src[1], fs);
885 tmp = i915_get_utemp(p);
886
887 i915_emit_arith(p,
888 A0_SGE,
889 tmp, A0_DEST_CHANNEL_ALL, 0,
890 src0,
891 src1, 0);
892
893 i915_emit_arith(p,
894 A0_SGE,
895 get_result_vector(p, &inst->Dst[0]),
896 A0_DEST_CHANNEL_ALL, 0,
897 src1,
898 src0, 0);
899
900 i915_emit_arith(p,
901 A0_MUL,
902 get_result_vector(p, &inst->Dst[0]),
903 A0_DEST_CHANNEL_ALL, 0,
904 get_result_vector(p, &inst->Dst[0]),
905 tmp, 0);
906
907 break;
908
909 case TGSI_OPCODE_SGE:
910 emit_simple_arith(p, inst, A0_SGE, 2, fs);
911 break;
912
913 case TGSI_OPCODE_SIN:
914 src0 = src_vector(p, &inst->Src[0], fs);
915 tmp = i915_get_utemp(p);
916
917 i915_emit_arith(p,
918 A0_MUL,
919 tmp, A0_DEST_CHANNEL_X, 0,
920 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
921
922 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
923
924 /*
925 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
926 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
927 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
928 * result = DP4 t1.wzyx, sin_constants
929 */
930 i915_emit_arith(p,
931 A0_MUL,
932 tmp, A0_DEST_CHANNEL_XY, 0,
933 swizzle(tmp, X, X, ONE, ONE),
934 swizzle(tmp, X, ONE, ONE, ONE), 0);
935
936 i915_emit_arith(p,
937 A0_MUL,
938 tmp, A0_DEST_CHANNEL_ALL, 0,
939 swizzle(tmp, X, Y, X, Y),
940 swizzle(tmp, X, X, ONE, ONE), 0);
941
942 i915_emit_arith(p,
943 A0_MUL,
944 tmp, A0_DEST_CHANNEL_ALL, 0,
945 swizzle(tmp, X, Y, Y, W),
946 swizzle(tmp, X, Z, ONE, ONE), 0);
947
948 i915_emit_arith(p,
949 A0_DP4,
950 get_result_vector(p, &inst->Dst[0]),
951 get_result_flags(inst), 0,
952 swizzle(tmp, W, Z, Y, X),
953 i915_emit_const4fv(p, sin_constants), 0);
954 break;
955
956 case TGSI_OPCODE_SLE:
957 /* like SGE, but swap reg0, reg1 */
958 emit_simple_arith_swap2(p, inst, A0_SGE, 2, fs);
959 break;
960
961 case TGSI_OPCODE_SLT:
962 emit_simple_arith(p, inst, A0_SLT, 2, fs);
963 break;
964
965 case TGSI_OPCODE_SGT:
966 /* like SLT, but swap reg0, reg1 */
967 emit_simple_arith_swap2(p, inst, A0_SLT, 2, fs);
968 break;
969
970 case TGSI_OPCODE_SNE:
971 /* if we're < or > then we're != */
972 src0 = src_vector(p, &inst->Src[0], fs);
973 src1 = src_vector(p, &inst->Src[1], fs);
974 tmp = i915_get_utemp(p);
975
976 i915_emit_arith(p,
977 A0_SLT,
978 tmp,
979 A0_DEST_CHANNEL_ALL, 0,
980 src0,
981 src1, 0);
982
983 i915_emit_arith(p,
984 A0_SLT,
985 get_result_vector(p, &inst->Dst[0]),
986 A0_DEST_CHANNEL_ALL, 0,
987 src1,
988 src0, 0);
989
990 i915_emit_arith(p,
991 A0_ADD,
992 get_result_vector(p, &inst->Dst[0]),
993 A0_DEST_CHANNEL_ALL, 0,
994 get_result_vector(p, &inst->Dst[0]),
995 tmp, 0);
996 break;
997
998 case TGSI_OPCODE_SSG:
999 /* compute (src>0) - (src<0) */
1000 src0 = src_vector(p, &inst->Src[0], fs);
1001 tmp = i915_get_utemp(p);
1002
1003 i915_emit_arith(p,
1004 A0_SLT,
1005 tmp,
1006 A0_DEST_CHANNEL_ALL, 0,
1007 src0,
1008 swizzle(src0, ZERO, ZERO, ZERO, ZERO), 0);
1009
1010 i915_emit_arith(p,
1011 A0_SLT,
1012 get_result_vector(p, &inst->Dst[0]),
1013 A0_DEST_CHANNEL_ALL, 0,
1014 swizzle(src0, ZERO, ZERO, ZERO, ZERO),
1015 src0, 0);
1016
1017 i915_emit_arith(p,
1018 A0_ADD,
1019 get_result_vector(p, &inst->Dst[0]),
1020 A0_DEST_CHANNEL_ALL, 0,
1021 get_result_vector(p, &inst->Dst[0]),
1022 negate(tmp, 1, 1, 1, 1), 0);
1023 break;
1024
1025 case TGSI_OPCODE_TEX:
1026 emit_tex(p, inst, T0_TEXLD, fs);
1027 break;
1028
1029 case TGSI_OPCODE_TRUNC:
1030 emit_simple_arith(p, inst, A0_TRC, 1, fs);
1031 break;
1032
1033 case TGSI_OPCODE_TXB:
1034 emit_tex(p, inst, T0_TEXLDB, fs);
1035 break;
1036
1037 case TGSI_OPCODE_TXP:
1038 emit_tex(p, inst, T0_TEXLDP, fs);
1039 break;
1040
1041 case TGSI_OPCODE_XPD:
1042 /* Cross product:
1043 * result.x = src0.y * src1.z - src0.z * src1.y;
1044 * result.y = src0.z * src1.x - src0.x * src1.z;
1045 * result.z = src0.x * src1.y - src0.y * src1.x;
1046 * result.w = undef;
1047 */
1048 src0 = src_vector(p, &inst->Src[0], fs);
1049 src1 = src_vector(p, &inst->Src[1], fs);
1050 tmp = i915_get_utemp(p);
1051
1052 i915_emit_arith(p,
1053 A0_MUL,
1054 tmp, A0_DEST_CHANNEL_ALL, 0,
1055 swizzle(src0, Z, X, Y, ONE),
1056 swizzle(src1, Y, Z, X, ONE), 0);
1057
1058 i915_emit_arith(p,
1059 A0_MAD,
1060 get_result_vector(p, &inst->Dst[0]),
1061 get_result_flags(inst), 0,
1062 swizzle(src0, Y, Z, X, ONE),
1063 swizzle(src1, Z, X, Y, ONE),
1064 negate(tmp, 1, 1, 1, 0));
1065 break;
1066
1067 default:
1068 i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
1069 p->error = 1;
1070 return;
1071 }
1072
1073 i915_release_utemps(p);
1074 }
1075
1076
i915_translate_token(struct i915_fp_compile * p,const union i915_full_token * token,struct i915_fragment_shader * fs)1077 static void i915_translate_token(struct i915_fp_compile *p,
1078 const union i915_full_token *token,
1079 struct i915_fragment_shader *fs)
1080 {
1081 struct i915_fragment_shader *ifs = p->shader;
1082 switch( token->Token.Type ) {
1083 case TGSI_TOKEN_TYPE_PROPERTY:
1084 /*
1085 * We only support one cbuf, but we still need to ignore the property
1086 * correctly so we don't hit the assert at the end of the switch case.
1087 */
1088 assert(token->FullProperty.Property.PropertyName ==
1089 TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS);
1090 break;
1091
1092 case TGSI_TOKEN_TYPE_DECLARATION:
1093 if (token->FullDeclaration.Declaration.File
1094 == TGSI_FILE_CONSTANT) {
1095 uint i;
1096 for (i = token->FullDeclaration.Range.First;
1097 i <= MIN2(token->FullDeclaration.Range.Last, I915_MAX_CONSTANT - 1);
1098 i++) {
1099 assert(ifs->constant_flags[i] == 0x0);
1100 ifs->constant_flags[i] = I915_CONSTFLAG_USER;
1101 ifs->num_constants = MAX2(ifs->num_constants, i + 1);
1102 }
1103 }
1104 else if (token->FullDeclaration.Declaration.File
1105 == TGSI_FILE_TEMPORARY) {
1106 uint i;
1107 for (i = token->FullDeclaration.Range.First;
1108 i <= token->FullDeclaration.Range.Last;
1109 i++) {
1110 if (i >= I915_MAX_TEMPORARY)
1111 debug_printf("Too many temps (%d)\n",i);
1112 else
1113 /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */
1114 p->temp_flag |= (1 << i); /* mark temp as used */
1115 }
1116 }
1117 break;
1118
1119 case TGSI_TOKEN_TYPE_IMMEDIATE:
1120 {
1121 const struct tgsi_full_immediate *imm
1122 = &token->FullImmediate;
1123 const uint pos = p->num_immediates++;
1124 uint j;
1125 assert( imm->Immediate.NrTokens <= 4 + 1 );
1126 for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
1127 p->immediates[pos][j] = imm->u[j].Float;
1128 }
1129 }
1130 break;
1131
1132 case TGSI_TOKEN_TYPE_INSTRUCTION:
1133 if (p->first_instruction) {
1134 /* resolve location of immediates */
1135 uint i, j;
1136 for (i = 0; i < p->num_immediates; i++) {
1137 /* find constant slot for this immediate */
1138 for (j = 0; j < I915_MAX_CONSTANT; j++) {
1139 if (ifs->constant_flags[j] == 0x0) {
1140 memcpy(ifs->constants[j],
1141 p->immediates[i],
1142 4 * sizeof(float));
1143 /*printf("immediate %d maps to const %d\n", i, j);*/
1144 ifs->constant_flags[j] = 0xf; /* all four comps used */
1145 p->immediates_map[i] = j;
1146 ifs->num_constants = MAX2(ifs->num_constants, j + 1);
1147 break;
1148 }
1149 }
1150 }
1151
1152 p->first_instruction = FALSE;
1153 }
1154
1155 i915_translate_instruction(p, &token->FullInstruction, fs);
1156 break;
1157
1158 default:
1159 assert( 0 );
1160 }
1161
1162 }
1163
1164 /**
1165 * Translate TGSI fragment shader into i915 hardware instructions.
1166 * \param p the translation state
1167 * \param tokens the TGSI token array
1168 */
1169 static void
i915_translate_instructions(struct i915_fp_compile * p,const struct i915_token_list * tokens,struct i915_fragment_shader * fs)1170 i915_translate_instructions(struct i915_fp_compile *p,
1171 const struct i915_token_list *tokens,
1172 struct i915_fragment_shader *fs)
1173 {
1174 int i;
1175 for(i = 0; i<tokens->NumTokens; i++) {
1176 i915_translate_token(p, &tokens->Tokens[i], fs);
1177 }
1178 }
1179
1180
1181 static struct i915_fp_compile *
i915_init_compile(struct i915_context * i915,struct i915_fragment_shader * ifs)1182 i915_init_compile(struct i915_context *i915,
1183 struct i915_fragment_shader *ifs)
1184 {
1185 struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
1186 int i;
1187
1188 p->shader = ifs;
1189
1190 /* Put new constants at end of const buffer, growing downward.
1191 * The problem is we don't know how many user-defined constants might
1192 * be specified with pipe->set_constant_buffer().
1193 * Should pre-scan the user's program to determine the highest-numbered
1194 * constant referenced.
1195 */
1196 ifs->num_constants = 0;
1197 memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags));
1198
1199 memset(&p->register_phases, 0, sizeof(p->register_phases));
1200
1201 for (i = 0; i < I915_TEX_UNITS; i++)
1202 ifs->generic_mapping[i] = -1;
1203
1204 p->first_instruction = TRUE;
1205
1206 p->nr_tex_indirect = 1; /* correct? */
1207 p->nr_tex_insn = 0;
1208 p->nr_alu_insn = 0;
1209 p->nr_decl_insn = 0;
1210
1211 p->csr = p->program;
1212 p->decl = p->declarations;
1213 p->decl_s = 0;
1214 p->decl_t = 0;
1215 p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
1216 p->utemp_flag = ~0x7;
1217
1218 /* initialize the first program word */
1219 *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
1220
1221 return p;
1222 }
1223
1224
1225 /* Copy compile results to the fragment program struct and destroy the
1226 * compilation context.
1227 */
1228 static void
i915_fini_compile(struct i915_context * i915,struct i915_fp_compile * p)1229 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
1230 {
1231 struct i915_fragment_shader *ifs = p->shader;
1232 unsigned long program_size = (unsigned long) (p->csr - p->program);
1233 unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
1234
1235 if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
1236 debug_printf("Exceeded max nr indirect texture lookups\n");
1237
1238 if (p->nr_tex_insn > I915_MAX_TEX_INSN)
1239 i915_program_error(p, "Exceeded max TEX instructions");
1240
1241 if (p->nr_alu_insn > I915_MAX_ALU_INSN)
1242 i915_program_error(p, "Exceeded max ALU instructions");
1243
1244 if (p->nr_decl_insn > I915_MAX_DECL_INSN)
1245 i915_program_error(p, "Exceeded max DECL instructions");
1246
1247 if (p->error) {
1248 p->NumNativeInstructions = 0;
1249 p->NumNativeAluInstructions = 0;
1250 p->NumNativeTexInstructions = 0;
1251 p->NumNativeTexIndirections = 0;
1252
1253 i915_use_passthrough_shader(ifs);
1254 }
1255 else {
1256 p->NumNativeInstructions
1257 = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
1258 p->NumNativeAluInstructions = p->nr_alu_insn;
1259 p->NumNativeTexInstructions = p->nr_tex_insn;
1260 p->NumNativeTexIndirections = p->nr_tex_indirect;
1261
1262 /* patch in the program length */
1263 p->declarations[0] |= program_size + decl_size - 2;
1264
1265 /* Copy compilation results to fragment program struct:
1266 */
1267 assert(!ifs->decl);
1268 assert(!ifs->program);
1269
1270 ifs->decl
1271 = (uint *) MALLOC(decl_size * sizeof(uint));
1272 ifs->program
1273 = (uint *) MALLOC(program_size * sizeof(uint));
1274
1275 if (ifs->decl) {
1276 ifs->decl_len = decl_size;
1277
1278 memcpy(ifs->decl,
1279 p->declarations,
1280 decl_size * sizeof(uint));
1281 }
1282
1283 if (ifs->program) {
1284 ifs->program_len = program_size;
1285
1286 memcpy(ifs->program,
1287 p->program,
1288 program_size * sizeof(uint));
1289 }
1290 }
1291
1292 /* Release the compilation struct:
1293 */
1294 FREE(p);
1295 }
1296
1297
1298
1299
1300
1301 /**
1302 * Rather than trying to intercept and jiggle depth writes during
1303 * emit, just move the value into its correct position at the end of
1304 * the program:
1305 */
1306 static void
i915_fixup_depth_write(struct i915_fp_compile * p)1307 i915_fixup_depth_write(struct i915_fp_compile *p)
1308 {
1309 /* XXX assuming pos/depth is always in output[0] */
1310 if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1311 const uint depth = UREG(REG_TYPE_OD, 0);
1312
1313 i915_emit_arith(p,
1314 A0_MOV, /* opcode */
1315 depth, /* dest reg */
1316 A0_DEST_CHANNEL_W, /* write mask */
1317 0, /* saturate? */
1318 swizzle(depth, X, Y, Z, Z), /* src0 */
1319 0, 0 /* src1, src2 */);
1320 }
1321 }
1322
1323
1324 void
i915_translate_fragment_program(struct i915_context * i915,struct i915_fragment_shader * fs)1325 i915_translate_fragment_program( struct i915_context *i915,
1326 struct i915_fragment_shader *fs)
1327 {
1328 struct i915_fp_compile *p;
1329 const struct tgsi_token *tokens = fs->state.tokens;
1330 struct i915_token_list* i_tokens;
1331
1332 #if 0
1333 tgsi_dump(tokens, 0);
1334 #endif
1335
1336 /* hw doesn't seem to like empty frag programs, even when the depth write
1337 * fixup gets emitted below - may that one is fishy, too? */
1338 if (fs->info.num_instructions == 1) {
1339 i915_use_passthrough_shader(fs);
1340
1341 return;
1342 }
1343
1344 p = i915_init_compile(i915, fs);
1345
1346 i_tokens = i915_optimize(tokens);
1347 i915_translate_instructions(p, i_tokens, fs);
1348 i915_fixup_depth_write(p);
1349
1350 i915_fini_compile(i915, p);
1351 i915_optimize_free(i_tokens);
1352
1353 #if 0
1354 i915_disassemble_program(NULL, fs->program, fs->program_len);
1355 #endif
1356 }
1357