projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge branch 'master' into r300g-glsl
[mesa.git] / src / gallium / drivers / i915 / i915_fpc_translate.c
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28
29 #include <stdarg.h>
30
31 #include "i915_reg.h"
32 #include "i915_context.h"
33 #include "i915_fpc.h"
34
35 #include "pipe/p_shader_tokens.h"
36 #include "util/u_math.h"
37 #include "util/u_memory.h"
38 #include "util/u_string.h"
39 #include "tgsi/tgsi_parse.h"
40 #include "tgsi/tgsi_dump.h"
41
42 #include "draw/draw_vertex.h"
43
44
45 /**
46 * Simple pass-through fragment shader to use when we don't have
47 * a real shader (or it fails to compile for some reason).
48 */
49 static unsigned passthrough[] =
50 {
51 _3DSTATE_PIXEL_SHADER_PROGRAM | ((2*3)-1),
52
53 /* declare input color:
54 */
55 (D0_DCL |
56 (REG_TYPE_T << D0_TYPE_SHIFT) |
57 (T_DIFFUSE << D0_NR_SHIFT) |
58 D0_CHANNEL_ALL),
59 0,
60 0,
61
62 /* move to output color:
63 */
64 (A0_MOV |
65 (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) |
66 A0_DEST_CHANNEL_ALL |
67 (REG_TYPE_T << A0_SRC0_TYPE_SHIFT) |
68 (T_DIFFUSE << A0_SRC0_NR_SHIFT)),
69 0x01230000, /* .xyzw */
70 0
71 };
72
73
74 /* 1, -1/3!, 1/5!, -1/7! */
75 static const float sin_constants[4] = { 1.0,
76 -1.0f / (3 * 2 * 1),
77 1.0f / (5 * 4 * 3 * 2 * 1),
78 -1.0f / (7 * 6 * 5 * 4 * 3 * 2 * 1)
79 };
80
81 /* 1, -1/2!, 1/4!, -1/6! */
82 static const float cos_constants[4] = { 1.0,
83 -1.0f / (2 * 1),
84 1.0f / (4 * 3 * 2 * 1),
85 -1.0f / (6 * 5 * 4 * 3 * 2 * 1)
86 };
87
88
89
90 /**
91 * component-wise negation of ureg
92 */
93 static INLINE int
94 negate(int reg, int x, int y, int z, int w)
95 {
96 /* Another neat thing about the UREG representation */
97 return reg ^ (((x & 1) << UREG_CHANNEL_X_NEGATE_SHIFT) |
98 ((y & 1) << UREG_CHANNEL_Y_NEGATE_SHIFT) |
99 ((z & 1) << UREG_CHANNEL_Z_NEGATE_SHIFT) |
100 ((w & 1) << UREG_CHANNEL_W_NEGATE_SHIFT));
101 }
102
103
104 /**
105 * In the event of a translation failure, we'll generate a simple color
106 * pass-through program.
107 */
108 static void
109 i915_use_passthrough_shader(struct i915_fragment_shader *fs)
110 {
111 fs->program = (uint *) MALLOC(sizeof(passthrough));
112 if (fs->program) {
113 memcpy(fs->program, passthrough, sizeof(passthrough));
114 fs->program_len = Elements(passthrough);
115 }
116 fs->num_constants = 0;
117 }
118
119
120 void
121 i915_program_error(struct i915_fp_compile *p, const char *msg, ...)
122 {
123 va_list args;
124 char buffer[1024];
125
126 debug_printf("i915_program_error: ");
127 va_start( args, msg );
128 util_vsnprintf( buffer, sizeof(buffer), msg, args );
129 va_end( args );
130 debug_printf(buffer);
131 debug_printf("\n");
132
133 p->error = 1;
134 }
135
136
137
138 /**
139 * Construct a ureg for the given source register. Will emit
140 * constants, apply swizzling and negation as needed.
141 */
142 static uint
143 src_vector(struct i915_fp_compile *p,
144 const struct tgsi_full_src_register *source)
145 {
146 uint index = source->SrcRegister.Index;
147 uint src = 0, sem_name, sem_ind;
148
149 switch (source->SrcRegister.File) {
150 case TGSI_FILE_TEMPORARY:
151 if (source->SrcRegister.Index >= I915_MAX_TEMPORARY) {
152 i915_program_error(p, "Exceeded max temporary reg");
153 return 0;
154 }
155 src = UREG(REG_TYPE_R, index);
156 break;
157 case TGSI_FILE_INPUT:
158 /* XXX: Packing COL1, FOGC into a single attribute works for
159 * texenv programs, but will fail for real fragment programs
160 * that use these attributes and expect them to be a full 4
161 * components wide. Could use a texcoord to pass these
162 * attributes if necessary, but that won't work in the general
163 * case.
164 *
165 * We also use a texture coordinate to pass wpos when possible.
166 */
167
168 sem_name = p->shader->info.input_semantic_name[index];
169 sem_ind = p->shader->info.input_semantic_index[index];
170
171 switch (sem_name) {
172 case TGSI_SEMANTIC_POSITION:
173 debug_printf("SKIP SEM POS\n");
174 /*
175 assert(p->wpos_tex != -1);
176 src = i915_emit_decl(p, REG_TYPE_T, p->wpos_tex, D0_CHANNEL_ALL);
177 */
178 break;
179 case TGSI_SEMANTIC_COLOR:
180 if (sem_ind == 0) {
181 src = i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
182 }
183 else {
184 /* secondary color */
185 assert(sem_ind == 1);
186 src = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_XYZ);
187 src = swizzle(src, X, Y, Z, ONE);
188 }
189 break;
190 case TGSI_SEMANTIC_FOG:
191 src = i915_emit_decl(p, REG_TYPE_T, T_FOG_W, D0_CHANNEL_W);
192 src = swizzle(src, W, W, W, W);
193 break;
194 case TGSI_SEMANTIC_GENERIC:
195 /* usually a texcoord */
196 src = i915_emit_decl(p, REG_TYPE_T, T_TEX0 + sem_ind, D0_CHANNEL_ALL);
197 break;
198 default:
199 i915_program_error(p, "Bad source->Index");
200 return 0;
201 }
202 break;
203
204 case TGSI_FILE_IMMEDIATE:
205 assert(index < p->num_immediates);
206 index = p->immediates_map[index];
207 /* fall-through */
208 case TGSI_FILE_CONSTANT:
209 src = UREG(REG_TYPE_CONST, index);
210 break;
211
212 default:
213 i915_program_error(p, "Bad source->File");
214 return 0;
215 }
216
217 if (source->SrcRegister.Extended) {
218 src = swizzle(src,
219 source->SrcRegisterExtSwz.ExtSwizzleX,
220 source->SrcRegisterExtSwz.ExtSwizzleY,
221 source->SrcRegisterExtSwz.ExtSwizzleZ,
222 source->SrcRegisterExtSwz.ExtSwizzleW);
223 }
224 else {
225 src = swizzle(src,
226 source->SrcRegister.SwizzleX,
227 source->SrcRegister.SwizzleY,
228 source->SrcRegister.SwizzleZ,
229 source->SrcRegister.SwizzleW);
230 }
231
232
233 /* There's both negate-all-components and per-component negation.
234 * Try to handle both here.
235 */
236 {
237 int nx = source->SrcRegisterExtSwz.NegateX;
238 int ny = source->SrcRegisterExtSwz.NegateY;
239 int nz = source->SrcRegisterExtSwz.NegateZ;
240 int nw = source->SrcRegisterExtSwz.NegateW;
241 if (source->SrcRegister.Negate) {
242 nx = !nx;
243 ny = !ny;
244 nz = !nz;
245 nw = !nw;
246 }
247 src = negate(src, nx, ny, nz, nw);
248 }
249
250 /* no abs() or post-abs negation */
251 #if 0
252 /* XXX assertions disabled to allow arbfplight.c to run */
253 /* XXX enable these assertions, or fix things */
254 assert(!source->SrcRegisterExtMod.Absolute);
255 assert(!source->SrcRegisterExtMod.Negate);
256 #endif
257 return src;
258 }
259
260
261 /**
262 * Construct a ureg for a destination register.
263 */
264 static uint
265 get_result_vector(struct i915_fp_compile *p,
266 const struct tgsi_full_dst_register *dest)
267 {
268 switch (dest->DstRegister.File) {
269 case TGSI_FILE_OUTPUT:
270 {
271 uint sem_name = p->shader->info.output_semantic_name[dest->DstRegister.Index];
272 switch (sem_name) {
273 case TGSI_SEMANTIC_POSITION:
274 return UREG(REG_TYPE_OD, 0);
275 case TGSI_SEMANTIC_COLOR:
276 return UREG(REG_TYPE_OC, 0);
277 default:
278 i915_program_error(p, "Bad inst->DstReg.Index/semantics");
279 return 0;
280 }
281 }
282 case TGSI_FILE_TEMPORARY:
283 return UREG(REG_TYPE_R, dest->DstRegister.Index);
284 default:
285 i915_program_error(p, "Bad inst->DstReg.File");
286 return 0;
287 }
288 }
289
290
291 /**
292 * Compute flags for saturation and writemask.
293 */
294 static uint
295 get_result_flags(const struct tgsi_full_instruction *inst)
296 {
297 const uint writeMask
298 = inst->FullDstRegisters[0].DstRegister.WriteMask;
299 uint flags = 0x0;
300
301 if (inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE)
302 flags |= A0_DEST_SATURATE;
303
304 if (writeMask & TGSI_WRITEMASK_X)
305 flags |= A0_DEST_CHANNEL_X;
306 if (writeMask & TGSI_WRITEMASK_Y)
307 flags |= A0_DEST_CHANNEL_Y;
308 if (writeMask & TGSI_WRITEMASK_Z)
309 flags |= A0_DEST_CHANNEL_Z;
310 if (writeMask & TGSI_WRITEMASK_W)
311 flags |= A0_DEST_CHANNEL_W;
312
313 return flags;
314 }
315
316
317 /**
318 * Convert TGSI_TEXTURE_x token to DO_SAMPLE_TYPE_x token
319 */
320 static uint
321 translate_tex_src_target(struct i915_fp_compile *p, uint tex)
322 {
323 switch (tex) {
324 case TGSI_TEXTURE_SHADOW1D:
325 /* fall-through */
326 case TGSI_TEXTURE_1D:
327 return D0_SAMPLE_TYPE_2D;
328
329 case TGSI_TEXTURE_SHADOW2D:
330 /* fall-through */
331 case TGSI_TEXTURE_2D:
332 return D0_SAMPLE_TYPE_2D;
333
334 case TGSI_TEXTURE_SHADOWRECT:
335 /* fall-through */
336 case TGSI_TEXTURE_RECT:
337 return D0_SAMPLE_TYPE_2D;
338
339 case TGSI_TEXTURE_3D:
340 return D0_SAMPLE_TYPE_VOLUME;
341
342 case TGSI_TEXTURE_CUBE:
343 return D0_SAMPLE_TYPE_CUBE;
344
345 default:
346 i915_program_error(p, "TexSrc type");
347 return 0;
348 }
349 }
350
351
352 /**
353 * Generate texel lookup instruction.
354 */
355 static void
356 emit_tex(struct i915_fp_compile *p,
357 const struct tgsi_full_instruction *inst,
358 uint opcode)
359 {
360 uint texture = inst->InstructionExtTexture.Texture;
361 uint unit = inst->FullSrcRegisters[1].SrcRegister.Index;
362 uint tex = translate_tex_src_target( p, texture );
363 uint sampler = i915_emit_decl(p, REG_TYPE_S, unit, tex);
364 uint coord = src_vector( p, &inst->FullSrcRegisters[0]);
365
366 i915_emit_texld( p,
367 get_result_vector( p, &inst->FullDstRegisters[0] ),
368 get_result_flags( inst ),
369 sampler,
370 coord,
371 opcode);
372 }
373
374
375 /**
376 * Generate a simple arithmetic instruction
377 * \param opcode the i915 opcode
378 * \param numArgs the number of input/src arguments
379 */
380 static void
381 emit_simple_arith(struct i915_fp_compile *p,
382 const struct tgsi_full_instruction *inst,
383 uint opcode, uint numArgs)
384 {
385 uint arg1, arg2, arg3;
386
387 assert(numArgs <= 3);
388
389 arg1 = (numArgs < 1) ? 0 : src_vector( p, &inst->FullSrcRegisters[0] );
390 arg2 = (numArgs < 2) ? 0 : src_vector( p, &inst->FullSrcRegisters[1] );
391 arg3 = (numArgs < 3) ? 0 : src_vector( p, &inst->FullSrcRegisters[2] );
392
393 i915_emit_arith( p,
394 opcode,
395 get_result_vector( p, &inst->FullDstRegisters[0]),
396 get_result_flags( inst ), 0,
397 arg1,
398 arg2,
399 arg3 );
400 }
401
402
403 /** As above, but swap the first two src regs */
404 static void
405 emit_simple_arith_swap2(struct i915_fp_compile *p,
406 const struct tgsi_full_instruction *inst,
407 uint opcode, uint numArgs)
408 {
409 struct tgsi_full_instruction inst2;
410
411 assert(numArgs == 2);
412
413 /* transpose first two registers */
414 inst2 = *inst;
415 inst2.FullSrcRegisters[0] = inst->FullSrcRegisters[1];
416 inst2.FullSrcRegisters[1] = inst->FullSrcRegisters[0];
417
418 emit_simple_arith(p, &inst2, opcode, numArgs);
419 }
420
421
422 #ifndef M_PI
423 #define M_PI 3.14159265358979323846
424 #endif
425
426 /*
427 * Translate TGSI instruction to i915 instruction.
428 *
429 * Possible concerns:
430 *
431 * SIN, COS -- could use another taylor step?
432 * LIT -- results seem a little different to sw mesa
433 * LOG -- different to mesa on negative numbers, but this is conformant.
434 */
435 static void
436 i915_translate_instruction(struct i915_fp_compile *p,
437 const struct tgsi_full_instruction *inst)
438 {
439 uint writemask;
440 uint src0, src1, src2, flags;
441 uint tmp = 0;
442
443 switch (inst->Instruction.Opcode) {
444 case TGSI_OPCODE_ABS:
445 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
446 i915_emit_arith(p,
447 A0_MAX,
448 get_result_vector(p, &inst->FullDstRegisters[0]),
449 get_result_flags(inst), 0,
450 src0, negate(src0, 1, 1, 1, 1), 0);
451 break;
452
453 case TGSI_OPCODE_ADD:
454 emit_simple_arith(p, inst, A0_ADD, 2);
455 break;
456
457 case TGSI_OPCODE_CMP:
458 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
459 src1 = src_vector(p, &inst->FullSrcRegisters[1]);
460 src2 = src_vector(p, &inst->FullSrcRegisters[2]);
461 i915_emit_arith(p, A0_CMP,
462 get_result_vector(p, &inst->FullDstRegisters[0]),
463 get_result_flags(inst),
464 0, src0, src2, src1); /* NOTE: order of src2, src1 */
465 break;
466
467 case TGSI_OPCODE_COS:
468 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
469 tmp = i915_get_utemp(p);
470
471 i915_emit_arith(p,
472 A0_MUL,
473 tmp, A0_DEST_CHANNEL_X, 0,
474 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
475
476 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
477
478 /* By choosing different taylor constants, could get rid of this mul:
479 */
480 i915_emit_arith(p,
481 A0_MUL,
482 tmp, A0_DEST_CHANNEL_X, 0,
483 tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0);
484
485 /*
486 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
487 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, 1
488 * t0 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
489 * result = DP4 t0, cos_constants
490 */
491 i915_emit_arith(p,
492 A0_MUL,
493 tmp, A0_DEST_CHANNEL_XY, 0,
494 swizzle(tmp, X, X, ONE, ONE),
495 swizzle(tmp, X, ONE, ONE, ONE), 0);
496
497 i915_emit_arith(p,
498 A0_MUL,
499 tmp, A0_DEST_CHANNEL_XYZ, 0,
500 swizzle(tmp, X, Y, X, ONE),
501 swizzle(tmp, X, X, ONE, ONE), 0);
502
503 i915_emit_arith(p,
504 A0_MUL,
505 tmp, A0_DEST_CHANNEL_XYZ, 0,
506 swizzle(tmp, X, X, Z, ONE),
507 swizzle(tmp, Z, ONE, ONE, ONE), 0);
508
509 i915_emit_arith(p,
510 A0_DP4,
511 get_result_vector(p, &inst->FullDstRegisters[0]),
512 get_result_flags(inst), 0,
513 swizzle(tmp, ONE, Z, Y, X),
514 i915_emit_const4fv(p, cos_constants), 0);
515 break;
516
517 case TGSI_OPCODE_DP3:
518 emit_simple_arith(p, inst, A0_DP3, 2);
519 break;
520
521 case TGSI_OPCODE_DP4:
522 emit_simple_arith(p, inst, A0_DP4, 2);
523 break;
524
525 case TGSI_OPCODE_DPH:
526 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
527 src1 = src_vector(p, &inst->FullSrcRegisters[1]);
528
529 i915_emit_arith(p,
530 A0_DP4,
531 get_result_vector(p, &inst->FullDstRegisters[0]),
532 get_result_flags(inst), 0,
533 swizzle(src0, X, Y, Z, ONE), src1, 0);
534 break;
535
536 case TGSI_OPCODE_DST:
537 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
538 src1 = src_vector(p, &inst->FullSrcRegisters[1]);
539
540 /* result[0] = 1 * 1;
541 * result[1] = a[1] * b[1];
542 * result[2] = a[2] * 1;
543 * result[3] = 1 * b[3];
544 */
545 i915_emit_arith(p,
546 A0_MUL,
547 get_result_vector(p, &inst->FullDstRegisters[0]),
548 get_result_flags(inst), 0,
549 swizzle(src0, ONE, Y, Z, ONE),
550 swizzle(src1, ONE, Y, ONE, W), 0);
551 break;
552
553 case TGSI_OPCODE_END:
554 /* no-op */
555 break;
556
557 case TGSI_OPCODE_EX2:
558 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
559
560 i915_emit_arith(p,
561 A0_EXP,
562 get_result_vector(p, &inst->FullDstRegisters[0]),
563 get_result_flags(inst), 0,
564 swizzle(src0, X, X, X, X), 0, 0);
565 break;
566
567 case TGSI_OPCODE_FLR:
568 emit_simple_arith(p, inst, A0_FLR, 1);
569 break;
570
571 case TGSI_OPCODE_FRC:
572 emit_simple_arith(p, inst, A0_FRC, 1);
573 break;
574
575 case TGSI_OPCODE_KIL:
576 /* kill if src[0].x < 0 || src[0].y < 0 ... */
577 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
578 tmp = i915_get_utemp(p);
579
580 i915_emit_texld(p,
581 tmp, /* dest reg: a dummy reg */
582 A0_DEST_CHANNEL_ALL, /* dest writemask */
583 0, /* sampler */
584 src0, /* coord*/
585 T0_TEXKILL); /* opcode */
586 break;
587
588 case TGSI_OPCODE_KILP:
589 assert(0); /* not tested yet */
590 break;
591
592 case TGSI_OPCODE_LG2:
593 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
594
595 i915_emit_arith(p,
596 A0_LOG,
597 get_result_vector(p, &inst->FullDstRegisters[0]),
598 get_result_flags(inst), 0,
599 swizzle(src0, X, X, X, X), 0, 0);
600 break;
601
602 case TGSI_OPCODE_LIT:
603 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
604 tmp = i915_get_utemp(p);
605
606 /* tmp = max( a.xyzw, a.00zw )
607 * XXX: Clamp tmp.w to -128..128
608 * tmp.y = log(tmp.y)
609 * tmp.y = tmp.w * tmp.y
610 * tmp.y = exp(tmp.y)
611 * result = cmp (a.11-x1, a.1x01, a.1xy1 )
612 */
613 i915_emit_arith(p, A0_MAX, tmp, A0_DEST_CHANNEL_ALL, 0,
614 src0, swizzle(src0, ZERO, ZERO, Z, W), 0);
615
616 i915_emit_arith(p, A0_LOG, tmp, A0_DEST_CHANNEL_Y, 0,
617 swizzle(tmp, Y, Y, Y, Y), 0, 0);
618
619 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_Y, 0,
620 swizzle(tmp, ZERO, Y, ZERO, ZERO),
621 swizzle(tmp, ZERO, W, ZERO, ZERO), 0);
622
623 i915_emit_arith(p, A0_EXP, tmp, A0_DEST_CHANNEL_Y, 0,
624 swizzle(tmp, Y, Y, Y, Y), 0, 0);
625
626 i915_emit_arith(p, A0_CMP,
627 get_result_vector(p, &inst->FullDstRegisters[0]),
628 get_result_flags(inst), 0,
629 negate(swizzle(tmp, ONE, ONE, X, ONE), 0, 0, 1, 0),
630 swizzle(tmp, ONE, X, ZERO, ONE),
631 swizzle(tmp, ONE, X, Y, ONE));
632
633 break;
634
635 case TGSI_OPCODE_LRP:
636 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
637 src1 = src_vector(p, &inst->FullSrcRegisters[1]);
638 src2 = src_vector(p, &inst->FullSrcRegisters[2]);
639 flags = get_result_flags(inst);
640 tmp = i915_get_utemp(p);
641
642 /* b*a + c*(1-a)
643 *
644 * b*a + c - ca
645 *
646 * tmp = b*a + c,
647 * result = (-c)*a + tmp
648 */
649 i915_emit_arith(p, A0_MAD, tmp,
650 flags & A0_DEST_CHANNEL_ALL, 0, src1, src0, src2);
651
652 i915_emit_arith(p, A0_MAD,
653 get_result_vector(p, &inst->FullDstRegisters[0]),
654 flags, 0, negate(src2, 1, 1, 1, 1), src0, tmp);
655 break;
656
657 case TGSI_OPCODE_MAD:
658 emit_simple_arith(p, inst, A0_MAD, 3);
659 break;
660
661 case TGSI_OPCODE_MAX:
662 emit_simple_arith(p, inst, A0_MAX, 2);
663 break;
664
665 case TGSI_OPCODE_MIN:
666 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
667 src1 = src_vector(p, &inst->FullSrcRegisters[1]);
668 tmp = i915_get_utemp(p);
669 flags = get_result_flags(inst);
670
671 i915_emit_arith(p,
672 A0_MAX,
673 tmp, flags & A0_DEST_CHANNEL_ALL, 0,
674 negate(src0, 1, 1, 1, 1),
675 negate(src1, 1, 1, 1, 1), 0);
676
677 i915_emit_arith(p,
678 A0_MOV,
679 get_result_vector(p, &inst->FullDstRegisters[0]),
680 flags, 0, negate(tmp, 1, 1, 1, 1), 0, 0);
681 break;
682
683 case TGSI_OPCODE_MOV:
684 case TGSI_OPCODE_SWZ:
685 emit_simple_arith(p, inst, A0_MOV, 1);
686 break;
687
688 case TGSI_OPCODE_MUL:
689 emit_simple_arith(p, inst, A0_MUL, 2);
690 break;
691
692 case TGSI_OPCODE_POW:
693 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
694 src1 = src_vector(p, &inst->FullSrcRegisters[1]);
695 tmp = i915_get_utemp(p);
696 flags = get_result_flags(inst);
697
698 /* XXX: masking on intermediate values, here and elsewhere.
699 */
700 i915_emit_arith(p,
701 A0_LOG,
702 tmp, A0_DEST_CHANNEL_X, 0,
703 swizzle(src0, X, X, X, X), 0, 0);
704
705 i915_emit_arith(p, A0_MUL, tmp, A0_DEST_CHANNEL_X, 0, tmp, src1, 0);
706
707 i915_emit_arith(p,
708 A0_EXP,
709 get_result_vector(p, &inst->FullDstRegisters[0]),
710 flags, 0, swizzle(tmp, X, X, X, X), 0, 0);
711 break;
712
713 case TGSI_OPCODE_RET:
714 /* XXX: no-op? */
715 break;
716
717 case TGSI_OPCODE_RCP:
718 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
719
720 i915_emit_arith(p,
721 A0_RCP,
722 get_result_vector(p, &inst->FullDstRegisters[0]),
723 get_result_flags(inst), 0,
724 swizzle(src0, X, X, X, X), 0, 0);
725 break;
726
727 case TGSI_OPCODE_RSQ:
728 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
729
730 i915_emit_arith(p,
731 A0_RSQ,
732 get_result_vector(p, &inst->FullDstRegisters[0]),
733 get_result_flags(inst), 0,
734 swizzle(src0, X, X, X, X), 0, 0);
735 break;
736
737 case TGSI_OPCODE_SCS:
738 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
739 tmp = i915_get_utemp(p);
740
741 /*
742 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
743 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
744 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
745 * scs.x = DP4 t1, sin_constants
746 * t1 = MUL t0.xxz1 t0.z111 ; x^6 x^4 x^2 1
747 * scs.y = DP4 t1, cos_constants
748 */
749 i915_emit_arith(p,
750 A0_MUL,
751 tmp, A0_DEST_CHANNEL_XY, 0,
752 swizzle(src0, X, X, ONE, ONE),
753 swizzle(src0, X, ONE, ONE, ONE), 0);
754
755 i915_emit_arith(p,
756 A0_MUL,
757 tmp, A0_DEST_CHANNEL_ALL, 0,
758 swizzle(tmp, X, Y, X, Y),
759 swizzle(tmp, X, X, ONE, ONE), 0);
760
761 writemask = inst->FullDstRegisters[0].DstRegister.WriteMask;
762
763 if (writemask & TGSI_WRITEMASK_Y) {
764 uint tmp1;
765
766 if (writemask & TGSI_WRITEMASK_X)
767 tmp1 = i915_get_utemp(p);
768 else
769 tmp1 = tmp;
770
771 i915_emit_arith(p,
772 A0_MUL,
773 tmp1, A0_DEST_CHANNEL_ALL, 0,
774 swizzle(tmp, X, Y, Y, W),
775 swizzle(tmp, X, Z, ONE, ONE), 0);
776
777 i915_emit_arith(p,
778 A0_DP4,
779 get_result_vector(p, &inst->FullDstRegisters[0]),
780 A0_DEST_CHANNEL_Y, 0,
781 swizzle(tmp1, W, Z, Y, X),
782 i915_emit_const4fv(p, sin_constants), 0);
783 }
784
785 if (writemask & TGSI_WRITEMASK_X) {
786 i915_emit_arith(p,
787 A0_MUL,
788 tmp, A0_DEST_CHANNEL_XYZ, 0,
789 swizzle(tmp, X, X, Z, ONE),
790 swizzle(tmp, Z, ONE, ONE, ONE), 0);
791
792 i915_emit_arith(p,
793 A0_DP4,
794 get_result_vector(p, &inst->FullDstRegisters[0]),
795 A0_DEST_CHANNEL_X, 0,
796 swizzle(tmp, ONE, Z, Y, X),
797 i915_emit_const4fv(p, cos_constants), 0);
798 }
799 break;
800
801 case TGSI_OPCODE_SGE:
802 emit_simple_arith(p, inst, A0_SGE, 2);
803 break;
804
805 case TGSI_OPCODE_SLE:
806 /* like SGE, but swap reg0, reg1 */
807 emit_simple_arith_swap2(p, inst, A0_SGE, 2);
808 break;
809
810 case TGSI_OPCODE_SIN:
811 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
812 tmp = i915_get_utemp(p);
813
814 i915_emit_arith(p,
815 A0_MUL,
816 tmp, A0_DEST_CHANNEL_X, 0,
817 src0, i915_emit_const1f(p, 1.0f / (float) (M_PI * 2.0)), 0);
818
819 i915_emit_arith(p, A0_MOD, tmp, A0_DEST_CHANNEL_X, 0, tmp, 0, 0);
820
821 /* By choosing different taylor constants, could get rid of this mul:
822 */
823 i915_emit_arith(p,
824 A0_MUL,
825 tmp, A0_DEST_CHANNEL_X, 0,
826 tmp, i915_emit_const1f(p, (float) (M_PI * 2.0)), 0);
827
828 /*
829 * t0.xy = MUL x.xx11, x.x1111 ; x^2, x, 1, 1
830 * t0 = MUL t0.xyxy t0.xx11 ; x^4, x^3, x^2, x
831 * t1 = MUL t0.xyyw t0.yz11 ; x^7 x^5 x^3 x
832 * result = DP4 t1.wzyx, sin_constants
833 */
834 i915_emit_arith(p,
835 A0_MUL,
836 tmp, A0_DEST_CHANNEL_XY, 0,
837 swizzle(tmp, X, X, ONE, ONE),
838 swizzle(tmp, X, ONE, ONE, ONE), 0);
839
840 i915_emit_arith(p,
841 A0_MUL,
842 tmp, A0_DEST_CHANNEL_ALL, 0,
843 swizzle(tmp, X, Y, X, Y),
844 swizzle(tmp, X, X, ONE, ONE), 0);
845
846 i915_emit_arith(p,
847 A0_MUL,
848 tmp, A0_DEST_CHANNEL_ALL, 0,
849 swizzle(tmp, X, Y, Y, W),
850 swizzle(tmp, X, Z, ONE, ONE), 0);
851
852 i915_emit_arith(p,
853 A0_DP4,
854 get_result_vector(p, &inst->FullDstRegisters[0]),
855 get_result_flags(inst), 0,
856 swizzle(tmp, W, Z, Y, X),
857 i915_emit_const4fv(p, sin_constants), 0);
858 break;
859
860 case TGSI_OPCODE_SLT:
861 emit_simple_arith(p, inst, A0_SLT, 2);
862 break;
863
864 case TGSI_OPCODE_SGT:
865 /* like SLT, but swap reg0, reg1 */
866 emit_simple_arith_swap2(p, inst, A0_SLT, 2);
867 break;
868
869 case TGSI_OPCODE_SUB:
870 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
871 src1 = src_vector(p, &inst->FullSrcRegisters[1]);
872
873 i915_emit_arith(p,
874 A0_ADD,
875 get_result_vector(p, &inst->FullDstRegisters[0]),
876 get_result_flags(inst), 0,
877 src0, negate(src1, 1, 1, 1, 1), 0);
878 break;
879
880 case TGSI_OPCODE_TEX:
881 emit_tex(p, inst, T0_TEXLD);
882 break;
883
884 case TGSI_OPCODE_TXB:
885 emit_tex(p, inst, T0_TEXLDB);
886 break;
887
888 case TGSI_OPCODE_TXP:
889 emit_tex(p, inst, T0_TEXLDP);
890 break;
891
892 case TGSI_OPCODE_XPD:
893 /* Cross product:
894 * result.x = src0.y * src1.z - src0.z * src1.y;
895 * result.y = src0.z * src1.x - src0.x * src1.z;
896 * result.z = src0.x * src1.y - src0.y * src1.x;
897 * result.w = undef;
898 */
899 src0 = src_vector(p, &inst->FullSrcRegisters[0]);
900 src1 = src_vector(p, &inst->FullSrcRegisters[1]);
901 tmp = i915_get_utemp(p);
902
903 i915_emit_arith(p,
904 A0_MUL,
905 tmp, A0_DEST_CHANNEL_ALL, 0,
906 swizzle(src0, Z, X, Y, ONE),
907 swizzle(src1, Y, Z, X, ONE), 0);
908
909 i915_emit_arith(p,
910 A0_MAD,
911 get_result_vector(p, &inst->FullDstRegisters[0]),
912 get_result_flags(inst), 0,
913 swizzle(src0, Y, Z, X, ONE),
914 swizzle(src1, Z, X, Y, ONE),
915 negate(tmp, 1, 1, 1, 0));
916 break;
917
918 default:
919 i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
920 p->error = 1;
921 return;
922 }
923
924 i915_release_utemps(p);
925 }
926
927
928 /**
929 * Translate TGSI fragment shader into i915 hardware instructions.
930 * \param p the translation state
931 * \param tokens the TGSI token array
932 */
933 static void
934 i915_translate_instructions(struct i915_fp_compile *p,
935 const struct tgsi_token *tokens)
936 {
937 struct i915_fragment_shader *ifs = p->shader;
938 struct tgsi_parse_context parse;
939
940 tgsi_parse_init( &parse, tokens );
941
942 while( !tgsi_parse_end_of_tokens( &parse ) ) {
943
944 tgsi_parse_token( &parse );
945
946 switch( parse.FullToken.Token.Type ) {
947 case TGSI_TOKEN_TYPE_DECLARATION:
948 if (parse.FullToken.FullDeclaration.Declaration.File
949 == TGSI_FILE_CONSTANT) {
950 uint i;
951 for (i = parse.FullToken.FullDeclaration.DeclarationRange.First;
952 i <= parse.FullToken.FullDeclaration.DeclarationRange.Last;
953 i++) {
954 assert(ifs->constant_flags[i] == 0x0);
955 ifs->constant_flags[i] = I915_CONSTFLAG_USER;
956 ifs->num_constants = MAX2(ifs->num_constants, i + 1);
957 }
958 }
959 else if (parse.FullToken.FullDeclaration.Declaration.File
960 == TGSI_FILE_TEMPORARY) {
961 uint i;
962 for (i = parse.FullToken.FullDeclaration.DeclarationRange.First;
963 i <= parse.FullToken.FullDeclaration.DeclarationRange.Last;
964 i++) {
965 assert(i < I915_MAX_TEMPORARY);
966 /* XXX just use shader->info->file_mask[TGSI_FILE_TEMPORARY] */
967 p->temp_flag |= (1 << i); /* mark temp as used */
968 }
969 }
970 break;
971
972 case TGSI_TOKEN_TYPE_IMMEDIATE:
973 {
974 const struct tgsi_full_immediate *imm
975 = &parse.FullToken.FullImmediate;
976 const uint pos = p->num_immediates++;
977 uint j;
978 assert( imm->Immediate.NrTokens <= 4 + 1 );
979 for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
980 p->immediates[pos][j] = imm->u[j].Float;
981 }
982 }
983 break;
984
985 case TGSI_TOKEN_TYPE_INSTRUCTION:
986 if (p->first_instruction) {
987 /* resolve location of immediates */
988 uint i, j;
989 for (i = 0; i < p->num_immediates; i++) {
990 /* find constant slot for this immediate */
991 for (j = 0; j < I915_MAX_CONSTANT; j++) {
992 if (ifs->constant_flags[j] == 0x0) {
993 memcpy(ifs->constants[j],
994 p->immediates[i],
995 4 * sizeof(float));
996 /*printf("immediate %d maps to const %d\n", i, j);*/
997 ifs->constant_flags[j] = 0xf; /* all four comps used */
998 p->immediates_map[i] = j;
999 ifs->num_constants = MAX2(ifs->num_constants, j + 1);
1000 break;
1001 }
1002 }
1003 }
1004
1005 p->first_instruction = FALSE;
1006 }
1007
1008 i915_translate_instruction(p, &parse.FullToken.FullInstruction);
1009 break;
1010
1011 default:
1012 assert( 0 );
1013 }
1014
1015 } /* while */
1016
1017 tgsi_parse_free (&parse);
1018 }
1019
1020
1021 static struct i915_fp_compile *
1022 i915_init_compile(struct i915_context *i915,
1023 struct i915_fragment_shader *ifs)
1024 {
1025 struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
1026
1027 p->shader = ifs;
1028
1029 /* Put new constants at end of const buffer, growing downward.
1030 * The problem is we don't know how many user-defined constants might
1031 * be specified with pipe->set_constant_buffer().
1032 * Should pre-scan the user's program to determine the highest-numbered
1033 * constant referenced.
1034 */
1035 ifs->num_constants = 0;
1036 memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags));
1037
1038 p->first_instruction = TRUE;
1039
1040 p->nr_tex_indirect = 1; /* correct? */
1041 p->nr_tex_insn = 0;
1042 p->nr_alu_insn = 0;
1043 p->nr_decl_insn = 0;
1044
1045 p->csr = p->program;
1046 p->decl = p->declarations;
1047 p->decl_s = 0;
1048 p->decl_t = 0;
1049 p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
1050 p->utemp_flag = ~0x7;
1051
1052 p->wpos_tex = -1;
1053
1054 /* initialize the first program word */
1055 *(p->decl++) = _3DSTATE_PIXEL_SHADER_PROGRAM;
1056
1057 return p;
1058 }
1059
1060
1061 /* Copy compile results to the fragment program struct and destroy the
1062 * compilation context.
1063 */
1064 static void
1065 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
1066 {
1067 struct i915_fragment_shader *ifs = p->shader;
1068 unsigned long program_size = (unsigned long) (p->csr - p->program);
1069 unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
1070
1071 if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
1072 i915_program_error(p, "Exceeded max nr indirect texture lookups");
1073
1074 if (p->nr_tex_insn > I915_MAX_TEX_INSN)
1075 i915_program_error(p, "Exceeded max TEX instructions");
1076
1077 if (p->nr_alu_insn > I915_MAX_ALU_INSN)
1078 i915_program_error(p, "Exceeded max ALU instructions");
1079
1080 if (p->nr_decl_insn > I915_MAX_DECL_INSN)
1081 i915_program_error(p, "Exceeded max DECL instructions");
1082
1083 if (p->error) {
1084 p->NumNativeInstructions = 0;
1085 p->NumNativeAluInstructions = 0;
1086 p->NumNativeTexInstructions = 0;
1087 p->NumNativeTexIndirections = 0;
1088
1089 i915_use_passthrough_shader(ifs);
1090 }
1091 else {
1092 p->NumNativeInstructions
1093 = p->nr_alu_insn + p->nr_tex_insn + p->nr_decl_insn;
1094 p->NumNativeAluInstructions = p->nr_alu_insn;
1095 p->NumNativeTexInstructions = p->nr_tex_insn;
1096 p->NumNativeTexIndirections = p->nr_tex_indirect;
1097
1098 /* patch in the program length */
1099 p->declarations[0] |= program_size + decl_size - 2;
1100
1101 /* Copy compilation results to fragment program struct:
1102 */
1103 assert(!ifs->program);
1104 ifs->program
1105 = (uint *) MALLOC((program_size + decl_size) * sizeof(uint));
1106 if (ifs->program) {
1107 ifs->program_len = program_size + decl_size;
1108
1109 memcpy(ifs->program,
1110 p->declarations,
1111 decl_size * sizeof(uint));
1112
1113 memcpy(ifs->program + decl_size,
1114 p->program,
1115 program_size * sizeof(uint));
1116 }
1117 }
1118
1119 /* Release the compilation struct:
1120 */
1121 FREE(p);
1122 }
1123
1124
1125 /**
1126 * Find an unused texture coordinate slot to use for fragment WPOS.
1127 * Update p->fp->wpos_tex with the result (-1 if no used texcoord slot is found).
1128 */
1129 static void
1130 i915_find_wpos_space(struct i915_fp_compile *p)
1131 {
1132 #if 0
1133 const uint inputs
1134 = p->shader->inputs_read | (1 << TGSI_ATTRIB_POS); /*XXX hack*/
1135 uint i;
1136
1137 p->wpos_tex = -1;
1138
1139 if (inputs & (1 << TGSI_ATTRIB_POS)) {
1140 for (i = 0; i < I915_TEX_UNITS; i++) {
1141 if ((inputs & (1 << (TGSI_ATTRIB_TEX0 + i))) == 0) {
1142 p->wpos_tex = i;
1143 return;
1144 }
1145 }
1146
1147 i915_program_error(p, "No free texcoord for wpos value");
1148 }
1149 #else
1150 if (p->shader->info.input_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1151 /* frag shader using the fragment position input */
1152 #if 0
1153 assert(0);
1154 #endif
1155 }
1156 #endif
1157 }
1158
1159
1160
1161
1162 /**
1163 * Rather than trying to intercept and jiggle depth writes during
1164 * emit, just move the value into its correct position at the end of
1165 * the program:
1166 */
1167 static void
1168 i915_fixup_depth_write(struct i915_fp_compile *p)
1169 {
1170 /* XXX assuming pos/depth is always in output[0] */
1171 if (p->shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
1172 const uint depth = UREG(REG_TYPE_OD, 0);
1173
1174 i915_emit_arith(p,
1175 A0_MOV, /* opcode */
1176 depth, /* dest reg */
1177 A0_DEST_CHANNEL_W, /* write mask */
1178 0, /* saturate? */
1179 swizzle(depth, X, Y, Z, Z), /* src0 */
1180 0, 0 /* src1, src2 */);
1181 }
1182 }
1183
1184
1185 void
1186 i915_translate_fragment_program( struct i915_context *i915,
1187 struct i915_fragment_shader *fs)
1188 {
1189 struct i915_fp_compile *p = i915_init_compile(i915, fs);
1190 const struct tgsi_token *tokens = fs->state.tokens;
1191
1192 i915_find_wpos_space(p);
1193
1194 #if 0
1195 tgsi_dump(tokens, 0);
1196 #endif
1197
1198 i915_translate_instructions(p, tokens);
1199 i915_fixup_depth_write(p);
1200
1201 i915_fini_compile(i915, p);
1202 }