projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge branch 'dri2'
[mesa.git] / src / mesa / drivers / dri / r300 / r300_fragprog.c
1 /*
2 * Copyright (C) 2005 Ben Skeggs.
3 *
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining
7 * a 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, sublicense, 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
16 * portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 */
27
28 /**
29 * \file
30 *
31 * \author Ben Skeggs <darktama@iinet.net.au>
32 *
33 * \author Jerome Glisse <j.glisse@gmail.com>
34 *
35 * \todo Depth write, WPOS/FOGC inputs
36 *
37 * \todo FogOption
38 *
39 * \todo Verify results of opcodes for accuracy, I've only checked them in
40 * specific cases.
41 */
42
43 #include "glheader.h"
44 #include "macros.h"
45 #include "enums.h"
46 #include "shader/prog_instruction.h"
47 #include "shader/prog_parameter.h"
48 #include "shader/prog_print.h"
49
50 #include "r300_context.h"
51 #include "r300_fragprog.h"
52 #include "r300_reg.h"
53 #include "r300_state.h"
54
55 /*
56 * Usefull macros and values
57 */
58 #define ERROR(fmt, args...) do { \
59 fprintf(stderr, "%s::%s(): " fmt "\n", \
60 __FILE__, __FUNCTION__, ##args); \
61 fp->error = GL_TRUE; \
62 } while(0)
63
64 #define PFS_INVAL 0xFFFFFFFF
65 #define COMPILE_STATE struct r300_pfs_compile_state *cs = fp->cs
66
67 #define SWIZZLE_XYZ 0
68 #define SWIZZLE_XXX 1
69 #define SWIZZLE_YYY 2
70 #define SWIZZLE_ZZZ 3
71 #define SWIZZLE_WWW 4
72 #define SWIZZLE_YZX 5
73 #define SWIZZLE_ZXY 6
74 #define SWIZZLE_WZY 7
75 #define SWIZZLE_111 8
76 #define SWIZZLE_000 9
77 #define SWIZZLE_HHH 10
78
79 #define swizzle(r, x, y, z, w) do_swizzle(fp, r, \
80 ((SWIZZLE_##x<<0)| \
81 (SWIZZLE_##y<<3)| \
82 (SWIZZLE_##z<<6)| \
83 (SWIZZLE_##w<<9)), \
84 0)
85
86 #define REG_TYPE_INPUT 0
87 #define REG_TYPE_OUTPUT 1
88 #define REG_TYPE_TEMP 2
89 #define REG_TYPE_CONST 3
90
91 #define REG_TYPE_SHIFT 0
92 #define REG_INDEX_SHIFT 2
93 #define REG_VSWZ_SHIFT 8
94 #define REG_SSWZ_SHIFT 13
95 #define REG_NEGV_SHIFT 18
96 #define REG_NEGS_SHIFT 19
97 #define REG_ABS_SHIFT 20
98 #define REG_NO_USE_SHIFT 21 // Hack for refcounting
99 #define REG_VALID_SHIFT 22 // Does the register contain a defined value?
100 #define REG_BUILTIN_SHIFT 23 // Is it a builtin (like all zero/all one)?
101
102 #define REG_TYPE_MASK (0x03 << REG_TYPE_SHIFT)
103 #define REG_INDEX_MASK (0x3F << REG_INDEX_SHIFT)
104 #define REG_VSWZ_MASK (0x1F << REG_VSWZ_SHIFT)
105 #define REG_SSWZ_MASK (0x1F << REG_SSWZ_SHIFT)
106 #define REG_NEGV_MASK (0x01 << REG_NEGV_SHIFT)
107 #define REG_NEGS_MASK (0x01 << REG_NEGS_SHIFT)
108 #define REG_ABS_MASK (0x01 << REG_ABS_SHIFT)
109 #define REG_NO_USE_MASK (0x01 << REG_NO_USE_SHIFT)
110 #define REG_VALID_MASK (0x01 << REG_VALID_SHIFT)
111 #define REG_BUILTIN_MASK (0x01 << REG_BUILTIN_SHIFT)
112
113 #define REG(type, index, vswz, sswz, nouse, valid, builtin) \
114 (((type << REG_TYPE_SHIFT) & REG_TYPE_MASK) | \
115 ((index << REG_INDEX_SHIFT) & REG_INDEX_MASK) | \
116 ((nouse << REG_NO_USE_SHIFT) & REG_NO_USE_MASK) | \
117 ((valid << REG_VALID_SHIFT) & REG_VALID_MASK) | \
118 ((builtin << REG_BUILTIN_SHIFT) & REG_BUILTIN_MASK) | \
119 ((vswz << REG_VSWZ_SHIFT) & REG_VSWZ_MASK) | \
120 ((sswz << REG_SSWZ_SHIFT) & REG_SSWZ_MASK))
121 #define REG_GET_TYPE(reg) \
122 ((reg & REG_TYPE_MASK) >> REG_TYPE_SHIFT)
123 #define REG_GET_INDEX(reg) \
124 ((reg & REG_INDEX_MASK) >> REG_INDEX_SHIFT)
125 #define REG_GET_VSWZ(reg) \
126 ((reg & REG_VSWZ_MASK) >> REG_VSWZ_SHIFT)
127 #define REG_GET_SSWZ(reg) \
128 ((reg & REG_SSWZ_MASK) >> REG_SSWZ_SHIFT)
129 #define REG_GET_NO_USE(reg) \
130 ((reg & REG_NO_USE_MASK) >> REG_NO_USE_SHIFT)
131 #define REG_GET_VALID(reg) \
132 ((reg & REG_VALID_MASK) >> REG_VALID_SHIFT)
133 #define REG_GET_BUILTIN(reg) \
134 ((reg & REG_BUILTIN_MASK) >> REG_BUILTIN_SHIFT)
135 #define REG_SET_TYPE(reg, type) \
136 reg = ((reg & ~REG_TYPE_MASK) | \
137 ((type << REG_TYPE_SHIFT) & REG_TYPE_MASK))
138 #define REG_SET_INDEX(reg, index) \
139 reg = ((reg & ~REG_INDEX_MASK) | \
140 ((index << REG_INDEX_SHIFT) & REG_INDEX_MASK))
141 #define REG_SET_VSWZ(reg, vswz) \
142 reg = ((reg & ~REG_VSWZ_MASK) | \
143 ((vswz << REG_VSWZ_SHIFT) & REG_VSWZ_MASK))
144 #define REG_SET_SSWZ(reg, sswz) \
145 reg = ((reg & ~REG_SSWZ_MASK) | \
146 ((sswz << REG_SSWZ_SHIFT) & REG_SSWZ_MASK))
147 #define REG_SET_NO_USE(reg, nouse) \
148 reg = ((reg & ~REG_NO_USE_MASK) | \
149 ((nouse << REG_NO_USE_SHIFT) & REG_NO_USE_MASK))
150 #define REG_SET_VALID(reg, valid) \
151 reg = ((reg & ~REG_VALID_MASK) | \
152 ((valid << REG_VALID_SHIFT) & REG_VALID_MASK))
153 #define REG_SET_BUILTIN(reg, builtin) \
154 reg = ((reg & ~REG_BUILTIN_MASK) | \
155 ((builtin << REG_BUILTIN_SHIFT) & REG_BUILTIN_MASK))
156 #define REG_ABS(reg) \
157 reg = (reg | REG_ABS_MASK)
158 #define REG_NEGV(reg) \
159 reg = (reg | REG_NEGV_MASK)
160 #define REG_NEGS(reg) \
161 reg = (reg | REG_NEGS_MASK)
162
163 /*
164 * Datas structures for fragment program generation
165 */
166
167 /* description of r300 native hw instructions */
168 static const struct {
169 const char *name;
170 int argc;
171 int v_op;
172 int s_op;
173 } r300_fpop[] = {
174 /* *INDENT-OFF* */
175 {"MAD", 3, R300_FPI0_OUTC_MAD, R300_FPI2_OUTA_MAD},
176 {"DP3", 2, R300_FPI0_OUTC_DP3, R300_FPI2_OUTA_DP4},
177 {"DP4", 2, R300_FPI0_OUTC_DP4, R300_FPI2_OUTA_DP4},
178 {"MIN", 2, R300_FPI0_OUTC_MIN, R300_FPI2_OUTA_MIN},
179 {"MAX", 2, R300_FPI0_OUTC_MAX, R300_FPI2_OUTA_MAX},
180 {"CMP", 3, R300_FPI0_OUTC_CMP, R300_FPI2_OUTA_CMP},
181 {"FRC", 1, R300_FPI0_OUTC_FRC, R300_FPI2_OUTA_FRC},
182 {"EX2", 1, R300_FPI0_OUTC_REPL_ALPHA, R300_FPI2_OUTA_EX2},
183 {"LG2", 1, R300_FPI0_OUTC_REPL_ALPHA, R300_FPI2_OUTA_LG2},
184 {"RCP", 1, R300_FPI0_OUTC_REPL_ALPHA, R300_FPI2_OUTA_RCP},
185 {"RSQ", 1, R300_FPI0_OUTC_REPL_ALPHA, R300_FPI2_OUTA_RSQ},
186 {"REPL_ALPHA", 1, R300_FPI0_OUTC_REPL_ALPHA, PFS_INVAL},
187 {"CMPH", 3, R300_FPI0_OUTC_CMPH, PFS_INVAL},
188 /* *INDENT-ON* */
189 };
190
191 /* vector swizzles r300 can support natively, with a couple of
192 * cases we handle specially
193 *
194 * REG_VSWZ/REG_SSWZ is an index into this table
195 */
196
197 /* mapping from SWIZZLE_* to r300 native values for scalar insns */
198 #define SWIZZLE_HALF 6
199
200 #define MAKE_SWZ3(x, y, z) (MAKE_SWIZZLE4(SWIZZLE_##x, \
201 SWIZZLE_##y, \
202 SWIZZLE_##z, \
203 SWIZZLE_ZERO))
204 /* native swizzles */
205 static const struct r300_pfs_swizzle {
206 GLuint hash; /* swizzle value this matches */
207 GLuint base; /* base value for hw swizzle */
208 GLuint stride; /* difference in base between arg0/1/2 */
209 GLuint flags;
210 } v_swiz[] = {
211 /* *INDENT-OFF* */
212 {MAKE_SWZ3(X, Y, Z), R300_FPI0_ARGC_SRC0C_XYZ, 4, SLOT_SRC_VECTOR},
213 {MAKE_SWZ3(X, X, X), R300_FPI0_ARGC_SRC0C_XXX, 4, SLOT_SRC_VECTOR},
214 {MAKE_SWZ3(Y, Y, Y), R300_FPI0_ARGC_SRC0C_YYY, 4, SLOT_SRC_VECTOR},
215 {MAKE_SWZ3(Z, Z, Z), R300_FPI0_ARGC_SRC0C_ZZZ, 4, SLOT_SRC_VECTOR},
216 {MAKE_SWZ3(W, W, W), R300_FPI0_ARGC_SRC0A, 1, SLOT_SRC_SCALAR},
217 {MAKE_SWZ3(Y, Z, X), R300_FPI0_ARGC_SRC0C_YZX, 1, SLOT_SRC_VECTOR},
218 {MAKE_SWZ3(Z, X, Y), R300_FPI0_ARGC_SRC0C_ZXY, 1, SLOT_SRC_VECTOR},
219 {MAKE_SWZ3(W, Z, Y), R300_FPI0_ARGC_SRC0CA_WZY, 1, SLOT_SRC_BOTH},
220 {MAKE_SWZ3(ONE, ONE, ONE), R300_FPI0_ARGC_ONE, 0, 0},
221 {MAKE_SWZ3(ZERO, ZERO, ZERO), R300_FPI0_ARGC_ZERO, 0, 0},
222 {MAKE_SWZ3(HALF, HALF, HALF), R300_FPI0_ARGC_HALF, 0, 0},
223 {PFS_INVAL, 0, 0, 0},
224 /* *INDENT-ON* */
225 };
226
227 /* used during matching of non-native swizzles */
228 #define SWZ_X_MASK (7 << 0)
229 #define SWZ_Y_MASK (7 << 3)
230 #define SWZ_Z_MASK (7 << 6)
231 #define SWZ_W_MASK (7 << 9)
232 static const struct {
233 GLuint hash; /* used to mask matching swizzle components */
234 int mask; /* actual outmask */
235 int count; /* count of components matched */
236 } s_mask[] = {
237 /* *INDENT-OFF* */
238 {SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK, 1 | 2 | 4, 3},
239 {SWZ_X_MASK | SWZ_Y_MASK, 1 | 2, 2},
240 {SWZ_X_MASK | SWZ_Z_MASK, 1 | 4, 2},
241 {SWZ_Y_MASK | SWZ_Z_MASK, 2 | 4, 2},
242 {SWZ_X_MASK, 1, 1},
243 {SWZ_Y_MASK, 2, 1},
244 {SWZ_Z_MASK, 4, 1},
245 {PFS_INVAL, PFS_INVAL, PFS_INVAL}
246 /* *INDENT-ON* */
247 };
248
249 static const struct {
250 int base; /* hw value of swizzle */
251 int stride; /* difference between SRC0/1/2 */
252 GLuint flags;
253 } s_swiz[] = {
254 /* *INDENT-OFF* */
255 {R300_FPI2_ARGA_SRC0C_X, 3, SLOT_SRC_VECTOR},
256 {R300_FPI2_ARGA_SRC0C_Y, 3, SLOT_SRC_VECTOR},
257 {R300_FPI2_ARGA_SRC0C_Z, 3, SLOT_SRC_VECTOR},
258 {R300_FPI2_ARGA_SRC0A, 1, SLOT_SRC_SCALAR},
259 {R300_FPI2_ARGA_ZERO, 0, 0},
260 {R300_FPI2_ARGA_ONE, 0, 0},
261 {R300_FPI2_ARGA_HALF, 0, 0}
262 /* *INDENT-ON* */
263 };
264
265 /* boiler-plate reg, for convenience */
266 static const GLuint undef = REG(REG_TYPE_TEMP,
267 0,
268 SWIZZLE_XYZ,
269 SWIZZLE_W,
270 GL_FALSE,
271 GL_FALSE,
272 GL_FALSE);
273
274 /* constant one source */
275 static const GLuint pfs_one = REG(REG_TYPE_CONST,
276 0,
277 SWIZZLE_111,
278 SWIZZLE_ONE,
279 GL_FALSE,
280 GL_TRUE,
281 GL_TRUE);
282
283 /* constant half source */
284 static const GLuint pfs_half = REG(REG_TYPE_CONST,
285 0,
286 SWIZZLE_HHH,
287 SWIZZLE_HALF,
288 GL_FALSE,
289 GL_TRUE,
290 GL_TRUE);
291
292 /* constant zero source */
293 static const GLuint pfs_zero = REG(REG_TYPE_CONST,
294 0,
295 SWIZZLE_000,
296 SWIZZLE_ZERO,
297 GL_FALSE,
298 GL_TRUE,
299 GL_TRUE);
300
301 /*
302 * Common functions prototypes
303 */
304 static void dump_program(struct r300_fragment_program *fp);
305 static void emit_arith(struct r300_fragment_program *fp, int op,
306 GLuint dest, int mask,
307 GLuint src0, GLuint src1, GLuint src2, int flags);
308
309 /**
310 * Get an R300 temporary that can be written to in the given slot.
311 */
312 static int get_hw_temp(struct r300_fragment_program *fp, int slot)
313 {
314 COMPILE_STATE;
315 int r;
316
317 for (r = 0; r < PFS_NUM_TEMP_REGS; ++r) {
318 if (cs->hwtemps[r].free >= 0 && cs->hwtemps[r].free <= slot)
319 break;
320 }
321
322 if (r >= PFS_NUM_TEMP_REGS) {
323 ERROR("Out of hardware temps\n");
324 return 0;
325 }
326 // Reserved is used to avoid the following scenario:
327 // R300 temporary X is first assigned to Mesa temporary Y during vector ops
328 // R300 temporary X is then assigned to Mesa temporary Z for further vector ops
329 // Then scalar ops on Mesa temporary Z are emitted and move back in time
330 // to overwrite the value of temporary Y.
331 // End scenario.
332 cs->hwtemps[r].reserved = cs->hwtemps[r].free;
333 cs->hwtemps[r].free = -1;
334
335 // Reset to some value that won't mess things up when the user
336 // tries to read from a temporary that hasn't been assigned a value yet.
337 // In the normal case, vector_valid and scalar_valid should be set to
338 // a sane value by the first emit that writes to this temporary.
339 cs->hwtemps[r].vector_valid = 0;
340 cs->hwtemps[r].scalar_valid = 0;
341
342 if (r > fp->max_temp_idx)
343 fp->max_temp_idx = r;
344
345 return r;
346 }
347
348 /**
349 * Get an R300 temporary that will act as a TEX destination register.
350 */
351 static int get_hw_temp_tex(struct r300_fragment_program *fp)
352 {
353 COMPILE_STATE;
354 int r;
355
356 for (r = 0; r < PFS_NUM_TEMP_REGS; ++r) {
357 if (cs->used_in_node & (1 << r))
358 continue;
359
360 // Note: Be very careful here
361 if (cs->hwtemps[r].free >= 0 && cs->hwtemps[r].free <= 0)
362 break;
363 }
364
365 if (r >= PFS_NUM_TEMP_REGS)
366 return get_hw_temp(fp, 0); /* Will cause an indirection */
367
368 cs->hwtemps[r].reserved = cs->hwtemps[r].free;
369 cs->hwtemps[r].free = -1;
370
371 // Reset to some value that won't mess things up when the user
372 // tries to read from a temporary that hasn't been assigned a value yet.
373 // In the normal case, vector_valid and scalar_valid should be set to
374 // a sane value by the first emit that writes to this temporary.
375 cs->hwtemps[r].vector_valid = cs->nrslots;
376 cs->hwtemps[r].scalar_valid = cs->nrslots;
377
378 if (r > fp->max_temp_idx)
379 fp->max_temp_idx = r;
380
381 return r;
382 }
383
384 /**
385 * Mark the given hardware register as free.
386 */
387 static void free_hw_temp(struct r300_fragment_program *fp, int idx)
388 {
389 COMPILE_STATE;
390
391 // Be very careful here. Consider sequences like
392 // MAD r0, r1,r2,r3
393 // TEX r4, ...
394 // The TEX instruction may be moved in front of the MAD instruction
395 // due to the way nodes work. We don't want to alias r1 and r4 in
396 // this case.
397 // I'm certain the register allocation could be further sanitized,
398 // but it's tricky because of stuff that can happen inside emit_tex
399 // and emit_arith.
400 cs->hwtemps[idx].free = cs->nrslots + 1;
401 }
402
403 /**
404 * Create a new Mesa temporary register.
405 */
406 static GLuint get_temp_reg(struct r300_fragment_program *fp)
407 {
408 COMPILE_STATE;
409 GLuint r = undef;
410 GLuint index;
411
412 index = ffs(~cs->temp_in_use);
413 if (!index) {
414 ERROR("Out of program temps\n");
415 return r;
416 }
417
418 cs->temp_in_use |= (1 << --index);
419 cs->temps[index].refcount = 0xFFFFFFFF;
420 cs->temps[index].reg = -1;
421
422 REG_SET_TYPE(r, REG_TYPE_TEMP);
423 REG_SET_INDEX(r, index);
424 REG_SET_VALID(r, GL_TRUE);
425 return r;
426 }
427
428 /**
429 * Create a new Mesa temporary register that will act as the destination
430 * register for a texture read.
431 */
432 static GLuint get_temp_reg_tex(struct r300_fragment_program *fp)
433 {
434 COMPILE_STATE;
435 GLuint r = undef;
436 GLuint index;
437
438 index = ffs(~cs->temp_in_use);
439 if (!index) {
440 ERROR("Out of program temps\n");
441 return r;
442 }
443
444 cs->temp_in_use |= (1 << --index);
445 cs->temps[index].refcount = 0xFFFFFFFF;
446 cs->temps[index].reg = get_hw_temp_tex(fp);
447
448 REG_SET_TYPE(r, REG_TYPE_TEMP);
449 REG_SET_INDEX(r, index);
450 REG_SET_VALID(r, GL_TRUE);
451 return r;
452 }
453
454 /**
455 * Free a Mesa temporary and the associated R300 temporary.
456 */
457 static void free_temp(struct r300_fragment_program *fp, GLuint r)
458 {
459 COMPILE_STATE;
460 GLuint index = REG_GET_INDEX(r);
461
462 if (!(cs->temp_in_use & (1 << index)))
463 return;
464
465 if (REG_GET_TYPE(r) == REG_TYPE_TEMP) {
466 free_hw_temp(fp, cs->temps[index].reg);
467 cs->temps[index].reg = -1;
468 cs->temp_in_use &= ~(1 << index);
469 } else if (REG_GET_TYPE(r) == REG_TYPE_INPUT) {
470 free_hw_temp(fp, cs->inputs[index].reg);
471 cs->inputs[index].reg = -1;
472 }
473 }
474
475 /**
476 * Emit a hardware constant/parameter.
477 *
478 * \p cp Stable pointer to an array of 4 floats.
479 * The pointer must be stable in the sense that it remains to be valid
480 * and hold the contents of the constant/parameter throughout the lifetime
481 * of the fragment program (actually, up until the next time the fragment
482 * program is translated).
483 */
484 static GLuint emit_const4fv(struct r300_fragment_program *fp,
485 const GLfloat * cp)
486 {
487 GLuint reg = undef;
488 int index;
489
490 for (index = 0; index < fp->const_nr; ++index) {
491 if (fp->constant[index] == cp)
492 break;
493 }
494
495 if (index >= fp->const_nr) {
496 if (index >= PFS_NUM_CONST_REGS) {
497 ERROR("Out of hw constants!\n");
498 return reg;
499 }
500
501 fp->const_nr++;
502 fp->constant[index] = cp;
503 }
504
505 REG_SET_TYPE(reg, REG_TYPE_CONST);
506 REG_SET_INDEX(reg, index);
507 REG_SET_VALID(reg, GL_TRUE);
508 return reg;
509 }
510
511 static inline GLuint negate(GLuint r)
512 {
513 REG_NEGS(r);
514 REG_NEGV(r);
515 return r;
516 }
517
518 /* Hack, to prevent clobbering sources used multiple times when
519 * emulating non-native instructions
520 */
521 static inline GLuint keep(GLuint r)
522 {
523 REG_SET_NO_USE(r, GL_TRUE);
524 return r;
525 }
526
527 static inline GLuint absolute(GLuint r)
528 {
529 REG_ABS(r);
530 return r;
531 }
532
533 static int swz_native(struct r300_fragment_program *fp,
534 GLuint src, GLuint * r, GLuint arbneg)
535 {
536 /* Native swizzle, handle negation */
537 src = (src & ~REG_NEGS_MASK) | (((arbneg >> 3) & 1) << REG_NEGS_SHIFT);
538
539 if ((arbneg & 0x7) == 0x0) {
540 src = src & ~REG_NEGV_MASK;
541 *r = src;
542 } else if ((arbneg & 0x7) == 0x7) {
543 src |= REG_NEGV_MASK;
544 *r = src;
545 } else {
546 if (!REG_GET_VALID(*r))
547 *r = get_temp_reg(fp);
548 src |= REG_NEGV_MASK;
549 emit_arith(fp,
550 PFS_OP_MAD,
551 *r, arbneg & 0x7, keep(src), pfs_one, pfs_zero, 0);
552 src = src & ~REG_NEGV_MASK;
553 emit_arith(fp,
554 PFS_OP_MAD,
555 *r,
556 (arbneg ^ 0x7) | WRITEMASK_W,
557 src, pfs_one, pfs_zero, 0);
558 }
559
560 return 3;
561 }
562
563 static int swz_emit_partial(struct r300_fragment_program *fp,
564 GLuint src,
565 GLuint * r, int mask, int mc, GLuint arbneg)
566 {
567 GLuint tmp;
568 GLuint wmask = 0;
569
570 if (!REG_GET_VALID(*r))
571 *r = get_temp_reg(fp);
572
573 /* A partial match, VSWZ/mask define what parts of the
574 * desired swizzle we match
575 */
576 if (mc + s_mask[mask].count == 3) {
577 wmask = WRITEMASK_W;
578 src |= ((arbneg >> 3) & 1) << REG_NEGS_SHIFT;
579 }
580
581 tmp = arbneg & s_mask[mask].mask;
582 if (tmp) {
583 tmp = tmp ^ s_mask[mask].mask;
584 if (tmp) {
585 emit_arith(fp,
586 PFS_OP_MAD,
587 *r,
588 arbneg & s_mask[mask].mask,
589 keep(src) | REG_NEGV_MASK,
590 pfs_one, pfs_zero, 0);
591 if (!wmask) {
592 REG_SET_NO_USE(src, GL_TRUE);
593 } else {
594 REG_SET_NO_USE(src, GL_FALSE);
595 }
596 emit_arith(fp,
597 PFS_OP_MAD,
598 *r, tmp | wmask, src, pfs_one, pfs_zero, 0);
599 } else {
600 if (!wmask) {
601 REG_SET_NO_USE(src, GL_TRUE);
602 } else {
603 REG_SET_NO_USE(src, GL_FALSE);
604 }
605 emit_arith(fp,
606 PFS_OP_MAD,
607 *r,
608 (arbneg & s_mask[mask].mask) | wmask,
609 src | REG_NEGV_MASK, pfs_one, pfs_zero, 0);
610 }
611 } else {
612 if (!wmask) {
613 REG_SET_NO_USE(src, GL_TRUE);
614 } else {
615 REG_SET_NO_USE(src, GL_FALSE);
616 }
617 emit_arith(fp, PFS_OP_MAD,
618 *r,
619 s_mask[mask].mask | wmask,
620 src, pfs_one, pfs_zero, 0);
621 }
622
623 return s_mask[mask].count;
624 }
625
626 static GLuint do_swizzle(struct r300_fragment_program *fp,
627 GLuint src, GLuint arbswz, GLuint arbneg)
628 {
629 GLuint r = undef;
630 GLuint vswz;
631 int c_mask = 0;
632 int v_match = 0;
633
634 /* If swizzling from something without an XYZW native swizzle,
635 * emit result to a temp, and do new swizzle from the temp.
636 */
637 #if 0
638 if (REG_GET_VSWZ(src) != SWIZZLE_XYZ || REG_GET_SSWZ(src) != SWIZZLE_W) {
639 GLuint temp = get_temp_reg(fp);
640 emit_arith(fp,
641 PFS_OP_MAD,
642 temp, WRITEMASK_XYZW, src, pfs_one, pfs_zero, 0);
643 src = temp;
644 }
645 #endif
646
647 if (REG_GET_VSWZ(src) != SWIZZLE_XYZ || REG_GET_SSWZ(src) != SWIZZLE_W) {
648 GLuint vsrcswz =
649 (v_swiz[REG_GET_VSWZ(src)].
650 hash & (SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK)) |
651 REG_GET_SSWZ(src) << 9;
652 GLint i;
653
654 GLuint newswz = 0;
655 GLuint offset;
656 for (i = 0; i < 4; ++i) {
657 offset = GET_SWZ(arbswz, i);
658
659 newswz |=
660 (offset <= 3) ? GET_SWZ(vsrcswz,
661 offset) << i *
662 3 : offset << i * 3;
663 }
664
665 arbswz = newswz & (SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK);
666 REG_SET_SSWZ(src, GET_SWZ(newswz, 3));
667 } else {
668 /* set scalar swizzling */
669 REG_SET_SSWZ(src, GET_SWZ(arbswz, 3));
670
671 }
672 do {
673 vswz = REG_GET_VSWZ(src);
674 do {
675 int chash;
676
677 REG_SET_VSWZ(src, vswz);
678 chash = v_swiz[REG_GET_VSWZ(src)].hash &
679 s_mask[c_mask].hash;
680
681 if (chash == (arbswz & s_mask[c_mask].hash)) {
682 if (s_mask[c_mask].count == 3) {
683 v_match += swz_native(fp,
684 src, &r, arbneg);
685 } else {
686 v_match += swz_emit_partial(fp,
687 src,
688 &r,
689 c_mask,
690 v_match,
691 arbneg);
692 }
693
694 if (v_match == 3)
695 return r;
696
697 /* Fill with something invalid.. all 0's was
698 * wrong before, matched SWIZZLE_X. So all
699 * 1's will be okay for now
700 */
701 arbswz |= (PFS_INVAL & s_mask[c_mask].hash);
702 }
703 } while (v_swiz[++vswz].hash != PFS_INVAL);
704 REG_SET_VSWZ(src, SWIZZLE_XYZ);
705 } while (s_mask[++c_mask].hash != PFS_INVAL);
706
707 ERROR("should NEVER get here\n");
708 return r;
709 }
710
711 static GLuint t_src(struct r300_fragment_program *fp,
712 struct prog_src_register fpsrc)
713 {
714 GLuint r = undef;
715
716 switch (fpsrc.File) {
717 case PROGRAM_TEMPORARY:
718 REG_SET_INDEX(r, fpsrc.Index);
719 REG_SET_VALID(r, GL_TRUE);
720 REG_SET_TYPE(r, REG_TYPE_TEMP);
721 break;
722 case PROGRAM_INPUT:
723 REG_SET_INDEX(r, fpsrc.Index);
724 REG_SET_VALID(r, GL_TRUE);
725 REG_SET_TYPE(r, REG_TYPE_INPUT);
726 break;
727 case PROGRAM_LOCAL_PARAM:
728 r = emit_const4fv(fp,
729 fp->mesa_program.Base.LocalParams[fpsrc.
730 Index]);
731 break;
732 case PROGRAM_ENV_PARAM:
733 r = emit_const4fv(fp,
734 fp->ctx->FragmentProgram.Parameters[fpsrc.
735 Index]);
736 break;
737 case PROGRAM_STATE_VAR:
738 case PROGRAM_NAMED_PARAM:
739 r = emit_const4fv(fp,
740 fp->mesa_program.Base.Parameters->
741 ParameterValues[fpsrc.Index]);
742 break;
743 default:
744 ERROR("unknown SrcReg->File %x\n", fpsrc.File);
745 return r;
746 }
747
748 /* no point swizzling ONE/ZERO/HALF constants... */
749 if (REG_GET_VSWZ(r) < SWIZZLE_111 || REG_GET_SSWZ(r) < SWIZZLE_ZERO)
750 r = do_swizzle(fp, r, fpsrc.Swizzle, fpsrc.NegateBase);
751 return r;
752 }
753
754 static GLuint t_scalar_src(struct r300_fragment_program *fp,
755 struct prog_src_register fpsrc)
756 {
757 struct prog_src_register src = fpsrc;
758 int sc = GET_SWZ(fpsrc.Swizzle, 0); /* X */
759
760 src.Swizzle = ((sc << 0) | (sc << 3) | (sc << 6) | (sc << 9));
761
762 return t_src(fp, src);
763 }
764
765 static GLuint t_dst(struct r300_fragment_program *fp,
766 struct prog_dst_register dest)
767 {
768 GLuint r = undef;
769
770 switch (dest.File) {
771 case PROGRAM_TEMPORARY:
772 REG_SET_INDEX(r, dest.Index);
773 REG_SET_VALID(r, GL_TRUE);
774 REG_SET_TYPE(r, REG_TYPE_TEMP);
775 return r;
776 case PROGRAM_OUTPUT:
777 REG_SET_TYPE(r, REG_TYPE_OUTPUT);
778 switch (dest.Index) {
779 case FRAG_RESULT_COLR:
780 case FRAG_RESULT_DEPR:
781 REG_SET_INDEX(r, dest.Index);
782 REG_SET_VALID(r, GL_TRUE);
783 return r;
784 default:
785 ERROR("Bad DstReg->Index 0x%x\n", dest.Index);
786 return r;
787 }
788 default:
789 ERROR("Bad DstReg->File 0x%x\n", dest.File);
790 return r;
791 }
792 }
793
794 static int t_hw_src(struct r300_fragment_program *fp, GLuint src, GLboolean tex)
795 {
796 COMPILE_STATE;
797 int idx;
798 int index = REG_GET_INDEX(src);
799
800 switch (REG_GET_TYPE(src)) {
801 case REG_TYPE_TEMP:
802 /* NOTE: if reg==-1 here, a source is being read that
803 * hasn't been written to. Undefined results.
804 */
805 if (cs->temps[index].reg == -1)
806 cs->temps[index].reg = get_hw_temp(fp, cs->nrslots);
807
808 idx = cs->temps[index].reg;
809
810 if (!REG_GET_NO_USE(src) && (--cs->temps[index].refcount == 0))
811 free_temp(fp, src);
812 break;
813 case REG_TYPE_INPUT:
814 idx = cs->inputs[index].reg;
815
816 if (!REG_GET_NO_USE(src) && (--cs->inputs[index].refcount == 0))
817 free_hw_temp(fp, cs->inputs[index].reg);
818 break;
819 case REG_TYPE_CONST:
820 return (index | SRC_CONST);
821 default:
822 ERROR("Invalid type for source reg\n");
823 return (0 | SRC_CONST);
824 }
825
826 if (!tex)
827 cs->used_in_node |= (1 << idx);
828
829 return idx;
830 }
831
832 static int t_hw_dst(struct r300_fragment_program *fp,
833 GLuint dest, GLboolean tex, int slot)
834 {
835 COMPILE_STATE;
836 int idx;
837 GLuint index = REG_GET_INDEX(dest);
838 assert(REG_GET_VALID(dest));
839
840 switch (REG_GET_TYPE(dest)) {
841 case REG_TYPE_TEMP:
842 if (cs->temps[REG_GET_INDEX(dest)].reg == -1) {
843 if (!tex) {
844 cs->temps[index].reg = get_hw_temp(fp, slot);
845 } else {
846 cs->temps[index].reg = get_hw_temp_tex(fp);
847 }
848 }
849 idx = cs->temps[index].reg;
850
851 if (!REG_GET_NO_USE(dest) && (--cs->temps[index].refcount == 0))
852 free_temp(fp, dest);
853
854 cs->dest_in_node |= (1 << idx);
855 cs->used_in_node |= (1 << idx);
856 break;
857 case REG_TYPE_OUTPUT:
858 switch (index) {
859 case FRAG_RESULT_COLR:
860 fp->node[fp->cur_node].flags |=
861 R300_PFS_NODE_OUTPUT_COLOR;
862 break;
863 case FRAG_RESULT_DEPR:
864 fp->node[fp->cur_node].flags |=
865 R300_PFS_NODE_OUTPUT_DEPTH;
866 break;
867 }
868 return index;
869 break;
870 default:
871 ERROR("invalid dest reg type %d\n", REG_GET_TYPE(dest));
872 return 0;
873 }
874
875 return idx;
876 }
877
878 static void emit_nop(struct r300_fragment_program *fp)
879 {
880 COMPILE_STATE;
881
882 if (cs->nrslots >= PFS_MAX_ALU_INST) {
883 ERROR("Out of ALU instruction slots\n");
884 return;
885 }
886
887 fp->alu.inst[cs->nrslots].inst0 = NOP_INST0;
888 fp->alu.inst[cs->nrslots].inst1 = NOP_INST1;
889 fp->alu.inst[cs->nrslots].inst2 = NOP_INST2;
890 fp->alu.inst[cs->nrslots].inst3 = NOP_INST3;
891 cs->nrslots++;
892 }
893
894 static void emit_tex(struct r300_fragment_program *fp,
895 struct prog_instruction *fpi, int opcode)
896 {
897 COMPILE_STATE;
898 GLuint coord = t_src(fp, fpi->SrcReg[0]);
899 GLuint dest = undef, rdest = undef;
900 GLuint din, uin;
901 int unit = fpi->TexSrcUnit;
902 int hwsrc, hwdest;
903 GLuint tempreg = 0;
904
905 uin = cs->used_in_node;
906 din = cs->dest_in_node;
907
908 /* Resolve source/dest to hardware registers */
909 if (opcode != R300_FPITX_OP_KIL) {
910 if (fpi->TexSrcTarget == TEXTURE_RECT_INDEX) {
911 /**
912 * Hardware uses [0..1]x[0..1] range for rectangle textures
913 * instead of [0..Width]x[0..Height].
914 * Add a scaling instruction.
915 *
916 * \todo Refactor this once we have proper rewriting/optimization
917 * support for programs.
918 */
919 gl_state_index tokens[STATE_LENGTH] = {
920 STATE_INTERNAL, STATE_R300_TEXRECT_FACTOR, 0, 0,
921 0
922 };
923 int factor_index;
924 GLuint factorreg;
925
926 tokens[2] = unit;
927 factor_index =
928 _mesa_add_state_reference(fp->mesa_program.Base.
929 Parameters, tokens);
930 factorreg =
931 emit_const4fv(fp,
932 fp->mesa_program.Base.Parameters->
933 ParameterValues[factor_index]);
934 tempreg = keep(get_temp_reg(fp));
935
936 emit_arith(fp, PFS_OP_MAD, tempreg, WRITEMASK_XYZW,
937 coord, factorreg, pfs_zero, 0);
938
939 /* Ensure correct node indirection */
940 uin = cs->used_in_node;
941 din = cs->dest_in_node;
942
943 hwsrc = t_hw_src(fp, tempreg, GL_TRUE);
944 } else {
945 hwsrc = t_hw_src(fp, coord, GL_TRUE);
946 }
947
948 dest = t_dst(fp, fpi->DstReg);
949
950 /* r300 doesn't seem to be able to do TEX->output reg */
951 if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
952 rdest = dest;
953 dest = get_temp_reg_tex(fp);
954 } else if (fpi->DstReg.WriteMask != WRITEMASK_XYZW) {
955 /* in case write mask isn't XYZW */
956 rdest = dest;
957 dest = get_temp_reg_tex(fp);
958 }
959 hwdest =
960 t_hw_dst(fp, dest, GL_TRUE,
961 fp->node[fp->cur_node].alu_offset);
962
963 /* Use a temp that hasn't been used in this node, rather
964 * than causing an indirection
965 */
966 if (uin & (1 << hwdest)) {
967 free_hw_temp(fp, hwdest);
968 hwdest = get_hw_temp_tex(fp);
969 cs->temps[REG_GET_INDEX(dest)].reg = hwdest;
970 }
971 } else {
972 hwdest = 0;
973 unit = 0;
974 hwsrc = t_hw_src(fp, coord, GL_TRUE);
975 }
976
977 /* Indirection if source has been written in this node, or if the
978 * dest has been read/written in this node
979 */
980 if ((REG_GET_TYPE(coord) != REG_TYPE_CONST &&
981 (din & (1 << hwsrc))) || (uin & (1 << hwdest))) {
982
983 /* Finish off current node */
984 if (fp->node[fp->cur_node].alu_offset == cs->nrslots)
985 emit_nop(fp);
986
987 fp->node[fp->cur_node].alu_end =
988 cs->nrslots - fp->node[fp->cur_node].alu_offset - 1;
989 assert(fp->node[fp->cur_node].alu_end >= 0);
990
991 if (++fp->cur_node >= PFS_MAX_TEX_INDIRECT) {
992 ERROR("too many levels of texture indirection\n");
993 return;
994 }
995
996 /* Start new node */
997 fp->node[fp->cur_node].tex_offset = fp->tex.length;
998 fp->node[fp->cur_node].alu_offset = cs->nrslots;
999 fp->node[fp->cur_node].tex_end = -1;
1000 fp->node[fp->cur_node].alu_end = -1;
1001 fp->node[fp->cur_node].flags = 0;
1002 cs->used_in_node = 0;
1003 cs->dest_in_node = 0;
1004 }
1005
1006 if (fp->cur_node == 0)
1007 fp->first_node_has_tex = 1;
1008
1009 fp->tex.inst[fp->tex.length++] = 0 | (hwsrc << R300_FPITX_SRC_SHIFT)
1010 | (hwdest << R300_FPITX_DST_SHIFT)
1011 | (unit << R300_FPITX_IMAGE_SHIFT)
1012 /* not entirely sure about this */
1013 | (opcode << R300_FPITX_OPCODE_SHIFT);
1014
1015 cs->dest_in_node |= (1 << hwdest);
1016 if (REG_GET_TYPE(coord) != REG_TYPE_CONST)
1017 cs->used_in_node |= (1 << hwsrc);
1018
1019 fp->node[fp->cur_node].tex_end++;
1020
1021 /* Copy from temp to output if needed */
1022 if (REG_GET_VALID(rdest)) {
1023 emit_arith(fp, PFS_OP_MAD, rdest, fpi->DstReg.WriteMask, dest,
1024 pfs_one, pfs_zero, 0);
1025 free_temp(fp, dest);
1026 }
1027
1028 /* Free temp register */
1029 if (tempreg != 0)
1030 free_temp(fp, tempreg);
1031 }
1032
1033 /**
1034 * Returns the first slot where we could possibly allow writing to dest,
1035 * according to register allocation.
1036 */
1037 static int get_earliest_allowed_write(struct r300_fragment_program *fp,
1038 GLuint dest, int mask)
1039 {
1040 COMPILE_STATE;
1041 int idx;
1042 int pos;
1043 GLuint index = REG_GET_INDEX(dest);
1044 assert(REG_GET_VALID(dest));
1045
1046 switch (REG_GET_TYPE(dest)) {
1047 case REG_TYPE_TEMP:
1048 if (cs->temps[index].reg == -1)
1049 return 0;
1050
1051 idx = cs->temps[index].reg;
1052 break;
1053 case REG_TYPE_OUTPUT:
1054 return 0;
1055 default:
1056 ERROR("invalid dest reg type %d\n", REG_GET_TYPE(dest));
1057 return 0;
1058 }
1059
1060 pos = cs->hwtemps[idx].reserved;
1061 if (mask & WRITEMASK_XYZ) {
1062 if (pos < cs->hwtemps[idx].vector_lastread)
1063 pos = cs->hwtemps[idx].vector_lastread;
1064 }
1065 if (mask & WRITEMASK_W) {
1066 if (pos < cs->hwtemps[idx].scalar_lastread)
1067 pos = cs->hwtemps[idx].scalar_lastread;
1068 }
1069
1070 return pos;
1071 }
1072
1073 /**
1074 * Allocates a slot for an ALU instruction that can consist of
1075 * a vertex part or a scalar part or both.
1076 *
1077 * Sources from src (src[0] to src[argc-1]) are added to the slot in the
1078 * appropriate position (vector and/or scalar), and their positions are
1079 * recorded in the srcpos array.
1080 *
1081 * This function emits instruction code for the source fetch and the
1082 * argument selection. It does not emit instruction code for the
1083 * opcode or the destination selection.
1084 *
1085 * @return the index of the slot
1086 */
1087 static int find_and_prepare_slot(struct r300_fragment_program *fp,
1088 GLboolean emit_vop,
1089 GLboolean emit_sop,
1090 int argc, GLuint * src, GLuint dest, int mask)
1091 {
1092 COMPILE_STATE;
1093 int hwsrc[3];
1094 int srcpos[3];
1095 unsigned int used;
1096 int tempused;
1097 int tempvsrc[3];
1098 int tempssrc[3];
1099 int pos;
1100 int regnr;
1101 int i, j;
1102
1103 // Determine instruction slots, whether sources are required on
1104 // vector or scalar side, and the smallest slot number where
1105 // all source registers are available
1106 used = 0;
1107 if (emit_vop)
1108 used |= SLOT_OP_VECTOR;
1109 if (emit_sop)
1110 used |= SLOT_OP_SCALAR;
1111
1112 pos = get_earliest_allowed_write(fp, dest, mask);
1113
1114 if (fp->node[fp->cur_node].alu_offset > pos)
1115 pos = fp->node[fp->cur_node].alu_offset;
1116 for (i = 0; i < argc; ++i) {
1117 if (!REG_GET_BUILTIN(src[i])) {
1118 if (emit_vop)
1119 used |= v_swiz[REG_GET_VSWZ(src[i])].flags << i;
1120 if (emit_sop)
1121 used |= s_swiz[REG_GET_SSWZ(src[i])].flags << i;
1122 }
1123
1124 hwsrc[i] = t_hw_src(fp, src[i], GL_FALSE); /* Note: sideeffects wrt refcounting! */
1125 regnr = hwsrc[i] & 31;
1126
1127 if (REG_GET_TYPE(src[i]) == REG_TYPE_TEMP) {
1128 if (used & (SLOT_SRC_VECTOR << i)) {
1129 if (cs->hwtemps[regnr].vector_valid > pos)
1130 pos = cs->hwtemps[regnr].vector_valid;
1131 }
1132 if (used & (SLOT_SRC_SCALAR << i)) {
1133 if (cs->hwtemps[regnr].scalar_valid > pos)
1134 pos = cs->hwtemps[regnr].scalar_valid;
1135 }
1136 }
1137 }
1138
1139 // Find a slot that fits
1140 for (;; ++pos) {
1141 if (cs->slot[pos].used & used & SLOT_OP_BOTH)
1142 continue;
1143
1144 if (pos >= cs->nrslots) {
1145 if (cs->nrslots >= PFS_MAX_ALU_INST) {
1146 ERROR("Out of ALU instruction slots\n");
1147 return -1;
1148 }
1149
1150 fp->alu.inst[pos].inst0 = NOP_INST0;
1151 fp->alu.inst[pos].inst1 = NOP_INST1;
1152 fp->alu.inst[pos].inst2 = NOP_INST2;
1153 fp->alu.inst[pos].inst3 = NOP_INST3;
1154
1155 cs->nrslots++;
1156 }
1157 // Note: When we need both parts (vector and scalar) of a source,
1158 // we always try to put them into the same position. This makes the
1159 // code easier to read, and it is optimal (i.e. one doesn't gain
1160 // anything by splitting the parts).
1161 // It also avoids headaches with swizzles that access both parts (i.e WXY)
1162 tempused = cs->slot[pos].used;
1163 for (i = 0; i < 3; ++i) {
1164 tempvsrc[i] = cs->slot[pos].vsrc[i];
1165 tempssrc[i] = cs->slot[pos].ssrc[i];
1166 }
1167
1168 for (i = 0; i < argc; ++i) {
1169 int flags = (used >> i) & SLOT_SRC_BOTH;
1170
1171 if (!flags) {
1172 srcpos[i] = 0;
1173 continue;
1174 }
1175
1176 for (j = 0; j < 3; ++j) {
1177 if ((tempused >> j) & flags & SLOT_SRC_VECTOR) {
1178 if (tempvsrc[j] != hwsrc[i])
1179 continue;
1180 }
1181
1182 if ((tempused >> j) & flags & SLOT_SRC_SCALAR) {
1183 if (tempssrc[j] != hwsrc[i])
1184 continue;
1185 }
1186
1187 break;
1188 }
1189
1190 if (j == 3)
1191 break;
1192
1193 srcpos[i] = j;
1194 tempused |= flags << j;
1195 if (flags & SLOT_SRC_VECTOR)
1196 tempvsrc[j] = hwsrc[i];
1197 if (flags & SLOT_SRC_SCALAR)
1198 tempssrc[j] = hwsrc[i];
1199 }
1200
1201 if (i == argc)
1202 break;
1203 }
1204
1205 // Found a slot, reserve it
1206 cs->slot[pos].used = tempused | (used & SLOT_OP_BOTH);
1207 for (i = 0; i < 3; ++i) {
1208 cs->slot[pos].vsrc[i] = tempvsrc[i];
1209 cs->slot[pos].ssrc[i] = tempssrc[i];
1210 }
1211
1212 for (i = 0; i < argc; ++i) {
1213 if (REG_GET_TYPE(src[i]) == REG_TYPE_TEMP) {
1214 int regnr = hwsrc[i] & 31;
1215
1216 if (used & (SLOT_SRC_VECTOR << i)) {
1217 if (cs->hwtemps[regnr].vector_lastread < pos)
1218 cs->hwtemps[regnr].vector_lastread =
1219 pos;
1220 }
1221 if (used & (SLOT_SRC_SCALAR << i)) {
1222 if (cs->hwtemps[regnr].scalar_lastread < pos)
1223 cs->hwtemps[regnr].scalar_lastread =
1224 pos;
1225 }
1226 }
1227 }
1228
1229 // Emit the source fetch code
1230 fp->alu.inst[pos].inst1 &= ~R300_FPI1_SRC_MASK;
1231 fp->alu.inst[pos].inst1 |=
1232 ((cs->slot[pos].vsrc[0] << R300_FPI1_SRC0C_SHIFT) |
1233 (cs->slot[pos].vsrc[1] << R300_FPI1_SRC1C_SHIFT) |
1234 (cs->slot[pos].vsrc[2] << R300_FPI1_SRC2C_SHIFT));
1235
1236 fp->alu.inst[pos].inst3 &= ~R300_FPI3_SRC_MASK;
1237 fp->alu.inst[pos].inst3 |=
1238 ((cs->slot[pos].ssrc[0] << R300_FPI3_SRC0A_SHIFT) |
1239 (cs->slot[pos].ssrc[1] << R300_FPI3_SRC1A_SHIFT) |
1240 (cs->slot[pos].ssrc[2] << R300_FPI3_SRC2A_SHIFT));
1241
1242 // Emit the argument selection code
1243 if (emit_vop) {
1244 int swz[3];
1245
1246 for (i = 0; i < 3; ++i) {
1247 if (i < argc) {
1248 swz[i] = (v_swiz[REG_GET_VSWZ(src[i])].base +
1249 (srcpos[i] *
1250 v_swiz[REG_GET_VSWZ(src[i])].
1251 stride)) | ((src[i] & REG_NEGV_MASK)
1252 ? ARG_NEG : 0) | ((src[i]
1253 &
1254 REG_ABS_MASK)
1255 ?
1256 ARG_ABS
1257 : 0);
1258 } else {
1259 swz[i] = R300_FPI0_ARGC_ZERO;
1260 }
1261 }
1262
1263 fp->alu.inst[pos].inst0 &=
1264 ~(R300_FPI0_ARG0C_MASK | R300_FPI0_ARG1C_MASK |
1265 R300_FPI0_ARG2C_MASK);
1266 fp->alu.inst[pos].inst0 |=
1267 (swz[0] << R300_FPI0_ARG0C_SHIFT) | (swz[1] <<
1268 R300_FPI0_ARG1C_SHIFT)
1269 | (swz[2] << R300_FPI0_ARG2C_SHIFT);
1270 }
1271
1272 if (emit_sop) {
1273 int swz[3];
1274
1275 for (i = 0; i < 3; ++i) {
1276 if (i < argc) {
1277 swz[i] = (s_swiz[REG_GET_SSWZ(src[i])].base +
1278 (srcpos[i] *
1279 s_swiz[REG_GET_SSWZ(src[i])].
1280 stride)) | ((src[i] & REG_NEGV_MASK)
1281 ? ARG_NEG : 0) | ((src[i]
1282 &
1283 REG_ABS_MASK)
1284 ?
1285 ARG_ABS
1286 : 0);
1287 } else {
1288 swz[i] = R300_FPI2_ARGA_ZERO;
1289 }
1290 }
1291
1292 fp->alu.inst[pos].inst2 &=
1293 ~(R300_FPI2_ARG0A_MASK | R300_FPI2_ARG1A_MASK |
1294 R300_FPI2_ARG2A_MASK);
1295 fp->alu.inst[pos].inst2 |=
1296 (swz[0] << R300_FPI2_ARG0A_SHIFT) | (swz[1] <<
1297 R300_FPI2_ARG1A_SHIFT)
1298 | (swz[2] << R300_FPI2_ARG2A_SHIFT);
1299 }
1300
1301 return pos;
1302 }
1303
1304 /**
1305 * Append an ALU instruction to the instruction list.
1306 */
1307 static void emit_arith(struct r300_fragment_program *fp,
1308 int op,
1309 GLuint dest,
1310 int mask,
1311 GLuint src0, GLuint src1, GLuint src2, int flags)
1312 {
1313 COMPILE_STATE;
1314 GLuint src[3] = { src0, src1, src2 };
1315 int hwdest;
1316 GLboolean emit_vop, emit_sop;
1317 int vop, sop, argc;
1318 int pos;
1319
1320 vop = r300_fpop[op].v_op;
1321 sop = r300_fpop[op].s_op;
1322 argc = r300_fpop[op].argc;
1323
1324 if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT &&
1325 REG_GET_INDEX(dest) == FRAG_RESULT_DEPR) {
1326 if (mask & WRITEMASK_Z) {
1327 mask = WRITEMASK_W;
1328 } else {
1329 return;
1330 }
1331 }
1332
1333 emit_vop = GL_FALSE;
1334 emit_sop = GL_FALSE;
1335 if ((mask & WRITEMASK_XYZ) || vop == R300_FPI0_OUTC_DP3)
1336 emit_vop = GL_TRUE;
1337 if ((mask & WRITEMASK_W) || vop == R300_FPI0_OUTC_REPL_ALPHA)
1338 emit_sop = GL_TRUE;
1339
1340 pos =
1341 find_and_prepare_slot(fp, emit_vop, emit_sop, argc, src, dest,
1342 mask);
1343 if (pos < 0)
1344 return;
1345
1346 hwdest = t_hw_dst(fp, dest, GL_FALSE, pos); /* Note: Side effects wrt register allocation */
1347
1348 if (flags & PFS_FLAG_SAT) {
1349 vop |= R300_FPI0_OUTC_SAT;
1350 sop |= R300_FPI2_OUTA_SAT;
1351 }
1352
1353 /* Throw the pieces together and get FPI0/1 */
1354 if (emit_vop) {
1355 fp->alu.inst[pos].inst0 |= vop;
1356
1357 fp->alu.inst[pos].inst1 |= hwdest << R300_FPI1_DSTC_SHIFT;
1358
1359 if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
1360 if (REG_GET_INDEX(dest) == FRAG_RESULT_COLR) {
1361 fp->alu.inst[pos].inst1 |=
1362 (mask & WRITEMASK_XYZ) <<
1363 R300_FPI1_DSTC_OUTPUT_MASK_SHIFT;
1364 } else
1365 assert(0);
1366 } else {
1367 fp->alu.inst[pos].inst1 |=
1368 (mask & WRITEMASK_XYZ) <<
1369 R300_FPI1_DSTC_REG_MASK_SHIFT;
1370
1371 cs->hwtemps[hwdest].vector_valid = pos + 1;
1372 }
1373 }
1374
1375 /* And now FPI2/3 */
1376 if (emit_sop) {
1377 fp->alu.inst[pos].inst2 |= sop;
1378
1379 if (mask & WRITEMASK_W) {
1380 if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
1381 if (REG_GET_INDEX(dest) == FRAG_RESULT_COLR) {
1382 fp->alu.inst[pos].inst3 |=
1383 (hwdest << R300_FPI3_DSTA_SHIFT) |
1384 R300_FPI3_DSTA_OUTPUT;
1385 } else if (REG_GET_INDEX(dest) ==
1386 FRAG_RESULT_DEPR) {
1387 fp->alu.inst[pos].inst3 |=
1388 R300_FPI3_DSTA_DEPTH;
1389 } else
1390 assert(0);
1391 } else {
1392 fp->alu.inst[pos].inst3 |=
1393 (hwdest << R300_FPI3_DSTA_SHIFT) |
1394 R300_FPI3_DSTA_REG;
1395
1396 cs->hwtemps[hwdest].scalar_valid = pos + 1;
1397 }
1398 }
1399 }
1400
1401 return;
1402 }
1403
1404 #if 0
1405 static GLuint get_attrib(struct r300_fragment_program *fp, GLuint attr)
1406 {
1407 struct gl_fragment_program *mp = &fp->mesa_program;
1408 GLuint r = undef;
1409
1410 if (!(mp->Base.InputsRead & (1 << attr))) {
1411 ERROR("Attribute %d was not provided!\n", attr);
1412 return undef;
1413 }
1414
1415 REG_SET_TYPE(r, REG_TYPE_INPUT);
1416 REG_SET_INDEX(r, attr);
1417 REG_SET_VALID(r, GL_TRUE);
1418 return r;
1419 }
1420 #endif
1421
1422 static GLfloat SinCosConsts[2][4] = {
1423 {
1424 1.273239545, // 4/PI
1425 -0.405284735, // -4/(PI*PI)
1426 3.141592654, // PI
1427 0.2225 // weight
1428 },
1429 {
1430 0.75,
1431 0.0,
1432 0.159154943, // 1/(2*PI)
1433 6.283185307 // 2*PI
1434 }
1435 };
1436
1437 /**
1438 * Emit a LIT instruction.
1439 * \p flags may be PFS_FLAG_SAT
1440 *
1441 * Definition of LIT (from ARB_fragment_program):
1442 * tmp = VectorLoad(op0);
1443 * if (tmp.x < 0) tmp.x = 0;
1444 * if (tmp.y < 0) tmp.y = 0;
1445 * if (tmp.w < -(128.0-epsilon)) tmp.w = -(128.0-epsilon);
1446 * else if (tmp.w > 128-epsilon) tmp.w = 128-epsilon;
1447 * result.x = 1.0;
1448 * result.y = tmp.x;
1449 * result.z = (tmp.x > 0) ? RoughApproxPower(tmp.y, tmp.w) : 0.0;
1450 * result.w = 1.0;
1451 *
1452 * The longest path of computation is the one leading to result.z,
1453 * consisting of 5 operations. This implementation of LIT takes
1454 * 5 slots. So unless there's some special undocumented opcode,
1455 * this implementation is potentially optimal. Unfortunately,
1456 * emit_arith is a bit too conservative because it doesn't understand
1457 * partial writes to the vector component.
1458 */
1459 static const GLfloat LitConst[4] =
1460 { 127.999999, 127.999999, 127.999999, -127.999999 };
1461
1462 static void emit_lit(struct r300_fragment_program *fp,
1463 GLuint dest, int mask, GLuint src, int flags)
1464 {
1465 COMPILE_STATE;
1466 GLuint cnst;
1467 int needTemporary;
1468 GLuint temp;
1469
1470 cnst = emit_const4fv(fp, LitConst);
1471
1472 needTemporary = 0;
1473 if ((mask & WRITEMASK_XYZW) != WRITEMASK_XYZW) {
1474 needTemporary = 1;
1475 } else if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
1476 // LIT is typically followed by DP3/DP4, so there's no point
1477 // in creating special code for this case
1478 needTemporary = 1;
1479 }
1480
1481 if (needTemporary) {
1482 temp = keep(get_temp_reg(fp));
1483 } else {
1484 temp = keep(dest);
1485 }
1486
1487 // Note: The order of emit_arith inside the slots is relevant,
1488 // because emit_arith only looks at scalar vs. vector when resolving
1489 // dependencies, and it does not consider individual vector components,
1490 // so swizzling between the two parts can create fake dependencies.
1491
1492 // First slot
1493 emit_arith(fp, PFS_OP_MAX, temp, WRITEMASK_XY,
1494 keep(src), pfs_zero, undef, 0);
1495 emit_arith(fp, PFS_OP_MAX, temp, WRITEMASK_W, src, cnst, undef, 0);
1496
1497 // Second slot
1498 emit_arith(fp, PFS_OP_MIN, temp, WRITEMASK_Z,
1499 swizzle(temp, W, W, W, W), cnst, undef, 0);
1500 emit_arith(fp, PFS_OP_LG2, temp, WRITEMASK_W,
1501 swizzle(temp, Y, Y, Y, Y), undef, undef, 0);
1502
1503 // Third slot
1504 // If desired, we saturate the y result here.
1505 // This does not affect the use as a condition variable in the CMP later
1506 emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_W,
1507 temp, swizzle(temp, Z, Z, Z, Z), pfs_zero, 0);
1508 emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_Y,
1509 swizzle(temp, X, X, X, X), pfs_one, pfs_zero, flags);
1510
1511 // Fourth slot
1512 emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_X,
1513 pfs_one, pfs_one, pfs_zero, 0);
1514 emit_arith(fp, PFS_OP_EX2, temp, WRITEMASK_W, temp, undef, undef, 0);
1515
1516 // Fifth slot
1517 emit_arith(fp, PFS_OP_CMP, temp, WRITEMASK_Z,
1518 pfs_zero, swizzle(temp, W, W, W, W),
1519 negate(swizzle(temp, Y, Y, Y, Y)), flags);
1520 emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_W, pfs_one, pfs_one,
1521 pfs_zero, 0);
1522
1523 if (needTemporary) {
1524 emit_arith(fp, PFS_OP_MAD, dest, mask,
1525 temp, pfs_one, pfs_zero, flags);
1526 free_temp(fp, temp);
1527 } else {
1528 // Decrease refcount of the destination
1529 t_hw_dst(fp, dest, GL_FALSE, cs->nrslots);
1530 }
1531 }
1532
1533 static GLboolean parse_program(struct r300_fragment_program *fp)
1534 {
1535 struct gl_fragment_program *mp = &fp->mesa_program;
1536 const struct prog_instruction *inst = mp->Base.Instructions;
1537 struct prog_instruction *fpi;
1538 GLuint src[3], dest, temp[2];
1539 int flags, mask = 0;
1540 int const_sin[2];
1541
1542 if (!inst || inst[0].Opcode == OPCODE_END) {
1543 ERROR("empty program?\n");
1544 return GL_FALSE;
1545 }
1546
1547 for (fpi = mp->Base.Instructions; fpi->Opcode != OPCODE_END; fpi++) {
1548 if (fpi->SaturateMode == SATURATE_ZERO_ONE)
1549 flags = PFS_FLAG_SAT;
1550 else
1551 flags = 0;
1552
1553 if (fpi->Opcode != OPCODE_KIL) {
1554 dest = t_dst(fp, fpi->DstReg);
1555 mask = fpi->DstReg.WriteMask;
1556 }
1557
1558 switch (fpi->Opcode) {
1559 case OPCODE_ABS:
1560 src[0] = t_src(fp, fpi->SrcReg[0]);
1561 emit_arith(fp, PFS_OP_MAD, dest, mask,
1562 absolute(src[0]), pfs_one, pfs_zero, flags);
1563 break;
1564 case OPCODE_ADD:
1565 src[0] = t_src(fp, fpi->SrcReg[0]);
1566 src[1] = t_src(fp, fpi->SrcReg[1]);
1567 emit_arith(fp, PFS_OP_MAD, dest, mask,
1568 src[0], pfs_one, src[1], flags);
1569 break;
1570 case OPCODE_CMP:
1571 src[0] = t_src(fp, fpi->SrcReg[0]);
1572 src[1] = t_src(fp, fpi->SrcReg[1]);
1573 src[2] = t_src(fp, fpi->SrcReg[2]);
1574 /* ARB_f_p - if src0.c < 0.0 ? src1.c : src2.c
1575 * r300 - if src2.c < 0.0 ? src1.c : src0.c
1576 */
1577 emit_arith(fp, PFS_OP_CMP, dest, mask,
1578 src[2], src[1], src[0], flags);
1579 break;
1580 case OPCODE_COS:
1581 /*
1582 * cos using a parabola (see SIN):
1583 * cos(x):
1584 * x = (x/(2*PI))+0.75
1585 * x = frac(x)
1586 * x = (x*2*PI)-PI
1587 * result = sin(x)
1588 */
1589 temp[0] = get_temp_reg(fp);
1590 const_sin[0] = emit_const4fv(fp, SinCosConsts[0]);
1591 const_sin[1] = emit_const4fv(fp, SinCosConsts[1]);
1592 src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
1593
1594 /* add 0.5*PI and do range reduction */
1595
1596 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_X,
1597 swizzle(src[0], X, X, X, X),
1598 swizzle(const_sin[1], Z, Z, Z, Z),
1599 swizzle(const_sin[1], X, X, X, X), 0);
1600
1601 emit_arith(fp, PFS_OP_FRC, temp[0], WRITEMASK_X,
1602 swizzle(temp[0], X, X, X, X),
1603 undef, undef, 0);
1604
1605 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(temp[0], X, X, X, X), swizzle(const_sin[1], W, W, W, W), //2*PI
1606 negate(swizzle(const_sin[0], Z, Z, Z, Z)), //-PI
1607 0);
1608
1609 /* SIN */
1610
1611 emit_arith(fp, PFS_OP_MAD, temp[0],
1612 WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
1613 Z, Z, Z,
1614 Z),
1615 const_sin[0], pfs_zero, 0);
1616
1617 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_X,
1618 swizzle(temp[0], Y, Y, Y, Y),
1619 absolute(swizzle(temp[0], Z, Z, Z, Z)),
1620 swizzle(temp[0], X, X, X, X), 0);
1621
1622 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Y,
1623 swizzle(temp[0], X, X, X, X),
1624 absolute(swizzle(temp[0], X, X, X, X)),
1625 negate(swizzle(temp[0], X, X, X, X)), 0);
1626
1627 emit_arith(fp, PFS_OP_MAD, dest, mask,
1628 swizzle(temp[0], Y, Y, Y, Y),
1629 swizzle(const_sin[0], W, W, W, W),
1630 swizzle(temp[0], X, X, X, X), flags);
1631
1632 free_temp(fp, temp[0]);
1633 break;
1634 case OPCODE_DP3:
1635 src[0] = t_src(fp, fpi->SrcReg[0]);
1636 src[1] = t_src(fp, fpi->SrcReg[1]);
1637 emit_arith(fp, PFS_OP_DP3, dest, mask,
1638 src[0], src[1], undef, flags);
1639 break;
1640 case OPCODE_DP4:
1641 src[0] = t_src(fp, fpi->SrcReg[0]);
1642 src[1] = t_src(fp, fpi->SrcReg[1]);
1643 emit_arith(fp, PFS_OP_DP4, dest, mask,
1644 src[0], src[1], undef, flags);
1645 break;
1646 case OPCODE_DPH:
1647 src[0] = t_src(fp, fpi->SrcReg[0]);
1648 src[1] = t_src(fp, fpi->SrcReg[1]);
1649 /* src0.xyz1 -> temp
1650 * DP4 dest, temp, src1
1651 */
1652 #if 0
1653 temp[0] = get_temp_reg(fp);
1654 src[0].s_swz = SWIZZLE_ONE;
1655 emit_arith(fp, PFS_OP_MAD, temp[0], mask,
1656 src[0], pfs_one, pfs_zero, 0);
1657 emit_arith(fp, PFS_OP_DP4, dest, mask,
1658 temp[0], src[1], undef, flags);
1659 free_temp(fp, temp[0]);
1660 #else
1661 emit_arith(fp, PFS_OP_DP4, dest, mask,
1662 swizzle(src[0], X, Y, Z, ONE), src[1],
1663 undef, flags);
1664 #endif
1665 break;
1666 case OPCODE_DST:
1667 src[0] = t_src(fp, fpi->SrcReg[0]);
1668 src[1] = t_src(fp, fpi->SrcReg[1]);
1669 /* dest.y = src0.y * src1.y */
1670 if (mask & WRITEMASK_Y)
1671 emit_arith(fp, PFS_OP_MAD, dest, WRITEMASK_Y,
1672 keep(src[0]), keep(src[1]),
1673 pfs_zero, flags);
1674 /* dest.z = src0.z */
1675 if (mask & WRITEMASK_Z)
1676 emit_arith(fp, PFS_OP_MAD, dest, WRITEMASK_Z,
1677 src[0], pfs_one, pfs_zero, flags);
1678 /* result.x = 1.0
1679 * result.w = src1.w */
1680 if (mask & WRITEMASK_XW) {
1681 REG_SET_VSWZ(src[1], SWIZZLE_111); /*Cheat */
1682 emit_arith(fp, PFS_OP_MAD, dest,
1683 mask & WRITEMASK_XW,
1684 src[1], pfs_one, pfs_zero, flags);
1685 }
1686 break;
1687 case OPCODE_EX2:
1688 src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
1689 emit_arith(fp, PFS_OP_EX2, dest, mask,
1690 src[0], undef, undef, flags);
1691 break;
1692 case OPCODE_FLR:
1693 src[0] = t_src(fp, fpi->SrcReg[0]);
1694 temp[0] = get_temp_reg(fp);
1695 /* FRC temp, src0
1696 * MAD dest, src0, 1.0, -temp
1697 */
1698 emit_arith(fp, PFS_OP_FRC, temp[0], mask,
1699 keep(src[0]), undef, undef, 0);
1700 emit_arith(fp, PFS_OP_MAD, dest, mask,
1701 src[0], pfs_one, negate(temp[0]), flags);
1702 free_temp(fp, temp[0]);
1703 break;
1704 case OPCODE_FRC:
1705 src[0] = t_src(fp, fpi->SrcReg[0]);
1706 emit_arith(fp, PFS_OP_FRC, dest, mask,
1707 src[0], undef, undef, flags);
1708 break;
1709 case OPCODE_KIL:
1710 emit_tex(fp, fpi, R300_FPITX_OP_KIL);
1711 break;
1712 case OPCODE_LG2:
1713 src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
1714 emit_arith(fp, PFS_OP_LG2, dest, mask,
1715 src[0], undef, undef, flags);
1716 break;
1717 case OPCODE_LIT:
1718 src[0] = t_src(fp, fpi->SrcReg[0]);
1719 emit_lit(fp, dest, mask, src[0], flags);
1720 break;
1721 case OPCODE_LRP:
1722 src[0] = t_src(fp, fpi->SrcReg[0]);
1723 src[1] = t_src(fp, fpi->SrcReg[1]);
1724 src[2] = t_src(fp, fpi->SrcReg[2]);
1725 /* result = tmp0tmp1 + (1 - tmp0)tmp2
1726 * = tmp0tmp1 + tmp2 + (-tmp0)tmp2
1727 * MAD temp, -tmp0, tmp2, tmp2
1728 * MAD result, tmp0, tmp1, temp
1729 */
1730 temp[0] = get_temp_reg(fp);
1731 emit_arith(fp, PFS_OP_MAD, temp[0], mask,
1732 negate(keep(src[0])), keep(src[2]), src[2],
1733 0);
1734 emit_arith(fp, PFS_OP_MAD, dest, mask,
1735 src[0], src[1], temp[0], flags);
1736 free_temp(fp, temp[0]);
1737 break;
1738 case OPCODE_MAD:
1739 src[0] = t_src(fp, fpi->SrcReg[0]);
1740 src[1] = t_src(fp, fpi->SrcReg[1]);
1741 src[2] = t_src(fp, fpi->SrcReg[2]);
1742 emit_arith(fp, PFS_OP_MAD, dest, mask,
1743 src[0], src[1], src[2], flags);
1744 break;
1745 case OPCODE_MAX:
1746 src[0] = t_src(fp, fpi->SrcReg[0]);
1747 src[1] = t_src(fp, fpi->SrcReg[1]);
1748 emit_arith(fp, PFS_OP_MAX, dest, mask,
1749 src[0], src[1], undef, flags);
1750 break;
1751 case OPCODE_MIN:
1752 src[0] = t_src(fp, fpi->SrcReg[0]);
1753 src[1] = t_src(fp, fpi->SrcReg[1]);
1754 emit_arith(fp, PFS_OP_MIN, dest, mask,
1755 src[0], src[1], undef, flags);
1756 break;
1757 case OPCODE_MOV:
1758 case OPCODE_SWZ:
1759 src[0] = t_src(fp, fpi->SrcReg[0]);
1760 emit_arith(fp, PFS_OP_MAD, dest, mask,
1761 src[0], pfs_one, pfs_zero, flags);
1762 break;
1763 case OPCODE_MUL:
1764 src[0] = t_src(fp, fpi->SrcReg[0]);
1765 src[1] = t_src(fp, fpi->SrcReg[1]);
1766 emit_arith(fp, PFS_OP_MAD, dest, mask,
1767 src[0], src[1], pfs_zero, flags);
1768 break;
1769 case OPCODE_POW:
1770 src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
1771 src[1] = t_scalar_src(fp, fpi->SrcReg[1]);
1772 temp[0] = get_temp_reg(fp);
1773 emit_arith(fp, PFS_OP_LG2, temp[0], WRITEMASK_W,
1774 src[0], undef, undef, 0);
1775 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_W,
1776 temp[0], src[1], pfs_zero, 0);
1777 emit_arith(fp, PFS_OP_EX2, dest, fpi->DstReg.WriteMask,
1778 temp[0], undef, undef, 0);
1779 free_temp(fp, temp[0]);
1780 break;
1781 case OPCODE_RCP:
1782 src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
1783 emit_arith(fp, PFS_OP_RCP, dest, mask,
1784 src[0], undef, undef, flags);
1785 break;
1786 case OPCODE_RSQ:
1787 src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
1788 emit_arith(fp, PFS_OP_RSQ, dest, mask,
1789 absolute(src[0]), pfs_zero, pfs_zero, flags);
1790 break;
1791 case OPCODE_SCS:
1792 /*
1793 * scs using a parabola :
1794 * scs(x):
1795 * result.x = sin(-abs(x)+0.5*PI) (cos)
1796 * result.y = sin(x) (sin)
1797 *
1798 */
1799 temp[0] = get_temp_reg(fp);
1800 temp[1] = get_temp_reg(fp);
1801 const_sin[0] = emit_const4fv(fp, SinCosConsts[0]);
1802 const_sin[1] = emit_const4fv(fp, SinCosConsts[1]);
1803 src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
1804
1805 /* x = -abs(x)+0.5*PI */
1806 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(const_sin[0], Z, Z, Z, Z), //PI
1807 pfs_half,
1808 negate(abs
1809 (swizzle(keep(src[0]), X, X, X, X))),
1810 0);
1811
1812 /* C*x (sin) */
1813 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_W,
1814 swizzle(const_sin[0], Y, Y, Y, Y),
1815 swizzle(keep(src[0]), X, X, X, X),
1816 pfs_zero, 0);
1817
1818 /* B*x, C*x (cos) */
1819 emit_arith(fp, PFS_OP_MAD, temp[0],
1820 WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
1821 Z, Z, Z,
1822 Z),
1823 const_sin[0], pfs_zero, 0);
1824
1825 /* B*x (sin) */
1826 emit_arith(fp, PFS_OP_MAD, temp[1], WRITEMASK_W,
1827 swizzle(const_sin[0], X, X, X, X),
1828 keep(src[0]), pfs_zero, 0);
1829
1830 /* y = B*x + C*x*abs(x) (sin) */
1831 emit_arith(fp, PFS_OP_MAD, temp[1], WRITEMASK_Z,
1832 absolute(src[0]),
1833 swizzle(temp[0], W, W, W, W),
1834 swizzle(temp[1], W, W, W, W), 0);
1835
1836 /* y = B*x + C*x*abs(x) (cos) */
1837 emit_arith(fp, PFS_OP_MAD, temp[1], WRITEMASK_W,
1838 swizzle(temp[0], Y, Y, Y, Y),
1839 absolute(swizzle(temp[0], Z, Z, Z, Z)),
1840 swizzle(temp[0], X, X, X, X), 0);
1841
1842 /* y*abs(y) - y (cos), y*abs(y) - y (sin) */
1843 emit_arith(fp, PFS_OP_MAD, temp[0],
1844 WRITEMASK_X | WRITEMASK_Y, swizzle(temp[1],
1845 W, Z, Y,
1846 X),
1847 absolute(swizzle(temp[1], W, Z, Y, X)),
1848 negate(swizzle(temp[1], W, Z, Y, X)), 0);
1849
1850 /* dest.xy = mad(temp.xy, P, temp2.wz) */
1851 emit_arith(fp, PFS_OP_MAD, dest,
1852 mask & (WRITEMASK_X | WRITEMASK_Y), temp[0],
1853 swizzle(const_sin[0], W, W, W, W),
1854 swizzle(temp[1], W, Z, Y, X), flags);
1855
1856 free_temp(fp, temp[0]);
1857 free_temp(fp, temp[1]);
1858 break;
1859 case OPCODE_SGE:
1860 src[0] = t_src(fp, fpi->SrcReg[0]);
1861 src[1] = t_src(fp, fpi->SrcReg[1]);
1862 temp[0] = get_temp_reg(fp);
1863 /* temp = src0 - src1
1864 * dest.c = (temp.c < 0.0) ? 0 : 1
1865 */
1866 emit_arith(fp, PFS_OP_MAD, temp[0], mask,
1867 src[0], pfs_one, negate(src[1]), 0);
1868 emit_arith(fp, PFS_OP_CMP, dest, mask,
1869 pfs_one, pfs_zero, temp[0], 0);
1870 free_temp(fp, temp[0]);
1871 break;
1872 case OPCODE_SIN:
1873 /*
1874 * using a parabola:
1875 * sin(x) = 4/pi * x + -4/(pi*pi) * x * abs(x)
1876 * extra precision is obtained by weighting against
1877 * itself squared.
1878 */
1879
1880 temp[0] = get_temp_reg(fp);
1881 const_sin[0] = emit_const4fv(fp, SinCosConsts[0]);
1882 const_sin[1] = emit_const4fv(fp, SinCosConsts[1]);
1883 src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
1884
1885 /* do range reduction */
1886
1887 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_X,
1888 swizzle(keep(src[0]), X, X, X, X),
1889 swizzle(const_sin[1], Z, Z, Z, Z),
1890 pfs_half, 0);
1891
1892 emit_arith(fp, PFS_OP_FRC, temp[0], WRITEMASK_X,
1893 swizzle(temp[0], X, X, X, X),
1894 undef, undef, 0);
1895
1896 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(temp[0], X, X, X, X), swizzle(const_sin[1], W, W, W, W), //2*PI
1897 negate(swizzle(const_sin[0], Z, Z, Z, Z)), //PI
1898 0);
1899
1900 /* SIN */
1901
1902 emit_arith(fp, PFS_OP_MAD, temp[0],
1903 WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
1904 Z, Z, Z,
1905 Z),
1906 const_sin[0], pfs_zero, 0);
1907
1908 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_X,
1909 swizzle(temp[0], Y, Y, Y, Y),
1910 absolute(swizzle(temp[0], Z, Z, Z, Z)),
1911 swizzle(temp[0], X, X, X, X), 0);
1912
1913 emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Y,
1914 swizzle(temp[0], X, X, X, X),
1915 absolute(swizzle(temp[0], X, X, X, X)),
1916 negate(swizzle(temp[0], X, X, X, X)), 0);
1917
1918 emit_arith(fp, PFS_OP_MAD, dest, mask,
1919 swizzle(temp[0], Y, Y, Y, Y),
1920 swizzle(const_sin[0], W, W, W, W),
1921 swizzle(temp[0], X, X, X, X), flags);
1922
1923 free_temp(fp, temp[0]);
1924 break;
1925 case OPCODE_SLT:
1926 src[0] = t_src(fp, fpi->SrcReg[0]);
1927 src[1] = t_src(fp, fpi->SrcReg[1]);
1928 temp[0] = get_temp_reg(fp);
1929 /* temp = src0 - src1
1930 * dest.c = (temp.c < 0.0) ? 1 : 0
1931 */
1932 emit_arith(fp, PFS_OP_MAD, temp[0], mask,
1933 src[0], pfs_one, negate(src[1]), 0);
1934 emit_arith(fp, PFS_OP_CMP, dest, mask,
1935 pfs_zero, pfs_one, temp[0], 0);
1936 free_temp(fp, temp[0]);
1937 break;
1938 case OPCODE_SUB:
1939 src[0] = t_src(fp, fpi->SrcReg[0]);
1940 src[1] = t_src(fp, fpi->SrcReg[1]);
1941 emit_arith(fp, PFS_OP_MAD, dest, mask,
1942 src[0], pfs_one, negate(src[1]), flags);
1943 break;
1944 case OPCODE_TEX:
1945 emit_tex(fp, fpi, R300_FPITX_OP_TEX);
1946 break;
1947 case OPCODE_TXB:
1948 emit_tex(fp, fpi, R300_FPITX_OP_TXB);
1949 break;
1950 case OPCODE_TXP:
1951 emit_tex(fp, fpi, R300_FPITX_OP_TXP);
1952 break;
1953 case OPCODE_XPD:{
1954 src[0] = t_src(fp, fpi->SrcReg[0]);
1955 src[1] = t_src(fp, fpi->SrcReg[1]);
1956 temp[0] = get_temp_reg(fp);
1957 /* temp = src0.zxy * src1.yzx */
1958 emit_arith(fp, PFS_OP_MAD, temp[0],
1959 WRITEMASK_XYZ, swizzle(keep(src[0]),
1960 Z, X, Y, W),
1961 swizzle(keep(src[1]), Y, Z, X, W),
1962 pfs_zero, 0);
1963 /* dest.xyz = src0.yzx * src1.zxy - temp
1964 * dest.w = undefined
1965 * */
1966 emit_arith(fp, PFS_OP_MAD, dest,
1967 mask & WRITEMASK_XYZ, swizzle(src[0],
1968 Y, Z,
1969 X, W),
1970 swizzle(src[1], Z, X, Y, W),
1971 negate(temp[0]), flags);
1972 /* cleanup */
1973 free_temp(fp, temp[0]);
1974 break;
1975 }
1976 default:
1977 ERROR("unknown fpi->Opcode %d\n", fpi->Opcode);
1978 break;
1979 }
1980
1981 if (fp->error)
1982 return GL_FALSE;
1983
1984 }
1985
1986 return GL_TRUE;
1987 }
1988
1989 static void insert_wpos(struct gl_program *prog)
1990 {
1991 static gl_state_index tokens[STATE_LENGTH] = {
1992 STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0
1993 };
1994 struct prog_instruction *fpi;
1995 GLuint window_index;
1996 int i = 0;
1997 GLuint tempregi = prog->NumTemporaries;
1998 /* should do something else if no temps left... */
1999 prog->NumTemporaries++;
2000
2001 fpi = _mesa_alloc_instructions(prog->NumInstructions + 3);
2002 _mesa_init_instructions(fpi, prog->NumInstructions + 3);
2003
2004 /* perspective divide */
2005 fpi[i].Opcode = OPCODE_RCP;
2006
2007 fpi[i].DstReg.File = PROGRAM_TEMPORARY;
2008 fpi[i].DstReg.Index = tempregi;
2009 fpi[i].DstReg.WriteMask = WRITEMASK_W;
2010 fpi[i].DstReg.CondMask = COND_TR;
2011
2012 fpi[i].SrcReg[0].File = PROGRAM_INPUT;
2013 fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
2014 fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW;
2015 i++;
2016
2017 fpi[i].Opcode = OPCODE_MUL;
2018
2019 fpi[i].DstReg.File = PROGRAM_TEMPORARY;
2020 fpi[i].DstReg.Index = tempregi;
2021 fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
2022 fpi[i].DstReg.CondMask = COND_TR;
2023
2024 fpi[i].SrcReg[0].File = PROGRAM_INPUT;
2025 fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
2026 fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
2027
2028 fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY;
2029 fpi[i].SrcReg[1].Index = tempregi;
2030 fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW;
2031 i++;
2032
2033 /* viewport transformation */
2034 window_index = _mesa_add_state_reference(prog->Parameters, tokens);
2035
2036 fpi[i].Opcode = OPCODE_MAD;
2037
2038 fpi[i].DstReg.File = PROGRAM_TEMPORARY;
2039 fpi[i].DstReg.Index = tempregi;
2040 fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
2041 fpi[i].DstReg.CondMask = COND_TR;
2042
2043 fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY;
2044 fpi[i].SrcReg[0].Index = tempregi;
2045 fpi[i].SrcReg[0].Swizzle =
2046 MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
2047
2048 fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR;
2049 fpi[i].SrcReg[1].Index = window_index;
2050 fpi[i].SrcReg[1].Swizzle =
2051 MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
2052
2053 fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR;
2054 fpi[i].SrcReg[2].Index = window_index;
2055 fpi[i].SrcReg[2].Swizzle =
2056 MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
2057 i++;
2058
2059 _mesa_copy_instructions(&fpi[i], prog->Instructions,
2060 prog->NumInstructions);
2061
2062 free(prog->Instructions);
2063
2064 prog->Instructions = fpi;
2065
2066 prog->NumInstructions += i;
2067 fpi = &prog->Instructions[prog->NumInstructions - 1];
2068
2069 assert(fpi->Opcode == OPCODE_END);
2070
2071 for (fpi = &prog->Instructions[3]; fpi->Opcode != OPCODE_END; fpi++) {
2072 for (i = 0; i < 3; i++)
2073 if (fpi->SrcReg[i].File == PROGRAM_INPUT &&
2074 fpi->SrcReg[i].Index == FRAG_ATTRIB_WPOS) {
2075 fpi->SrcReg[i].File = PROGRAM_TEMPORARY;
2076 fpi->SrcReg[i].Index = tempregi;
2077 }
2078 }
2079 }
2080
2081 /* - Init structures
2082 * - Determine what hwregs each input corresponds to
2083 */
2084 static void init_program(r300ContextPtr r300, struct r300_fragment_program *fp)
2085 {
2086 struct r300_pfs_compile_state *cs = NULL;
2087 struct gl_fragment_program *mp = &fp->mesa_program;
2088 struct prog_instruction *fpi;
2089 GLuint InputsRead = mp->Base.InputsRead;
2090 GLuint temps_used = 0; /* for fp->temps[] */
2091 int i, j;
2092
2093 /* New compile, reset tracking data */
2094 fp->optimization =
2095 driQueryOptioni(&r300->radeon.optionCache, "fp_optimization");
2096 fp->translated = GL_FALSE;
2097 fp->error = GL_FALSE;
2098 fp->cs = cs = &(R300_CONTEXT(fp->ctx)->state.pfs_compile);
2099 fp->tex.length = 0;
2100 fp->cur_node = 0;
2101 fp->first_node_has_tex = 0;
2102 fp->const_nr = 0;
2103 fp->max_temp_idx = 0;
2104 fp->node[0].alu_end = -1;
2105 fp->node[0].tex_end = -1;
2106
2107 _mesa_memset(cs, 0, sizeof(*fp->cs));
2108 for (i = 0; i < PFS_MAX_ALU_INST; i++) {
2109 for (j = 0; j < 3; j++) {
2110 cs->slot[i].vsrc[j] = SRC_CONST;
2111 cs->slot[i].ssrc[j] = SRC_CONST;
2112 }
2113 }
2114
2115 /* Work out what temps the Mesa inputs correspond to, this must match
2116 * what setup_rs_unit does, which shouldn't be a problem as rs_unit
2117 * configures itself based on the fragprog's InputsRead
2118 *
2119 * NOTE: this depends on get_hw_temp() allocating registers in order,
2120 * starting from register 0.
2121 */
2122
2123 /* Texcoords come first */
2124 for (i = 0; i < fp->ctx->Const.MaxTextureUnits; i++) {
2125 if (InputsRead & (FRAG_BIT_TEX0 << i)) {
2126 cs->inputs[FRAG_ATTRIB_TEX0 + i].refcount = 0;
2127 cs->inputs[FRAG_ATTRIB_TEX0 + i].reg =
2128 get_hw_temp(fp, 0);
2129 }
2130 }
2131 InputsRead &= ~FRAG_BITS_TEX_ANY;
2132
2133 /* fragment position treated as a texcoord */
2134 if (InputsRead & FRAG_BIT_WPOS) {
2135 cs->inputs[FRAG_ATTRIB_WPOS].refcount = 0;
2136 cs->inputs[FRAG_ATTRIB_WPOS].reg = get_hw_temp(fp, 0);
2137 insert_wpos(&mp->Base);
2138 }
2139 InputsRead &= ~FRAG_BIT_WPOS;
2140
2141 /* Then primary colour */
2142 if (InputsRead & FRAG_BIT_COL0) {
2143 cs->inputs[FRAG_ATTRIB_COL0].refcount = 0;
2144 cs->inputs[FRAG_ATTRIB_COL0].reg = get_hw_temp(fp, 0);
2145 }
2146 InputsRead &= ~FRAG_BIT_COL0;
2147
2148 /* Secondary color */
2149 if (InputsRead & FRAG_BIT_COL1) {
2150 cs->inputs[FRAG_ATTRIB_COL1].refcount = 0;
2151 cs->inputs[FRAG_ATTRIB_COL1].reg = get_hw_temp(fp, 0);
2152 }
2153 InputsRead &= ~FRAG_BIT_COL1;
2154
2155 /* Anything else */
2156 if (InputsRead) {
2157 WARN_ONCE("Don't know how to handle inputs 0x%x\n", InputsRead);
2158 /* force read from hwreg 0 for now */
2159 for (i = 0; i < 32; i++)
2160 if (InputsRead & (1 << i))
2161 cs->inputs[i].reg = 0;
2162 }
2163
2164 /* Pre-parse the mesa program, grabbing refcounts on input/temp regs.
2165 * That way, we can free up the reg when it's no longer needed
2166 */
2167 if (!mp->Base.Instructions) {
2168 ERROR("No instructions found in program\n");
2169 return;
2170 }
2171
2172 for (fpi = mp->Base.Instructions; fpi->Opcode != OPCODE_END; fpi++) {
2173 int idx;
2174
2175 for (i = 0; i < 3; i++) {
2176 idx = fpi->SrcReg[i].Index;
2177 switch (fpi->SrcReg[i].File) {
2178 case PROGRAM_TEMPORARY:
2179 if (!(temps_used & (1 << idx))) {
2180 cs->temps[idx].reg = -1;
2181 cs->temps[idx].refcount = 1;
2182 temps_used |= (1 << idx);
2183 } else
2184 cs->temps[idx].refcount++;
2185 break;
2186 case PROGRAM_INPUT:
2187 cs->inputs[idx].refcount++;
2188 break;
2189 default:
2190 break;
2191 }
2192 }
2193
2194 idx = fpi->DstReg.Index;
2195 if (fpi->DstReg.File == PROGRAM_TEMPORARY) {
2196 if (!(temps_used & (1 << idx))) {
2197 cs->temps[idx].reg = -1;
2198 cs->temps[idx].refcount = 1;
2199 temps_used |= (1 << idx);
2200 } else
2201 cs->temps[idx].refcount++;
2202 }
2203 }
2204 cs->temp_in_use = temps_used;
2205 }
2206
2207 static void update_params(struct r300_fragment_program *fp)
2208 {
2209 struct gl_fragment_program *mp = &fp->mesa_program;
2210
2211 /* Ask Mesa nicely to fill in ParameterValues for us */
2212 if (mp->Base.Parameters)
2213 _mesa_load_state_parameters(fp->ctx, mp->Base.Parameters);
2214 }
2215
2216 void r300TranslateFragmentShader(r300ContextPtr r300,
2217 struct r300_fragment_program *fp)
2218 {
2219 struct r300_pfs_compile_state *cs = NULL;
2220
2221 if (!fp->translated) {
2222
2223 init_program(r300, fp);
2224 cs = fp->cs;
2225
2226 if (parse_program(fp) == GL_FALSE) {
2227 dump_program(fp);
2228 return;
2229 }
2230
2231 /* Finish off */
2232 fp->node[fp->cur_node].alu_end =
2233 cs->nrslots - fp->node[fp->cur_node].alu_offset - 1;
2234 if (fp->node[fp->cur_node].tex_end < 0)
2235 fp->node[fp->cur_node].tex_end = 0;
2236 fp->alu_offset = 0;
2237 fp->alu_end = cs->nrslots - 1;
2238 fp->tex_offset = 0;
2239 fp->tex_end = fp->tex.length ? fp->tex.length - 1 : 0;
2240 assert(fp->node[fp->cur_node].alu_end >= 0);
2241 assert(fp->alu_end >= 0);
2242
2243 fp->translated = GL_TRUE;
2244 if (RADEON_DEBUG & DEBUG_PIXEL)
2245 dump_program(fp);
2246 r300UpdateStateParameters(fp->ctx, _NEW_PROGRAM);
2247 }
2248
2249 update_params(fp);
2250 }
2251
2252 /* just some random things... */
2253 static void dump_program(struct r300_fragment_program *fp)
2254 {
2255 int n, i, j;
2256 static int pc = 0;
2257
2258 fprintf(stderr, "pc=%d*************************************\n", pc++);
2259
2260 fprintf(stderr, "Mesa program:\n");
2261 fprintf(stderr, "-------------\n");
2262 _mesa_print_program(&fp->mesa_program.Base);
2263 fflush(stdout);
2264
2265 fprintf(stderr, "Hardware program\n");
2266 fprintf(stderr, "----------------\n");
2267
2268 for (n = 0; n < (fp->cur_node + 1); n++) {
2269 fprintf(stderr, "NODE %d: alu_offset: %d, tex_offset: %d, "
2270 "alu_end: %d, tex_end: %d\n", n,
2271 fp->node[n].alu_offset,
2272 fp->node[n].tex_offset,
2273 fp->node[n].alu_end, fp->node[n].tex_end);
2274
2275 if (fp->tex.length) {
2276 fprintf(stderr, " TEX:\n");
2277 for (i = fp->node[n].tex_offset;
2278 i <= fp->node[n].tex_offset + fp->node[n].tex_end;
2279 ++i) {
2280 const char *instr;
2281
2282 switch ((fp->tex.
2283 inst[i] >> R300_FPITX_OPCODE_SHIFT) &
2284 15) {
2285 case R300_FPITX_OP_TEX:
2286 instr = "TEX";
2287 break;
2288 case R300_FPITX_OP_KIL:
2289 instr = "KIL";
2290 break;
2291 case R300_FPITX_OP_TXP:
2292 instr = "TXP";
2293 break;
2294 case R300_FPITX_OP_TXB:
2295 instr = "TXB";
2296 break;
2297 default:
2298 instr = "UNKNOWN";
2299 }
2300
2301 fprintf(stderr,
2302 " %s t%i, %c%i, texture[%i] (%08x)\n",
2303 instr,
2304 (fp->tex.
2305 inst[i] >> R300_FPITX_DST_SHIFT) & 31,
2306 (fp->tex.
2307 inst[i] & R300_FPITX_SRC_CONST) ? 'c' :
2308 't',
2309 (fp->tex.
2310 inst[i] >> R300_FPITX_SRC_SHIFT) & 31,
2311 (fp->tex.
2312 inst[i] & R300_FPITX_IMAGE_MASK) >>
2313 R300_FPITX_IMAGE_SHIFT,
2314 fp->tex.inst[i]);
2315 }
2316 }
2317
2318 for (i = fp->node[n].alu_offset;
2319 i <= fp->node[n].alu_offset + fp->node[n].alu_end; ++i) {
2320 char srcc[3][10], dstc[20];
2321 char srca[3][10], dsta[20];
2322 char argc[3][20];
2323 char arga[3][20];
2324 char flags[5], tmp[10];
2325
2326 for (j = 0; j < 3; ++j) {
2327 int regc = fp->alu.inst[i].inst1 >> (j * 6);
2328 int rega = fp->alu.inst[i].inst3 >> (j * 6);
2329
2330 sprintf(srcc[j], "%c%i",
2331 (regc & 32) ? 'c' : 't', regc & 31);
2332 sprintf(srca[j], "%c%i",
2333 (rega & 32) ? 'c' : 't', rega & 31);
2334 }
2335
2336 dstc[0] = 0;
2337 sprintf(flags, "%s%s%s",
2338 (fp->alu.inst[i].
2339 inst1 & R300_FPI1_DSTC_REG_X) ? "x" : "",
2340 (fp->alu.inst[i].
2341 inst1 & R300_FPI1_DSTC_REG_Y) ? "y" : "",
2342 (fp->alu.inst[i].
2343 inst1 & R300_FPI1_DSTC_REG_Z) ? "z" : "");
2344 if (flags[0] != 0) {
2345 sprintf(dstc, "t%i.%s ",
2346 (fp->alu.inst[i].
2347 inst1 >> R300_FPI1_DSTC_SHIFT) & 31,
2348 flags);
2349 }
2350 sprintf(flags, "%s%s%s",
2351 (fp->alu.inst[i].
2352 inst1 & R300_FPI1_DSTC_OUTPUT_X) ? "x" : "",
2353 (fp->alu.inst[i].
2354 inst1 & R300_FPI1_DSTC_OUTPUT_Y) ? "y" : "",
2355 (fp->alu.inst[i].
2356 inst1 & R300_FPI1_DSTC_OUTPUT_Z) ? "z" : "");
2357 if (flags[0] != 0) {
2358 sprintf(tmp, "o%i.%s",
2359 (fp->alu.inst[i].
2360 inst1 >> R300_FPI1_DSTC_SHIFT) & 31,
2361 flags);
2362 strcat(dstc, tmp);
2363 }
2364
2365 dsta[0] = 0;
2366 if (fp->alu.inst[i].inst3 & R300_FPI3_DSTA_REG) {
2367 sprintf(dsta, "t%i.w ",
2368 (fp->alu.inst[i].
2369 inst3 >> R300_FPI3_DSTA_SHIFT) & 31);
2370 }
2371 if (fp->alu.inst[i].inst3 & R300_FPI3_DSTA_OUTPUT) {
2372 sprintf(tmp, "o%i.w ",
2373 (fp->alu.inst[i].
2374 inst3 >> R300_FPI3_DSTA_SHIFT) & 31);
2375 strcat(dsta, tmp);
2376 }
2377 if (fp->alu.inst[i].inst3 & R300_FPI3_DSTA_DEPTH) {
2378 strcat(dsta, "Z");
2379 }
2380
2381 fprintf(stderr,
2382 "%3i: xyz: %3s %3s %3s -> %-20s (%08x)\n"
2383 " w: %3s %3s %3s -> %-20s (%08x)\n", i,
2384 srcc[0], srcc[1], srcc[2], dstc,
2385 fp->alu.inst[i].inst1, srca[0], srca[1],
2386 srca[2], dsta, fp->alu.inst[i].inst3);
2387
2388 for (j = 0; j < 3; ++j) {
2389 int regc = fp->alu.inst[i].inst0 >> (j * 7);
2390 int rega = fp->alu.inst[i].inst2 >> (j * 7);
2391 int d;
2392 char buf[20];
2393
2394 d = regc & 31;
2395 if (d < 12) {
2396 switch (d % 4) {
2397 case R300_FPI0_ARGC_SRC0C_XYZ:
2398 sprintf(buf, "%s.xyz",
2399 srcc[d / 4]);
2400 break;
2401 case R300_FPI0_ARGC_SRC0C_XXX:
2402 sprintf(buf, "%s.xxx",
2403 srcc[d / 4]);
2404 break;
2405 case R300_FPI0_ARGC_SRC0C_YYY:
2406 sprintf(buf, "%s.yyy",
2407 srcc[d / 4]);
2408 break;
2409 case R300_FPI0_ARGC_SRC0C_ZZZ:
2410 sprintf(buf, "%s.zzz",
2411 srcc[d / 4]);
2412 break;
2413 }
2414 } else if (d < 15) {
2415 sprintf(buf, "%s.www", srca[d - 12]);
2416 } else if (d == 20) {
2417 sprintf(buf, "0.0");
2418 } else if (d == 21) {
2419 sprintf(buf, "1.0");
2420 } else if (d == 22) {
2421 sprintf(buf, "0.5");
2422 } else if (d >= 23 && d < 32) {
2423 d -= 23;
2424 switch (d / 3) {
2425 case 0:
2426 sprintf(buf, "%s.yzx",
2427 srcc[d % 3]);
2428 break;
2429 case 1:
2430 sprintf(buf, "%s.zxy",
2431 srcc[d % 3]);
2432 break;
2433 case 2:
2434 sprintf(buf, "%s.Wzy",
2435 srcc[d % 3]);
2436 break;
2437 }
2438 } else {
2439 sprintf(buf, "%i", d);
2440 }
2441
2442 sprintf(argc[j], "%s%s%s%s",
2443 (regc & 32) ? "-" : "",
2444 (regc & 64) ? "|" : "",
2445 buf, (regc & 64) ? "|" : "");
2446
2447 d = rega & 31;
2448 if (d < 9) {
2449 sprintf(buf, "%s.%c", srcc[d / 3],
2450 'x' + (char)(d % 3));
2451 } else if (d < 12) {
2452 sprintf(buf, "%s.w", srca[d - 9]);
2453 } else if (d == 16) {
2454 sprintf(buf, "0.0");
2455 } else if (d == 17) {
2456 sprintf(buf, "1.0");
2457 } else if (d == 18) {
2458 sprintf(buf, "0.5");
2459 } else {
2460 sprintf(buf, "%i", d);
2461 }
2462
2463 sprintf(arga[j], "%s%s%s%s",
2464 (rega & 32) ? "-" : "",
2465 (rega & 64) ? "|" : "",
2466 buf, (rega & 64) ? "|" : "");
2467 }
2468
2469 fprintf(stderr, " xyz: %8s %8s %8s op: %08x\n"
2470 " w: %8s %8s %8s op: %08x\n",
2471 argc[0], argc[1], argc[2],
2472 fp->alu.inst[i].inst0, arga[0], arga[1],
2473 arga[2], fp->alu.inst[i].inst2);
2474 }
2475 }
2476 }