i965: Don't make consumers of brw_DO()/brw_WHILE() track loop start.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_clip_unfilled.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
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 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 */
31
32 #include "main/glheader.h"
33 #include "main/macros.h"
34 #include "main/enums.h"
35 #include "program/program.h"
36
37 #include "intel_batchbuffer.h"
38
39 #include "brw_defines.h"
40 #include "brw_context.h"
41 #include "brw_eu.h"
42 #include "brw_clip.h"
43
44
45
46 /* This is performed against the original triangles, so no indirection
47 * required:
48 BZZZT!
49 */
50 static void compute_tri_direction( struct brw_clip_compile *c )
51 {
52 struct brw_compile *p = &c->func;
53 struct brw_reg e = c->reg.tmp0;
54 struct brw_reg f = c->reg.tmp1;
55 GLuint hpos_offset = brw_vert_result_to_offset(&c->vue_map,
56 VERT_RESULT_HPOS);
57 struct brw_reg v0 = byte_offset(c->reg.vertex[0], hpos_offset);
58 struct brw_reg v1 = byte_offset(c->reg.vertex[1], hpos_offset);
59 struct brw_reg v2 = byte_offset(c->reg.vertex[2], hpos_offset);
60
61
62 struct brw_reg v0n = get_tmp(c);
63 struct brw_reg v1n = get_tmp(c);
64 struct brw_reg v2n = get_tmp(c);
65
66 /* Convert to NDC.
67 * NOTE: We can't modify the original vertex coordinates,
68 * as it may impact further operations.
69 * So, we have to keep normalized coordinates in temp registers.
70 *
71 * TBD-KC
72 * Try to optimize unnecessary MOV's.
73 */
74 brw_MOV(p, v0n, v0);
75 brw_MOV(p, v1n, v1);
76 brw_MOV(p, v2n, v2);
77
78 brw_clip_project_position(c, v0n);
79 brw_clip_project_position(c, v1n);
80 brw_clip_project_position(c, v2n);
81
82 /* Calculate the vectors of two edges of the triangle:
83 */
84 brw_ADD(p, e, v0n, negate(v2n));
85 brw_ADD(p, f, v1n, negate(v2n));
86
87 /* Take their crossproduct:
88 */
89 brw_set_access_mode(p, BRW_ALIGN_16);
90 brw_MUL(p, vec4(brw_null_reg()), brw_swizzle(e, 1,2,0,3), brw_swizzle(f,2,0,1,3));
91 brw_MAC(p, vec4(e), negate(brw_swizzle(e, 2,0,1,3)), brw_swizzle(f,1,2,0,3));
92 brw_set_access_mode(p, BRW_ALIGN_1);
93
94 brw_MUL(p, c->reg.dir, c->reg.dir, vec4(e));
95 }
96
97
98 static void cull_direction( struct brw_clip_compile *c )
99 {
100 struct brw_compile *p = &c->func;
101 GLuint conditional;
102
103 assert (!(c->key.fill_ccw == CLIP_CULL &&
104 c->key.fill_cw == CLIP_CULL));
105
106 if (c->key.fill_ccw == CLIP_CULL)
107 conditional = BRW_CONDITIONAL_GE;
108 else
109 conditional = BRW_CONDITIONAL_L;
110
111 brw_CMP(p,
112 vec1(brw_null_reg()),
113 conditional,
114 get_element(c->reg.dir, 2),
115 brw_imm_f(0));
116
117 brw_IF(p, BRW_EXECUTE_1);
118 {
119 brw_clip_kill_thread(c);
120 }
121 brw_ENDIF(p);
122 }
123
124
125
126 static void copy_bfc( struct brw_clip_compile *c )
127 {
128 struct brw_compile *p = &c->func;
129 GLuint conditional;
130
131 /* Do we have any colors to copy?
132 */
133 if (!(brw_clip_have_vert_result(c, VERT_RESULT_COL0) &&
134 brw_clip_have_vert_result(c, VERT_RESULT_BFC0)) &&
135 !(brw_clip_have_vert_result(c, VERT_RESULT_COL1) &&
136 brw_clip_have_vert_result(c, VERT_RESULT_BFC1)))
137 return;
138
139 /* In some wierd degnerate cases we can end up testing the
140 * direction twice, once for culling and once for bfc copying. Oh
141 * well, that's what you get for setting wierd GL state.
142 */
143 if (c->key.copy_bfc_ccw)
144 conditional = BRW_CONDITIONAL_GE;
145 else
146 conditional = BRW_CONDITIONAL_L;
147
148 brw_CMP(p,
149 vec1(brw_null_reg()),
150 conditional,
151 get_element(c->reg.dir, 2),
152 brw_imm_f(0));
153
154 brw_IF(p, BRW_EXECUTE_1);
155 {
156 GLuint i;
157
158 for (i = 0; i < 3; i++) {
159 if (brw_clip_have_vert_result(c, VERT_RESULT_COL0) &&
160 brw_clip_have_vert_result(c, VERT_RESULT_BFC0))
161 brw_MOV(p,
162 byte_offset(c->reg.vertex[i],
163 brw_vert_result_to_offset(&c->vue_map,
164 VERT_RESULT_COL0)),
165 byte_offset(c->reg.vertex[i],
166 brw_vert_result_to_offset(&c->vue_map,
167 VERT_RESULT_BFC0)));
168
169 if (brw_clip_have_vert_result(c, VERT_RESULT_COL1) &&
170 brw_clip_have_vert_result(c, VERT_RESULT_BFC1))
171 brw_MOV(p,
172 byte_offset(c->reg.vertex[i],
173 brw_vert_result_to_offset(&c->vue_map,
174 VERT_RESULT_COL1)),
175 byte_offset(c->reg.vertex[i],
176 brw_vert_result_to_offset(&c->vue_map,
177 VERT_RESULT_BFC1)));
178 }
179 }
180 brw_ENDIF(p);
181 }
182
183
184
185
186 /*
187 GLfloat iz = 1.0 / dir.z;
188 GLfloat ac = dir.x * iz;
189 GLfloat bc = dir.y * iz;
190 offset = ctx->Polygon.OffsetUnits * DEPTH_SCALE;
191 offset += MAX2( abs(ac), abs(bc) ) * ctx->Polygon.OffsetFactor;
192 offset *= MRD;
193 */
194 static void compute_offset( struct brw_clip_compile *c )
195 {
196 struct brw_compile *p = &c->func;
197 struct brw_reg off = c->reg.offset;
198 struct brw_reg dir = c->reg.dir;
199
200 brw_math_invert(p, get_element(off, 2), get_element(dir, 2));
201 brw_MUL(p, vec2(off), dir, get_element(off, 2));
202
203 brw_CMP(p,
204 vec1(brw_null_reg()),
205 BRW_CONDITIONAL_GE,
206 brw_abs(get_element(off, 0)),
207 brw_abs(get_element(off, 1)));
208
209 brw_SEL(p, vec1(off), brw_abs(get_element(off, 0)), brw_abs(get_element(off, 1)));
210 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
211
212 brw_MUL(p, vec1(off), off, brw_imm_f(c->key.offset_factor));
213 brw_ADD(p, vec1(off), off, brw_imm_f(c->key.offset_units));
214 }
215
216
217 static void merge_edgeflags( struct brw_clip_compile *c )
218 {
219 struct brw_compile *p = &c->func;
220 struct brw_reg tmp0 = get_element_ud(c->reg.tmp0, 0);
221
222 brw_AND(p, tmp0, get_element_ud(c->reg.R0, 2), brw_imm_ud(PRIM_MASK));
223 brw_CMP(p,
224 vec1(brw_null_reg()),
225 BRW_CONDITIONAL_EQ,
226 tmp0,
227 brw_imm_ud(_3DPRIM_POLYGON));
228
229 /* Get away with using reg.vertex because we know that this is not
230 * a _3DPRIM_TRISTRIP_REVERSE:
231 */
232 brw_IF(p, BRW_EXECUTE_1);
233 {
234 brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ);
235 brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<8));
236 brw_MOV(p, byte_offset(c->reg.vertex[0],
237 brw_vert_result_to_offset(&c->vue_map,
238 VERT_RESULT_EDGE)),
239 brw_imm_f(0));
240 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
241
242 brw_set_conditionalmod(p, BRW_CONDITIONAL_EQ);
243 brw_AND(p, vec1(brw_null_reg()), get_element_ud(c->reg.R0, 2), brw_imm_ud(1<<9));
244 brw_MOV(p, byte_offset(c->reg.vertex[2],
245 brw_vert_result_to_offset(&c->vue_map,
246 VERT_RESULT_EDGE)),
247 brw_imm_f(0));
248 brw_set_predicate_control(p, BRW_PREDICATE_NONE);
249 }
250 brw_ENDIF(p);
251 }
252
253
254
255 static void apply_one_offset( struct brw_clip_compile *c,
256 struct brw_indirect vert )
257 {
258 struct brw_compile *p = &c->func;
259 GLuint ndc_offset = brw_vert_result_to_offset(&c->vue_map,
260 BRW_VERT_RESULT_NDC);
261 struct brw_reg z = deref_1f(vert, ndc_offset +
262 2 * type_sz(BRW_REGISTER_TYPE_F));
263
264 brw_ADD(p, z, z, vec1(c->reg.offset));
265 }
266
267
268
269 /***********************************************************************
270 * Output clipped polygon as an unfilled primitive:
271 */
272 static void emit_lines(struct brw_clip_compile *c,
273 bool do_offset)
274 {
275 struct brw_compile *p = &c->func;
276 struct brw_indirect v0 = brw_indirect(0, 0);
277 struct brw_indirect v1 = brw_indirect(1, 0);
278 struct brw_indirect v0ptr = brw_indirect(2, 0);
279 struct brw_indirect v1ptr = brw_indirect(3, 0);
280
281 /* Need a seperate loop for offset:
282 */
283 if (do_offset) {
284 brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
285 brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
286
287 brw_DO(p, BRW_EXECUTE_1);
288 {
289 brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
290 brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
291
292 apply_one_offset(c, v0);
293
294 brw_set_conditionalmod(p, BRW_CONDITIONAL_G);
295 brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
296 }
297 brw_WHILE(p);
298 }
299
300 /* v1ptr = &inlist[nr_verts]
301 * *v1ptr = v0
302 */
303 brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
304 brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
305 brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v0ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
306 brw_ADD(p, get_addr_reg(v1ptr), get_addr_reg(v1ptr), retype(c->reg.nr_verts, BRW_REGISTER_TYPE_UW));
307 brw_MOV(p, deref_1uw(v1ptr, 0), deref_1uw(v0ptr, 0));
308
309 brw_DO(p, BRW_EXECUTE_1);
310 {
311 brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
312 brw_MOV(p, get_addr_reg(v1), deref_1uw(v0ptr, 2));
313 brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
314
315 /* draw edge if edgeflag != 0 */
316 brw_CMP(p,
317 vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
318 deref_1f(v0, brw_vert_result_to_offset(&c->vue_map,
319 VERT_RESULT_EDGE)),
320 brw_imm_f(0));
321 brw_IF(p, BRW_EXECUTE_1);
322 {
323 brw_clip_emit_vue(c, v0, 1, 0,
324 (_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
325 | URB_WRITE_PRIM_START);
326 brw_clip_emit_vue(c, v1, 1, 0,
327 (_3DPRIM_LINESTRIP << URB_WRITE_PRIM_TYPE_SHIFT)
328 | URB_WRITE_PRIM_END);
329 }
330 brw_ENDIF(p);
331
332 brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
333 brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
334 }
335 brw_WHILE(p);
336 }
337
338
339
340 static void emit_points(struct brw_clip_compile *c,
341 bool do_offset )
342 {
343 struct brw_compile *p = &c->func;
344
345 struct brw_indirect v0 = brw_indirect(0, 0);
346 struct brw_indirect v0ptr = brw_indirect(2, 0);
347
348 brw_MOV(p, c->reg.loopcount, c->reg.nr_verts);
349 brw_MOV(p, get_addr_reg(v0ptr), brw_address(c->reg.inlist));
350
351 brw_DO(p, BRW_EXECUTE_1);
352 {
353 brw_MOV(p, get_addr_reg(v0), deref_1uw(v0ptr, 0));
354 brw_ADD(p, get_addr_reg(v0ptr), get_addr_reg(v0ptr), brw_imm_uw(2));
355
356 /* draw if edgeflag != 0
357 */
358 brw_CMP(p,
359 vec1(brw_null_reg()), BRW_CONDITIONAL_NZ,
360 deref_1f(v0, brw_vert_result_to_offset(&c->vue_map,
361 VERT_RESULT_EDGE)),
362 brw_imm_f(0));
363 brw_IF(p, BRW_EXECUTE_1);
364 {
365 if (do_offset)
366 apply_one_offset(c, v0);
367
368 brw_clip_emit_vue(c, v0, 1, 0,
369 (_3DPRIM_POINTLIST << URB_WRITE_PRIM_TYPE_SHIFT)
370 | URB_WRITE_PRIM_START | URB_WRITE_PRIM_END);
371 }
372 brw_ENDIF(p);
373
374 brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
375 brw_ADD(p, c->reg.loopcount, c->reg.loopcount, brw_imm_d(-1));
376 }
377 brw_WHILE(p);
378 }
379
380
381
382
383
384
385
386 static void emit_primitives( struct brw_clip_compile *c,
387 GLuint mode,
388 bool do_offset )
389 {
390 switch (mode) {
391 case CLIP_FILL:
392 brw_clip_tri_emit_polygon(c);
393 break;
394
395 case CLIP_LINE:
396 emit_lines(c, do_offset);
397 break;
398
399 case CLIP_POINT:
400 emit_points(c, do_offset);
401 break;
402
403 case CLIP_CULL:
404 assert(0);
405 break;
406 }
407 }
408
409
410
411 static void emit_unfilled_primitives( struct brw_clip_compile *c )
412 {
413 struct brw_compile *p = &c->func;
414
415 /* Direction culling has already been done.
416 */
417 if (c->key.fill_ccw != c->key.fill_cw &&
418 c->key.fill_ccw != CLIP_CULL &&
419 c->key.fill_cw != CLIP_CULL)
420 {
421 brw_CMP(p,
422 vec1(brw_null_reg()),
423 BRW_CONDITIONAL_GE,
424 get_element(c->reg.dir, 2),
425 brw_imm_f(0));
426
427 brw_IF(p, BRW_EXECUTE_1);
428 {
429 emit_primitives(c, c->key.fill_ccw, c->key.offset_ccw);
430 }
431 brw_ELSE(p);
432 {
433 emit_primitives(c, c->key.fill_cw, c->key.offset_cw);
434 }
435 brw_ENDIF(p);
436 }
437 else if (c->key.fill_cw != CLIP_CULL) {
438 emit_primitives(c, c->key.fill_cw, c->key.offset_cw);
439 }
440 else if (c->key.fill_ccw != CLIP_CULL) {
441 emit_primitives(c, c->key.fill_ccw, c->key.offset_ccw);
442 }
443 }
444
445
446
447
448 static void check_nr_verts( struct brw_clip_compile *c )
449 {
450 struct brw_compile *p = &c->func;
451
452 brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_L, c->reg.nr_verts, brw_imm_d(3));
453 brw_IF(p, BRW_EXECUTE_1);
454 {
455 brw_clip_kill_thread(c);
456 }
457 brw_ENDIF(p);
458 }
459
460
461 void brw_emit_unfilled_clip( struct brw_clip_compile *c )
462 {
463 struct brw_compile *p = &c->func;
464
465 c->need_direction = ((c->key.offset_ccw || c->key.offset_cw) ||
466 (c->key.fill_ccw != c->key.fill_cw) ||
467 c->key.fill_ccw == CLIP_CULL ||
468 c->key.fill_cw == CLIP_CULL ||
469 c->key.copy_bfc_cw ||
470 c->key.copy_bfc_ccw);
471
472 brw_clip_tri_alloc_regs(c, 3 + c->key.nr_userclip + 6);
473 brw_clip_tri_init_vertices(c);
474 brw_clip_init_ff_sync(c);
475
476 assert(brw_clip_have_vert_result(c, VERT_RESULT_EDGE));
477
478 if (c->key.fill_ccw == CLIP_CULL &&
479 c->key.fill_cw == CLIP_CULL) {
480 brw_clip_kill_thread(c);
481 return;
482 }
483
484 merge_edgeflags(c);
485
486 /* Need to use the inlist indirection here:
487 */
488 if (c->need_direction)
489 compute_tri_direction(c);
490
491 if (c->key.fill_ccw == CLIP_CULL ||
492 c->key.fill_cw == CLIP_CULL)
493 cull_direction(c);
494
495 if (c->key.offset_ccw ||
496 c->key.offset_cw)
497 compute_offset(c);
498
499 if (c->key.copy_bfc_ccw ||
500 c->key.copy_bfc_cw)
501 copy_bfc(c);
502
503 /* Need to do this whether we clip or not:
504 */
505 if (c->key.do_flat_shading)
506 brw_clip_tri_flat_shade(c);
507
508 brw_clip_init_clipmask(c);
509 brw_CMP(p, vec1(brw_null_reg()), BRW_CONDITIONAL_NZ, c->reg.planemask, brw_imm_ud(0));
510 brw_IF(p, BRW_EXECUTE_1);
511 {
512 brw_clip_init_planes(c);
513 brw_clip_tri(c);
514 check_nr_verts(c);
515 }
516 brw_ENDIF(p);
517
518 emit_unfilled_primitives(c);
519 brw_clip_kill_thread(c);
520 }
521
522
523