r300g: fix some bugs with zbuffer compression (v4)
[mesa.git] / src / gallium / drivers / r300 / r300_context.c
1 /*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
22
23 #include "draw/draw_context.h"
24
25 #include "util/u_memory.h"
26 #include "util/u_sampler.h"
27 #include "util/u_simple_list.h"
28 #include "util/u_upload_mgr.h"
29
30 #include "r300_cb.h"
31 #include "r300_context.h"
32 #include "r300_emit.h"
33 #include "r300_hyperz.h"
34 #include "r300_screen.h"
35 #include "r300_screen_buffer.h"
36 #include "r300_winsys.h"
37
38 static void r300_update_num_contexts(struct r300_screen *r300screen,
39 int diff)
40 {
41 if (diff > 0) {
42 p_atomic_inc(&r300screen->num_contexts);
43
44 if (r300screen->num_contexts > 1)
45 util_slab_set_thread_safety(&r300screen->pool_buffers,
46 UTIL_SLAB_MULTITHREADED);
47 } else {
48 p_atomic_dec(&r300screen->num_contexts);
49
50 if (r300screen->num_contexts <= 1)
51 util_slab_set_thread_safety(&r300screen->pool_buffers,
52 UTIL_SLAB_SINGLETHREADED);
53 }
54 }
55
56 static void r300_release_referenced_objects(struct r300_context *r300)
57 {
58 struct pipe_framebuffer_state *fb =
59 (struct pipe_framebuffer_state*)r300->fb_state.state;
60 struct r300_textures_state *textures =
61 (struct r300_textures_state*)r300->textures_state.state;
62 struct r300_query *query, *temp;
63 unsigned i;
64
65 /* Framebuffer state. */
66 util_unreference_framebuffer_state(fb);
67
68 /* Textures. */
69 for (i = 0; i < textures->sampler_view_count; i++)
70 pipe_sampler_view_reference(
71 (struct pipe_sampler_view**)&textures->sampler_views[i], NULL);
72
73 /* The special dummy texture for texkill. */
74 if (r300->texkill_sampler) {
75 pipe_sampler_view_reference(
76 (struct pipe_sampler_view**)&r300->texkill_sampler,
77 NULL);
78 }
79
80 /* The dummy VBO. */
81 pipe_resource_reference(&r300->dummy_vb, NULL);
82
83 /* The SWTCL VBO. */
84 pipe_resource_reference(&r300->vbo, NULL);
85
86 /* Vertex buffers. */
87 for (i = 0; i < r300->vertex_buffer_count; i++) {
88 pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
89 pipe_resource_reference(&r300->valid_vertex_buffer[i], NULL);
90 }
91
92 /* If there are any queries pending or not destroyed, remove them now. */
93 foreach_s(query, temp, &r300->query_list) {
94 remove_from_list(query);
95 FREE(query);
96 }
97
98 r300->context.delete_depth_stencil_alpha_state(&r300->context,
99 r300->dsa_decompress_zmask);
100 }
101
102 static void r300_destroy_context(struct pipe_context* context)
103 {
104 struct r300_context* r300 = r300_context(context);
105
106 if (r300->blitter)
107 util_blitter_destroy(r300->blitter);
108 if (r300->draw)
109 draw_destroy(r300->draw);
110
111 if (r300->upload_vb)
112 u_upload_destroy(r300->upload_vb);
113 if (r300->upload_ib)
114 u_upload_destroy(r300->upload_ib);
115
116 if (r300->tran.translate_cache)
117 translate_cache_destroy(r300->tran.translate_cache);
118
119 /* XXX: This function assumes r300->query_list was initialized */
120 r300_release_referenced_objects(r300);
121
122 if (r300->cs)
123 r300->rws->cs_destroy(r300->cs);
124
125 /* XXX: No way to tell if this was initialized or not? */
126 util_slab_destroy(&r300->pool_transfers);
127
128 r300_update_num_contexts(r300->screen, -1);
129
130 /* Free the structs allocated in r300_setup_atoms() */
131 if (r300->aa_state.state) {
132 FREE(r300->aa_state.state);
133 FREE(r300->blend_color_state.state);
134 FREE(r300->clip_state.state);
135 FREE(r300->fb_state.state);
136 FREE(r300->gpu_flush.state);
137 FREE(r300->hyperz_state.state);
138 FREE(r300->invariant_state.state);
139 FREE(r300->rs_block_state.state);
140 FREE(r300->scissor_state.state);
141 FREE(r300->textures_state.state);
142 FREE(r300->vap_invariant_state.state);
143 FREE(r300->viewport_state.state);
144 FREE(r300->ztop_state.state);
145 FREE(r300->fs_constants.state);
146 FREE(r300->vs_constants.state);
147 if (!r300->screen->caps.has_tcl) {
148 FREE(r300->vertex_stream_state.state);
149 }
150 }
151 FREE(r300);
152 }
153
154 void r300_flush_cb(void *data)
155 {
156 struct r300_context* const cs_context_copy = data;
157
158 cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
159 }
160
161 #define R300_INIT_ATOM(atomname, atomsize) \
162 do { \
163 r300->atomname.name = #atomname; \
164 r300->atomname.state = NULL; \
165 r300->atomname.size = atomsize; \
166 r300->atomname.emit = r300_emit_##atomname; \
167 r300->atomname.dirty = FALSE; \
168 } while (0)
169
170 #define R300_ALLOC_ATOM(atomname, statetype) \
171 do { \
172 r300->atomname.state = CALLOC_STRUCT(statetype); \
173 if (r300->atomname.state == NULL) \
174 return FALSE; \
175 } while (0)
176
177 static boolean r300_setup_atoms(struct r300_context* r300)
178 {
179 boolean is_rv350 = r300->screen->caps.is_rv350;
180 boolean is_r500 = r300->screen->caps.is_r500;
181 boolean has_tcl = r300->screen->caps.has_tcl;
182 boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
183 boolean drm_2_6_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0);
184 boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
185 boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0;
186
187 /* Create the actual atom list.
188 *
189 * Some atoms never change size, others change every emit - those have
190 * the size of 0 here.
191 *
192 * NOTE: The framebuffer state is split into these atoms:
193 * - gpu_flush (unpipelined regs)
194 * - aa_state (unpipelined regs)
195 * - fb_state (unpipelined regs)
196 * - hyperz_state (unpipelined regs followed by pipelined ones)
197 * - fb_state_pipelined (pipelined regs)
198 * The motivation behind this is to be able to emit a strict
199 * subset of the regs, and to have reasonable register ordering. */
200 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
201 R300_INIT_ATOM(gpu_flush, 9);
202 R300_INIT_ATOM(aa_state, 4);
203 R300_INIT_ATOM(fb_state, 0);
204 R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
205 /* ZB (unpipelined), SC. */
206 R300_INIT_ATOM(ztop_state, 2);
207 /* ZB, FG. */
208 R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
209 /* RB3D. */
210 R300_INIT_ATOM(blend_state, 8);
211 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
212 /* SC. */
213 R300_INIT_ATOM(scissor_state, 3);
214 /* GB, FG, GA, SU, SC, RB3D. */
215 R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0));
216 /* VAP. */
217 R300_INIT_ATOM(viewport_state, 9);
218 R300_INIT_ATOM(pvs_flush, 2);
219 R300_INIT_ATOM(vap_invariant_state, 9);
220 R300_INIT_ATOM(vertex_stream_state, 0);
221 R300_INIT_ATOM(vs_state, 0);
222 R300_INIT_ATOM(vs_constants, 0);
223 R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
224 /* VAP, RS, GA, GB, SU, SC. */
225 R300_INIT_ATOM(rs_block_state, 0);
226 R300_INIT_ATOM(rs_state, 0);
227 /* SC, US. */
228 R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0));
229 /* US. */
230 R300_INIT_ATOM(fs, 0);
231 R300_INIT_ATOM(fs_rc_constant_state, 0);
232 R300_INIT_ATOM(fs_constants, 0);
233 /* TX. */
234 R300_INIT_ATOM(texture_cache_inval, 2);
235 R300_INIT_ATOM(textures_state, 0);
236 if (can_hyperz) {
237 /* HiZ Clear */
238 if (has_hiz_ram)
239 R300_INIT_ATOM(hiz_clear, 0);
240 /* zmask clear */
241 R300_INIT_ATOM(zmask_clear, 4);
242 }
243 /* ZB (unpipelined), SU. */
244 R300_INIT_ATOM(query_start, 4);
245
246 /* Replace emission functions for r500. */
247 if (is_r500) {
248 r300->fs.emit = r500_emit_fs;
249 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
250 r300->fs_constants.emit = r500_emit_fs_constants;
251 }
252
253 /* Some non-CSO atoms need explicit space to store the state locally. */
254 R300_ALLOC_ATOM(aa_state, r300_aa_state);
255 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
256 R300_ALLOC_ATOM(clip_state, r300_clip_state);
257 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
258 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
259 R300_ALLOC_ATOM(textures_state, r300_textures_state);
260 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
261 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
262 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
263 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
264 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
265 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
266 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
267 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
268 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
269 if (!r300->screen->caps.has_tcl) {
270 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
271 }
272
273 /* Some non-CSO atoms don't use the state pointer. */
274 r300->fb_state_pipelined.allow_null_state = TRUE;
275 r300->fs_rc_constant_state.allow_null_state = TRUE;
276 r300->pvs_flush.allow_null_state = TRUE;
277 r300->query_start.allow_null_state = TRUE;
278 r300->texture_cache_inval.allow_null_state = TRUE;
279
280 /* Some states must be marked as dirty here to properly set up
281 * hardware in the first command stream. */
282 r300_mark_atom_dirty(r300, &r300->invariant_state);
283 r300_mark_atom_dirty(r300, &r300->pvs_flush);
284 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
285 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
286 r300_mark_atom_dirty(r300, &r300->textures_state);
287
288 return TRUE;
289 }
290
291 /* Not every state tracker calls every driver function before the first draw
292 * call and we must initialize the command buffers somehow. */
293 static void r300_init_states(struct pipe_context *pipe)
294 {
295 struct r300_context *r300 = r300_context(pipe);
296 struct pipe_blend_color bc = {{0}};
297 struct pipe_clip_state cs = {{{0}}};
298 struct pipe_scissor_state ss = {0};
299 struct r300_clip_state *clip =
300 (struct r300_clip_state*)r300->clip_state.state;
301 struct r300_gpu_flush *gpuflush =
302 (struct r300_gpu_flush*)r300->gpu_flush.state;
303 struct r300_vap_invariant_state *vap_invariant =
304 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
305 struct r300_invariant_state *invariant =
306 (struct r300_invariant_state*)r300->invariant_state.state;
307
308 CB_LOCALS;
309
310 pipe->set_blend_color(pipe, &bc);
311 pipe->set_scissor_state(pipe, &ss);
312
313 /* Initialize the clip state. */
314 if (r300->screen->caps.has_tcl) {
315 pipe->set_clip_state(pipe, &cs);
316 } else {
317 BEGIN_CB(clip->cb, 2);
318 OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
319 END_CB;
320 }
321
322 /* Initialize the GPU flush. */
323 {
324 BEGIN_CB(gpuflush->cb_flush_clean, 6);
325
326 /* Flush and free renderbuffer caches. */
327 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
328 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
329 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
330 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
331 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
332 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
333
334 /* Wait until the GPU is idle.
335 * This fixes random pixels sometimes appearing probably caused
336 * by incomplete rendering. */
337 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
338 END_CB;
339 }
340
341 /* Initialize the VAP invariant state. */
342 {
343 BEGIN_CB(vap_invariant->cb, 9);
344 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
345 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
346 OUT_CB_32F(1.0);
347 OUT_CB_32F(1.0);
348 OUT_CB_32F(1.0);
349 OUT_CB_32F(1.0);
350 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
351 END_CB;
352 }
353
354 /* Initialize the invariant state. */
355 {
356 BEGIN_CB(invariant->cb, r300->invariant_state.size);
357 OUT_CB_REG(R300_GB_SELECT, 0);
358 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
359 OUT_CB_REG(R300_GA_ROUND_MODE, 1);
360 OUT_CB_REG(R300_GA_OFFSET, 0);
361 OUT_CB_REG(R300_SU_TEX_WRAP, 0);
362 OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
363 OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
364 OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
365
366 if (r300->screen->caps.is_rv350) {
367 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
368 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
369 }
370 END_CB;
371 }
372
373 /* Initialize the hyperz state. */
374 {
375 struct r300_hyperz_state *hyperz =
376 (struct r300_hyperz_state*)r300->hyperz_state.state;
377 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
378 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
379 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
380 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
381 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
382 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
383
384 if (r300->screen->caps.is_r500 ||
385 (r300->screen->caps.is_rv350 &&
386 r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) {
387 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
388 }
389 END_CB;
390 }
391 }
392
393 struct pipe_context* r300_create_context(struct pipe_screen* screen,
394 void *priv)
395 {
396 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
397 struct r300_screen* r300screen = r300_screen(screen);
398 struct r300_winsys_screen *rws = r300screen->rws;
399
400 if (!r300)
401 return NULL;
402
403 r300_update_num_contexts(r300screen, 1);
404
405 r300->rws = rws;
406 r300->screen = r300screen;
407
408 r300->context.winsys = (struct pipe_winsys*)rws;
409 r300->context.screen = screen;
410 r300->context.priv = priv;
411
412 r300->context.destroy = r300_destroy_context;
413
414 make_empty_list(&r300->query_list);
415
416 util_slab_create(&r300->pool_transfers,
417 sizeof(struct pipe_transfer), 64,
418 UTIL_SLAB_SINGLETHREADED);
419
420 r300->cs = rws->cs_create(rws);
421 if (r300->cs == NULL)
422 goto fail;
423
424 if (!r300screen->caps.has_tcl) {
425 /* Create a Draw. This is used for SW TCL. */
426 r300->draw = draw_create(&r300->context);
427 if (r300->draw == NULL)
428 goto fail;
429 /* Enable our renderer. */
430 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
431 /* Disable converting points/lines to triangles. */
432 draw_wide_line_threshold(r300->draw, 10000000.f);
433 draw_wide_point_threshold(r300->draw, 10000000.f);
434 }
435
436 if (!r300_setup_atoms(r300))
437 goto fail;
438
439 r300_init_blit_functions(r300);
440 r300_init_flush_functions(r300);
441 r300_init_query_functions(r300);
442 r300_init_state_functions(r300);
443 r300_init_resource_functions(r300);
444
445 r300->blitter = util_blitter_create(&r300->context);
446 if (r300->blitter == NULL)
447 goto fail;
448
449 /* Render functions must be initialized after blitter. */
450 r300_init_render_functions(r300);
451
452 rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
453
454 /* setup hyper-z mm */
455 if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
456 if (!r300_hyperz_init_mm(r300))
457 goto fail;
458
459 r300->upload_ib = u_upload_create(&r300->context,
460 64 * 1024, 16,
461 PIPE_BIND_INDEX_BUFFER);
462
463 if (r300->upload_ib == NULL)
464 goto fail;
465
466 r300->upload_vb = u_upload_create(&r300->context,
467 1024 * 1024, 16,
468 PIPE_BIND_VERTEX_BUFFER);
469 if (r300->upload_vb == NULL)
470 goto fail;
471
472 r300->tran.translate_cache = translate_cache_create();
473 if (r300->tran.translate_cache == NULL)
474 goto fail;
475
476 r300_init_states(&r300->context);
477
478 /* The KIL opcode needs the first texture unit to be enabled
479 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
480 * dummy texture there. */
481 if (!r300->screen->caps.is_r500) {
482 struct pipe_resource *tex;
483 struct pipe_resource rtempl = {{0}};
484 struct pipe_sampler_view vtempl = {{0}};
485
486 rtempl.target = PIPE_TEXTURE_2D;
487 rtempl.format = PIPE_FORMAT_I8_UNORM;
488 rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
489 rtempl.usage = PIPE_USAGE_IMMUTABLE;
490 rtempl.width0 = 1;
491 rtempl.height0 = 1;
492 rtempl.depth0 = 1;
493 tex = screen->resource_create(screen, &rtempl);
494
495 u_sampler_view_default_template(&vtempl, tex, tex->format);
496
497 r300->texkill_sampler = (struct r300_sampler_view*)
498 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
499
500 pipe_resource_reference(&tex, NULL);
501 }
502
503 {
504 struct pipe_resource vb = {};
505 vb.target = PIPE_BUFFER;
506 vb.format = PIPE_FORMAT_R8_UNORM;
507 vb.bind = PIPE_BIND_VERTEX_BUFFER;
508 vb.usage = PIPE_USAGE_IMMUTABLE;
509 vb.width0 = sizeof(float) * 16;
510 vb.height0 = 1;
511 vb.depth0 = 1;
512
513 r300->dummy_vb = screen->resource_create(screen, &vb);
514 }
515
516 {
517 struct pipe_depth_stencil_alpha_state dsa = {};
518 dsa.depth.writemask = 1;
519
520 r300->dsa_decompress_zmask =
521 r300->context.create_depth_stencil_alpha_state(&r300->context,
522 &dsa);
523 }
524
525 return &r300->context;
526
527 fail:
528 r300_destroy_context(&r300->context);
529 return NULL;
530 }
531
532 void r300_finish(struct r300_context *r300)
533 {
534 struct pipe_framebuffer_state *fb;
535 unsigned i;
536
537 /* This is a preliminary implementation of glFinish.
538 *
539 * The ideal implementation should use something like EmitIrqLocked and
540 * WaitIrq, or better, real fences.
541 */
542 if (r300->fb_state.state) {
543 fb = r300->fb_state.state;
544
545 for (i = 0; i < fb->nr_cbufs; i++) {
546 if (fb->cbufs[i]->texture) {
547 r300->rws->buffer_wait(r300->rws,
548 r300_texture(fb->cbufs[i]->texture)->buffer);
549 return;
550 }
551 }
552 if (fb->zsbuf && fb->zsbuf->texture) {
553 r300->rws->buffer_wait(r300->rws,
554 r300_texture(fb->zsbuf->texture)->buffer);
555 }
556 }
557 }