Merge branch 'draw-instanced'
[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 #ifdef HAVE_LLVM
39 #include "gallivm/lp_bld_init.h"
40 #endif
41
42 static void r300_update_num_contexts(struct r300_screen *r300screen,
43 int diff)
44 {
45 if (diff > 0) {
46 p_atomic_inc(&r300screen->num_contexts);
47
48 if (r300screen->num_contexts > 1)
49 util_slab_set_thread_safety(&r300screen->pool_buffers,
50 UTIL_SLAB_MULTITHREADED);
51 } else {
52 p_atomic_dec(&r300screen->num_contexts);
53
54 if (r300screen->num_contexts <= 1)
55 util_slab_set_thread_safety(&r300screen->pool_buffers,
56 UTIL_SLAB_SINGLETHREADED);
57 }
58 }
59
60 static void r300_release_referenced_objects(struct r300_context *r300)
61 {
62 struct pipe_framebuffer_state *fb =
63 (struct pipe_framebuffer_state*)r300->fb_state.state;
64 struct r300_textures_state *textures =
65 (struct r300_textures_state*)r300->textures_state.state;
66 struct r300_query *query, *temp;
67 unsigned i;
68
69 /* Framebuffer state. */
70 util_unreference_framebuffer_state(fb);
71
72 /* Textures. */
73 for (i = 0; i < textures->sampler_view_count; i++)
74 pipe_sampler_view_reference(
75 (struct pipe_sampler_view**)&textures->sampler_views[i], NULL);
76
77 /* The special dummy texture for texkill. */
78 if (r300->texkill_sampler) {
79 pipe_sampler_view_reference(
80 (struct pipe_sampler_view**)&r300->texkill_sampler,
81 NULL);
82 }
83
84 /* The dummy VBO. */
85 pipe_resource_reference(&r300->dummy_vb, NULL);
86
87 /* The SWTCL VBO. */
88 pipe_resource_reference(&r300->vbo, NULL);
89
90 /* Vertex buffers. */
91 for (i = 0; i < r300->vertex_buffer_count; i++) {
92 pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
93 pipe_resource_reference(&r300->valid_vertex_buffer[i], NULL);
94 }
95
96 /* If there are any queries pending or not destroyed, remove them now. */
97 foreach_s(query, temp, &r300->query_list) {
98 remove_from_list(query);
99 FREE(query);
100 }
101 }
102
103 static void r300_destroy_context(struct pipe_context* context)
104 {
105 struct r300_context* r300 = r300_context(context);
106
107 if (r300->blitter)
108 util_blitter_destroy(r300->blitter);
109 if (r300->draw) {
110 draw_destroy(r300->draw);
111
112 #ifdef HAVE_LLVM
113 gallivm_destroy(r300->gallivm);
114 #endif
115 }
116
117 if (r300->upload_vb)
118 u_upload_destroy(r300->upload_vb);
119 if (r300->upload_ib)
120 u_upload_destroy(r300->upload_ib);
121
122 if (r300->tran.translate_cache)
123 translate_cache_destroy(r300->tran.translate_cache);
124
125 /* XXX: This function assumes r300->query_list was initialized */
126 r300_release_referenced_objects(r300);
127
128 if (r300->zmask_mm)
129 r300_hyperz_destroy_mm(r300);
130
131 if (r300->cs)
132 r300->rws->cs_destroy(r300->cs);
133
134 /* XXX: No way to tell if this was initialized or not? */
135 util_slab_destroy(&r300->pool_transfers);
136
137 r300_update_num_contexts(r300->screen, -1);
138
139 /* Free the structs allocated in r300_setup_atoms() */
140 if (r300->aa_state.state) {
141 FREE(r300->aa_state.state);
142 FREE(r300->blend_color_state.state);
143 FREE(r300->clip_state.state);
144 FREE(r300->fb_state.state);
145 FREE(r300->gpu_flush.state);
146 FREE(r300->hyperz_state.state);
147 FREE(r300->invariant_state.state);
148 FREE(r300->rs_block_state.state);
149 FREE(r300->scissor_state.state);
150 FREE(r300->textures_state.state);
151 FREE(r300->vap_invariant_state.state);
152 FREE(r300->viewport_state.state);
153 FREE(r300->ztop_state.state);
154 FREE(r300->fs_constants.state);
155 FREE(r300->vs_constants.state);
156 if (!r300->screen->caps.has_tcl) {
157 FREE(r300->vertex_stream_state.state);
158 }
159 }
160 FREE(r300);
161 }
162
163 void r300_flush_cb(void *data)
164 {
165 struct r300_context* const cs_context_copy = data;
166
167 cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
168 }
169
170 #define R300_INIT_ATOM(atomname, atomsize) \
171 do { \
172 r300->atomname.name = #atomname; \
173 r300->atomname.state = NULL; \
174 r300->atomname.size = atomsize; \
175 r300->atomname.emit = r300_emit_##atomname; \
176 r300->atomname.dirty = FALSE; \
177 } while (0)
178
179 #define R300_ALLOC_ATOM(atomname, statetype) \
180 do { \
181 r300->atomname.state = CALLOC_STRUCT(statetype); \
182 if (r300->atomname.state == NULL) \
183 return FALSE; \
184 } while (0)
185
186 static boolean r300_setup_atoms(struct r300_context* r300)
187 {
188 boolean is_rv350 = r300->screen->caps.is_rv350;
189 boolean is_r500 = r300->screen->caps.is_r500;
190 boolean has_tcl = r300->screen->caps.has_tcl;
191 boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
192 boolean drm_2_6_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0);
193 boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
194 boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0;
195
196 /* Create the actual atom list.
197 *
198 * Some atoms never change size, others change every emit - those have
199 * the size of 0 here.
200 *
201 * NOTE: The framebuffer state is split into these atoms:
202 * - gpu_flush (unpipelined regs)
203 * - aa_state (unpipelined regs)
204 * - fb_state (unpipelined regs)
205 * - hyperz_state (unpipelined regs followed by pipelined ones)
206 * - fb_state_pipelined (pipelined regs)
207 * The motivation behind this is to be able to emit a strict
208 * subset of the regs, and to have reasonable register ordering. */
209 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
210 R300_INIT_ATOM(gpu_flush, 9);
211 R300_INIT_ATOM(aa_state, 4);
212 R300_INIT_ATOM(fb_state, 0);
213 R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
214 /* ZB (unpipelined), SC. */
215 R300_INIT_ATOM(ztop_state, 2);
216 /* ZB, FG. */
217 R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
218 /* RB3D. */
219 R300_INIT_ATOM(blend_state, 8);
220 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
221 /* SC. */
222 R300_INIT_ATOM(scissor_state, 3);
223 /* GB, FG, GA, SU, SC, RB3D. */
224 R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0));
225 /* VAP. */
226 R300_INIT_ATOM(viewport_state, 9);
227 R300_INIT_ATOM(pvs_flush, 2);
228 R300_INIT_ATOM(vap_invariant_state, 9);
229 R300_INIT_ATOM(vertex_stream_state, 0);
230 R300_INIT_ATOM(vs_state, 0);
231 R300_INIT_ATOM(vs_constants, 0);
232 R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
233 /* VAP, RS, GA, GB, SU, SC. */
234 R300_INIT_ATOM(rs_block_state, 0);
235 R300_INIT_ATOM(rs_state, 0);
236 /* SC, US. */
237 R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0));
238 /* US. */
239 R300_INIT_ATOM(fs, 0);
240 R300_INIT_ATOM(fs_rc_constant_state, 0);
241 R300_INIT_ATOM(fs_constants, 0);
242 /* TX. */
243 R300_INIT_ATOM(texture_cache_inval, 2);
244 R300_INIT_ATOM(textures_state, 0);
245 if (can_hyperz) {
246 /* HiZ Clear */
247 if (has_hiz_ram)
248 R300_INIT_ATOM(hiz_clear, 0);
249 /* zmask clear */
250 R300_INIT_ATOM(zmask_clear, 0);
251 }
252 /* ZB (unpipelined), SU. */
253 R300_INIT_ATOM(query_start, 4);
254
255 /* Replace emission functions for r500. */
256 if (is_r500) {
257 r300->fs.emit = r500_emit_fs;
258 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
259 r300->fs_constants.emit = r500_emit_fs_constants;
260 }
261
262 /* Some non-CSO atoms need explicit space to store the state locally. */
263 R300_ALLOC_ATOM(aa_state, r300_aa_state);
264 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
265 R300_ALLOC_ATOM(clip_state, r300_clip_state);
266 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
267 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
268 R300_ALLOC_ATOM(textures_state, r300_textures_state);
269 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
270 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
271 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
272 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
273 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
274 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
275 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
276 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
277 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
278 if (!r300->screen->caps.has_tcl) {
279 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
280 }
281
282 /* Some non-CSO atoms don't use the state pointer. */
283 r300->fb_state_pipelined.allow_null_state = TRUE;
284 r300->fs_rc_constant_state.allow_null_state = TRUE;
285 r300->pvs_flush.allow_null_state = TRUE;
286 r300->query_start.allow_null_state = TRUE;
287 r300->texture_cache_inval.allow_null_state = TRUE;
288
289 /* Some states must be marked as dirty here to properly set up
290 * hardware in the first command stream. */
291 r300_mark_atom_dirty(r300, &r300->invariant_state);
292 r300_mark_atom_dirty(r300, &r300->pvs_flush);
293 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
294 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
295 r300_mark_atom_dirty(r300, &r300->textures_state);
296
297 return TRUE;
298 }
299
300 /* Not every state tracker calls every driver function before the first draw
301 * call and we must initialize the command buffers somehow. */
302 static void r300_init_states(struct pipe_context *pipe)
303 {
304 struct r300_context *r300 = r300_context(pipe);
305 struct pipe_blend_color bc = {{0}};
306 struct pipe_clip_state cs = {{{0}}};
307 struct pipe_scissor_state ss = {0};
308 struct r300_clip_state *clip =
309 (struct r300_clip_state*)r300->clip_state.state;
310 struct r300_gpu_flush *gpuflush =
311 (struct r300_gpu_flush*)r300->gpu_flush.state;
312 struct r300_vap_invariant_state *vap_invariant =
313 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
314 struct r300_invariant_state *invariant =
315 (struct r300_invariant_state*)r300->invariant_state.state;
316
317 CB_LOCALS;
318
319 pipe->set_blend_color(pipe, &bc);
320 pipe->set_scissor_state(pipe, &ss);
321
322 /* Initialize the clip state. */
323 if (r300->screen->caps.has_tcl) {
324 pipe->set_clip_state(pipe, &cs);
325 } else {
326 BEGIN_CB(clip->cb, 2);
327 OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
328 END_CB;
329 }
330
331 /* Initialize the GPU flush. */
332 {
333 BEGIN_CB(gpuflush->cb_flush_clean, 6);
334
335 /* Flush and free renderbuffer caches. */
336 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
337 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
338 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
339 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
340 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
341 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
342
343 /* Wait until the GPU is idle.
344 * This fixes random pixels sometimes appearing probably caused
345 * by incomplete rendering. */
346 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
347 END_CB;
348 }
349
350 /* Initialize the VAP invariant state. */
351 {
352 BEGIN_CB(vap_invariant->cb, 9);
353 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
354 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
355 OUT_CB_32F(1.0);
356 OUT_CB_32F(1.0);
357 OUT_CB_32F(1.0);
358 OUT_CB_32F(1.0);
359 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
360 END_CB;
361 }
362
363 /* Initialize the invariant state. */
364 {
365 BEGIN_CB(invariant->cb, r300->invariant_state.size);
366 OUT_CB_REG(R300_GB_SELECT, 0);
367 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
368 OUT_CB_REG(R300_GA_ROUND_MODE, 1);
369 OUT_CB_REG(R300_GA_OFFSET, 0);
370 OUT_CB_REG(R300_SU_TEX_WRAP, 0);
371 OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
372 OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
373 OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
374
375 if (r300->screen->caps.is_rv350) {
376 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
377 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
378 }
379 END_CB;
380 }
381
382 /* Initialize the hyperz state. */
383 {
384 struct r300_hyperz_state *hyperz =
385 (struct r300_hyperz_state*)r300->hyperz_state.state;
386 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
387 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
388 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
389 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
390 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
391 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
392
393 if (r300->screen->caps.is_r500 ||
394 (r300->screen->caps.is_rv350 &&
395 r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) {
396 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
397 }
398 END_CB;
399 }
400 }
401
402 struct pipe_context* r300_create_context(struct pipe_screen* screen,
403 void *priv)
404 {
405 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
406 struct r300_screen* r300screen = r300_screen(screen);
407 struct r300_winsys_screen *rws = r300screen->rws;
408
409 if (!r300)
410 return NULL;
411
412 r300_update_num_contexts(r300screen, 1);
413
414 r300->rws = rws;
415 r300->screen = r300screen;
416
417 r300->context.winsys = (struct pipe_winsys*)rws;
418 r300->context.screen = screen;
419 r300->context.priv = priv;
420
421 r300->context.destroy = r300_destroy_context;
422
423 make_empty_list(&r300->query_list);
424
425 util_slab_create(&r300->pool_transfers,
426 sizeof(struct pipe_transfer), 64,
427 UTIL_SLAB_SINGLETHREADED);
428
429 r300->cs = rws->cs_create(rws);
430 if (r300->cs == NULL)
431 goto fail;
432
433 if (!r300screen->caps.has_tcl) {
434 /* Create a Draw. This is used for SW TCL. */
435 #ifdef HAVE_LLVM
436 r300->gallivm = gallivm_create();
437 r300->draw = draw_create_gallivm(&r300->context, r300->gallivm);
438 #else
439 r300->draw = draw_create(&r300->context);
440 #endif
441 if (r300->draw == NULL)
442 goto fail;
443 /* Enable our renderer. */
444 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
445 /* Disable converting points/lines to triangles. */
446 draw_wide_line_threshold(r300->draw, 10000000.f);
447 draw_wide_point_threshold(r300->draw, 10000000.f);
448 }
449
450 if (!r300_setup_atoms(r300))
451 goto fail;
452
453 r300_init_blit_functions(r300);
454 r300_init_flush_functions(r300);
455 r300_init_query_functions(r300);
456 r300_init_state_functions(r300);
457 r300_init_resource_functions(r300);
458
459 r300->blitter = util_blitter_create(&r300->context);
460 if (r300->blitter == NULL)
461 goto fail;
462
463 /* Render functions must be initialized after blitter. */
464 r300_init_render_functions(r300);
465
466 rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
467
468 /* setup hyper-z mm */
469 if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
470 if (!r300_hyperz_init_mm(r300))
471 goto fail;
472
473 r300->upload_ib = u_upload_create(&r300->context,
474 64 * 1024, 16,
475 PIPE_BIND_INDEX_BUFFER);
476
477 if (r300->upload_ib == NULL)
478 goto fail;
479
480 r300->upload_vb = u_upload_create(&r300->context,
481 1024 * 1024, 16,
482 PIPE_BIND_VERTEX_BUFFER);
483 if (r300->upload_vb == NULL)
484 goto fail;
485
486 r300->tran.translate_cache = translate_cache_create();
487 if (r300->tran.translate_cache == NULL)
488 goto fail;
489
490 r300_init_states(&r300->context);
491
492 /* The KIL opcode needs the first texture unit to be enabled
493 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
494 * dummy texture there. */
495 if (!r300->screen->caps.is_r500) {
496 struct pipe_resource *tex;
497 struct pipe_resource rtempl = {{0}};
498 struct pipe_sampler_view vtempl = {{0}};
499
500 rtempl.target = PIPE_TEXTURE_2D;
501 rtempl.format = PIPE_FORMAT_I8_UNORM;
502 rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
503 rtempl.usage = PIPE_USAGE_IMMUTABLE;
504 rtempl.width0 = 1;
505 rtempl.height0 = 1;
506 rtempl.depth0 = 1;
507 tex = screen->resource_create(screen, &rtempl);
508
509 u_sampler_view_default_template(&vtempl, tex, tex->format);
510
511 r300->texkill_sampler = (struct r300_sampler_view*)
512 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
513
514 pipe_resource_reference(&tex, NULL);
515 }
516
517 {
518 struct pipe_resource vb = {};
519 vb.target = PIPE_BUFFER;
520 vb.format = PIPE_FORMAT_R8_UNORM;
521 vb.bind = PIPE_BIND_VERTEX_BUFFER;
522 vb.usage = PIPE_USAGE_IMMUTABLE;
523 vb.width0 = sizeof(float) * 16;
524 vb.height0 = 1;
525 vb.depth0 = 1;
526
527 r300->dummy_vb = screen->resource_create(screen, &vb);
528 }
529
530 return &r300->context;
531
532 fail:
533 r300_destroy_context(&r300->context);
534 return NULL;
535 }
536
537 void r300_finish(struct r300_context *r300)
538 {
539 struct pipe_framebuffer_state *fb;
540 unsigned i;
541
542 /* This is a preliminary implementation of glFinish.
543 *
544 * The ideal implementation should use something like EmitIrqLocked and
545 * WaitIrq, or better, real fences.
546 */
547 if (r300->fb_state.state) {
548 fb = r300->fb_state.state;
549
550 for (i = 0; i < fb->nr_cbufs; i++) {
551 if (fb->cbufs[i]->texture) {
552 r300->rws->buffer_wait(r300->rws,
553 r300_texture(fb->cbufs[i]->texture)->buffer);
554 return;
555 }
556 }
557 if (fb->zsbuf && fb->zsbuf->texture) {
558 r300->rws->buffer_wait(r300->rws,
559 r300_texture(fb->zsbuf->texture)->buffer);
560 }
561 }
562 }