Merge remote branch 'origin/master' into pipe-video
[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 #include <inttypes.h>
39
40 static void r300_update_num_contexts(struct r300_screen *r300screen,
41 int diff)
42 {
43 if (diff > 0) {
44 p_atomic_inc(&r300screen->num_contexts);
45
46 if (r300screen->num_contexts > 1)
47 util_mempool_set_thread_safety(&r300screen->pool_buffers,
48 UTIL_MEMPOOL_MULTITHREADED);
49 } else {
50 p_atomic_dec(&r300screen->num_contexts);
51
52 if (r300screen->num_contexts <= 1)
53 util_mempool_set_thread_safety(&r300screen->pool_buffers,
54 UTIL_MEMPOOL_SINGLETHREADED);
55 }
56 }
57
58 static void r300_release_referenced_objects(struct r300_context *r300)
59 {
60 struct pipe_framebuffer_state *fb =
61 (struct pipe_framebuffer_state*)r300->fb_state.state;
62 struct r300_textures_state *textures =
63 (struct r300_textures_state*)r300->textures_state.state;
64 struct r300_query *query, *temp;
65 unsigned i;
66
67 /* Framebuffer state. */
68 util_unreference_framebuffer_state(fb);
69
70 /* Textures. */
71 for (i = 0; i < textures->sampler_view_count; i++)
72 pipe_sampler_view_reference(
73 (struct pipe_sampler_view**)&textures->sampler_views[i], NULL);
74
75 /* The special dummy texture for texkill. */
76 if (r300->texkill_sampler) {
77 pipe_sampler_view_reference(
78 (struct pipe_sampler_view**)&r300->texkill_sampler,
79 NULL);
80 }
81
82 /* The SWTCL VBO. */
83 pipe_resource_reference(&r300->vbo, NULL);
84
85 /* Vertex buffers. */
86 for (i = 0; i < r300->vertex_buffer_count; i++) {
87 pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
88 }
89
90 /* If there are any queries pending or not destroyed, remove them now. */
91 foreach_s(query, temp, &r300->query_list) {
92 remove_from_list(query);
93 FREE(query);
94 }
95 }
96
97 static void r300_destroy_context(struct pipe_context* context)
98 {
99 struct r300_context* r300 = r300_context(context);
100 struct r300_atom *atom;
101
102 if (r300->blitter)
103 util_blitter_destroy(r300->blitter);
104 if (r300->draw)
105 draw_destroy(r300->draw);
106
107 /* Print stats, if enabled. */
108 if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) {
109 fprintf(stderr, "r300: Stats for context %p:\n", r300);
110 fprintf(stderr, " : Flushes: %" PRIu64 "\n", r300->flush_counter);
111 foreach(atom, &r300->atom_list) {
112 fprintf(stderr, " : %s: %" PRIu64 " emits\n",
113 atom->name, atom->counter);
114 }
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_mempool_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 insert_at_tail(&r300->atom_list, &r300->atomname); \
178 } while (0)
179
180 static void r300_setup_atoms(struct r300_context* r300)
181 {
182 boolean is_rv350 = r300->screen->caps.is_rv350;
183 boolean is_r500 = r300->screen->caps.is_r500;
184 boolean has_tcl = r300->screen->caps.has_tcl;
185 boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
186 boolean drm_2_6_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0);
187 boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
188 boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0;
189
190 /* Create the actual atom list.
191 *
192 * Each atom is examined and emitted in the order it appears here, which
193 * can affect performance and conformance if not handled with care.
194 *
195 * Some atoms never change size, others change every emit - those have
196 * the size of 0 here.
197 *
198 * NOTE: The framebuffer state is split into these atoms:
199 * - gpu_flush (unpipelined regs)
200 * - aa_state (unpipelined regs)
201 * - fb_state (unpipelined regs)
202 * - hyperz_state (unpipelined regs followed by pipelined ones)
203 * - fb_state_pipelined (pipelined regs)
204 * The motivation behind this is to be able to emit a strict
205 * subset of the regs, and to have reasonable register ordering. */
206 make_empty_list(&r300->atom_list);
207 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
208 R300_INIT_ATOM(gpu_flush, 9);
209 R300_INIT_ATOM(aa_state, 4);
210 R300_INIT_ATOM(fb_state, 0);
211 R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
212 /* ZB (unpipelined), SC. */
213 R300_INIT_ATOM(ztop_state, 2);
214 /* ZB, FG. */
215 R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
216 /* RB3D. */
217 R300_INIT_ATOM(blend_state, 8);
218 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
219 /* SC. */
220 R300_INIT_ATOM(scissor_state, 3);
221 /* GB, FG, GA, SU, SC, RB3D. */
222 R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0));
223 /* VAP. */
224 R300_INIT_ATOM(viewport_state, 9);
225 R300_INIT_ATOM(pvs_flush, 2);
226 R300_INIT_ATOM(vap_invariant_state, 9);
227 R300_INIT_ATOM(vertex_stream_state, 0);
228 R300_INIT_ATOM(vs_state, 0);
229 R300_INIT_ATOM(vs_constants, 0);
230 R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
231 /* VAP, RS, GA, GB, SU, SC. */
232 R300_INIT_ATOM(rs_block_state, 0);
233 R300_INIT_ATOM(rs_state, 0);
234 /* SC, US. */
235 R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0));
236 /* US. */
237 R300_INIT_ATOM(fs, 0);
238 R300_INIT_ATOM(fs_rc_constant_state, 0);
239 R300_INIT_ATOM(fs_constants, 0);
240 /* TX. */
241 R300_INIT_ATOM(texture_cache_inval, 2);
242 R300_INIT_ATOM(textures_state, 0);
243 if (can_hyperz) {
244 /* HiZ Clear */
245 if (has_hiz_ram)
246 R300_INIT_ATOM(hiz_clear, 0);
247 /* zmask clear */
248 R300_INIT_ATOM(zmask_clear, 0);
249 }
250 /* ZB (unpipelined), SU. */
251 R300_INIT_ATOM(query_start, 4);
252
253 /* Replace emission functions for r500. */
254 if (is_r500) {
255 r300->fs.emit = r500_emit_fs;
256 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
257 r300->fs_constants.emit = r500_emit_fs_constants;
258 }
259
260 /* Some non-CSO atoms need explicit space to store the state locally. */
261 r300->aa_state.state = CALLOC_STRUCT(r300_aa_state);
262 r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
263 r300->clip_state.state = CALLOC_STRUCT(r300_clip_state);
264 r300->fb_state.state = CALLOC_STRUCT(pipe_framebuffer_state);
265 r300->gpu_flush.state = CALLOC_STRUCT(pipe_framebuffer_state);
266 r300->hyperz_state.state = CALLOC_STRUCT(r300_hyperz_state);
267 r300->invariant_state.state = CALLOC_STRUCT(r300_invariant_state);
268 r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
269 r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
270 r300->textures_state.state = CALLOC_STRUCT(r300_textures_state);
271 r300->vap_invariant_state.state = CALLOC_STRUCT(r300_vap_invariant_state);
272 r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
273 r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
274 r300->fs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
275 r300->vs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
276 if (!r300->screen->caps.has_tcl) {
277 r300->vertex_stream_state.state = CALLOC_STRUCT(r300_vertex_stream_state);
278 }
279
280 /* Some non-CSO atoms don't use the state pointer. */
281 r300->fb_state_pipelined.allow_null_state = TRUE;
282 r300->fs_rc_constant_state.allow_null_state = TRUE;
283 r300->pvs_flush.allow_null_state = TRUE;
284 r300->query_start.allow_null_state = TRUE;
285 r300->texture_cache_inval.allow_null_state = TRUE;
286
287 /* Some states must be marked as dirty here to properly set up
288 * hardware in the first command stream. */
289 r300->invariant_state.dirty = TRUE;
290 r300->pvs_flush.dirty = TRUE;
291 r300->vap_invariant_state.dirty = TRUE;
292 r300->texture_cache_inval.dirty = TRUE;
293 r300->textures_state.dirty = TRUE;
294 }
295
296 /* Not every state tracker calls every driver function before the first draw
297 * call and we must initialize the command buffers somehow. */
298 static void r300_init_states(struct pipe_context *pipe)
299 {
300 struct r300_context *r300 = r300_context(pipe);
301 struct pipe_blend_color bc = {{0}};
302 struct pipe_clip_state cs = {{{0}}};
303 struct pipe_scissor_state ss = {0};
304 struct r300_clip_state *clip =
305 (struct r300_clip_state*)r300->clip_state.state;
306 struct r300_gpu_flush *gpuflush =
307 (struct r300_gpu_flush*)r300->gpu_flush.state;
308 struct r300_vap_invariant_state *vap_invariant =
309 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
310 struct r300_invariant_state *invariant =
311 (struct r300_invariant_state*)r300->invariant_state.state;
312
313 CB_LOCALS;
314
315 pipe->set_blend_color(pipe, &bc);
316 pipe->set_scissor_state(pipe, &ss);
317
318 /* Initialize the clip state. */
319 if (r300_context(pipe)->screen->caps.has_tcl) {
320 pipe->set_clip_state(pipe, &cs);
321 } else {
322 BEGIN_CB(clip->cb, 2);
323 OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
324 END_CB;
325 }
326
327 /* Initialize the GPU flush. */
328 {
329 BEGIN_CB(gpuflush->cb_flush_clean, 6);
330
331 /* Flush and free renderbuffer caches. */
332 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
333 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
334 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
335 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
336 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
337 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
338
339 /* Wait until the GPU is idle.
340 * This fixes random pixels sometimes appearing probably caused
341 * by incomplete rendering. */
342 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
343 END_CB;
344 }
345
346 /* Initialize the VAP invariant state. */
347 {
348 BEGIN_CB(vap_invariant->cb, 9);
349 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
350 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
351 OUT_CB_32F(1.0);
352 OUT_CB_32F(1.0);
353 OUT_CB_32F(1.0);
354 OUT_CB_32F(1.0);
355 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
356 END_CB;
357 }
358
359 /* Initialize the invariant state. */
360 {
361 BEGIN_CB(invariant->cb, r300->invariant_state.size);
362 OUT_CB_REG(R300_GB_SELECT, 0);
363 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
364 OUT_CB_REG(R300_GA_ROUND_MODE, 1);
365 OUT_CB_REG(R300_GA_OFFSET, 0);
366 OUT_CB_REG(R300_SU_TEX_WRAP, 0);
367 OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
368 OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
369 OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
370
371 if (r300->screen->caps.is_rv350) {
372 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
373 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
374 }
375 END_CB;
376 }
377
378 /* Initialize the hyperz state. */
379 {
380 struct r300_hyperz_state *hyperz =
381 (struct r300_hyperz_state*)r300->hyperz_state.state;
382 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
383 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
384 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
385 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
386 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
387 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
388
389 if (r300->screen->caps.is_r500 ||
390 (r300->screen->caps.is_rv350 &&
391 r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) {
392 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
393 }
394 END_CB;
395 }
396 }
397
398 struct pipe_context* r300_create_context(struct pipe_screen* screen,
399 void *priv)
400 {
401 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
402 struct r300_screen* r300screen = r300_screen(screen);
403 struct r300_winsys_screen *rws = r300screen->rws;
404
405 if (!r300)
406 return NULL;
407
408 r300_update_num_contexts(r300screen, 1);
409
410 r300->rws = rws;
411 r300->screen = r300screen;
412
413 r300->context.winsys = (struct pipe_winsys*)rws;
414 r300->context.screen = screen;
415 r300->context.priv = priv;
416
417 r300->context.destroy = r300_destroy_context;
418
419 make_empty_list(&r300->query_list);
420
421 util_mempool_create(&r300->pool_transfers,
422 sizeof(struct pipe_transfer), 64,
423 UTIL_MEMPOOL_SINGLETHREADED);
424
425 r300->cs = rws->cs_create(rws);
426 if (r300->cs == NULL)
427 goto fail;
428
429 if (!r300screen->caps.has_tcl) {
430 /* Create a Draw. This is used for SW TCL. */
431 r300->draw = draw_create(&r300->context);
432 /* Enable our renderer. */
433 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
434 /* Disable converting points/lines to triangles. */
435 draw_wide_line_threshold(r300->draw, 10000000.f);
436 draw_wide_point_threshold(r300->draw, 10000000.f);
437 }
438
439 r300_setup_atoms(r300);
440
441 r300_init_blit_functions(r300);
442 r300_init_flush_functions(r300);
443 r300_init_query_functions(r300);
444 r300_init_state_functions(r300);
445 r300_init_resource_functions(r300);
446
447 r300->blitter = util_blitter_create(&r300->context);
448 if (r300->blitter == NULL)
449 goto fail;
450
451 /* Render functions must be initialized after blitter. */
452 r300_init_render_functions(r300);
453
454 rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
455
456 /* setup hyper-z mm */
457 if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
458 if (!r300_hyperz_init_mm(r300))
459 goto fail;
460
461 r300->upload_ib = u_upload_create(&r300->context,
462 32 * 1024, 16,
463 PIPE_BIND_INDEX_BUFFER);
464
465 if (r300->upload_ib == NULL)
466 goto fail;
467
468 r300->upload_vb = u_upload_create(&r300->context,
469 128 * 1024, 16,
470 PIPE_BIND_VERTEX_BUFFER);
471 if (r300->upload_vb == NULL)
472 goto fail;
473
474 r300->tran.translate_cache = translate_cache_create();
475 if (r300->tran.translate_cache == NULL)
476 goto fail;
477
478 r300_init_states(&r300->context);
479
480 /* The KIL opcode needs the first texture unit to be enabled
481 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
482 * dummy texture there. */
483 if (!r300->screen->caps.is_r500) {
484 struct pipe_resource *tex;
485 struct pipe_resource rtempl = {{0}};
486 struct pipe_sampler_view vtempl = {{0}};
487
488 rtempl.target = PIPE_TEXTURE_2D;
489 rtempl.format = PIPE_FORMAT_I8_UNORM;
490 rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
491 rtempl.width0 = 1;
492 rtempl.height0 = 1;
493 rtempl.depth0 = 1;
494 tex = screen->resource_create(screen, &rtempl);
495
496 u_sampler_view_default_template(&vtempl, tex, tex->format);
497
498 r300->texkill_sampler = (struct r300_sampler_view*)
499 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
500
501 pipe_resource_reference(&tex, NULL);
502 }
503
504 return &r300->context;
505
506 fail:
507 r300_destroy_context(&r300->context);
508 return NULL;
509 }
510
511 void r300_finish(struct r300_context *r300)
512 {
513 struct pipe_framebuffer_state *fb;
514 unsigned i;
515
516 /* This is a preliminary implementation of glFinish.
517 *
518 * The ideal implementation should use something like EmitIrqLocked and
519 * WaitIrq, or better, real fences.
520 */
521 if (r300->fb_state.state) {
522 fb = r300->fb_state.state;
523
524 for (i = 0; i < fb->nr_cbufs; i++) {
525 if (fb->cbufs[i]->texture) {
526 r300->rws->buffer_wait(r300->rws,
527 r300_texture(fb->cbufs[i]->texture)->buffer);
528 return;
529 }
530 }
531 if (fb->zsbuf && fb->zsbuf->texture) {
532 r300->rws->buffer_wait(r300->rws,
533 r300_texture(fb->zsbuf->texture)->buffer);
534 }
535 }
536 }