Merge branch 'master' of git://anongit.freedesktop.org/mesa/mesa
[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 #include "os/os_time.h"
30 #include "vl/vl_decoder.h"
31 #include "vl/vl_video_buffer.h"
32
33 #include "r300_cb.h"
34 #include "r300_context.h"
35 #include "r300_emit.h"
36 #include "r300_screen.h"
37 #include "r300_screen_buffer.h"
38
39 static void r300_update_num_contexts(struct r300_screen *r300screen,
40 int diff)
41 {
42 pipe_mutex_lock(r300screen->num_contexts_mutex);
43 if (diff > 0) {
44 r300screen->num_contexts++;
45
46 if (r300screen->num_contexts > 1)
47 util_slab_set_thread_safety(&r300screen->pool_buffers,
48 UTIL_SLAB_MULTITHREADED);
49 } else {
50 r300screen->num_contexts--;
51
52 if (r300screen->num_contexts <= 1)
53 util_slab_set_thread_safety(&r300screen->pool_buffers,
54 UTIL_SLAB_SINGLETHREADED);
55 }
56 pipe_mutex_unlock(r300screen->num_contexts_mutex);
57 }
58
59 static void r300_release_referenced_objects(struct r300_context *r300)
60 {
61 struct pipe_framebuffer_state *fb =
62 (struct pipe_framebuffer_state*)r300->fb_state.state;
63 struct r300_textures_state *textures =
64 (struct r300_textures_state*)r300->textures_state.state;
65 struct r300_query *query, *temp;
66 unsigned i;
67
68 /* Framebuffer state. */
69 util_unreference_framebuffer_state(fb);
70
71 /* Textures. */
72 for (i = 0; i < textures->sampler_view_count; i++)
73 pipe_sampler_view_reference(
74 (struct pipe_sampler_view**)&textures->sampler_views[i], NULL);
75
76 /* The special dummy texture for texkill. */
77 if (r300->texkill_sampler) {
78 pipe_sampler_view_reference(
79 (struct pipe_sampler_view**)&r300->texkill_sampler,
80 NULL);
81 }
82
83 /* Manually-created vertex buffers. */
84 pipe_resource_reference(&r300->dummy_vb, NULL);
85 pipe_resource_reference(&r300->vbo, NULL);
86
87 /* If there are any queries pending or not destroyed, remove them now. */
88 foreach_s(query, temp, &r300->query_list) {
89 remove_from_list(query);
90 FREE(query);
91 }
92
93 r300->context.delete_depth_stencil_alpha_state(&r300->context,
94 r300->dsa_decompress_zmask);
95 }
96
97 static void r300_destroy_context(struct pipe_context* context)
98 {
99 struct r300_context* r300 = r300_context(context);
100
101 if (r300->cs && r300->hyperz_enabled) {
102 r300->rws->cs_request_feature(r300->cs, RADEON_FID_R300_HYPERZ_ACCESS, FALSE);
103 }
104
105 if (r300->blitter)
106 util_blitter_destroy(r300->blitter);
107 if (r300->draw)
108 draw_destroy(r300->draw);
109
110 if (r300->vbuf_mgr)
111 u_vbuf_mgr_destroy(r300->vbuf_mgr);
112
113 /* XXX: This function assumes r300->query_list was initialized */
114 r300_release_referenced_objects(r300);
115
116 if (r300->cs)
117 r300->rws->cs_destroy(r300->cs);
118
119 /* XXX: No way to tell if this was initialized or not? */
120 util_slab_destroy(&r300->pool_transfers);
121
122 r300_update_num_contexts(r300->screen, -1);
123
124 /* Free the structs allocated in r300_setup_atoms() */
125 if (r300->aa_state.state) {
126 FREE(r300->aa_state.state);
127 FREE(r300->blend_color_state.state);
128 FREE(r300->clip_state.state);
129 FREE(r300->fb_state.state);
130 FREE(r300->gpu_flush.state);
131 FREE(r300->hyperz_state.state);
132 FREE(r300->invariant_state.state);
133 FREE(r300->rs_block_state.state);
134 FREE(r300->scissor_state.state);
135 FREE(r300->textures_state.state);
136 FREE(r300->vap_invariant_state.state);
137 FREE(r300->viewport_state.state);
138 FREE(r300->ztop_state.state);
139 FREE(r300->fs_constants.state);
140 FREE(r300->vs_constants.state);
141 if (!r300->screen->caps.has_tcl) {
142 FREE(r300->vertex_stream_state.state);
143 }
144 }
145 FREE(r300);
146 }
147
148 static void r300_flush_callback(void *data, unsigned flags)
149 {
150 struct r300_context* const cs_context_copy = data;
151
152 r300_flush(&cs_context_copy->context, flags, NULL);
153 }
154
155 #define R300_INIT_ATOM(atomname, atomsize) \
156 do { \
157 r300->atomname.name = #atomname; \
158 r300->atomname.state = NULL; \
159 r300->atomname.size = atomsize; \
160 r300->atomname.emit = r300_emit_##atomname; \
161 r300->atomname.dirty = FALSE; \
162 } while (0)
163
164 #define R300_ALLOC_ATOM(atomname, statetype) \
165 do { \
166 r300->atomname.state = CALLOC_STRUCT(statetype); \
167 if (r300->atomname.state == NULL) \
168 return FALSE; \
169 } while (0)
170
171 static boolean r300_setup_atoms(struct r300_context* r300)
172 {
173 boolean is_rv350 = r300->screen->caps.is_rv350;
174 boolean is_r500 = r300->screen->caps.is_r500;
175 boolean has_tcl = r300->screen->caps.has_tcl;
176 boolean drm_2_6_0 = r300->screen->info.drm_minor >= 6;
177
178 /* Create the actual atom list.
179 *
180 * Some atoms never change size, others change every emit - those have
181 * the size of 0 here.
182 *
183 * NOTE: The framebuffer state is split into these atoms:
184 * - gpu_flush (unpipelined regs)
185 * - aa_state (unpipelined regs)
186 * - fb_state (unpipelined regs)
187 * - hyperz_state (unpipelined regs followed by pipelined ones)
188 * - fb_state_pipelined (pipelined regs)
189 * The motivation behind this is to be able to emit a strict
190 * subset of the regs, and to have reasonable register ordering. */
191 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
192 R300_INIT_ATOM(gpu_flush, 9);
193 R300_INIT_ATOM(aa_state, 4);
194 R300_INIT_ATOM(fb_state, 0);
195 R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
196 /* ZB (unpipelined), SC. */
197 R300_INIT_ATOM(ztop_state, 2);
198 /* ZB, FG. */
199 R300_INIT_ATOM(dsa_state, is_r500 ? (drm_2_6_0 ? 10 : 8) : 6);
200 /* RB3D. */
201 R300_INIT_ATOM(blend_state, 8);
202 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
203 /* SC. */
204 R300_INIT_ATOM(scissor_state, 3);
205 /* GB, FG, GA, SU, SC, RB3D. */
206 R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0) + (is_r500 ? 4 : 0));
207 /* VAP. */
208 R300_INIT_ATOM(viewport_state, 9);
209 R300_INIT_ATOM(pvs_flush, 2);
210 R300_INIT_ATOM(vap_invariant_state, is_r500 ? 11 : 9);
211 R300_INIT_ATOM(vertex_stream_state, 0);
212 R300_INIT_ATOM(vs_state, 0);
213 R300_INIT_ATOM(vs_constants, 0);
214 R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
215 /* VAP, RS, GA, GB, SU, SC. */
216 R300_INIT_ATOM(rs_block_state, 0);
217 R300_INIT_ATOM(rs_state, 0);
218 /* SC, US. */
219 R300_INIT_ATOM(fb_state_pipelined, 8);
220 /* US. */
221 R300_INIT_ATOM(fs, 0);
222 R300_INIT_ATOM(fs_rc_constant_state, 0);
223 R300_INIT_ATOM(fs_constants, 0);
224 /* TX. */
225 R300_INIT_ATOM(texture_cache_inval, 2);
226 R300_INIT_ATOM(textures_state, 0);
227 /* HiZ Clear */
228 R300_INIT_ATOM(hiz_clear, r300->screen->caps.hiz_ram > 0 ? 4 : 0);
229 /* zmask clear */
230 R300_INIT_ATOM(zmask_clear, r300->screen->caps.zmask_ram > 0 ? 4 : 0);
231 /* ZB (unpipelined), SU. */
232 R300_INIT_ATOM(query_start, 4);
233
234 /* Replace emission functions for r500. */
235 if (is_r500) {
236 r300->fs.emit = r500_emit_fs;
237 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
238 r300->fs_constants.emit = r500_emit_fs_constants;
239 }
240
241 /* Some non-CSO atoms need explicit space to store the state locally. */
242 R300_ALLOC_ATOM(aa_state, r300_aa_state);
243 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
244 R300_ALLOC_ATOM(clip_state, r300_clip_state);
245 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
246 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
247 R300_ALLOC_ATOM(textures_state, r300_textures_state);
248 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
249 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
250 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
251 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
252 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
253 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
254 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
255 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
256 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
257 if (!r300->screen->caps.has_tcl) {
258 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
259 }
260
261 /* Some non-CSO atoms don't use the state pointer. */
262 r300->fb_state_pipelined.allow_null_state = TRUE;
263 r300->fs_rc_constant_state.allow_null_state = TRUE;
264 r300->pvs_flush.allow_null_state = TRUE;
265 r300->query_start.allow_null_state = TRUE;
266 r300->texture_cache_inval.allow_null_state = TRUE;
267
268 /* Some states must be marked as dirty here to properly set up
269 * hardware in the first command stream. */
270 r300_mark_atom_dirty(r300, &r300->invariant_state);
271 r300_mark_atom_dirty(r300, &r300->pvs_flush);
272 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
273 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
274 r300_mark_atom_dirty(r300, &r300->textures_state);
275
276 return TRUE;
277 }
278
279 /* Not every state tracker calls every driver function before the first draw
280 * call and we must initialize the command buffers somehow. */
281 static void r300_init_states(struct pipe_context *pipe)
282 {
283 struct r300_context *r300 = r300_context(pipe);
284 struct pipe_blend_color bc = {{0}};
285 struct pipe_clip_state cs = {{{0}}};
286 struct pipe_scissor_state ss = {0};
287 struct r300_clip_state *clip =
288 (struct r300_clip_state*)r300->clip_state.state;
289 struct r300_gpu_flush *gpuflush =
290 (struct r300_gpu_flush*)r300->gpu_flush.state;
291 struct r300_vap_invariant_state *vap_invariant =
292 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
293 struct r300_invariant_state *invariant =
294 (struct r300_invariant_state*)r300->invariant_state.state;
295
296 CB_LOCALS;
297
298 pipe->set_blend_color(pipe, &bc);
299 pipe->set_scissor_state(pipe, &ss);
300
301 /* Initialize the clip state. */
302 if (r300->screen->caps.has_tcl) {
303 pipe->set_clip_state(pipe, &cs);
304 } else {
305 BEGIN_CB(clip->cb, 2);
306 OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
307 END_CB;
308 }
309
310 /* Initialize the GPU flush. */
311 {
312 BEGIN_CB(gpuflush->cb_flush_clean, 6);
313
314 /* Flush and free renderbuffer caches. */
315 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
316 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
317 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
318 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
319 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
320 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
321
322 /* Wait until the GPU is idle.
323 * This fixes random pixels sometimes appearing probably caused
324 * by incomplete rendering. */
325 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
326 END_CB;
327 }
328
329 /* Initialize the VAP invariant state. */
330 {
331 BEGIN_CB(vap_invariant->cb, r300->vap_invariant_state.size);
332 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
333 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
334 OUT_CB_32F(1.0);
335 OUT_CB_32F(1.0);
336 OUT_CB_32F(1.0);
337 OUT_CB_32F(1.0);
338 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
339
340 if (r300->screen->caps.is_r500) {
341 OUT_CB_REG(R500_VAP_TEX_TO_COLOR_CNTL, 0);
342 }
343 END_CB;
344 }
345
346 /* Initialize the invariant state. */
347 {
348 BEGIN_CB(invariant->cb, r300->invariant_state.size);
349 OUT_CB_REG(R300_GB_SELECT, 0);
350 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
351 OUT_CB_REG(R300_GA_OFFSET, 0);
352 OUT_CB_REG(R300_SU_TEX_WRAP, 0);
353 OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
354 OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
355 OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
356 OUT_CB_REG(R300_SC_SCREENDOOR, 0xffffff);
357
358 if (r300->screen->caps.is_rv350) {
359 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
360 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
361 }
362
363 if (r300->screen->caps.is_r500) {
364 OUT_CB_REG(R500_GA_COLOR_CONTROL_PS3, 0);
365 OUT_CB_REG(R500_SU_TEX_WRAP_PS3, 0);
366 }
367 END_CB;
368 }
369
370 /* Initialize the hyperz state. */
371 {
372 struct r300_hyperz_state *hyperz =
373 (struct r300_hyperz_state*)r300->hyperz_state.state;
374 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
375 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
376 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
377 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
378 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
379 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
380
381 if (r300->screen->caps.is_r500 ||
382 (r300->screen->caps.is_rv350 &&
383 r300->screen->info.drm_minor >= 6)) {
384 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
385 }
386 END_CB;
387 }
388 }
389
390 struct pipe_context* r300_create_context(struct pipe_screen* screen,
391 void *priv)
392 {
393 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
394 struct r300_screen* r300screen = r300_screen(screen);
395 struct radeon_winsys *rws = r300screen->rws;
396
397 if (!r300)
398 return NULL;
399
400 r300_update_num_contexts(r300screen, 1);
401
402 r300->rws = rws;
403 r300->screen = r300screen;
404
405 r300->context.winsys = (struct pipe_winsys*)rws;
406 r300->context.screen = screen;
407 r300->context.priv = priv;
408
409 r300->context.destroy = r300_destroy_context;
410
411 make_empty_list(&r300->query_list);
412
413 util_slab_create(&r300->pool_transfers,
414 sizeof(struct pipe_transfer), 64,
415 UTIL_SLAB_SINGLETHREADED);
416
417 r300->cs = rws->cs_create(rws);
418 if (r300->cs == NULL)
419 goto fail;
420
421 if (!r300screen->caps.has_tcl) {
422 /* Create a Draw. This is used for SW TCL. */
423 r300->draw = draw_create(&r300->context);
424 if (r300->draw == NULL)
425 goto fail;
426 /* Enable our renderer. */
427 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
428 /* Disable converting points/lines to triangles. */
429 draw_wide_line_threshold(r300->draw, 10000000.f);
430 draw_wide_point_threshold(r300->draw, 10000000.f);
431 }
432
433 if (!r300_setup_atoms(r300))
434 goto fail;
435
436 r300_init_blit_functions(r300);
437 r300_init_flush_functions(r300);
438 r300_init_query_functions(r300);
439 r300_init_state_functions(r300);
440 r300_init_resource_functions(r300);
441
442 r300->context.create_video_decoder = vl_create_decoder;
443 r300->context.create_video_buffer = vl_video_buffer_create;
444
445 r300->vbuf_mgr = u_vbuf_mgr_create(&r300->context, 1024 * 1024, 16,
446 PIPE_BIND_VERTEX_BUFFER |
447 PIPE_BIND_INDEX_BUFFER,
448 U_VERTEX_FETCH_DWORD_ALIGNED);
449 if (!r300->vbuf_mgr)
450 goto fail;
451
452 r300->blitter = util_blitter_create(&r300->context);
453 if (r300->blitter == NULL)
454 goto fail;
455
456 /* Render functions must be initialized after blitter. */
457 r300_init_render_functions(r300);
458 r300_init_states(&r300->context);
459
460 rws->cs_set_flush_callback(r300->cs, r300_flush_callback, r300);
461
462 /* The KIL opcode needs the first texture unit to be enabled
463 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
464 * dummy texture there. */
465 if (!r300->screen->caps.is_r500) {
466 struct pipe_resource *tex;
467 struct pipe_resource rtempl = {{0}};
468 struct pipe_sampler_view vtempl = {{0}};
469
470 rtempl.target = PIPE_TEXTURE_2D;
471 rtempl.format = PIPE_FORMAT_I8_UNORM;
472 rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
473 rtempl.usage = PIPE_USAGE_IMMUTABLE;
474 rtempl.width0 = 1;
475 rtempl.height0 = 1;
476 rtempl.depth0 = 1;
477 tex = screen->resource_create(screen, &rtempl);
478
479 u_sampler_view_default_template(&vtempl, tex, tex->format);
480
481 r300->texkill_sampler = (struct r300_sampler_view*)
482 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
483
484 pipe_resource_reference(&tex, NULL);
485 }
486
487 {
488 struct pipe_resource vb;
489 memset(&vb, 0, sizeof(vb));
490 vb.target = PIPE_BUFFER;
491 vb.format = PIPE_FORMAT_R8_UNORM;
492 vb.bind = PIPE_BIND_VERTEX_BUFFER;
493 vb.usage = PIPE_USAGE_IMMUTABLE;
494 vb.width0 = sizeof(float) * 16;
495 vb.height0 = 1;
496 vb.depth0 = 1;
497
498 r300->dummy_vb = screen->resource_create(screen, &vb);
499 }
500
501 {
502 struct pipe_depth_stencil_alpha_state dsa;
503 memset(&dsa, 0, sizeof(dsa));
504 dsa.depth.writemask = 1;
505
506 r300->dsa_decompress_zmask =
507 r300->context.create_depth_stencil_alpha_state(&r300->context,
508 &dsa);
509 }
510
511 r300->hyperz_time_of_last_flush = os_time_get();
512
513 /* Print driver info. */
514 #ifdef DEBUG
515 {
516 #else
517 if (DBG_ON(r300, DBG_INFO)) {
518 #endif
519 fprintf(stderr,
520 "r300: DRM version: %d.%d.%d, Name: %s, ID: 0x%04x, GB: %d, Z: %d\n"
521 "r300: GART size: %d MB, VRAM size: %d MB\n"
522 "r300: AA compression RAM: %s, Z compression RAM: %s, HiZ RAM: %s\n",
523 r300->screen->info.drm_major,
524 r300->screen->info.drm_minor,
525 r300->screen->info.drm_patchlevel,
526 screen->get_name(screen),
527 r300->screen->info.pci_id,
528 r300->screen->info.r300_num_gb_pipes,
529 r300->screen->info.r300_num_z_pipes,
530 r300->screen->info.gart_size >> 20,
531 r300->screen->info.vram_size >> 20,
532 "YES", /* XXX really? */
533 r300->screen->caps.zmask_ram ? "YES" : "NO",
534 r300->screen->caps.hiz_ram ? "YES" : "NO");
535 }
536
537 return &r300->context;
538
539 fail:
540 r300_destroy_context(&r300->context);
541 return NULL;
542 }