r300g: implement fences using dummy relocations
[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 /* If there are any queries pending or not destroyed, remove them now. */
87 foreach_s(query, temp, &r300->query_list) {
88 remove_from_list(query);
89 FREE(query);
90 }
91
92 r300->context.delete_depth_stencil_alpha_state(&r300->context,
93 r300->dsa_decompress_zmask);
94 }
95
96 static void r300_destroy_context(struct pipe_context* context)
97 {
98 struct r300_context* r300 = r300_context(context);
99
100 if (r300->blitter)
101 util_blitter_destroy(r300->blitter);
102 if (r300->draw)
103 draw_destroy(r300->draw);
104
105 if (r300->vbuf_mgr)
106 u_vbuf_mgr_destroy(r300->vbuf_mgr);
107
108 /* XXX: This function assumes r300->query_list was initialized */
109 r300_release_referenced_objects(r300);
110
111 if (r300->cs)
112 r300->rws->cs_destroy(r300->cs);
113
114 /* XXX: No way to tell if this was initialized or not? */
115 util_slab_destroy(&r300->pool_transfers);
116
117 r300_update_num_contexts(r300->screen, -1);
118
119 /* Free the structs allocated in r300_setup_atoms() */
120 if (r300->aa_state.state) {
121 FREE(r300->aa_state.state);
122 FREE(r300->blend_color_state.state);
123 FREE(r300->clip_state.state);
124 FREE(r300->fb_state.state);
125 FREE(r300->gpu_flush.state);
126 FREE(r300->hyperz_state.state);
127 FREE(r300->invariant_state.state);
128 FREE(r300->rs_block_state.state);
129 FREE(r300->scissor_state.state);
130 FREE(r300->textures_state.state);
131 FREE(r300->vap_invariant_state.state);
132 FREE(r300->viewport_state.state);
133 FREE(r300->ztop_state.state);
134 FREE(r300->fs_constants.state);
135 FREE(r300->vs_constants.state);
136 if (!r300->screen->caps.has_tcl) {
137 FREE(r300->vertex_stream_state.state);
138 }
139 }
140 FREE(r300);
141 }
142
143 void r300_flush_cb(void *data)
144 {
145 struct r300_context* const cs_context_copy = data;
146
147 cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
148 }
149
150 #define R300_INIT_ATOM(atomname, atomsize) \
151 do { \
152 r300->atomname.name = #atomname; \
153 r300->atomname.state = NULL; \
154 r300->atomname.size = atomsize; \
155 r300->atomname.emit = r300_emit_##atomname; \
156 r300->atomname.dirty = FALSE; \
157 } while (0)
158
159 #define R300_ALLOC_ATOM(atomname, statetype) \
160 do { \
161 r300->atomname.state = CALLOC_STRUCT(statetype); \
162 if (r300->atomname.state == NULL) \
163 return FALSE; \
164 } while (0)
165
166 static boolean r300_setup_atoms(struct r300_context* r300)
167 {
168 boolean is_rv350 = r300->screen->caps.is_rv350;
169 boolean is_r500 = r300->screen->caps.is_r500;
170 boolean has_tcl = r300->screen->caps.has_tcl;
171 boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
172 boolean drm_2_6_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0);
173 boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
174 boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0;
175
176 /* Create the actual atom list.
177 *
178 * Some atoms never change size, others change every emit - those have
179 * the size of 0 here.
180 *
181 * NOTE: The framebuffer state is split into these atoms:
182 * - gpu_flush (unpipelined regs)
183 * - aa_state (unpipelined regs)
184 * - fb_state (unpipelined regs)
185 * - hyperz_state (unpipelined regs followed by pipelined ones)
186 * - fb_state_pipelined (pipelined regs)
187 * The motivation behind this is to be able to emit a strict
188 * subset of the regs, and to have reasonable register ordering. */
189 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
190 R300_INIT_ATOM(gpu_flush, 9);
191 R300_INIT_ATOM(aa_state, 4);
192 R300_INIT_ATOM(fb_state, 0);
193 R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
194 /* ZB (unpipelined), SC. */
195 R300_INIT_ATOM(ztop_state, 2);
196 /* ZB, FG. */
197 R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
198 /* RB3D. */
199 R300_INIT_ATOM(blend_state, 8);
200 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
201 /* SC. */
202 R300_INIT_ATOM(scissor_state, 3);
203 /* GB, FG, GA, SU, SC, RB3D. */
204 R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0));
205 /* VAP. */
206 R300_INIT_ATOM(viewport_state, 9);
207 R300_INIT_ATOM(pvs_flush, 2);
208 R300_INIT_ATOM(vap_invariant_state, 9);
209 R300_INIT_ATOM(vertex_stream_state, 0);
210 R300_INIT_ATOM(vs_state, 0);
211 R300_INIT_ATOM(vs_constants, 0);
212 R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
213 /* VAP, RS, GA, GB, SU, SC. */
214 R300_INIT_ATOM(rs_block_state, 0);
215 R300_INIT_ATOM(rs_state, 0);
216 /* SC, US. */
217 R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0));
218 /* US. */
219 R300_INIT_ATOM(fs, 0);
220 R300_INIT_ATOM(fs_rc_constant_state, 0);
221 R300_INIT_ATOM(fs_constants, 0);
222 /* TX. */
223 R300_INIT_ATOM(texture_cache_inval, 2);
224 R300_INIT_ATOM(textures_state, 0);
225 if (can_hyperz) {
226 /* HiZ Clear */
227 if (has_hiz_ram)
228 R300_INIT_ATOM(hiz_clear, 0);
229 /* zmask clear */
230 R300_INIT_ATOM(zmask_clear, 4);
231 }
232 /* ZB (unpipelined), SU. */
233 R300_INIT_ATOM(query_start, 4);
234
235 /* Replace emission functions for r500. */
236 if (is_r500) {
237 r300->fs.emit = r500_emit_fs;
238 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
239 r300->fs_constants.emit = r500_emit_fs_constants;
240 }
241
242 /* Some non-CSO atoms need explicit space to store the state locally. */
243 R300_ALLOC_ATOM(aa_state, r300_aa_state);
244 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
245 R300_ALLOC_ATOM(clip_state, r300_clip_state);
246 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
247 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
248 R300_ALLOC_ATOM(textures_state, r300_textures_state);
249 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
250 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
251 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
252 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
253 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
254 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
255 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
256 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
257 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
258 if (!r300->screen->caps.has_tcl) {
259 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
260 }
261
262 /* Some non-CSO atoms don't use the state pointer. */
263 r300->fb_state_pipelined.allow_null_state = TRUE;
264 r300->fs_rc_constant_state.allow_null_state = TRUE;
265 r300->pvs_flush.allow_null_state = TRUE;
266 r300->query_start.allow_null_state = TRUE;
267 r300->texture_cache_inval.allow_null_state = TRUE;
268
269 /* Some states must be marked as dirty here to properly set up
270 * hardware in the first command stream. */
271 r300_mark_atom_dirty(r300, &r300->invariant_state);
272 r300_mark_atom_dirty(r300, &r300->pvs_flush);
273 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
274 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
275 r300_mark_atom_dirty(r300, &r300->textures_state);
276
277 return TRUE;
278 }
279
280 /* Not every state tracker calls every driver function before the first draw
281 * call and we must initialize the command buffers somehow. */
282 static void r300_init_states(struct pipe_context *pipe)
283 {
284 struct r300_context *r300 = r300_context(pipe);
285 struct pipe_blend_color bc = {{0}};
286 struct pipe_clip_state cs = {{{0}}};
287 struct pipe_scissor_state ss = {0};
288 struct r300_clip_state *clip =
289 (struct r300_clip_state*)r300->clip_state.state;
290 struct r300_gpu_flush *gpuflush =
291 (struct r300_gpu_flush*)r300->gpu_flush.state;
292 struct r300_vap_invariant_state *vap_invariant =
293 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
294 struct r300_invariant_state *invariant =
295 (struct r300_invariant_state*)r300->invariant_state.state;
296
297 CB_LOCALS;
298
299 pipe->set_blend_color(pipe, &bc);
300 pipe->set_scissor_state(pipe, &ss);
301
302 /* Initialize the clip state. */
303 if (r300->screen->caps.has_tcl) {
304 pipe->set_clip_state(pipe, &cs);
305 } else {
306 BEGIN_CB(clip->cb, 2);
307 OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
308 END_CB;
309 }
310
311 /* Initialize the GPU flush. */
312 {
313 BEGIN_CB(gpuflush->cb_flush_clean, 6);
314
315 /* Flush and free renderbuffer caches. */
316 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
317 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
318 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
319 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
320 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
321 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
322
323 /* Wait until the GPU is idle.
324 * This fixes random pixels sometimes appearing probably caused
325 * by incomplete rendering. */
326 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
327 END_CB;
328 }
329
330 /* Initialize the VAP invariant state. */
331 {
332 BEGIN_CB(vap_invariant->cb, 9);
333 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
334 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
335 OUT_CB_32F(1.0);
336 OUT_CB_32F(1.0);
337 OUT_CB_32F(1.0);
338 OUT_CB_32F(1.0);
339 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
340 END_CB;
341 }
342
343 /* Initialize the invariant state. */
344 {
345 BEGIN_CB(invariant->cb, r300->invariant_state.size);
346 OUT_CB_REG(R300_GB_SELECT, 0);
347 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
348 OUT_CB_REG(R300_GA_ROUND_MODE, 1);
349 OUT_CB_REG(R300_GA_OFFSET, 0);
350 OUT_CB_REG(R300_SU_TEX_WRAP, 0);
351 OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
352 OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
353 OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
354
355 if (r300->screen->caps.is_rv350) {
356 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
357 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
358 }
359 END_CB;
360 }
361
362 /* Initialize the hyperz state. */
363 {
364 struct r300_hyperz_state *hyperz =
365 (struct r300_hyperz_state*)r300->hyperz_state.state;
366 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
367 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
368 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
369 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
370 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
371 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
372
373 if (r300->screen->caps.is_r500 ||
374 (r300->screen->caps.is_rv350 &&
375 r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) {
376 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
377 }
378 END_CB;
379 }
380 }
381
382 struct pipe_context* r300_create_context(struct pipe_screen* screen,
383 void *priv)
384 {
385 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
386 struct r300_screen* r300screen = r300_screen(screen);
387 struct r300_winsys_screen *rws = r300screen->rws;
388
389 if (!r300)
390 return NULL;
391
392 r300_update_num_contexts(r300screen, 1);
393
394 r300->rws = rws;
395 r300->screen = r300screen;
396
397 r300->context.winsys = (struct pipe_winsys*)rws;
398 r300->context.screen = screen;
399 r300->context.priv = priv;
400
401 r300->context.destroy = r300_destroy_context;
402
403 make_empty_list(&r300->query_list);
404
405 util_slab_create(&r300->pool_transfers,
406 sizeof(struct pipe_transfer), 64,
407 UTIL_SLAB_SINGLETHREADED);
408
409 r300->cs = rws->cs_create(rws);
410 if (r300->cs == NULL)
411 goto fail;
412
413 if (!r300screen->caps.has_tcl) {
414 /* Create a Draw. This is used for SW TCL. */
415 r300->draw = draw_create(&r300->context);
416 if (r300->draw == NULL)
417 goto fail;
418 /* Enable our renderer. */
419 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
420 /* Disable converting points/lines to triangles. */
421 draw_wide_line_threshold(r300->draw, 10000000.f);
422 draw_wide_point_threshold(r300->draw, 10000000.f);
423 }
424
425 if (!r300_setup_atoms(r300))
426 goto fail;
427
428 r300_init_blit_functions(r300);
429 r300_init_flush_functions(r300);
430 r300_init_query_functions(r300);
431 r300_init_state_functions(r300);
432 r300_init_resource_functions(r300);
433
434 r300->vbuf_mgr = u_vbuf_mgr_create(&r300->context, 1024 * 1024, 16,
435 PIPE_BIND_VERTEX_BUFFER |
436 PIPE_BIND_INDEX_BUFFER,
437 U_VERTEX_FETCH_DWORD_ALIGNED);
438 if (!r300->vbuf_mgr)
439 goto fail;
440
441 r300->blitter = util_blitter_create(&r300->context);
442 if (r300->blitter == NULL)
443 goto fail;
444
445 /* Render functions must be initialized after blitter. */
446 r300_init_render_functions(r300);
447
448 rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
449
450 /* setup hyper-z mm */
451 if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
452 if (!r300_hyperz_init_mm(r300))
453 goto fail;
454
455 r300_init_states(&r300->context);
456
457 /* The KIL opcode needs the first texture unit to be enabled
458 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
459 * dummy texture there. */
460 if (!r300->screen->caps.is_r500) {
461 struct pipe_resource *tex;
462 struct pipe_resource rtempl = {{0}};
463 struct pipe_sampler_view vtempl = {{0}};
464
465 rtempl.target = PIPE_TEXTURE_2D;
466 rtempl.format = PIPE_FORMAT_I8_UNORM;
467 rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
468 rtempl.usage = PIPE_USAGE_IMMUTABLE;
469 rtempl.width0 = 1;
470 rtempl.height0 = 1;
471 rtempl.depth0 = 1;
472 tex = screen->resource_create(screen, &rtempl);
473
474 u_sampler_view_default_template(&vtempl, tex, tex->format);
475
476 r300->texkill_sampler = (struct r300_sampler_view*)
477 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
478
479 pipe_resource_reference(&tex, NULL);
480 }
481
482 {
483 struct pipe_resource vb = {};
484 vb.target = PIPE_BUFFER;
485 vb.format = PIPE_FORMAT_R8_UNORM;
486 vb.bind = PIPE_BIND_VERTEX_BUFFER;
487 vb.usage = PIPE_USAGE_IMMUTABLE;
488 vb.width0 = sizeof(float) * 16;
489 vb.height0 = 1;
490 vb.depth0 = 1;
491
492 r300->dummy_vb = screen->resource_create(screen, &vb);
493 }
494
495 {
496 struct pipe_depth_stencil_alpha_state dsa = {};
497 dsa.depth.writemask = 1;
498
499 r300->dsa_decompress_zmask =
500 r300->context.create_depth_stencil_alpha_state(&r300->context,
501 &dsa);
502 }
503
504 /* Print driver info. */
505 #ifdef NDEBUG
506 if (DBG_ON(r300, DBG_INFO)) {
507 #else
508 {
509 #endif
510 fprintf(stderr,
511 "r300: DRM version: %d.%d.%d, Name: %s, ID: 0x%04x, GB: %d, Z: %d\n"
512 "r300: GART size: %d MB, VRAM size: %d MB\n"
513 "r300: AA compression: %s, Z compression: %s, HiZ: %s\n",
514 rws->get_value(rws, R300_VID_DRM_MAJOR),
515 rws->get_value(rws, R300_VID_DRM_MINOR),
516 rws->get_value(rws, R300_VID_DRM_PATCHLEVEL),
517 screen->get_name(screen),
518 rws->get_value(rws, R300_VID_PCI_ID),
519 rws->get_value(rws, R300_VID_GB_PIPES),
520 rws->get_value(rws, R300_VID_Z_PIPES),
521 rws->get_value(rws, R300_VID_GART_SIZE) >> 20,
522 rws->get_value(rws, R300_VID_VRAM_SIZE) >> 20,
523 rws->get_value(rws, R300_CAN_AACOMPRESS) ? "YES" : "NO",
524 rws->get_value(rws, R300_CAN_HYPERZ) ? "YES" : "NO",
525 rws->get_value(rws, R300_CAN_HYPERZ) &&
526 r300->screen->caps.hiz_ram ? "YES" : "NO");
527 }
528
529 return &r300->context;
530
531 fail:
532 r300_destroy_context(&r300->context);
533 return NULL;
534 }