gallium: make user vertex buffers optional
[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 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 /* Manually-created vertex buffers. */
83 pipe_resource_reference(&r300->dummy_vb.buffer, NULL);
84 pipe_resource_reference(&r300->vbo, NULL);
85
86 r300->context.delete_depth_stencil_alpha_state(&r300->context,
87 r300->dsa_decompress_zmask);
88 }
89
90 static void r300_destroy_context(struct pipe_context* context)
91 {
92 struct r300_context* r300 = r300_context(context);
93
94 if (r300->cs && r300->hyperz_enabled) {
95 r300->rws->cs_request_feature(r300->cs, RADEON_FID_R300_HYPERZ_ACCESS, FALSE);
96 }
97
98 if (r300->blitter)
99 util_blitter_destroy(r300->blitter);
100 if (r300->draw)
101 draw_destroy(r300->draw);
102
103 u_upload_destroy(r300->uploader);
104
105 /* XXX: This function assumes r300->query_list was initialized */
106 r300_release_referenced_objects(r300);
107
108 if (r300->cs)
109 r300->rws->cs_destroy(r300->cs);
110
111 /* XXX: No way to tell if this was initialized or not? */
112 util_slab_destroy(&r300->pool_transfers);
113
114 r300_update_num_contexts(r300->screen, -1);
115
116 /* Free the structs allocated in r300_setup_atoms() */
117 if (r300->aa_state.state) {
118 FREE(r300->aa_state.state);
119 FREE(r300->blend_color_state.state);
120 FREE(r300->clip_state.state);
121 FREE(r300->fb_state.state);
122 FREE(r300->gpu_flush.state);
123 FREE(r300->hyperz_state.state);
124 FREE(r300->invariant_state.state);
125 FREE(r300->rs_block_state.state);
126 FREE(r300->scissor_state.state);
127 FREE(r300->textures_state.state);
128 FREE(r300->vap_invariant_state.state);
129 FREE(r300->viewport_state.state);
130 FREE(r300->ztop_state.state);
131 FREE(r300->fs_constants.state);
132 FREE(r300->vs_constants.state);
133 if (!r300->screen->caps.has_tcl) {
134 FREE(r300->vertex_stream_state.state);
135 }
136 }
137 FREE(r300);
138 }
139
140 static void r300_flush_callback(void *data, unsigned flags)
141 {
142 struct r300_context* const cs_context_copy = data;
143
144 r300_flush(&cs_context_copy->context, flags, NULL);
145 }
146
147 #define R300_INIT_ATOM(atomname, atomsize) \
148 do { \
149 r300->atomname.name = #atomname; \
150 r300->atomname.state = NULL; \
151 r300->atomname.size = atomsize; \
152 r300->atomname.emit = r300_emit_##atomname; \
153 r300->atomname.dirty = FALSE; \
154 } while (0)
155
156 #define R300_ALLOC_ATOM(atomname, statetype) \
157 do { \
158 r300->atomname.state = CALLOC_STRUCT(statetype); \
159 if (r300->atomname.state == NULL) \
160 return FALSE; \
161 } while (0)
162
163 static boolean r300_setup_atoms(struct r300_context* r300)
164 {
165 boolean is_rv350 = r300->screen->caps.is_rv350;
166 boolean is_r500 = r300->screen->caps.is_r500;
167 boolean has_tcl = r300->screen->caps.has_tcl;
168 boolean drm_2_6_0 = r300->screen->info.drm_minor >= 6;
169
170 /* Create the actual atom list.
171 *
172 * Some atoms never change size, others change every emit - those have
173 * the size of 0 here.
174 *
175 * NOTE: The framebuffer state is split into these atoms:
176 * - gpu_flush (unpipelined regs)
177 * - aa_state (unpipelined regs)
178 * - fb_state (unpipelined regs)
179 * - hyperz_state (unpipelined regs followed by pipelined ones)
180 * - fb_state_pipelined (pipelined regs)
181 * The motivation behind this is to be able to emit a strict
182 * subset of the regs, and to have reasonable register ordering. */
183 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
184 R300_INIT_ATOM(gpu_flush, 9);
185 R300_INIT_ATOM(aa_state, 4);
186 R300_INIT_ATOM(fb_state, 0);
187 R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
188 /* ZB (unpipelined), SC. */
189 R300_INIT_ATOM(ztop_state, 2);
190 /* ZB, FG. */
191 R300_INIT_ATOM(dsa_state, is_r500 ? (drm_2_6_0 ? 10 : 8) : 6);
192 /* RB3D. */
193 R300_INIT_ATOM(blend_state, 8);
194 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
195 /* SC. */
196 R300_INIT_ATOM(scissor_state, 3);
197 /* GB, FG, GA, SU, SC, RB3D. */
198 R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0) + (is_r500 ? 4 : 0));
199 /* VAP. */
200 R300_INIT_ATOM(viewport_state, 9);
201 R300_INIT_ATOM(pvs_flush, 2);
202 R300_INIT_ATOM(vap_invariant_state, is_r500 ? 11 : 9);
203 R300_INIT_ATOM(vertex_stream_state, 0);
204 R300_INIT_ATOM(vs_state, 0);
205 R300_INIT_ATOM(vs_constants, 0);
206 R300_INIT_ATOM(clip_state, has_tcl ? 3 + (6 * 4) : 0);
207 /* VAP, RS, GA, GB, SU, SC. */
208 R300_INIT_ATOM(rs_block_state, 0);
209 R300_INIT_ATOM(rs_state, 0);
210 /* SC, US. */
211 R300_INIT_ATOM(fb_state_pipelined, 8);
212 /* US. */
213 R300_INIT_ATOM(fs, 0);
214 R300_INIT_ATOM(fs_rc_constant_state, 0);
215 R300_INIT_ATOM(fs_constants, 0);
216 /* TX. */
217 R300_INIT_ATOM(texture_cache_inval, 2);
218 R300_INIT_ATOM(textures_state, 0);
219 /* HiZ Clear */
220 R300_INIT_ATOM(hiz_clear, r300->screen->caps.hiz_ram > 0 ? 4 : 0);
221 /* zmask clear */
222 R300_INIT_ATOM(zmask_clear, r300->screen->caps.zmask_ram > 0 ? 4 : 0);
223 /* ZB (unpipelined), SU. */
224 R300_INIT_ATOM(query_start, 4);
225
226 /* Replace emission functions for r500. */
227 if (is_r500) {
228 r300->fs.emit = r500_emit_fs;
229 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
230 r300->fs_constants.emit = r500_emit_fs_constants;
231 }
232
233 /* Some non-CSO atoms need explicit space to store the state locally. */
234 R300_ALLOC_ATOM(aa_state, r300_aa_state);
235 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
236 R300_ALLOC_ATOM(clip_state, r300_clip_state);
237 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
238 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
239 R300_ALLOC_ATOM(textures_state, r300_textures_state);
240 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
241 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
242 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
243 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
244 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
245 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
246 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
247 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
248 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
249 if (!r300->screen->caps.has_tcl) {
250 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
251 }
252
253 /* Some non-CSO atoms don't use the state pointer. */
254 r300->fb_state_pipelined.allow_null_state = TRUE;
255 r300->fs_rc_constant_state.allow_null_state = TRUE;
256 r300->pvs_flush.allow_null_state = TRUE;
257 r300->query_start.allow_null_state = TRUE;
258 r300->texture_cache_inval.allow_null_state = TRUE;
259
260 /* Some states must be marked as dirty here to properly set up
261 * hardware in the first command stream. */
262 r300_mark_atom_dirty(r300, &r300->invariant_state);
263 r300_mark_atom_dirty(r300, &r300->pvs_flush);
264 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
265 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
266 r300_mark_atom_dirty(r300, &r300->textures_state);
267
268 return TRUE;
269 }
270
271 /* Not every state tracker calls every driver function before the first draw
272 * call and we must initialize the command buffers somehow. */
273 static void r300_init_states(struct pipe_context *pipe)
274 {
275 struct r300_context *r300 = r300_context(pipe);
276 struct pipe_blend_color bc = {{0}};
277 struct pipe_clip_state cs = {{{0}}};
278 struct pipe_scissor_state ss = {0};
279 struct r300_gpu_flush *gpuflush =
280 (struct r300_gpu_flush*)r300->gpu_flush.state;
281 struct r300_vap_invariant_state *vap_invariant =
282 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
283 struct r300_invariant_state *invariant =
284 (struct r300_invariant_state*)r300->invariant_state.state;
285
286 CB_LOCALS;
287
288 pipe->set_blend_color(pipe, &bc);
289 pipe->set_clip_state(pipe, &cs);
290 pipe->set_scissor_state(pipe, &ss);
291
292 /* Initialize the GPU flush. */
293 {
294 BEGIN_CB(gpuflush->cb_flush_clean, 6);
295
296 /* Flush and free renderbuffer caches. */
297 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
298 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
299 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
300 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
301 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
302 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
303
304 /* Wait until the GPU is idle.
305 * This fixes random pixels sometimes appearing probably caused
306 * by incomplete rendering. */
307 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
308 END_CB;
309 }
310
311 /* Initialize the VAP invariant state. */
312 {
313 BEGIN_CB(vap_invariant->cb, r300->vap_invariant_state.size);
314 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
315 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
316 OUT_CB_32F(1.0);
317 OUT_CB_32F(1.0);
318 OUT_CB_32F(1.0);
319 OUT_CB_32F(1.0);
320 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
321
322 if (r300->screen->caps.is_r500) {
323 OUT_CB_REG(R500_VAP_TEX_TO_COLOR_CNTL, 0);
324 }
325 END_CB;
326 }
327
328 /* Initialize the invariant state. */
329 {
330 BEGIN_CB(invariant->cb, r300->invariant_state.size);
331 OUT_CB_REG(R300_GB_SELECT, 0);
332 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
333 OUT_CB_REG(R300_GA_OFFSET, 0);
334 OUT_CB_REG(R300_SU_TEX_WRAP, 0);
335 OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
336 OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
337 OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
338 OUT_CB_REG(R300_SC_SCREENDOOR, 0xffffff);
339
340 if (r300->screen->caps.is_rv350) {
341 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
342 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
343 }
344
345 if (r300->screen->caps.is_r500) {
346 OUT_CB_REG(R500_GA_COLOR_CONTROL_PS3, 0);
347 OUT_CB_REG(R500_SU_TEX_WRAP_PS3, 0);
348 }
349 END_CB;
350 }
351
352 /* Initialize the hyperz state. */
353 {
354 struct r300_hyperz_state *hyperz =
355 (struct r300_hyperz_state*)r300->hyperz_state.state;
356 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
357 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
358 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
359 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
360 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
361 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
362
363 if (r300->screen->caps.is_r500 ||
364 (r300->screen->caps.is_rv350 &&
365 r300->screen->info.drm_minor >= 6)) {
366 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
367 }
368 END_CB;
369 }
370 }
371
372 struct pipe_context* r300_create_context(struct pipe_screen* screen,
373 void *priv)
374 {
375 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
376 struct r300_screen* r300screen = r300_screen(screen);
377 struct radeon_winsys *rws = r300screen->rws;
378
379 if (!r300)
380 return NULL;
381
382 r300_update_num_contexts(r300screen, 1);
383
384 r300->rws = rws;
385 r300->screen = r300screen;
386
387 r300->context.screen = screen;
388 r300->context.priv = priv;
389
390 r300->context.destroy = r300_destroy_context;
391
392 util_slab_create(&r300->pool_transfers,
393 sizeof(struct pipe_transfer), 64,
394 UTIL_SLAB_SINGLETHREADED);
395
396 r300->cs = rws->cs_create(rws);
397 if (r300->cs == NULL)
398 goto fail;
399
400 if (!r300screen->caps.has_tcl) {
401 /* Create a Draw. This is used for SW TCL. */
402 r300->draw = draw_create(&r300->context);
403 if (r300->draw == NULL)
404 goto fail;
405 /* Enable our renderer. */
406 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
407 /* Disable converting points/lines to triangles. */
408 draw_wide_line_threshold(r300->draw, 10000000.f);
409 draw_wide_point_threshold(r300->draw, 10000000.f);
410 draw_wide_point_sprites(r300->draw, FALSE);
411 draw_enable_line_stipple(r300->draw, TRUE);
412 draw_enable_point_sprites(r300->draw, FALSE);
413 }
414
415 if (!r300_setup_atoms(r300))
416 goto fail;
417
418 r300_init_blit_functions(r300);
419 r300_init_flush_functions(r300);
420 r300_init_query_functions(r300);
421 r300_init_state_functions(r300);
422 r300_init_resource_functions(r300);
423 r300_init_render_functions(r300);
424 r300_init_states(&r300->context);
425
426 r300->context.create_video_decoder = vl_create_decoder;
427 r300->context.create_video_buffer = vl_video_buffer_create;
428
429 r300->uploader = u_upload_create(&r300->context, 256 * 1024, 4,
430 PIPE_BIND_INDEX_BUFFER);
431
432 r300->blitter = util_blitter_create(&r300->context);
433 if (r300->blitter == NULL)
434 goto fail;
435 r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
436
437 rws->cs_set_flush_callback(r300->cs, r300_flush_callback, r300);
438
439 /* The KIL opcode needs the first texture unit to be enabled
440 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
441 * dummy texture there. */
442 if (!r300->screen->caps.is_r500) {
443 struct pipe_resource *tex;
444 struct pipe_resource rtempl = {{0}};
445 struct pipe_sampler_view vtempl = {{0}};
446
447 rtempl.target = PIPE_TEXTURE_2D;
448 rtempl.format = PIPE_FORMAT_I8_UNORM;
449 rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
450 rtempl.usage = PIPE_USAGE_IMMUTABLE;
451 rtempl.width0 = 1;
452 rtempl.height0 = 1;
453 rtempl.depth0 = 1;
454 tex = screen->resource_create(screen, &rtempl);
455
456 u_sampler_view_default_template(&vtempl, tex, tex->format);
457
458 r300->texkill_sampler = (struct r300_sampler_view*)
459 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
460
461 pipe_resource_reference(&tex, NULL);
462 }
463
464 {
465 struct pipe_resource vb;
466 memset(&vb, 0, sizeof(vb));
467 vb.target = PIPE_BUFFER;
468 vb.format = PIPE_FORMAT_R8_UNORM;
469 vb.bind = PIPE_BIND_VERTEX_BUFFER;
470 vb.usage = PIPE_USAGE_IMMUTABLE;
471 vb.width0 = sizeof(float) * 16;
472 vb.height0 = 1;
473 vb.depth0 = 1;
474
475 r300->dummy_vb.buffer = screen->resource_create(screen, &vb);
476 }
477
478 {
479 struct pipe_depth_stencil_alpha_state dsa;
480 memset(&dsa, 0, sizeof(dsa));
481 dsa.depth.writemask = 1;
482
483 r300->dsa_decompress_zmask =
484 r300->context.create_depth_stencil_alpha_state(&r300->context,
485 &dsa);
486 }
487
488 r300->hyperz_time_of_last_flush = os_time_get();
489
490 /* Print driver info. */
491 #ifdef DEBUG
492 {
493 #else
494 if (DBG_ON(r300, DBG_INFO)) {
495 #endif
496 fprintf(stderr,
497 "r300: DRM version: %d.%d.%d, Name: %s, ID: 0x%04x, GB: %d, Z: %d\n"
498 "r300: GART size: %d MB, VRAM size: %d MB\n"
499 "r300: AA compression RAM: %s, Z compression RAM: %s, HiZ RAM: %s\n",
500 r300->screen->info.drm_major,
501 r300->screen->info.drm_minor,
502 r300->screen->info.drm_patchlevel,
503 screen->get_name(screen),
504 r300->screen->info.pci_id,
505 r300->screen->info.r300_num_gb_pipes,
506 r300->screen->info.r300_num_z_pipes,
507 r300->screen->info.gart_size >> 20,
508 r300->screen->info.vram_size >> 20,
509 "YES", /* XXX really? */
510 r300->screen->caps.zmask_ram ? "YES" : "NO",
511 r300->screen->caps.hiz_ram ? "YES" : "NO");
512 }
513
514 return &r300->context;
515
516 fail:
517 r300_destroy_context(&r300->context);
518 return NULL;
519 }