u_vbuf: make use of the new CAPs to determine what to do
[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 if (r300->vbuf_mgr)
104 u_vbuf_destroy(r300->vbuf_mgr);
105
106 /* XXX: This function assumes r300->query_list was initialized */
107 r300_release_referenced_objects(r300);
108
109 if (r300->cs)
110 r300->rws->cs_destroy(r300->cs);
111
112 /* XXX: No way to tell if this was initialized or not? */
113 util_slab_destroy(&r300->pool_transfers);
114
115 r300_update_num_contexts(r300->screen, -1);
116
117 /* Free the structs allocated in r300_setup_atoms() */
118 if (r300->aa_state.state) {
119 FREE(r300->aa_state.state);
120 FREE(r300->blend_color_state.state);
121 FREE(r300->clip_state.state);
122 FREE(r300->fb_state.state);
123 FREE(r300->gpu_flush.state);
124 FREE(r300->hyperz_state.state);
125 FREE(r300->invariant_state.state);
126 FREE(r300->rs_block_state.state);
127 FREE(r300->scissor_state.state);
128 FREE(r300->textures_state.state);
129 FREE(r300->vap_invariant_state.state);
130 FREE(r300->viewport_state.state);
131 FREE(r300->ztop_state.state);
132 FREE(r300->fs_constants.state);
133 FREE(r300->vs_constants.state);
134 if (!r300->screen->caps.has_tcl) {
135 FREE(r300->vertex_stream_state.state);
136 }
137 }
138 FREE(r300);
139 }
140
141 static void r300_flush_callback(void *data, unsigned flags)
142 {
143 struct r300_context* const cs_context_copy = data;
144
145 r300_flush(&cs_context_copy->context, flags, NULL);
146 }
147
148 #define R300_INIT_ATOM(atomname, atomsize) \
149 do { \
150 r300->atomname.name = #atomname; \
151 r300->atomname.state = NULL; \
152 r300->atomname.size = atomsize; \
153 r300->atomname.emit = r300_emit_##atomname; \
154 r300->atomname.dirty = FALSE; \
155 } while (0)
156
157 #define R300_ALLOC_ATOM(atomname, statetype) \
158 do { \
159 r300->atomname.state = CALLOC_STRUCT(statetype); \
160 if (r300->atomname.state == NULL) \
161 return FALSE; \
162 } while (0)
163
164 static boolean r300_setup_atoms(struct r300_context* r300)
165 {
166 boolean is_rv350 = r300->screen->caps.is_rv350;
167 boolean is_r500 = r300->screen->caps.is_r500;
168 boolean has_tcl = r300->screen->caps.has_tcl;
169 boolean drm_2_6_0 = r300->screen->info.drm_minor >= 6;
170
171 /* Create the actual atom list.
172 *
173 * Some atoms never change size, others change every emit - those have
174 * the size of 0 here.
175 *
176 * NOTE: The framebuffer state is split into these atoms:
177 * - gpu_flush (unpipelined regs)
178 * - aa_state (unpipelined regs)
179 * - fb_state (unpipelined regs)
180 * - hyperz_state (unpipelined regs followed by pipelined ones)
181 * - fb_state_pipelined (pipelined regs)
182 * The motivation behind this is to be able to emit a strict
183 * subset of the regs, and to have reasonable register ordering. */
184 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
185 R300_INIT_ATOM(gpu_flush, 9);
186 R300_INIT_ATOM(aa_state, 4);
187 R300_INIT_ATOM(fb_state, 0);
188 R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
189 /* ZB (unpipelined), SC. */
190 R300_INIT_ATOM(ztop_state, 2);
191 /* ZB, FG. */
192 R300_INIT_ATOM(dsa_state, is_r500 ? (drm_2_6_0 ? 10 : 8) : 6);
193 /* RB3D. */
194 R300_INIT_ATOM(blend_state, 8);
195 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
196 /* SC. */
197 R300_INIT_ATOM(scissor_state, 3);
198 /* GB, FG, GA, SU, SC, RB3D. */
199 R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0) + (is_r500 ? 4 : 0));
200 /* VAP. */
201 R300_INIT_ATOM(viewport_state, 9);
202 R300_INIT_ATOM(pvs_flush, 2);
203 R300_INIT_ATOM(vap_invariant_state, is_r500 ? 11 : 9);
204 R300_INIT_ATOM(vertex_stream_state, 0);
205 R300_INIT_ATOM(vs_state, 0);
206 R300_INIT_ATOM(vs_constants, 0);
207 R300_INIT_ATOM(clip_state, has_tcl ? 3 + (6 * 4) : 0);
208 /* VAP, RS, GA, GB, SU, SC. */
209 R300_INIT_ATOM(rs_block_state, 0);
210 R300_INIT_ATOM(rs_state, 0);
211 /* SC, US. */
212 R300_INIT_ATOM(fb_state_pipelined, 8);
213 /* US. */
214 R300_INIT_ATOM(fs, 0);
215 R300_INIT_ATOM(fs_rc_constant_state, 0);
216 R300_INIT_ATOM(fs_constants, 0);
217 /* TX. */
218 R300_INIT_ATOM(texture_cache_inval, 2);
219 R300_INIT_ATOM(textures_state, 0);
220 /* HiZ Clear */
221 R300_INIT_ATOM(hiz_clear, r300->screen->caps.hiz_ram > 0 ? 4 : 0);
222 /* zmask clear */
223 R300_INIT_ATOM(zmask_clear, r300->screen->caps.zmask_ram > 0 ? 4 : 0);
224 /* ZB (unpipelined), SU. */
225 R300_INIT_ATOM(query_start, 4);
226
227 /* Replace emission functions for r500. */
228 if (is_r500) {
229 r300->fs.emit = r500_emit_fs;
230 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
231 r300->fs_constants.emit = r500_emit_fs_constants;
232 }
233
234 /* Some non-CSO atoms need explicit space to store the state locally. */
235 R300_ALLOC_ATOM(aa_state, r300_aa_state);
236 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
237 R300_ALLOC_ATOM(clip_state, r300_clip_state);
238 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
239 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
240 R300_ALLOC_ATOM(textures_state, r300_textures_state);
241 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
242 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
243 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
244 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
245 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
246 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
247 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
248 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
249 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
250 if (!r300->screen->caps.has_tcl) {
251 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
252 }
253
254 /* Some non-CSO atoms don't use the state pointer. */
255 r300->fb_state_pipelined.allow_null_state = TRUE;
256 r300->fs_rc_constant_state.allow_null_state = TRUE;
257 r300->pvs_flush.allow_null_state = TRUE;
258 r300->query_start.allow_null_state = TRUE;
259 r300->texture_cache_inval.allow_null_state = TRUE;
260
261 /* Some states must be marked as dirty here to properly set up
262 * hardware in the first command stream. */
263 r300_mark_atom_dirty(r300, &r300->invariant_state);
264 r300_mark_atom_dirty(r300, &r300->pvs_flush);
265 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
266 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
267 r300_mark_atom_dirty(r300, &r300->textures_state);
268
269 return TRUE;
270 }
271
272 /* Not every state tracker calls every driver function before the first draw
273 * call and we must initialize the command buffers somehow. */
274 static void r300_init_states(struct pipe_context *pipe)
275 {
276 struct r300_context *r300 = r300_context(pipe);
277 struct pipe_blend_color bc = {{0}};
278 struct pipe_clip_state cs = {{{0}}};
279 struct pipe_scissor_state ss = {0};
280 struct r300_gpu_flush *gpuflush =
281 (struct r300_gpu_flush*)r300->gpu_flush.state;
282 struct r300_vap_invariant_state *vap_invariant =
283 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
284 struct r300_invariant_state *invariant =
285 (struct r300_invariant_state*)r300->invariant_state.state;
286
287 CB_LOCALS;
288
289 pipe->set_blend_color(pipe, &bc);
290 pipe->set_clip_state(pipe, &cs);
291 pipe->set_scissor_state(pipe, &ss);
292
293 /* Initialize the GPU flush. */
294 {
295 BEGIN_CB(gpuflush->cb_flush_clean, 6);
296
297 /* Flush and free renderbuffer caches. */
298 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
299 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
300 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
301 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
302 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
303 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
304
305 /* Wait until the GPU is idle.
306 * This fixes random pixels sometimes appearing probably caused
307 * by incomplete rendering. */
308 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
309 END_CB;
310 }
311
312 /* Initialize the VAP invariant state. */
313 {
314 BEGIN_CB(vap_invariant->cb, r300->vap_invariant_state.size);
315 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
316 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
317 OUT_CB_32F(1.0);
318 OUT_CB_32F(1.0);
319 OUT_CB_32F(1.0);
320 OUT_CB_32F(1.0);
321 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
322
323 if (r300->screen->caps.is_r500) {
324 OUT_CB_REG(R500_VAP_TEX_TO_COLOR_CNTL, 0);
325 }
326 END_CB;
327 }
328
329 /* Initialize the invariant state. */
330 {
331 BEGIN_CB(invariant->cb, r300->invariant_state.size);
332 OUT_CB_REG(R300_GB_SELECT, 0);
333 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
334 OUT_CB_REG(R300_GA_OFFSET, 0);
335 OUT_CB_REG(R300_SU_TEX_WRAP, 0);
336 OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
337 OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
338 OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
339 OUT_CB_REG(R300_SC_SCREENDOOR, 0xffffff);
340
341 if (r300->screen->caps.is_rv350) {
342 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
343 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
344 }
345
346 if (r300->screen->caps.is_r500) {
347 OUT_CB_REG(R500_GA_COLOR_CONTROL_PS3, 0);
348 OUT_CB_REG(R500_SU_TEX_WRAP_PS3, 0);
349 }
350 END_CB;
351 }
352
353 /* Initialize the hyperz state. */
354 {
355 struct r300_hyperz_state *hyperz =
356 (struct r300_hyperz_state*)r300->hyperz_state.state;
357 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
358 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
359 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
360 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
361 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
362 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
363
364 if (r300->screen->caps.is_r500 ||
365 (r300->screen->caps.is_rv350 &&
366 r300->screen->info.drm_minor >= 6)) {
367 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
368 }
369 END_CB;
370 }
371 }
372
373 struct pipe_context* r300_create_context(struct pipe_screen* screen,
374 void *priv)
375 {
376 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
377 struct r300_screen* r300screen = r300_screen(screen);
378 struct radeon_winsys *rws = r300screen->rws;
379
380 if (!r300)
381 return NULL;
382
383 r300_update_num_contexts(r300screen, 1);
384
385 r300->rws = rws;
386 r300->screen = r300screen;
387
388 r300->context.screen = screen;
389 r300->context.priv = priv;
390
391 r300->context.destroy = r300_destroy_context;
392
393 util_slab_create(&r300->pool_transfers,
394 sizeof(struct pipe_transfer), 64,
395 UTIL_SLAB_SINGLETHREADED);
396
397 r300->cs = rws->cs_create(rws);
398 if (r300->cs == NULL)
399 goto fail;
400
401 if (!r300screen->caps.has_tcl) {
402 /* Create a Draw. This is used for SW TCL. */
403 r300->draw = draw_create(&r300->context);
404 if (r300->draw == NULL)
405 goto fail;
406 /* Enable our renderer. */
407 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
408 /* Disable converting points/lines to triangles. */
409 draw_wide_line_threshold(r300->draw, 10000000.f);
410 draw_wide_point_threshold(r300->draw, 10000000.f);
411 draw_wide_point_sprites(r300->draw, FALSE);
412 draw_enable_line_stipple(r300->draw, TRUE);
413 draw_enable_point_sprites(r300->draw, FALSE);
414 }
415
416 if (!r300_setup_atoms(r300))
417 goto fail;
418
419 r300_init_blit_functions(r300);
420 r300_init_flush_functions(r300);
421 r300_init_query_functions(r300);
422 r300_init_state_functions(r300);
423 r300_init_resource_functions(r300);
424 r300_init_render_functions(r300);
425 r300_init_states(&r300->context);
426
427 r300->context.create_video_decoder = vl_create_decoder;
428 r300->context.create_video_buffer = vl_video_buffer_create;
429
430 if (r300->screen->caps.has_tcl) {
431 struct u_vbuf_caps caps;
432
433 u_vbuf_get_caps(screen, &caps);
434 caps.format_fixed32 = 0;
435
436 r300->vbuf_mgr = u_vbuf_create(&r300->context, &caps, 1024 * 1024, 4,
437 PIPE_BIND_VERTEX_BUFFER |
438 PIPE_BIND_INDEX_BUFFER);
439 if (!r300->vbuf_mgr)
440 goto fail;
441 }
442
443 r300->blitter = util_blitter_create(&r300->context);
444 if (r300->blitter == NULL)
445 goto fail;
446 r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
447
448 rws->cs_set_flush_callback(r300->cs, r300_flush_callback, r300);
449
450 /* The KIL opcode needs the first texture unit to be enabled
451 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
452 * dummy texture there. */
453 if (!r300->screen->caps.is_r500) {
454 struct pipe_resource *tex;
455 struct pipe_resource rtempl = {{0}};
456 struct pipe_sampler_view vtempl = {{0}};
457
458 rtempl.target = PIPE_TEXTURE_2D;
459 rtempl.format = PIPE_FORMAT_I8_UNORM;
460 rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
461 rtempl.usage = PIPE_USAGE_IMMUTABLE;
462 rtempl.width0 = 1;
463 rtempl.height0 = 1;
464 rtempl.depth0 = 1;
465 tex = screen->resource_create(screen, &rtempl);
466
467 u_sampler_view_default_template(&vtempl, tex, tex->format);
468
469 r300->texkill_sampler = (struct r300_sampler_view*)
470 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
471
472 pipe_resource_reference(&tex, NULL);
473 }
474
475 {
476 struct pipe_resource vb;
477 memset(&vb, 0, sizeof(vb));
478 vb.target = PIPE_BUFFER;
479 vb.format = PIPE_FORMAT_R8_UNORM;
480 vb.bind = PIPE_BIND_VERTEX_BUFFER;
481 vb.usage = PIPE_USAGE_IMMUTABLE;
482 vb.width0 = sizeof(float) * 16;
483 vb.height0 = 1;
484 vb.depth0 = 1;
485
486 r300->dummy_vb.buffer = screen->resource_create(screen, &vb);
487 }
488
489 {
490 struct pipe_depth_stencil_alpha_state dsa;
491 memset(&dsa, 0, sizeof(dsa));
492 dsa.depth.writemask = 1;
493
494 r300->dsa_decompress_zmask =
495 r300->context.create_depth_stencil_alpha_state(&r300->context,
496 &dsa);
497 }
498
499 r300->hyperz_time_of_last_flush = os_time_get();
500
501 /* Print driver info. */
502 #ifdef DEBUG
503 {
504 #else
505 if (DBG_ON(r300, DBG_INFO)) {
506 #endif
507 fprintf(stderr,
508 "r300: DRM version: %d.%d.%d, Name: %s, ID: 0x%04x, GB: %d, Z: %d\n"
509 "r300: GART size: %d MB, VRAM size: %d MB\n"
510 "r300: AA compression RAM: %s, Z compression RAM: %s, HiZ RAM: %s\n",
511 r300->screen->info.drm_major,
512 r300->screen->info.drm_minor,
513 r300->screen->info.drm_patchlevel,
514 screen->get_name(screen),
515 r300->screen->info.pci_id,
516 r300->screen->info.r300_num_gb_pipes,
517 r300->screen->info.r300_num_z_pipes,
518 r300->screen->info.gart_size >> 20,
519 r300->screen->info.vram_size >> 20,
520 "YES", /* XXX really? */
521 r300->screen->caps.zmask_ram ? "YES" : "NO",
522 r300->screen->caps.hiz_ram ? "YES" : "NO");
523 }
524
525 return &r300->context;
526
527 fail:
528 r300_destroy_context(&r300->context);
529 return NULL;
530 }