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 static void r300_update_num_contexts(struct r300_screen *r300screen,
39 int diff)
40 {
41 pipe_mutex_lock(r300screen->num_contexts_mutex);
42 if (diff > 0) {
43 r300screen->num_contexts++;
44
45 if (r300screen->num_contexts > 1)
46 util_slab_set_thread_safety(&r300screen->pool_buffers,
47 UTIL_SLAB_MULTITHREADED);
48 } else {
49 r300screen->num_contexts--;
50
51 if (r300screen->num_contexts <= 1)
52 util_slab_set_thread_safety(&r300screen->pool_buffers,
53 UTIL_SLAB_SINGLETHREADED);
54 }
55 pipe_mutex_unlock(r300screen->num_contexts_mutex);
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 dummy VBO. */
83 pipe_resource_reference(&r300->dummy_vb, NULL);
84
85 /* The SWTCL VBO. */
86 pipe_resource_reference(&r300->vbo, NULL);
87
88 /* If there are any queries pending or not destroyed, remove them now. */
89 foreach_s(query, temp, &r300->query_list) {
90 remove_from_list(query);
91 FREE(query);
92 }
93
94 r300->context.delete_depth_stencil_alpha_state(&r300->context,
95 r300->dsa_decompress_zmask);
96 }
97
98 static void r300_destroy_context(struct pipe_context* context)
99 {
100 struct r300_context* r300 = r300_context(context);
101
102 if (r300->blitter)
103 util_blitter_destroy(r300->blitter);
104 if (r300->draw)
105 draw_destroy(r300->draw);
106
107 if (r300->vbuf_mgr)
108 u_vbuf_mgr_destroy(r300->vbuf_mgr);
109
110 /* XXX: This function assumes r300->query_list was initialized */
111 r300_release_referenced_objects(r300);
112
113 if (r300->cs)
114 r300->rws->cs_destroy(r300->cs);
115
116 /* XXX: No way to tell if this was initialized or not? */
117 util_slab_destroy(&r300->pool_transfers);
118
119 r300_update_num_contexts(r300->screen, -1);
120
121 /* Free the structs allocated in r300_setup_atoms() */
122 if (r300->aa_state.state) {
123 FREE(r300->aa_state.state);
124 FREE(r300->blend_color_state.state);
125 FREE(r300->clip_state.state);
126 FREE(r300->fb_state.state);
127 FREE(r300->gpu_flush.state);
128 FREE(r300->hyperz_state.state);
129 FREE(r300->invariant_state.state);
130 FREE(r300->rs_block_state.state);
131 FREE(r300->scissor_state.state);
132 FREE(r300->textures_state.state);
133 FREE(r300->vap_invariant_state.state);
134 FREE(r300->viewport_state.state);
135 FREE(r300->ztop_state.state);
136 FREE(r300->fs_constants.state);
137 FREE(r300->vs_constants.state);
138 if (!r300->screen->caps.has_tcl) {
139 FREE(r300->vertex_stream_state.state);
140 }
141 }
142 FREE(r300);
143 }
144
145 void r300_flush_cb(void *data)
146 {
147 struct r300_context* const cs_context_copy = data;
148
149 cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
150 }
151
152 #define R300_INIT_ATOM(atomname, atomsize) \
153 do { \
154 r300->atomname.name = #atomname; \
155 r300->atomname.state = NULL; \
156 r300->atomname.size = atomsize; \
157 r300->atomname.emit = r300_emit_##atomname; \
158 r300->atomname.dirty = FALSE; \
159 } while (0)
160
161 #define R300_ALLOC_ATOM(atomname, statetype) \
162 do { \
163 r300->atomname.state = CALLOC_STRUCT(statetype); \
164 if (r300->atomname.state == NULL) \
165 return FALSE; \
166 } while (0)
167
168 static boolean r300_setup_atoms(struct r300_context* r300)
169 {
170 boolean is_rv350 = r300->screen->caps.is_rv350;
171 boolean is_r500 = r300->screen->caps.is_r500;
172 boolean has_tcl = r300->screen->caps.has_tcl;
173 boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
174 boolean drm_2_6_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0);
175 boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
176 boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0;
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 ? 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, 18 + (is_rv350 ? 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, 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, 5 + (drm_2_3_0 ? 3 : 0));
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 if (can_hyperz) {
228 /* HiZ Clear */
229 if (has_hiz_ram)
230 R300_INIT_ATOM(hiz_clear, 0);
231 /* zmask clear */
232 R300_INIT_ATOM(zmask_clear, 4);
233 }
234 /* ZB (unpipelined), SU. */
235 R300_INIT_ATOM(query_start, 4);
236
237 /* Replace emission functions for r500. */
238 if (is_r500) {
239 r300->fs.emit = r500_emit_fs;
240 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
241 r300->fs_constants.emit = r500_emit_fs_constants;
242 }
243
244 /* Some non-CSO atoms need explicit space to store the state locally. */
245 R300_ALLOC_ATOM(aa_state, r300_aa_state);
246 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
247 R300_ALLOC_ATOM(clip_state, r300_clip_state);
248 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
249 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
250 R300_ALLOC_ATOM(textures_state, r300_textures_state);
251 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
252 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
253 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
254 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
255 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
256 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
257 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
258 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
259 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
260 if (!r300->screen->caps.has_tcl) {
261 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
262 }
263
264 /* Some non-CSO atoms don't use the state pointer. */
265 r300->fb_state_pipelined.allow_null_state = TRUE;
266 r300->fs_rc_constant_state.allow_null_state = TRUE;
267 r300->pvs_flush.allow_null_state = TRUE;
268 r300->query_start.allow_null_state = TRUE;
269 r300->texture_cache_inval.allow_null_state = TRUE;
270
271 /* Some states must be marked as dirty here to properly set up
272 * hardware in the first command stream. */
273 r300_mark_atom_dirty(r300, &r300->invariant_state);
274 r300_mark_atom_dirty(r300, &r300->pvs_flush);
275 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
276 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
277 r300_mark_atom_dirty(r300, &r300->textures_state);
278
279 return TRUE;
280 }
281
282 /* Not every state tracker calls every driver function before the first draw
283 * call and we must initialize the command buffers somehow. */
284 static void r300_init_states(struct pipe_context *pipe)
285 {
286 struct r300_context *r300 = r300_context(pipe);
287 struct pipe_blend_color bc = {{0}};
288 struct pipe_clip_state cs = {{{0}}};
289 struct pipe_scissor_state ss = {0};
290 struct r300_clip_state *clip =
291 (struct r300_clip_state*)r300->clip_state.state;
292 struct r300_gpu_flush *gpuflush =
293 (struct r300_gpu_flush*)r300->gpu_flush.state;
294 struct r300_vap_invariant_state *vap_invariant =
295 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
296 struct r300_invariant_state *invariant =
297 (struct r300_invariant_state*)r300->invariant_state.state;
298
299 CB_LOCALS;
300
301 pipe->set_blend_color(pipe, &bc);
302 pipe->set_scissor_state(pipe, &ss);
303
304 /* Initialize the clip state. */
305 if (r300->screen->caps.has_tcl) {
306 pipe->set_clip_state(pipe, &cs);
307 } else {
308 BEGIN_CB(clip->cb, 2);
309 OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
310 END_CB;
311 }
312
313 /* Initialize the GPU flush. */
314 {
315 BEGIN_CB(gpuflush->cb_flush_clean, 6);
316
317 /* Flush and free renderbuffer caches. */
318 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
319 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
320 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
321 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
322 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
323 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
324
325 /* Wait until the GPU is idle.
326 * This fixes random pixels sometimes appearing probably caused
327 * by incomplete rendering. */
328 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
329 END_CB;
330 }
331
332 /* Initialize the VAP invariant state. */
333 {
334 BEGIN_CB(vap_invariant->cb, 9);
335 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
336 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
337 OUT_CB_32F(1.0);
338 OUT_CB_32F(1.0);
339 OUT_CB_32F(1.0);
340 OUT_CB_32F(1.0);
341 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
342 END_CB;
343 }
344
345 /* Initialize the invariant state. */
346 {
347 BEGIN_CB(invariant->cb, r300->invariant_state.size);
348 OUT_CB_REG(R300_GB_SELECT, 0);
349 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
350 OUT_CB_REG(R300_GA_ROUND_MODE, 1);
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 END_CB;
363 }
364
365 /* Initialize the hyperz state. */
366 {
367 struct r300_hyperz_state *hyperz =
368 (struct r300_hyperz_state*)r300->hyperz_state.state;
369 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
370 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
371 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
372 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
373 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
374 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
375
376 if (r300->screen->caps.is_r500 ||
377 (r300->screen->caps.is_rv350 &&
378 r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) {
379 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
380 }
381 END_CB;
382 }
383 }
384
385 struct pipe_context* r300_create_context(struct pipe_screen* screen,
386 void *priv)
387 {
388 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
389 struct r300_screen* r300screen = r300_screen(screen);
390 struct r300_winsys_screen *rws = r300screen->rws;
391
392 if (!r300)
393 return NULL;
394
395 r300_update_num_contexts(r300screen, 1);
396
397 r300->rws = rws;
398 r300->screen = r300screen;
399
400 r300->context.winsys = (struct pipe_winsys*)rws;
401 r300->context.screen = screen;
402 r300->context.priv = priv;
403
404 r300->context.destroy = r300_destroy_context;
405
406 make_empty_list(&r300->query_list);
407
408 util_slab_create(&r300->pool_transfers,
409 sizeof(struct pipe_transfer), 64,
410 UTIL_SLAB_SINGLETHREADED);
411
412 r300->cs = rws->cs_create(rws);
413 if (r300->cs == NULL)
414 goto fail;
415
416 if (!r300screen->caps.has_tcl) {
417 /* Create a Draw. This is used for SW TCL. */
418 r300->draw = draw_create(&r300->context);
419 if (r300->draw == NULL)
420 goto fail;
421 /* Enable our renderer. */
422 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
423 /* Disable converting points/lines to triangles. */
424 draw_wide_line_threshold(r300->draw, 10000000.f);
425 draw_wide_point_threshold(r300->draw, 10000000.f);
426 }
427
428 if (!r300_setup_atoms(r300))
429 goto fail;
430
431 r300_init_blit_functions(r300);
432 r300_init_flush_functions(r300);
433 r300_init_query_functions(r300);
434 r300_init_state_functions(r300);
435 r300_init_resource_functions(r300);
436
437 r300->vbuf_mgr = u_vbuf_mgr_create(&r300->context, 1024 * 1024, 16,
438 PIPE_BIND_VERTEX_BUFFER |
439 PIPE_BIND_INDEX_BUFFER,
440 U_VERTEX_FETCH_DWORD_ALIGNED);
441 if (!r300->vbuf_mgr)
442 goto fail;
443
444 r300->blitter = util_blitter_create(&r300->context);
445 if (r300->blitter == NULL)
446 goto fail;
447
448 /* Render functions must be initialized after blitter. */
449 r300_init_render_functions(r300);
450
451 rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
452
453 /* setup hyper-z mm */
454 if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
455 if (!r300_hyperz_init_mm(r300))
456 goto fail;
457
458 r300_init_states(&r300->context);
459
460 /* The KIL opcode needs the first texture unit to be enabled
461 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
462 * dummy texture there. */
463 if (!r300->screen->caps.is_r500) {
464 struct pipe_resource *tex;
465 struct pipe_resource rtempl = {{0}};
466 struct pipe_sampler_view vtempl = {{0}};
467
468 rtempl.target = PIPE_TEXTURE_2D;
469 rtempl.format = PIPE_FORMAT_I8_UNORM;
470 rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
471 rtempl.usage = PIPE_USAGE_IMMUTABLE;
472 rtempl.width0 = 1;
473 rtempl.height0 = 1;
474 rtempl.depth0 = 1;
475 tex = screen->resource_create(screen, &rtempl);
476
477 u_sampler_view_default_template(&vtempl, tex, tex->format);
478
479 r300->texkill_sampler = (struct r300_sampler_view*)
480 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
481
482 pipe_resource_reference(&tex, NULL);
483 }
484
485 {
486 struct pipe_resource vb;
487 memset(&vb, 0, sizeof(vb));
488 vb.target = PIPE_BUFFER;
489 vb.format = PIPE_FORMAT_R8_UNORM;
490 vb.bind = PIPE_BIND_VERTEX_BUFFER;
491 vb.usage = PIPE_USAGE_IMMUTABLE;
492 vb.width0 = sizeof(float) * 16;
493 vb.height0 = 1;
494 vb.depth0 = 1;
495
496 r300->dummy_vb = screen->resource_create(screen, &vb);
497 }
498
499 {
500 struct pipe_depth_stencil_alpha_state dsa;
501 memset(&dsa, 0, sizeof(dsa));
502 dsa.depth.writemask = 1;
503
504 r300->dsa_decompress_zmask =
505 r300->context.create_depth_stencil_alpha_state(&r300->context,
506 &dsa);
507 }
508
509 /* Print driver info. */
510 #ifdef NDEBUG
511 if (DBG_ON(r300, DBG_INFO)) {
512 #else
513 {
514 #endif
515 fprintf(stderr,
516 "r300: DRM version: %d.%d.%d, Name: %s, ID: 0x%04x, GB: %d, Z: %d\n"
517 "r300: GART size: %d MB, VRAM size: %d MB\n"
518 "r300: AA compression: %s, Z compression: %s, HiZ: %s\n",
519 rws->get_value(rws, R300_VID_DRM_MAJOR),
520 rws->get_value(rws, R300_VID_DRM_MINOR),
521 rws->get_value(rws, R300_VID_DRM_PATCHLEVEL),
522 screen->get_name(screen),
523 rws->get_value(rws, R300_VID_PCI_ID),
524 rws->get_value(rws, R300_VID_GB_PIPES),
525 rws->get_value(rws, R300_VID_Z_PIPES),
526 rws->get_value(rws, R300_VID_GART_SIZE) >> 20,
527 rws->get_value(rws, R300_VID_VRAM_SIZE) >> 20,
528 rws->get_value(rws, R300_CAN_AACOMPRESS) ? "YES" : "NO",
529 rws->get_value(rws, R300_CAN_HYPERZ) ? "YES" : "NO",
530 rws->get_value(rws, R300_CAN_HYPERZ) &&
531 r300->screen->caps.hiz_ram ? "YES" : "NO");
532 }
533
534 return &r300->context;
535
536 fail:
537 r300_destroy_context(&r300->context);
538 return NULL;
539 }