r300: use the new parent/child pools for transfers (v2)
[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/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 #include "compiler/radeon_regalloc.h"
39
40 #include <inttypes.h>
41
42 static void r300_release_referenced_objects(struct r300_context *r300)
43 {
44 struct pipe_framebuffer_state *fb =
45 (struct pipe_framebuffer_state*)r300->fb_state.state;
46 struct r300_textures_state *textures =
47 (struct r300_textures_state*)r300->textures_state.state;
48 unsigned i;
49
50 /* Framebuffer state. */
51 util_unreference_framebuffer_state(fb);
52
53 /* Textures. */
54 for (i = 0; i < textures->sampler_view_count; i++)
55 pipe_sampler_view_reference(
56 (struct pipe_sampler_view**)&textures->sampler_views[i], NULL);
57
58 /* The special dummy texture for texkill. */
59 if (r300->texkill_sampler) {
60 pipe_sampler_view_reference(
61 (struct pipe_sampler_view**)&r300->texkill_sampler,
62 NULL);
63 }
64
65 /* Manually-created vertex buffers. */
66 pipe_resource_reference(&r300->dummy_vb.buffer, NULL);
67 pb_reference(&r300->vbo, NULL);
68
69 r300->context.delete_depth_stencil_alpha_state(&r300->context,
70 r300->dsa_decompress_zmask);
71 }
72
73 static void r300_destroy_context(struct pipe_context* context)
74 {
75 struct r300_context* r300 = r300_context(context);
76
77 if (r300->cs && r300->hyperz_enabled) {
78 r300->rws->cs_request_feature(r300->cs, RADEON_FID_R300_HYPERZ_ACCESS, FALSE);
79 }
80 if (r300->cs && r300->cmask_access) {
81 r300->rws->cs_request_feature(r300->cs, RADEON_FID_R300_CMASK_ACCESS, FALSE);
82 }
83
84 if (r300->blitter)
85 util_blitter_destroy(r300->blitter);
86 if (r300->draw)
87 draw_destroy(r300->draw);
88
89 if (r300->uploader)
90 u_upload_destroy(r300->uploader);
91
92 /* XXX: This function assumes r300->query_list was initialized */
93 r300_release_referenced_objects(r300);
94
95 if (r300->cs)
96 r300->rws->cs_destroy(r300->cs);
97 if (r300->ctx)
98 r300->rws->ctx_destroy(r300->ctx);
99
100 rc_destroy_regalloc_state(&r300->fs_regalloc_state);
101
102 /* XXX: No way to tell if this was initialized or not? */
103 slab_destroy_child(&r300->pool_transfers);
104
105 /* Free the structs allocated in r300_setup_atoms() */
106 if (r300->aa_state.state) {
107 FREE(r300->aa_state.state);
108 FREE(r300->blend_color_state.state);
109 FREE(r300->clip_state.state);
110 FREE(r300->fb_state.state);
111 FREE(r300->gpu_flush.state);
112 FREE(r300->hyperz_state.state);
113 FREE(r300->invariant_state.state);
114 FREE(r300->rs_block_state.state);
115 FREE(r300->sample_mask.state);
116 FREE(r300->scissor_state.state);
117 FREE(r300->textures_state.state);
118 FREE(r300->vap_invariant_state.state);
119 FREE(r300->viewport_state.state);
120 FREE(r300->ztop_state.state);
121 FREE(r300->fs_constants.state);
122 FREE(r300->vs_constants.state);
123 if (!r300->screen->caps.has_tcl) {
124 FREE(r300->vertex_stream_state.state);
125 }
126 }
127 FREE(r300);
128 }
129
130 static void r300_flush_callback(void *data, unsigned flags,
131 struct pipe_fence_handle **fence)
132 {
133 struct r300_context* const cs_context_copy = data;
134
135 r300_flush(&cs_context_copy->context, flags, fence);
136 }
137
138 #define R300_INIT_ATOM(atomname, atomsize) \
139 do { \
140 r300->atomname.name = #atomname; \
141 r300->atomname.state = NULL; \
142 r300->atomname.size = atomsize; \
143 r300->atomname.emit = r300_emit_##atomname; \
144 r300->atomname.dirty = FALSE; \
145 } while (0)
146
147 #define R300_ALLOC_ATOM(atomname, statetype) \
148 do { \
149 r300->atomname.state = CALLOC_STRUCT(statetype); \
150 if (r300->atomname.state == NULL) \
151 return FALSE; \
152 } while (0)
153
154 static boolean r300_setup_atoms(struct r300_context* r300)
155 {
156 boolean is_rv350 = r300->screen->caps.is_rv350;
157 boolean is_r500 = r300->screen->caps.is_r500;
158 boolean has_tcl = r300->screen->caps.has_tcl;
159
160 /* Create the actual atom list.
161 *
162 * Some atoms never change size, others change every emit - those have
163 * the size of 0 here.
164 *
165 * NOTE: The framebuffer state is split into these atoms:
166 * - gpu_flush (unpipelined regs)
167 * - aa_state (unpipelined regs)
168 * - fb_state (unpipelined regs)
169 * - hyperz_state (unpipelined regs followed by pipelined ones)
170 * - fb_state_pipelined (pipelined regs)
171 * The motivation behind this is to be able to emit a strict
172 * subset of the regs, and to have reasonable register ordering. */
173 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
174 R300_INIT_ATOM(gpu_flush, 9);
175 R300_INIT_ATOM(aa_state, 4);
176 R300_INIT_ATOM(fb_state, 0);
177 R300_INIT_ATOM(hyperz_state, is_r500 || is_rv350 ? 10 : 8);
178 /* ZB (unpipelined), SC. */
179 R300_INIT_ATOM(ztop_state, 2);
180 /* ZB, FG. */
181 R300_INIT_ATOM(dsa_state, is_r500 ? 10 : 6);
182 /* RB3D. */
183 R300_INIT_ATOM(blend_state, 8);
184 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
185 /* SC. */
186 R300_INIT_ATOM(sample_mask, 2);
187 R300_INIT_ATOM(scissor_state, 3);
188 /* GB, FG, GA, SU, SC, RB3D. */
189 R300_INIT_ATOM(invariant_state, 14 + (is_rv350 ? 4 : 0) + (is_r500 ? 4 : 0));
190 /* VAP. */
191 R300_INIT_ATOM(viewport_state, 9);
192 R300_INIT_ATOM(pvs_flush, 2);
193 R300_INIT_ATOM(vap_invariant_state, is_r500 || !has_tcl ? 11 : 9);
194 R300_INIT_ATOM(vertex_stream_state, 0);
195 R300_INIT_ATOM(vs_state, 0);
196 R300_INIT_ATOM(vs_constants, 0);
197 R300_INIT_ATOM(clip_state, has_tcl ? 3 + (6 * 4) : 0);
198 /* VAP, RS, GA, GB, SU, SC. */
199 R300_INIT_ATOM(rs_block_state, 0);
200 R300_INIT_ATOM(rs_state, 0);
201 /* SC, US. */
202 R300_INIT_ATOM(fb_state_pipelined, 8);
203 /* US. */
204 R300_INIT_ATOM(fs, 0);
205 R300_INIT_ATOM(fs_rc_constant_state, 0);
206 R300_INIT_ATOM(fs_constants, 0);
207 /* TX. */
208 R300_INIT_ATOM(texture_cache_inval, 2);
209 R300_INIT_ATOM(textures_state, 0);
210 /* Clear commands */
211 R300_INIT_ATOM(hiz_clear, r300->screen->caps.hiz_ram > 0 ? 4 : 0);
212 R300_INIT_ATOM(zmask_clear, r300->screen->caps.zmask_ram > 0 ? 4 : 0);
213 R300_INIT_ATOM(cmask_clear, 4);
214 /* ZB (unpipelined), SU. */
215 R300_INIT_ATOM(query_start, 4);
216
217 /* Replace emission functions for r500. */
218 if (is_r500) {
219 r300->fs.emit = r500_emit_fs;
220 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
221 r300->fs_constants.emit = r500_emit_fs_constants;
222 }
223
224 /* Some non-CSO atoms need explicit space to store the state locally. */
225 R300_ALLOC_ATOM(aa_state, r300_aa_state);
226 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
227 R300_ALLOC_ATOM(clip_state, r300_clip_state);
228 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
229 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
230 R300_ALLOC_ATOM(textures_state, r300_textures_state);
231 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
232 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
233 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
234 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
235 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
236 r300->sample_mask.state = malloc(4);
237 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
238 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
239 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
240 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
241 if (!r300->screen->caps.has_tcl) {
242 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
243 }
244
245 /* Some non-CSO atoms don't use the state pointer. */
246 r300->fb_state_pipelined.allow_null_state = TRUE;
247 r300->fs_rc_constant_state.allow_null_state = TRUE;
248 r300->pvs_flush.allow_null_state = TRUE;
249 r300->query_start.allow_null_state = TRUE;
250 r300->texture_cache_inval.allow_null_state = TRUE;
251
252 /* Some states must be marked as dirty here to properly set up
253 * hardware in the first command stream. */
254 r300_mark_atom_dirty(r300, &r300->invariant_state);
255 r300_mark_atom_dirty(r300, &r300->pvs_flush);
256 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
257 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
258 r300_mark_atom_dirty(r300, &r300->textures_state);
259
260 return TRUE;
261 }
262
263 /* Not every state tracker calls every driver function before the first draw
264 * call and we must initialize the command buffers somehow. */
265 static void r300_init_states(struct pipe_context *pipe)
266 {
267 struct r300_context *r300 = r300_context(pipe);
268 struct pipe_blend_color bc = {{0}};
269 struct pipe_clip_state cs = {{{0}}};
270 struct pipe_scissor_state ss = {0};
271 struct r300_gpu_flush *gpuflush =
272 (struct r300_gpu_flush*)r300->gpu_flush.state;
273 struct r300_vap_invariant_state *vap_invariant =
274 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
275 struct r300_invariant_state *invariant =
276 (struct r300_invariant_state*)r300->invariant_state.state;
277
278 CB_LOCALS;
279
280 pipe->set_blend_color(pipe, &bc);
281 pipe->set_clip_state(pipe, &cs);
282 pipe->set_scissor_states(pipe, 0, 1, &ss);
283 pipe->set_sample_mask(pipe, ~0);
284
285 /* Initialize the GPU flush. */
286 {
287 BEGIN_CB(gpuflush->cb_flush_clean, 6);
288
289 /* Flush and free renderbuffer caches. */
290 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
291 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
292 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
293 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
294 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
295 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
296
297 /* Wait until the GPU is idle.
298 * This fixes random pixels sometimes appearing probably caused
299 * by incomplete rendering. */
300 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
301 END_CB;
302 }
303
304 /* Initialize the VAP invariant state. */
305 {
306 BEGIN_CB(vap_invariant->cb, r300->vap_invariant_state.size);
307 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
308 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
309 OUT_CB_32F(1.0);
310 OUT_CB_32F(1.0);
311 OUT_CB_32F(1.0);
312 OUT_CB_32F(1.0);
313 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
314
315 if (r300->screen->caps.is_r500) {
316 OUT_CB_REG(R500_VAP_TEX_TO_COLOR_CNTL, 0);
317 } else if (!r300->screen->caps.has_tcl) {
318 /* RSxxx:
319 * Static VAP setup since r300_emit_vs_state() is never called.
320 */
321 OUT_CB_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(10) |
322 R300_PVS_NUM_CNTLRS(5) |
323 R300_PVS_NUM_FPUS(2) |
324 R300_PVS_VF_MAX_VTX_NUM(5));
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
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 || r300->screen->caps.is_rv350) {
364 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
365 }
366 END_CB;
367 }
368 }
369
370 struct pipe_context* r300_create_context(struct pipe_screen* screen,
371 void *priv, unsigned flags)
372 {
373 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
374 struct r300_screen* r300screen = r300_screen(screen);
375 struct radeon_winsys *rws = r300screen->rws;
376
377 if (!r300)
378 return NULL;
379
380 r300->rws = rws;
381 r300->screen = r300screen;
382
383 r300->context.screen = screen;
384 r300->context.priv = priv;
385
386 r300->context.destroy = r300_destroy_context;
387
388 slab_create_child(&r300->pool_transfers, &r300screen->pool_transfers);
389
390 r300->ctx = rws->ctx_create(rws);
391 if (!r300->ctx)
392 goto fail;
393
394 r300->cs = rws->cs_create(r300->ctx, RING_GFX, r300_flush_callback, r300);
395 if (r300->cs == NULL)
396 goto fail;
397
398 if (!r300screen->caps.has_tcl) {
399 /* Create a Draw. This is used for SW TCL. */
400 r300->draw = draw_create(&r300->context);
401 if (r300->draw == NULL)
402 goto fail;
403 /* Enable our renderer. */
404 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
405 /* Disable converting points/lines to triangles. */
406 draw_wide_line_threshold(r300->draw, 10000000.f);
407 draw_wide_point_threshold(r300->draw, 10000000.f);
408 draw_wide_point_sprites(r300->draw, FALSE);
409 draw_enable_line_stipple(r300->draw, TRUE);
410 draw_enable_point_sprites(r300->draw, FALSE);
411 }
412
413 if (!r300_setup_atoms(r300))
414 goto fail;
415
416 r300_init_blit_functions(r300);
417 r300_init_flush_functions(r300);
418 r300_init_query_functions(r300);
419 r300_init_state_functions(r300);
420 r300_init_resource_functions(r300);
421 r300_init_render_functions(r300);
422 r300_init_states(&r300->context);
423
424 r300->context.create_video_codec = vl_create_decoder;
425 r300->context.create_video_buffer = vl_video_buffer_create;
426
427 r300->uploader = u_upload_create(&r300->context, 256 * 1024,
428 PIPE_BIND_CUSTOM, PIPE_USAGE_STREAM);
429
430 r300->blitter = util_blitter_create(&r300->context);
431 if (r300->blitter == NULL)
432 goto fail;
433 r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
434
435 /* The KIL opcode needs the first texture unit to be enabled
436 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
437 * dummy texture there. */
438 if (!r300->screen->caps.is_r500) {
439 struct pipe_resource *tex;
440 struct pipe_resource rtempl = {{0}};
441 struct pipe_sampler_view vtempl = {{0}};
442
443 rtempl.target = PIPE_TEXTURE_2D;
444 rtempl.format = PIPE_FORMAT_I8_UNORM;
445 rtempl.usage = PIPE_USAGE_IMMUTABLE;
446 rtempl.width0 = 1;
447 rtempl.height0 = 1;
448 rtempl.depth0 = 1;
449 tex = screen->resource_create(screen, &rtempl);
450
451 u_sampler_view_default_template(&vtempl, tex, tex->format);
452
453 r300->texkill_sampler = (struct r300_sampler_view*)
454 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
455
456 pipe_resource_reference(&tex, NULL);
457 }
458
459 if (r300screen->caps.has_tcl) {
460 struct pipe_resource vb;
461 memset(&vb, 0, sizeof(vb));
462 vb.target = PIPE_BUFFER;
463 vb.format = PIPE_FORMAT_R8_UNORM;
464 vb.usage = PIPE_USAGE_DEFAULT;
465 vb.width0 = sizeof(float) * 16;
466 vb.height0 = 1;
467 vb.depth0 = 1;
468
469 r300->dummy_vb.buffer = screen->resource_create(screen, &vb);
470 r300->context.set_vertex_buffers(&r300->context, 0, 1, &r300->dummy_vb);
471 }
472
473 {
474 struct pipe_depth_stencil_alpha_state dsa;
475 memset(&dsa, 0, sizeof(dsa));
476 dsa.depth.writemask = 1;
477
478 r300->dsa_decompress_zmask =
479 r300->context.create_depth_stencil_alpha_state(&r300->context,
480 &dsa);
481 }
482
483 r300->hyperz_time_of_last_flush = os_time_get();
484
485 /* Register allocator state */
486 rc_init_regalloc_state(&r300->fs_regalloc_state);
487
488 /* Print driver info. */
489 #ifdef DEBUG
490 {
491 #else
492 if (DBG_ON(r300, DBG_INFO)) {
493 #endif
494 fprintf(stderr,
495 "r300: DRM version: %d.%d.%d, Name: %s, ID: 0x%04x, GB: %d, Z: %d\n"
496 "r300: GART size: %"PRIu64" MB, VRAM size: %"PRIu64" MB\n"
497 "r300: AA compression RAM: %s, Z compression RAM: %s, HiZ RAM: %s\n",
498 r300->screen->info.drm_major,
499 r300->screen->info.drm_minor,
500 r300->screen->info.drm_patchlevel,
501 screen->get_name(screen),
502 r300->screen->info.pci_id,
503 r300->screen->info.r300_num_gb_pipes,
504 r300->screen->info.r300_num_z_pipes,
505 r300->screen->info.gart_size >> 20,
506 r300->screen->info.vram_size >> 20,
507 "YES", /* XXX really? */
508 r300->screen->caps.zmask_ram ? "YES" : "NO",
509 r300->screen->caps.hiz_ram ? "YES" : "NO");
510 }
511
512 return &r300->context;
513
514 fail:
515 r300_destroy_context(&r300->context);
516 return NULL;
517 }