radeonsi: don't emit PA_SC_LINE_STIPPLE after every rasterizer state change
[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
98 rc_destroy_regalloc_state(&r300->fs_regalloc_state);
99
100 /* XXX: No way to tell if this was initialized or not? */
101 util_slab_destroy(&r300->pool_transfers);
102
103 /* Free the structs allocated in r300_setup_atoms() */
104 if (r300->aa_state.state) {
105 FREE(r300->aa_state.state);
106 FREE(r300->blend_color_state.state);
107 FREE(r300->clip_state.state);
108 FREE(r300->fb_state.state);
109 FREE(r300->gpu_flush.state);
110 FREE(r300->hyperz_state.state);
111 FREE(r300->invariant_state.state);
112 FREE(r300->rs_block_state.state);
113 FREE(r300->sample_mask.state);
114 FREE(r300->scissor_state.state);
115 FREE(r300->textures_state.state);
116 FREE(r300->vap_invariant_state.state);
117 FREE(r300->viewport_state.state);
118 FREE(r300->ztop_state.state);
119 FREE(r300->fs_constants.state);
120 FREE(r300->vs_constants.state);
121 if (!r300->screen->caps.has_tcl) {
122 FREE(r300->vertex_stream_state.state);
123 }
124 }
125 FREE(r300);
126 }
127
128 static void r300_flush_callback(void *data, unsigned flags,
129 struct pipe_fence_handle **fence)
130 {
131 struct r300_context* const cs_context_copy = data;
132
133 r300_flush(&cs_context_copy->context, flags, fence);
134 }
135
136 #define R300_INIT_ATOM(atomname, atomsize) \
137 do { \
138 r300->atomname.name = #atomname; \
139 r300->atomname.state = NULL; \
140 r300->atomname.size = atomsize; \
141 r300->atomname.emit = r300_emit_##atomname; \
142 r300->atomname.dirty = FALSE; \
143 } while (0)
144
145 #define R300_ALLOC_ATOM(atomname, statetype) \
146 do { \
147 r300->atomname.state = CALLOC_STRUCT(statetype); \
148 if (r300->atomname.state == NULL) \
149 return FALSE; \
150 } while (0)
151
152 static boolean r300_setup_atoms(struct r300_context* r300)
153 {
154 boolean is_rv350 = r300->screen->caps.is_rv350;
155 boolean is_r500 = r300->screen->caps.is_r500;
156 boolean has_tcl = r300->screen->caps.has_tcl;
157 boolean drm_2_6_0 = r300->screen->info.drm_minor >= 6;
158
159 /* Create the actual atom list.
160 *
161 * Some atoms never change size, others change every emit - those have
162 * the size of 0 here.
163 *
164 * NOTE: The framebuffer state is split into these atoms:
165 * - gpu_flush (unpipelined regs)
166 * - aa_state (unpipelined regs)
167 * - fb_state (unpipelined regs)
168 * - hyperz_state (unpipelined regs followed by pipelined ones)
169 * - fb_state_pipelined (pipelined regs)
170 * The motivation behind this is to be able to emit a strict
171 * subset of the regs, and to have reasonable register ordering. */
172 /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
173 R300_INIT_ATOM(gpu_flush, 9);
174 R300_INIT_ATOM(aa_state, 4);
175 R300_INIT_ATOM(fb_state, 0);
176 R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
177 /* ZB (unpipelined), SC. */
178 R300_INIT_ATOM(ztop_state, 2);
179 /* ZB, FG. */
180 R300_INIT_ATOM(dsa_state, is_r500 ? (drm_2_6_0 ? 10 : 8) : 6);
181 /* RB3D. */
182 R300_INIT_ATOM(blend_state, 8);
183 R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
184 /* SC. */
185 R300_INIT_ATOM(sample_mask, 2);
186 R300_INIT_ATOM(scissor_state, 3);
187 /* GB, FG, GA, SU, SC, RB3D. */
188 R300_INIT_ATOM(invariant_state, 14 + (is_rv350 ? 4 : 0) + (is_r500 ? 4 : 0));
189 /* VAP. */
190 R300_INIT_ATOM(viewport_state, 9);
191 R300_INIT_ATOM(pvs_flush, 2);
192 R300_INIT_ATOM(vap_invariant_state, is_r500 ? 11 : 9);
193 R300_INIT_ATOM(vertex_stream_state, 0);
194 R300_INIT_ATOM(vs_state, 0);
195 R300_INIT_ATOM(vs_constants, 0);
196 R300_INIT_ATOM(clip_state, has_tcl ? 3 + (6 * 4) : 0);
197 /* VAP, RS, GA, GB, SU, SC. */
198 R300_INIT_ATOM(rs_block_state, 0);
199 R300_INIT_ATOM(rs_state, 0);
200 /* SC, US. */
201 R300_INIT_ATOM(fb_state_pipelined, 8);
202 /* US. */
203 R300_INIT_ATOM(fs, 0);
204 R300_INIT_ATOM(fs_rc_constant_state, 0);
205 R300_INIT_ATOM(fs_constants, 0);
206 /* TX. */
207 R300_INIT_ATOM(texture_cache_inval, 2);
208 R300_INIT_ATOM(textures_state, 0);
209 /* Clear commands */
210 R300_INIT_ATOM(hiz_clear, r300->screen->caps.hiz_ram > 0 ? 4 : 0);
211 R300_INIT_ATOM(zmask_clear, r300->screen->caps.zmask_ram > 0 ? 4 : 0);
212 R300_INIT_ATOM(cmask_clear, 4);
213 /* ZB (unpipelined), SU. */
214 R300_INIT_ATOM(query_start, 4);
215
216 /* Replace emission functions for r500. */
217 if (is_r500) {
218 r300->fs.emit = r500_emit_fs;
219 r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
220 r300->fs_constants.emit = r500_emit_fs_constants;
221 }
222
223 /* Some non-CSO atoms need explicit space to store the state locally. */
224 R300_ALLOC_ATOM(aa_state, r300_aa_state);
225 R300_ALLOC_ATOM(blend_color_state, r300_blend_color_state);
226 R300_ALLOC_ATOM(clip_state, r300_clip_state);
227 R300_ALLOC_ATOM(hyperz_state, r300_hyperz_state);
228 R300_ALLOC_ATOM(invariant_state, r300_invariant_state);
229 R300_ALLOC_ATOM(textures_state, r300_textures_state);
230 R300_ALLOC_ATOM(vap_invariant_state, r300_vap_invariant_state);
231 R300_ALLOC_ATOM(viewport_state, r300_viewport_state);
232 R300_ALLOC_ATOM(ztop_state, r300_ztop_state);
233 R300_ALLOC_ATOM(fb_state, pipe_framebuffer_state);
234 R300_ALLOC_ATOM(gpu_flush, pipe_framebuffer_state);
235 r300->sample_mask.state = malloc(4);
236 R300_ALLOC_ATOM(scissor_state, pipe_scissor_state);
237 R300_ALLOC_ATOM(rs_block_state, r300_rs_block);
238 R300_ALLOC_ATOM(fs_constants, r300_constant_buffer);
239 R300_ALLOC_ATOM(vs_constants, r300_constant_buffer);
240 if (!r300->screen->caps.has_tcl) {
241 R300_ALLOC_ATOM(vertex_stream_state, r300_vertex_stream_state);
242 }
243
244 /* Some non-CSO atoms don't use the state pointer. */
245 r300->fb_state_pipelined.allow_null_state = TRUE;
246 r300->fs_rc_constant_state.allow_null_state = TRUE;
247 r300->pvs_flush.allow_null_state = TRUE;
248 r300->query_start.allow_null_state = TRUE;
249 r300->texture_cache_inval.allow_null_state = TRUE;
250
251 /* Some states must be marked as dirty here to properly set up
252 * hardware in the first command stream. */
253 r300_mark_atom_dirty(r300, &r300->invariant_state);
254 r300_mark_atom_dirty(r300, &r300->pvs_flush);
255 r300_mark_atom_dirty(r300, &r300->vap_invariant_state);
256 r300_mark_atom_dirty(r300, &r300->texture_cache_inval);
257 r300_mark_atom_dirty(r300, &r300->textures_state);
258
259 return TRUE;
260 }
261
262 /* Not every state tracker calls every driver function before the first draw
263 * call and we must initialize the command buffers somehow. */
264 static void r300_init_states(struct pipe_context *pipe)
265 {
266 struct r300_context *r300 = r300_context(pipe);
267 struct pipe_blend_color bc = {{0}};
268 struct pipe_clip_state cs = {{{0}}};
269 struct pipe_scissor_state ss = {0};
270 struct r300_gpu_flush *gpuflush =
271 (struct r300_gpu_flush*)r300->gpu_flush.state;
272 struct r300_vap_invariant_state *vap_invariant =
273 (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
274 struct r300_invariant_state *invariant =
275 (struct r300_invariant_state*)r300->invariant_state.state;
276
277 CB_LOCALS;
278
279 pipe->set_blend_color(pipe, &bc);
280 pipe->set_clip_state(pipe, &cs);
281 pipe->set_scissor_states(pipe, 0, 1, &ss);
282 pipe->set_sample_mask(pipe, ~0);
283
284 /* Initialize the GPU flush. */
285 {
286 BEGIN_CB(gpuflush->cb_flush_clean, 6);
287
288 /* Flush and free renderbuffer caches. */
289 OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
290 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
291 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
292 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
293 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
294 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
295
296 /* Wait until the GPU is idle.
297 * This fixes random pixels sometimes appearing probably caused
298 * by incomplete rendering. */
299 OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
300 END_CB;
301 }
302
303 /* Initialize the VAP invariant state. */
304 {
305 BEGIN_CB(vap_invariant->cb, r300->vap_invariant_state.size);
306 OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
307 OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
308 OUT_CB_32F(1.0);
309 OUT_CB_32F(1.0);
310 OUT_CB_32F(1.0);
311 OUT_CB_32F(1.0);
312 OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
313
314 if (r300->screen->caps.is_r500) {
315 OUT_CB_REG(R500_VAP_TEX_TO_COLOR_CNTL, 0);
316 }
317 END_CB;
318 }
319
320 /* Initialize the invariant state. */
321 {
322 BEGIN_CB(invariant->cb, r300->invariant_state.size);
323 OUT_CB_REG(R300_GB_SELECT, 0);
324 OUT_CB_REG(R300_FG_FOG_BLEND, 0);
325 OUT_CB_REG(R300_GA_OFFSET, 0);
326 OUT_CB_REG(R300_SU_TEX_WRAP, 0);
327 OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
328 OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
329 OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
330
331 if (r300->screen->caps.is_rv350) {
332 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
333 OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
334 }
335
336 if (r300->screen->caps.is_r500) {
337 OUT_CB_REG(R500_GA_COLOR_CONTROL_PS3, 0);
338 OUT_CB_REG(R500_SU_TEX_WRAP_PS3, 0);
339 }
340 END_CB;
341 }
342
343 /* Initialize the hyperz state. */
344 {
345 struct r300_hyperz_state *hyperz =
346 (struct r300_hyperz_state*)r300->hyperz_state.state;
347 BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
348 OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
349 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
350 OUT_CB_REG(R300_ZB_BW_CNTL, 0);
351 OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
352 OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
353
354 if (r300->screen->caps.is_r500 ||
355 (r300->screen->caps.is_rv350 &&
356 r300->screen->info.drm_minor >= 6)) {
357 OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
358 }
359 END_CB;
360 }
361 }
362
363 struct pipe_context* r300_create_context(struct pipe_screen* screen,
364 void *priv)
365 {
366 struct r300_context* r300 = CALLOC_STRUCT(r300_context);
367 struct r300_screen* r300screen = r300_screen(screen);
368 struct radeon_winsys *rws = r300screen->rws;
369
370 if (!r300)
371 return NULL;
372
373 r300->rws = rws;
374 r300->screen = r300screen;
375
376 r300->context.screen = screen;
377 r300->context.priv = priv;
378
379 r300->context.destroy = r300_destroy_context;
380
381 util_slab_create(&r300->pool_transfers,
382 sizeof(struct pipe_transfer), 64,
383 UTIL_SLAB_SINGLETHREADED);
384
385 r300->cs = rws->cs_create(rws, RING_GFX, r300_flush_callback, r300, NULL);
386 if (r300->cs == NULL)
387 goto fail;
388
389 if (!r300screen->caps.has_tcl) {
390 /* Create a Draw. This is used for SW TCL. */
391 r300->draw = draw_create(&r300->context);
392 if (r300->draw == NULL)
393 goto fail;
394 /* Enable our renderer. */
395 draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
396 /* Disable converting points/lines to triangles. */
397 draw_wide_line_threshold(r300->draw, 10000000.f);
398 draw_wide_point_threshold(r300->draw, 10000000.f);
399 draw_wide_point_sprites(r300->draw, FALSE);
400 draw_enable_line_stipple(r300->draw, TRUE);
401 draw_enable_point_sprites(r300->draw, FALSE);
402 }
403
404 if (!r300_setup_atoms(r300))
405 goto fail;
406
407 r300_init_blit_functions(r300);
408 r300_init_flush_functions(r300);
409 r300_init_query_functions(r300);
410 r300_init_state_functions(r300);
411 r300_init_resource_functions(r300);
412 r300_init_render_functions(r300);
413 r300_init_states(&r300->context);
414
415 r300->context.create_video_codec = vl_create_decoder;
416 r300->context.create_video_buffer = vl_video_buffer_create;
417
418 r300->uploader = u_upload_create(&r300->context, 256 * 1024, 4,
419 PIPE_BIND_CUSTOM);
420
421 r300->blitter = util_blitter_create(&r300->context);
422 if (r300->blitter == NULL)
423 goto fail;
424 r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
425
426 /* The KIL opcode needs the first texture unit to be enabled
427 * on r3xx-r4xx. In order to calm down the CS checker, we bind this
428 * dummy texture there. */
429 if (!r300->screen->caps.is_r500) {
430 struct pipe_resource *tex;
431 struct pipe_resource rtempl = {{0}};
432 struct pipe_sampler_view vtempl = {{0}};
433
434 rtempl.target = PIPE_TEXTURE_2D;
435 rtempl.format = PIPE_FORMAT_I8_UNORM;
436 rtempl.usage = PIPE_USAGE_IMMUTABLE;
437 rtempl.width0 = 1;
438 rtempl.height0 = 1;
439 rtempl.depth0 = 1;
440 tex = screen->resource_create(screen, &rtempl);
441
442 u_sampler_view_default_template(&vtempl, tex, tex->format);
443
444 r300->texkill_sampler = (struct r300_sampler_view*)
445 r300->context.create_sampler_view(&r300->context, tex, &vtempl);
446
447 pipe_resource_reference(&tex, NULL);
448 }
449
450 if (r300screen->caps.has_tcl) {
451 struct pipe_resource vb;
452 memset(&vb, 0, sizeof(vb));
453 vb.target = PIPE_BUFFER;
454 vb.format = PIPE_FORMAT_R8_UNORM;
455 vb.usage = PIPE_USAGE_DEFAULT;
456 vb.width0 = sizeof(float) * 16;
457 vb.height0 = 1;
458 vb.depth0 = 1;
459
460 r300->dummy_vb.buffer = screen->resource_create(screen, &vb);
461 r300->context.set_vertex_buffers(&r300->context, 0, 1, &r300->dummy_vb);
462 }
463
464 {
465 struct pipe_depth_stencil_alpha_state dsa;
466 memset(&dsa, 0, sizeof(dsa));
467 dsa.depth.writemask = 1;
468
469 r300->dsa_decompress_zmask =
470 r300->context.create_depth_stencil_alpha_state(&r300->context,
471 &dsa);
472 }
473
474 r300->hyperz_time_of_last_flush = os_time_get();
475
476 /* Register allocator state */
477 rc_init_regalloc_state(&r300->fs_regalloc_state);
478
479 /* Print driver info. */
480 #ifdef DEBUG
481 {
482 #else
483 if (DBG_ON(r300, DBG_INFO)) {
484 #endif
485 fprintf(stderr,
486 "r300: DRM version: %d.%d.%d, Name: %s, ID: 0x%04x, GB: %d, Z: %d\n"
487 "r300: GART size: %"PRIu64" MB, VRAM size: %"PRIu64" MB\n"
488 "r300: AA compression RAM: %s, Z compression RAM: %s, HiZ RAM: %s\n",
489 r300->screen->info.drm_major,
490 r300->screen->info.drm_minor,
491 r300->screen->info.drm_patchlevel,
492 screen->get_name(screen),
493 r300->screen->info.pci_id,
494 r300->screen->info.r300_num_gb_pipes,
495 r300->screen->info.r300_num_z_pipes,
496 r300->screen->info.gart_size >> 20,
497 r300->screen->info.vram_size >> 20,
498 "YES", /* XXX really? */
499 r300->screen->caps.zmask_ram ? "YES" : "NO",
500 r300->screen->caps.hiz_ram ? "YES" : "NO");
501 }
502
503 return &r300->context;
504
505 fail:
506 r300_destroy_context(&r300->context);
507 return NULL;
508 }