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 #include <inttypes.h>
  39
  40 static void r300_update_num_contexts(struct r300_screen *r300screen,
  41                                      int diff)
  42 {
  43     if (diff > 0) {
  44         p_atomic_inc(&r300screen->num_contexts);
  45
  46         if (r300screen->num_contexts > 1)
  47             util_mempool_set_thread_safety(&r300screen->pool_buffers,
  48                                            UTIL_MEMPOOL_MULTITHREADED);
  49     } else {
  50         p_atomic_dec(&r300screen->num_contexts);
  51
  52         if (r300screen->num_contexts <= 1)
  53             util_mempool_set_thread_safety(&r300screen->pool_buffers,
  54                                            UTIL_MEMPOOL_SINGLETHREADED);
  55     }
  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_assign_framebuffer_state(fb, NULL);
  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 SWTCL VBO. */
  83     pipe_resource_reference(&r300->vbo, NULL);
  84
  85     /* Vertex buffers. */
  86     for (i = 0; i < r300->vertex_buffer_count; i++) {
  87         pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
  88     }
  89
  90     /* If there are any queries pending or not destroyed, remove them now. */
  91     foreach_s(query, temp, &r300->query_list) {
  92         remove_from_list(query);
  93         FREE(query);
  94     }
  95 }
  96
  97 static void r300_destroy_context(struct pipe_context* context)
  98 {
  99     struct r300_context* r300 = r300_context(context);
 100     struct r300_atom *atom;
 101
 102     util_blitter_destroy(r300->blitter);
 103     draw_destroy(r300->draw);
 104
 105     /* Print stats, if enabled. */
 106     if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) {
 107         fprintf(stderr, "r300: Stats for context %p:\n", r300);
 108         fprintf(stderr, "    : Flushes: %" PRIu64 "\n", r300->flush_counter);
 109         foreach(atom, &r300->atom_list) {
 110             fprintf(stderr, "    : %s: %" PRIu64 " emits\n",
 111                 atom->name, atom->counter);
 112         }
 113     }
 114
 115     u_upload_destroy(r300->upload_vb);
 116     u_upload_destroy(r300->upload_ib);
 117
 118     /* setup hyper-z mm */
 119     if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
 120         r300_hyperz_destroy_mm(r300);
 121
 122     translate_cache_destroy(r300->tran.translate_cache);
 123
 124     r300_release_referenced_objects(r300);
 125
 126     r300->rws->cs_destroy(r300->cs);
 127
 128     util_mempool_destroy(&r300->pool_transfers);
 129
 130     r300_update_num_contexts(r300->screen, -1);
 131
 132     FREE(r300->aa_state.state);
 133     FREE(r300->blend_color_state.state);
 134     FREE(r300->clip_state.state);
 135     FREE(r300->fb_state.state);
 136     FREE(r300->gpu_flush.state);
 137     FREE(r300->hyperz_state.state);
 138     FREE(r300->invariant_state.state);
 139     FREE(r300->rs_block_state.state);
 140     FREE(r300->scissor_state.state);
 141     FREE(r300->textures_state.state);
 142     FREE(r300->vap_invariant_state.state);
 143     FREE(r300->viewport_state.state);
 144     FREE(r300->ztop_state.state);
 145     FREE(r300->fs_constants.state);
 146     FREE(r300->vs_constants.state);
 147     if (!r300->screen->caps.has_tcl) {
 148         FREE(r300->vertex_stream_state.state);
 149     }
 150     FREE(r300);
 151 }
 152
 153 void r300_flush_cb(void *data)
 154 {
 155     struct r300_context* const cs_context_copy = data;
 156
 157     cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
 158 }
 159
 160 #define R300_INIT_ATOM(atomname, atomsize) \
 161     r300->atomname.name = #atomname; \
 162     r300->atomname.state = NULL; \
 163     r300->atomname.size = atomsize; \
 164     r300->atomname.emit = r300_emit_##atomname; \
 165     r300->atomname.dirty = FALSE; \
 166     insert_at_tail(&r300->atom_list, &r300->atomname);
 167
 168 static void 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 has_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      * Each atom is examined and emitted in the order it appears here, which
 181      * can affect performance and conformance if not handled with care.
 182      *
 183      * Some atoms never change size, others change every emit - those have
 184      * the size of 0 here.
 185      *
 186      * NOTE: The framebuffer state is split into these atoms:
 187      * - gpu_flush          (unpipelined regs)
 188      * - aa_state           (unpipelined regs)
 189      * - fb_state           (unpipelined regs)
 190      * - hyperz_state       (unpipelined regs followed by pipelined ones)
 191      * - fb_state_pipelined (pipelined regs)
 192      * The motivation behind this is to be able to emit a strict
 193      * subset of the regs, and to have reasonable register ordering. */
 194     make_empty_list(&r300->atom_list);
 195     /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
 196     R300_INIT_ATOM(gpu_flush, 9);
 197     R300_INIT_ATOM(aa_state, 4);
 198     R300_INIT_ATOM(fb_state, 0);
 199     R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
 200     /* ZB (unpipelined), SC. */
 201     R300_INIT_ATOM(ztop_state, 2);
 202     /* ZB, FG. */
 203     R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
 204     /* RB3D. */
 205     R300_INIT_ATOM(blend_state, 8);
 206     R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
 207     /* SC. */
 208     R300_INIT_ATOM(scissor_state, 3);
 209     /* GB, FG, GA, SU, SC, RB3D. */
 210     R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0));
 211     /* VAP. */
 212     R300_INIT_ATOM(viewport_state, 9);
 213     R300_INIT_ATOM(pvs_flush, 2);
 214     R300_INIT_ATOM(vap_invariant_state, 9);
 215     R300_INIT_ATOM(vertex_stream_state, 0);
 216     R300_INIT_ATOM(vs_state, 0);
 217     R300_INIT_ATOM(vs_constants, 0);
 218     R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
 219     /* VAP, RS, GA, GB, SU, SC. */
 220     R300_INIT_ATOM(rs_block_state, 0);
 221     R300_INIT_ATOM(rs_state, 0);
 222     /* SC, US. */
 223     R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0));
 224     /* US. */
 225     R300_INIT_ATOM(fs, 0);
 226     R300_INIT_ATOM(fs_rc_constant_state, 0);
 227     R300_INIT_ATOM(fs_constants, 0);
 228     /* TX. */
 229     R300_INIT_ATOM(texture_cache_inval, 2);
 230     R300_INIT_ATOM(textures_state, 0);
 231     if (has_hyperz) {
 232         /* HiZ Clear */
 233         if (has_hiz_ram)
 234             R300_INIT_ATOM(hiz_clear, 0);
 235         /* zmask clear */
 236         R300_INIT_ATOM(zmask_clear, 0);
 237     }
 238     /* ZB (unpipelined), SU. */
 239     R300_INIT_ATOM(query_start, 4);
 240
 241     /* Replace emission functions for r500. */
 242     if (is_r500) {
 243         r300->fs.emit = r500_emit_fs;
 244         r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
 245         r300->fs_constants.emit = r500_emit_fs_constants;
 246     }
 247
 248     /* Some non-CSO atoms need explicit space to store the state locally. */
 249     r300->aa_state.state = CALLOC_STRUCT(r300_aa_state);
 250     r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
 251     r300->clip_state.state = CALLOC_STRUCT(r300_clip_state);
 252     r300->fb_state.state = CALLOC_STRUCT(pipe_framebuffer_state);
 253     r300->gpu_flush.state = CALLOC_STRUCT(pipe_framebuffer_state);
 254     r300->hyperz_state.state = CALLOC_STRUCT(r300_hyperz_state);
 255     r300->invariant_state.state = CALLOC_STRUCT(r300_invariant_state);
 256     r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
 257     r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
 258     r300->textures_state.state = CALLOC_STRUCT(r300_textures_state);
 259     r300->vap_invariant_state.state = CALLOC_STRUCT(r300_vap_invariant_state);
 260     r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
 261     r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
 262     r300->fs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
 263     r300->vs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
 264     if (!r300->screen->caps.has_tcl) {
 265         r300->vertex_stream_state.state = CALLOC_STRUCT(r300_vertex_stream_state);
 266     }
 267
 268     /* Some non-CSO atoms don't use the state pointer. */
 269     r300->fb_state_pipelined.allow_null_state = TRUE;
 270     r300->fs_rc_constant_state.allow_null_state = TRUE;
 271     r300->pvs_flush.allow_null_state = TRUE;
 272     r300->query_start.allow_null_state = TRUE;
 273     r300->texture_cache_inval.allow_null_state = TRUE;
 274
 275     /* Some states must be marked as dirty here to properly set up
 276      * hardware in the first command stream. */
 277     r300->invariant_state.dirty = TRUE;
 278     r300->pvs_flush.dirty = TRUE;
 279     r300->vap_invariant_state.dirty = TRUE;
 280     r300->texture_cache_inval.dirty = TRUE;
 281     r300->textures_state.dirty = TRUE;
 282 }
 283
 284 /* Not every state tracker calls every driver function before the first draw
 285  * call and we must initialize the command buffers somehow. */
 286 static void r300_init_states(struct pipe_context *pipe)
 287 {
 288     struct r300_context *r300 = r300_context(pipe);
 289     struct pipe_blend_color bc = {{0}};
 290     struct pipe_clip_state cs = {{{0}}};
 291     struct pipe_scissor_state ss = {0};
 292     struct r300_clip_state *clip =
 293             (struct r300_clip_state*)r300->clip_state.state;
 294     struct r300_gpu_flush *gpuflush =
 295             (struct r300_gpu_flush*)r300->gpu_flush.state;
 296     struct r300_vap_invariant_state *vap_invariant =
 297             (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
 298     struct r300_invariant_state *invariant =
 299             (struct r300_invariant_state*)r300->invariant_state.state;
 300
 301     CB_LOCALS;
 302
 303     pipe->set_blend_color(pipe, &bc);
 304     pipe->set_scissor_state(pipe, &ss);
 305
 306     /* Initialize the clip state. */
 307     if (r300_context(pipe)->screen->caps.has_tcl) {
 308         pipe->set_clip_state(pipe, &cs);
 309     } else {
 310         BEGIN_CB(clip->cb, 2);
 311         OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
 312         END_CB;
 313     }
 314
 315     /* Initialize the GPU flush. */
 316     {
 317         BEGIN_CB(gpuflush->cb_flush_clean, 6);
 318
 319         /* Flush and free renderbuffer caches. */
 320         OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
 321             R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
 322             R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
 323         OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
 324             R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
 325             R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
 326
 327         /* Wait until the GPU is idle.
 328          * This fixes random pixels sometimes appearing probably caused
 329          * by incomplete rendering. */
 330         OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
 331         END_CB;
 332     }
 333
 334     /* Initialize the VAP invariant state. */
 335     {
 336         BEGIN_CB(vap_invariant->cb, 9);
 337         OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
 338         OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
 339         OUT_CB_32F(1.0);
 340         OUT_CB_32F(1.0);
 341         OUT_CB_32F(1.0);
 342         OUT_CB_32F(1.0);
 343         OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
 344         END_CB;
 345     }
 346
 347     /* Initialize the invariant state. */
 348     {
 349         BEGIN_CB(invariant->cb, r300->invariant_state.size);
 350         OUT_CB_REG(R300_GB_SELECT, 0);
 351         OUT_CB_REG(R300_FG_FOG_BLEND, 0);
 352         OUT_CB_REG(R300_GA_ROUND_MODE, 1);
 353         OUT_CB_REG(R300_GA_OFFSET, 0);
 354         OUT_CB_REG(R300_SU_TEX_WRAP, 0);
 355         OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
 356         OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
 357         OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
 358
 359         if (r300->screen->caps.is_rv350) {
 360             OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
 361             OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
 362         }
 363         END_CB;
 364     }
 365
 366     /* Initialize the hyperz state. */
 367     {
 368         struct r300_hyperz_state *hyperz =
 369             (struct r300_hyperz_state*)r300->hyperz_state.state;
 370         BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
 371         OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
 372                    R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
 373         OUT_CB_REG(R300_ZB_BW_CNTL, 0);
 374         OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
 375         OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
 376
 377         if (r300->screen->caps.is_r500 ||
 378             (r300->screen->caps.is_rv350 &&
 379              r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) {
 380             OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
 381         }
 382         END_CB;
 383     }
 384 }
 385
 386 struct pipe_context* r300_create_context(struct pipe_screen* screen,
 387                                          void *priv)
 388 {
 389     struct r300_context* r300 = CALLOC_STRUCT(r300_context);
 390     struct r300_screen* r300screen = r300_screen(screen);
 391     struct r300_winsys_screen *rws = r300screen->rws;
 392
 393     if (!r300)
 394         return NULL;
 395
 396     r300_update_num_contexts(r300screen, 1);
 397
 398     r300->rws = rws;
 399     r300->screen = r300screen;
 400
 401     r300->context.winsys = (struct pipe_winsys*)rws;
 402     r300->context.screen = screen;
 403     r300->context.priv = priv;
 404
 405     r300->context.destroy = r300_destroy_context;
 406
 407     r300->cs = rws->cs_create(rws);
 408
 409     util_mempool_create(&r300->pool_transfers,
 410                         sizeof(struct pipe_transfer), 64,
 411                         UTIL_MEMPOOL_SINGLETHREADED);
 412
 413     if (!r300screen->caps.has_tcl) {
 414         /* Create a Draw. This is used for SW TCL. */
 415         r300->draw = draw_create(&r300->context);
 416         /* Enable our renderer. */
 417         draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
 418         /* Enable Draw's clipping. */
 419         draw_set_driver_clipping(r300->draw, FALSE);
 420         /* Disable converting points/lines to triangles. */
 421         draw_wide_line_threshold(r300->draw, 10000000.f);
 422         draw_wide_point_threshold(r300->draw, 10000000.f);
 423     }
 424
 425     r300_setup_atoms(r300);
 426
 427     make_empty_list(&r300->query_list);
 428
 429     r300_init_blit_functions(r300);
 430     r300_init_flush_functions(r300);
 431     r300_init_query_functions(r300);
 432     r300_init_state_functions(r300);
 433     r300_init_resource_functions(r300);
 434
 435     r300->blitter = util_blitter_create(&r300->context);
 436
 437     /* Render functions must be initialized after blitter. */
 438     r300_init_render_functions(r300);
 439
 440     rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
 441
 442     /* setup hyper-z mm */
 443     if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
 444         r300_hyperz_init_mm(r300);
 445
 446     r300->upload_ib = u_upload_create(&r300->context,
 447                                       32 * 1024, 16,
 448                                       PIPE_BIND_INDEX_BUFFER);
 449
 450     if (r300->upload_ib == NULL)
 451         goto no_upload_ib;
 452
 453     r300->upload_vb = u_upload_create(&r300->context,
 454                                       128 * 1024, 16,
 455                                       PIPE_BIND_VERTEX_BUFFER);
 456     if (r300->upload_vb == NULL)
 457         goto no_upload_vb;
 458
 459     r300->tran.translate_cache = translate_cache_create();
 460
 461     r300_init_states(&r300->context);
 462
 463     /* The KIL opcode needs the first texture unit to be enabled
 464      * on r3xx-r4xx. In order to calm down the CS checker, we bind this
 465      * dummy texture there. */
 466     if (!r300->screen->caps.is_r500) {
 467         struct pipe_resource *tex;
 468         struct pipe_resource rtempl = {{0}};
 469         struct pipe_sampler_view vtempl = {{0}};
 470
 471         rtempl.target = PIPE_TEXTURE_2D;
 472         rtempl.format = PIPE_FORMAT_I8_UNORM;
 473         rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
 474         rtempl.width0 = 1;
 475         rtempl.height0 = 1;
 476         rtempl.depth0 = 1;
 477         tex = screen->resource_create(screen, &rtempl);
 478
 479         u_sampler_view_default_template(&vtempl, tex, tex->format);
 480
 481         r300->texkill_sampler = (struct r300_sampler_view*)
 482             r300->context.create_sampler_view(&r300->context, tex, &vtempl);
 483
 484         pipe_resource_reference(&tex, NULL);
 485     }
 486
 487     return &r300->context;
 488
 489  no_upload_ib:
 490     u_upload_destroy(r300->upload_ib);
 491  no_upload_vb:
 492     FREE(r300);
 493     return NULL;
 494 }
 495
 496 void r300_finish(struct r300_context *r300)
 497 {
 498     struct pipe_framebuffer_state *fb;
 499     unsigned i;
 500
 501     /* This is a preliminary implementation of glFinish.
 502      *
 503      * The ideal implementation should use something like EmitIrqLocked and
 504      * WaitIrq, or better, real fences.
 505      */
 506     if (r300->fb_state.state) {
 507         fb = r300->fb_state.state;
 508
 509         for (i = 0; i < fb->nr_cbufs; i++) {
 510             if (fb->cbufs[i]->texture) {
 511                 r300->rws->buffer_wait(r300->rws,
 512                     r300_texture(fb->cbufs[i]->texture)->buffer);
 513                 return;
 514             }
 515         }
 516         if (fb->zsbuf && fb->zsbuf->texture) {
 517             r300->rws->buffer_wait(r300->rws,
 518                 r300_texture(fb->zsbuf->texture)->buffer);
 519         }
 520     }
 521 }