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_screen.h"
  34 #include "r300_screen_buffer.h"
  35 #include "r300_winsys.h"
  36
  37 #include <inttypes.h>
  38
  39 static void r300_update_num_contexts(struct r300_screen *r300screen,
  40                                      int diff)
  41 {
  42     if (diff > 0) {
  43         p_atomic_inc(&r300screen->num_contexts);
  44
  45         if (r300screen->num_contexts > 1)
  46             util_mempool_set_thread_safety(&r300screen->pool_buffers,
  47                                            UTIL_MEMPOOL_MULTITHREADED);
  48     } else {
  49         p_atomic_dec(&r300screen->num_contexts);
  50
  51         if (r300screen->num_contexts <= 1)
  52             util_mempool_set_thread_safety(&r300screen->pool_buffers,
  53                                            UTIL_MEMPOOL_SINGLETHREADED);
  54     }
  55 }
  56
  57 static void r300_release_referenced_objects(struct r300_context *r300)
  58 {
  59     struct pipe_framebuffer_state *fb =
  60             (struct pipe_framebuffer_state*)r300->fb_state.state;
  61     struct r300_textures_state *textures =
  62             (struct r300_textures_state*)r300->textures_state.state;
  63     struct r300_query *query, *temp;
  64     unsigned i;
  65
  66     /* Framebuffer state. */
  67     util_assign_framebuffer_state(fb, NULL);
  68
  69     /* Textures. */
  70     for (i = 0; i < textures->sampler_view_count; i++)
  71         pipe_sampler_view_reference(
  72                 (struct pipe_sampler_view**)&textures->sampler_views[i], NULL);
  73
  74     /* The special dummy texture for texkill. */
  75     if (r300->texkill_sampler) {
  76         pipe_sampler_view_reference(
  77                 (struct pipe_sampler_view**)&r300->texkill_sampler,
  78                 NULL);
  79     }
  80
  81     /* The SWTCL VBO. */
  82     pipe_resource_reference(&r300->vbo, NULL);
  83
  84     /* Vertex buffers. */
  85     for (i = 0; i < r300->vertex_buffer_count; i++) {
  86         pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
  87     }
  88
  89     /* If there are any queries pending or not destroyed, remove them now. */
  90     foreach_s(query, temp, &r300->query_list) {
  91         remove_from_list(query);
  92         FREE(query);
  93     }
  94 }
  95
  96 static void r300_destroy_context(struct pipe_context* context)
  97 {
  98     struct r300_context* r300 = r300_context(context);
  99     struct r300_atom *atom;
 100
 101     util_blitter_destroy(r300->blitter);
 102     draw_destroy(r300->draw);
 103
 104     /* Print stats, if enabled. */
 105     if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) {
 106         fprintf(stderr, "r300: Stats for context %p:\n", r300);
 107         fprintf(stderr, "    : Flushes: %" PRIu64 "\n", r300->flush_counter);
 108         foreach(atom, &r300->atom_list) {
 109             fprintf(stderr, "    : %s: %" PRIu64 " emits\n",
 110                 atom->name, atom->counter);
 111         }
 112     }
 113
 114     u_upload_destroy(r300->upload_vb);
 115     u_upload_destroy(r300->upload_ib);
 116
 117     translate_cache_destroy(r300->tran.translate_cache);
 118
 119     r300_release_referenced_objects(r300);
 120
 121     r300->rws->cs_destroy(r300->cs);
 122
 123     util_mempool_destroy(&r300->pool_transfers);
 124
 125     r300_update_num_contexts(r300->screen, -1);
 126
 127     FREE(r300->aa_state.state);
 128     FREE(r300->blend_color_state.state);
 129     FREE(r300->clip_state.state);
 130     FREE(r300->fb_state.state);
 131     FREE(r300->gpu_flush.state);
 132     FREE(r300->hyperz_state.state);
 133     FREE(r300->invariant_state.state);
 134     FREE(r300->rs_block_state.state);
 135     FREE(r300->scissor_state.state);
 136     FREE(r300->textures_state.state);
 137     FREE(r300->vap_invariant_state.state);
 138     FREE(r300->viewport_state.state);
 139     FREE(r300->ztop_state.state);
 140     FREE(r300->fs_constants.state);
 141     FREE(r300->vs_constants.state);
 142     if (!r300->screen->caps.has_tcl) {
 143         FREE(r300->vertex_stream_state.state);
 144     }
 145     FREE(r300);
 146 }
 147
 148 void r300_flush_cb(void *data)
 149 {
 150     struct r300_context* const cs_context_copy = data;
 151
 152     cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
 153 }
 154
 155 #define R300_INIT_ATOM(atomname, atomsize) \
 156     r300->atomname.name = #atomname; \
 157     r300->atomname.state = NULL; \
 158     r300->atomname.size = atomsize; \
 159     r300->atomname.emit = r300_emit_##atomname; \
 160     r300->atomname.dirty = FALSE; \
 161     insert_at_tail(&r300->atom_list, &r300->atomname);
 162
 163 static void r300_setup_atoms(struct r300_context* r300)
 164 {
 165     boolean is_rv350 = r300->screen->caps.is_rv350;
 166     boolean is_r500 = r300->screen->caps.is_r500;
 167     boolean has_tcl = r300->screen->caps.has_tcl;
 168     boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
 169
 170     /* Create the actual atom list.
 171      *
 172      * Each atom is examined and emitted in the order it appears here, which
 173      * can affect performance and conformance if not handled with care.
 174      *
 175      * Some atoms never change size, others change every emit - those have
 176      * the size of 0 here.
 177      *
 178      * NOTE: The framebuffer state is split into these atoms:
 179      * - gpu_flush          (unpipelined regs)
 180      * - aa_state           (unpipelined regs)
 181      * - fb_state           (unpipelined regs)
 182      * - hyperz_state       (unpipelined regs followed by pipelined ones)
 183      * - fb_state_pipelined (pipelined regs)
 184      * The motivation behind this is to be able to emit a strict
 185      * subset of the regs, and to have reasonable register ordering. */
 186     make_empty_list(&r300->atom_list);
 187     /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
 188     R300_INIT_ATOM(gpu_flush, 9);
 189     R300_INIT_ATOM(aa_state, 4);
 190     R300_INIT_ATOM(fb_state, 0);
 191     /* ZB (unpipelined), SC. */
 192     R300_INIT_ATOM(hyperz_state, 6);
 193     R300_INIT_ATOM(ztop_state, 2);
 194     /* ZB, FG. */
 195     R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
 196     /* RB3D. */
 197     R300_INIT_ATOM(blend_state, 8);
 198     R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
 199     /* SC. */
 200     R300_INIT_ATOM(scissor_state, 3);
 201     /* GB, FG, GA, SU, SC, RB3D. */
 202     R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0));
 203     /* VAP. */
 204     R300_INIT_ATOM(viewport_state, 9);
 205     R300_INIT_ATOM(pvs_flush, 2);
 206     R300_INIT_ATOM(vap_invariant_state, 9);
 207     R300_INIT_ATOM(vertex_stream_state, 0);
 208     R300_INIT_ATOM(vs_state, 0);
 209     R300_INIT_ATOM(vs_constants, 0);
 210     R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
 211     /* VAP, RS, GA, GB, SU, SC. */
 212     R300_INIT_ATOM(rs_block_state, 0);
 213     R300_INIT_ATOM(rs_state, 0);
 214     /* SC, US. */
 215     R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0));
 216     /* US. */
 217     R300_INIT_ATOM(fs, 0);
 218     R300_INIT_ATOM(fs_rc_constant_state, 0);
 219     R300_INIT_ATOM(fs_constants, 0);
 220     /* TX. */
 221     R300_INIT_ATOM(texture_cache_inval, 2);
 222     R300_INIT_ATOM(textures_state, 0);
 223     /* ZB (unpipelined), SU. */
 224     R300_INIT_ATOM(query_start, 4);
 225
 226     /* Replace emission functions for r500. */
 227     if (is_r500) {
 228         r300->fs.emit = r500_emit_fs;
 229         r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
 230         r300->fs_constants.emit = r500_emit_fs_constants;
 231     }
 232
 233     /* Some non-CSO atoms need explicit space to store the state locally. */
 234     r300->aa_state.state = CALLOC_STRUCT(r300_aa_state);
 235     r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
 236     r300->clip_state.state = CALLOC_STRUCT(r300_clip_state);
 237     r300->fb_state.state = CALLOC_STRUCT(pipe_framebuffer_state);
 238     r300->gpu_flush.state = CALLOC_STRUCT(pipe_framebuffer_state);
 239     r300->hyperz_state.state = CALLOC_STRUCT(r300_hyperz_state);
 240     r300->invariant_state.state = CALLOC_STRUCT(r300_invariant_state);
 241     r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
 242     r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
 243     r300->textures_state.state = CALLOC_STRUCT(r300_textures_state);
 244     r300->vap_invariant_state.state = CALLOC_STRUCT(r300_vap_invariant_state);
 245     r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
 246     r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
 247     r300->fs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
 248     r300->vs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
 249     if (!r300->screen->caps.has_tcl) {
 250         r300->vertex_stream_state.state = CALLOC_STRUCT(r300_vertex_stream_state);
 251     }
 252
 253     /* Some non-CSO atoms don't use the state pointer. */
 254     r300->fb_state_pipelined.allow_null_state = TRUE;
 255     r300->fs_rc_constant_state.allow_null_state = TRUE;
 256     r300->pvs_flush.allow_null_state = TRUE;
 257     r300->query_start.allow_null_state = TRUE;
 258     r300->texture_cache_inval.allow_null_state = TRUE;
 259
 260     /* Some states must be marked as dirty here to properly set up
 261      * hardware in the first command stream. */
 262     r300->invariant_state.dirty = TRUE;
 263     r300->pvs_flush.dirty = TRUE;
 264     r300->vap_invariant_state.dirty = TRUE;
 265     r300->texture_cache_inval.dirty = TRUE;
 266     r300->textures_state.dirty = TRUE;
 267 }
 268
 269 /* Not every state tracker calls every driver function before the first draw
 270  * call and we must initialize the command buffers somehow. */
 271 static void r300_init_states(struct pipe_context *pipe)
 272 {
 273     struct r300_context *r300 = r300_context(pipe);
 274     struct pipe_blend_color bc = {{0}};
 275     struct pipe_clip_state cs = {{{0}}};
 276     struct pipe_scissor_state ss = {0};
 277     struct r300_clip_state *clip =
 278             (struct r300_clip_state*)r300->clip_state.state;
 279     struct r300_gpu_flush *gpuflush =
 280             (struct r300_gpu_flush*)r300->gpu_flush.state;
 281     struct r300_vap_invariant_state *vap_invariant =
 282             (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
 283     struct r300_invariant_state *invariant =
 284             (struct r300_invariant_state*)r300->invariant_state.state;
 285     struct r300_hyperz_state *hyperz =
 286             (struct r300_hyperz_state*)r300->hyperz_state.state;
 287     CB_LOCALS;
 288
 289     pipe->set_blend_color(pipe, &bc);
 290     pipe->set_scissor_state(pipe, &ss);
 291
 292     /* Initialize the clip state. */
 293     if (r300_context(pipe)->screen->caps.has_tcl) {
 294         pipe->set_clip_state(pipe, &cs);
 295     } else {
 296         BEGIN_CB(clip->cb, 2);
 297         OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
 298         END_CB;
 299     }
 300
 301     /* Initialize the GPU flush. */
 302     {
 303         BEGIN_CB(gpuflush->cb_flush_clean, 6);
 304
 305         /* Flush and free renderbuffer caches. */
 306         OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
 307             R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
 308             R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
 309         OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
 310             R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
 311             R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
 312
 313         /* Wait until the GPU is idle.
 314          * This fixes random pixels sometimes appearing probably caused
 315          * by incomplete rendering. */
 316         OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
 317         END_CB;
 318     }
 319
 320     /* Initialize the VAP invariant state. */
 321     {
 322         BEGIN_CB(vap_invariant->cb, 9);
 323         OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
 324         OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
 325         OUT_CB_32F(1.0);
 326         OUT_CB_32F(1.0);
 327         OUT_CB_32F(1.0);
 328         OUT_CB_32F(1.0);
 329         OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
 330         END_CB;
 331     }
 332
 333     /* Initialize the invariant state. */
 334     {
 335         BEGIN_CB(invariant->cb, r300->invariant_state.size);
 336         OUT_CB_REG(R300_GB_SELECT, 0);
 337         OUT_CB_REG(R300_FG_FOG_BLEND, 0);
 338         OUT_CB_REG(R300_GA_ROUND_MODE, 1);
 339         OUT_CB_REG(R300_GA_OFFSET, 0);
 340         OUT_CB_REG(R300_SU_TEX_WRAP, 0);
 341         OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
 342         OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
 343         OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
 344
 345         if (r300->screen->caps.is_rv350) {
 346             OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
 347             OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
 348         }
 349         END_CB;
 350     }
 351
 352     /* Initialize the hyperz state. */
 353     {
 354         BEGIN_CB(&hyperz->cb_begin, r300->hyperz_state.size);
 355         OUT_CB_REG(R300_ZB_BW_CNTL, 0);
 356         OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
 357         OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
 358         END_CB;
 359     }
 360 }
 361
 362 struct pipe_context* r300_create_context(struct pipe_screen* screen,
 363                                          void *priv)
 364 {
 365     struct r300_context* r300 = CALLOC_STRUCT(r300_context);
 366     struct r300_screen* r300screen = r300_screen(screen);
 367     struct r300_winsys_screen *rws = r300screen->rws;
 368
 369     if (!r300)
 370         return NULL;
 371
 372     r300_update_num_contexts(r300screen, 1);
 373
 374     r300->rws = rws;
 375     r300->screen = r300screen;
 376
 377     r300->context.winsys = (struct pipe_winsys*)rws;
 378     r300->context.screen = screen;
 379     r300->context.priv = priv;
 380
 381     r300->context.destroy = r300_destroy_context;
 382
 383     r300->cs = rws->cs_create(rws);
 384
 385     util_mempool_create(&r300->pool_transfers,
 386                         sizeof(struct pipe_transfer), 64,
 387                         UTIL_MEMPOOL_SINGLETHREADED);
 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         /* Enable our renderer. */
 393         draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
 394         /* Enable Draw's clipping. */
 395         draw_set_driver_clipping(r300->draw, FALSE);
 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     }
 400
 401     r300_setup_atoms(r300);
 402
 403     make_empty_list(&r300->query_list);
 404
 405     r300_init_blit_functions(r300);
 406     r300_init_flush_functions(r300);
 407     r300_init_query_functions(r300);
 408     r300_init_state_functions(r300);
 409     r300_init_resource_functions(r300);
 410
 411     r300->blitter = util_blitter_create(&r300->context);
 412
 413     /* Render functions must be initialized after blitter. */
 414     r300_init_render_functions(r300);
 415
 416     rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
 417
 418     r300->upload_ib = u_upload_create(&r300->context,
 419                                       32 * 1024, 16,
 420                                       PIPE_BIND_INDEX_BUFFER);
 421
 422     if (r300->upload_ib == NULL)
 423         goto no_upload_ib;
 424
 425     r300->upload_vb = u_upload_create(&r300->context,
 426                                       128 * 1024, 16,
 427                                       PIPE_BIND_VERTEX_BUFFER);
 428     if (r300->upload_vb == NULL)
 429         goto no_upload_vb;
 430
 431     r300->tran.translate_cache = translate_cache_create();
 432
 433     r300_init_states(&r300->context);
 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.bind = PIPE_BIND_SAMPLER_VIEW;
 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     return &r300->context;
 460
 461  no_upload_ib:
 462     u_upload_destroy(r300->upload_ib);
 463  no_upload_vb:
 464     FREE(r300);
 465     return NULL;
 466 }
 467
 468 void r300_finish(struct r300_context *r300)
 469 {
 470     struct pipe_framebuffer_state *fb;
 471     unsigned i;
 472
 473     /* This is a preliminary implementation of glFinish.
 474      *
 475      * The ideal implementation should use something like EmitIrqLocked and
 476      * WaitIrq, or better, real fences.
 477      */
 478     if (r300->fb_state.state) {
 479         fb = r300->fb_state.state;
 480
 481         for (i = 0; i < fb->nr_cbufs; i++) {
 482             if (fb->cbufs[i]->texture) {
 483                 r300->rws->buffer_wait(r300->rws,
 484                     r300_texture(fb->cbufs[i]->texture)->buffer);
 485                 return;
 486             }
 487         }
 488         if (fb->zsbuf && fb->zsbuf->texture) {
 489             r300->rws->buffer_wait(r300->rws,
 490                 r300_texture(fb->zsbuf->texture)->buffer);
 491         }
 492     }
 493 }