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_release_referenced_objects(struct r300_context *r300)
  40 {
  41     struct pipe_framebuffer_state *fb =
  42             (struct pipe_framebuffer_state*)r300->fb_state.state;
  43     struct r300_textures_state *textures =
  44             (struct r300_textures_state*)r300->textures_state.state;
  45     struct r300_query *query, *temp;
  46     unsigned i;
  47
  48     /* Framebuffer state. */
  49     util_assign_framebuffer_state(fb, NULL);
  50
  51     /* Textures. */
  52     for (i = 0; i < textures->sampler_view_count; i++)
  53         pipe_sampler_view_reference(
  54                 (struct pipe_sampler_view**)&textures->sampler_views[i], NULL);
  55
  56     /* The special dummy texture for texkill. */
  57     if (r300->texkill_sampler) {
  58         pipe_sampler_view_reference(
  59                 (struct pipe_sampler_view**)&r300->texkill_sampler,
  60                 NULL);
  61     }
  62
  63     /* The SWTCL VBO. */
  64     pipe_resource_reference(&r300->vbo, NULL);
  65
  66     /* Vertex buffers. */
  67     for (i = 0; i < r300->vertex_buffer_count; i++) {
  68         pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL);
  69     }
  70
  71     /* If there are any queries pending or not destroyed, remove them now. */
  72     foreach_s(query, temp, &r300->query_list) {
  73         remove_from_list(query);
  74         FREE(query);
  75     }
  76 }
  77
  78 static void r300_destroy_context(struct pipe_context* context)
  79 {
  80     struct r300_context* r300 = r300_context(context);
  81     struct r300_atom *atom;
  82
  83     util_blitter_destroy(r300->blitter);
  84     draw_destroy(r300->draw);
  85
  86     /* Print stats, if enabled. */
  87     if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) {
  88         fprintf(stderr, "r300: Stats for context %p:\n", r300);
  89         fprintf(stderr, "    : Flushes: %" PRIu64 "\n", r300->flush_counter);
  90         foreach(atom, &r300->atom_list) {
  91             fprintf(stderr, "    : %s: %" PRIu64 " emits\n",
  92                 atom->name, atom->counter);
  93         }
  94     }
  95
  96     u_upload_destroy(r300->upload_vb);
  97     u_upload_destroy(r300->upload_ib);
  98
  99     translate_cache_destroy(r300->tran.translate_cache);
 100
 101     r300_release_referenced_objects(r300);
 102
 103     r300->rws->cs_destroy(r300->cs);
 104
 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->scissor_state.state);
 114     FREE(r300->textures_state.state);
 115     FREE(r300->vap_invariant_state.state);
 116     FREE(r300->viewport_state.state);
 117     FREE(r300->ztop_state.state);
 118     FREE(r300->fs_constants.state);
 119     FREE(r300->vs_constants.state);
 120     if (!r300->screen->caps.has_tcl) {
 121         FREE(r300->vertex_stream_state.state);
 122     }
 123     FREE(r300);
 124 }
 125
 126 void r300_flush_cb(void *data)
 127 {
 128     struct r300_context* const cs_context_copy = data;
 129
 130     cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
 131 }
 132
 133 #define R300_INIT_ATOM(atomname, atomsize) \
 134     r300->atomname.name = #atomname; \
 135     r300->atomname.state = NULL; \
 136     r300->atomname.size = atomsize; \
 137     r300->atomname.emit = r300_emit_##atomname; \
 138     r300->atomname.dirty = FALSE; \
 139     insert_at_tail(&r300->atom_list, &r300->atomname);
 140
 141 static void r300_setup_atoms(struct r300_context* r300)
 142 {
 143     boolean is_rv350 = r300->screen->caps.is_rv350;
 144     boolean is_r500 = r300->screen->caps.is_r500;
 145     boolean has_tcl = r300->screen->caps.has_tcl;
 146     boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
 147
 148     /* Create the actual atom list.
 149      *
 150      * Each atom is examined and emitted in the order it appears here, which
 151      * can affect performance and conformance if not handled with care.
 152      *
 153      * Some atoms never change size, others change every emit - those have
 154      * the size of 0 here.
 155      *
 156      * NOTE: The framebuffer state is split into these atoms:
 157      * - gpu_flush          (unpipelined regs)
 158      * - aa_state           (unpipelined regs)
 159      * - fb_state           (unpipelined regs)
 160      * - hyperz_state       (unpipelined regs followed by pipelined ones)
 161      * - fb_state_pipelined (pipelined regs)
 162      * The motivation behind this is to be able to emit a strict
 163      * subset of the regs, and to have reasonable register ordering. */
 164     make_empty_list(&r300->atom_list);
 165     /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */
 166     R300_INIT_ATOM(gpu_flush, 9);
 167     R300_INIT_ATOM(aa_state, 4);
 168     R300_INIT_ATOM(fb_state, 0);
 169     /* ZB (unpipelined), SC. */
 170     R300_INIT_ATOM(hyperz_state, 6);
 171     R300_INIT_ATOM(ztop_state, 2);
 172     /* ZB, FG. */
 173     R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
 174     /* RB3D. */
 175     R300_INIT_ATOM(blend_state, 8);
 176     R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
 177     /* SC. */
 178     R300_INIT_ATOM(scissor_state, 3);
 179     /* GB, FG, GA, SU, SC, RB3D. */
 180     R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0));
 181     /* VAP. */
 182     R300_INIT_ATOM(viewport_state, 9);
 183     R300_INIT_ATOM(pvs_flush, 2);
 184     R300_INIT_ATOM(vap_invariant_state, 9);
 185     R300_INIT_ATOM(vertex_stream_state, 0);
 186     R300_INIT_ATOM(vs_state, 0);
 187     R300_INIT_ATOM(vs_constants, 0);
 188     R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
 189     /* VAP, RS, GA, GB, SU, SC. */
 190     R300_INIT_ATOM(rs_block_state, 0);
 191     R300_INIT_ATOM(rs_state, 0);
 192     /* SC, US. */
 193     R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0));
 194     /* US. */
 195     R300_INIT_ATOM(fs, 0);
 196     R300_INIT_ATOM(fs_rc_constant_state, 0);
 197     R300_INIT_ATOM(fs_constants, 0);
 198     /* TX. */
 199     R300_INIT_ATOM(texture_cache_inval, 2);
 200     R300_INIT_ATOM(textures_state, 0);
 201     /* ZB (unpipelined), SU. */
 202     R300_INIT_ATOM(query_start, 4);
 203
 204     /* Replace emission functions for r500. */
 205     if (is_r500) {
 206         r300->fs.emit = r500_emit_fs;
 207         r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state;
 208         r300->fs_constants.emit = r500_emit_fs_constants;
 209     }
 210
 211     /* Some non-CSO atoms need explicit space to store the state locally. */
 212     r300->aa_state.state = CALLOC_STRUCT(r300_aa_state);
 213     r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
 214     r300->clip_state.state = CALLOC_STRUCT(r300_clip_state);
 215     r300->fb_state.state = CALLOC_STRUCT(pipe_framebuffer_state);
 216     r300->gpu_flush.state = CALLOC_STRUCT(pipe_framebuffer_state);
 217     r300->hyperz_state.state = CALLOC_STRUCT(r300_hyperz_state);
 218     r300->invariant_state.state = CALLOC_STRUCT(r300_invariant_state);
 219     r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
 220     r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
 221     r300->textures_state.state = CALLOC_STRUCT(r300_textures_state);
 222     r300->vap_invariant_state.state = CALLOC_STRUCT(r300_vap_invariant_state);
 223     r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
 224     r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
 225     r300->fs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
 226     r300->vs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
 227     if (!r300->screen->caps.has_tcl) {
 228         r300->vertex_stream_state.state = CALLOC_STRUCT(r300_vertex_stream_state);
 229     }
 230
 231     /* Some non-CSO atoms don't use the state pointer. */
 232     r300->fb_state_pipelined.allow_null_state = TRUE;
 233     r300->fs_rc_constant_state.allow_null_state = TRUE;
 234     r300->pvs_flush.allow_null_state = TRUE;
 235     r300->query_start.allow_null_state = TRUE;
 236     r300->texture_cache_inval.allow_null_state = TRUE;
 237
 238     /* Some states must be marked as dirty here to properly set up
 239      * hardware in the first command stream. */
 240     r300->invariant_state.dirty = TRUE;
 241     r300->pvs_flush.dirty = TRUE;
 242     r300->vap_invariant_state.dirty = TRUE;
 243     r300->texture_cache_inval.dirty = TRUE;
 244     r300->textures_state.dirty = TRUE;
 245 }
 246
 247 /* Not every state tracker calls every driver function before the first draw
 248  * call and we must initialize the command buffers somehow. */
 249 static void r300_init_states(struct pipe_context *pipe)
 250 {
 251     struct r300_context *r300 = r300_context(pipe);
 252     struct pipe_blend_color bc = {{0}};
 253     struct pipe_clip_state cs = {{{0}}};
 254     struct pipe_scissor_state ss = {0};
 255     struct r300_clip_state *clip =
 256             (struct r300_clip_state*)r300->clip_state.state;
 257     struct r300_gpu_flush *gpuflush =
 258             (struct r300_gpu_flush*)r300->gpu_flush.state;
 259     struct r300_vap_invariant_state *vap_invariant =
 260             (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
 261     struct r300_invariant_state *invariant =
 262             (struct r300_invariant_state*)r300->invariant_state.state;
 263     struct r300_hyperz_state *hyperz =
 264             (struct r300_hyperz_state*)r300->hyperz_state.state;
 265     CB_LOCALS;
 266
 267     pipe->set_blend_color(pipe, &bc);
 268     pipe->set_scissor_state(pipe, &ss);
 269
 270     /* Initialize the clip state. */
 271     if (r300_context(pipe)->screen->caps.has_tcl) {
 272         pipe->set_clip_state(pipe, &cs);
 273     } else {
 274         BEGIN_CB(clip->cb, 2);
 275         OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
 276         END_CB;
 277     }
 278
 279     /* Initialize the GPU flush. */
 280     {
 281         BEGIN_CB(gpuflush->cb_flush_clean, 6);
 282
 283         /* Flush and free renderbuffer caches. */
 284         OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT,
 285             R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
 286             R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
 287         OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
 288             R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
 289             R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
 290
 291         /* Wait until the GPU is idle.
 292          * This fixes random pixels sometimes appearing probably caused
 293          * by incomplete rendering. */
 294         OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN);
 295         END_CB;
 296     }
 297
 298     /* Initialize the VAP invariant state. */
 299     {
 300         BEGIN_CB(vap_invariant->cb, 9);
 301         OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
 302         OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
 303         OUT_CB_32F(1.0);
 304         OUT_CB_32F(1.0);
 305         OUT_CB_32F(1.0);
 306         OUT_CB_32F(1.0);
 307         OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
 308         END_CB;
 309     }
 310
 311     /* Initialize the invariant state. */
 312     {
 313         BEGIN_CB(invariant->cb, r300->invariant_state.size);
 314         OUT_CB_REG(R300_GB_SELECT, 0);
 315         OUT_CB_REG(R300_FG_FOG_BLEND, 0);
 316         OUT_CB_REG(R300_GA_ROUND_MODE, 1);
 317         OUT_CB_REG(R300_GA_OFFSET, 0);
 318         OUT_CB_REG(R300_SU_TEX_WRAP, 0);
 319         OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF);
 320         OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0);
 321         OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525);
 322
 323         if (r300->screen->caps.is_rv350) {
 324             OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
 325             OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
 326         }
 327         END_CB;
 328     }
 329
 330     /* Initialize the hyperz state. */
 331     {
 332         BEGIN_CB(&hyperz->cb_begin, r300->hyperz_state.size);
 333         OUT_CB_REG(R300_ZB_BW_CNTL, 0);
 334         OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
 335         OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
 336         END_CB;
 337     }
 338 }
 339
 340 struct pipe_context* r300_create_context(struct pipe_screen* screen,
 341                                          void *priv)
 342 {
 343     struct r300_context* r300 = CALLOC_STRUCT(r300_context);
 344     struct r300_screen* r300screen = r300_screen(screen);
 345     struct r300_winsys_screen *rws = r300screen->rws;
 346
 347     if (!r300)
 348         return NULL;
 349
 350     r300->rws = rws;
 351     r300->screen = r300screen;
 352
 353     r300->context.winsys = (struct pipe_winsys*)rws;
 354     r300->context.screen = screen;
 355     r300->context.priv = priv;
 356
 357     r300->context.destroy = r300_destroy_context;
 358
 359     r300->cs = rws->cs_create(rws);
 360
 361     if (!r300screen->caps.has_tcl) {
 362         /* Create a Draw. This is used for SW TCL. */
 363         r300->draw = draw_create(&r300->context);
 364         /* Enable our renderer. */
 365         draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
 366         /* Enable Draw's clipping. */
 367         draw_set_driver_clipping(r300->draw, FALSE);
 368         /* Disable converting points/lines to triangles. */
 369         draw_wide_line_threshold(r300->draw, 10000000.f);
 370         draw_wide_point_threshold(r300->draw, 10000000.f);
 371     }
 372
 373     r300_setup_atoms(r300);
 374
 375     make_empty_list(&r300->query_list);
 376
 377     r300_init_blit_functions(r300);
 378     r300_init_flush_functions(r300);
 379     r300_init_query_functions(r300);
 380     r300_init_state_functions(r300);
 381     r300_init_resource_functions(r300);
 382
 383     r300->blitter = util_blitter_create(&r300->context);
 384
 385     /* Render functions must be initialized after blitter. */
 386     r300_init_render_functions(r300);
 387
 388     rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
 389
 390     r300->upload_ib = u_upload_create(&r300->context,
 391                                       32 * 1024, 16,
 392                                       PIPE_BIND_INDEX_BUFFER);
 393
 394     if (r300->upload_ib == NULL)
 395         goto no_upload_ib;
 396
 397     r300->upload_vb = u_upload_create(&r300->context,
 398                                       128 * 1024, 16,
 399                                       PIPE_BIND_VERTEX_BUFFER);
 400     if (r300->upload_vb == NULL)
 401         goto no_upload_vb;
 402
 403     r300->tran.translate_cache = translate_cache_create();
 404
 405     r300_init_states(&r300->context);
 406
 407     /* The KIL opcode needs the first texture unit to be enabled
 408      * on r3xx-r4xx. In order to calm down the CS checker, we bind this
 409      * dummy texture there. */
 410     if (!r300->screen->caps.is_r500) {
 411         struct pipe_resource *tex;
 412         struct pipe_resource rtempl = {{0}};
 413         struct pipe_sampler_view vtempl = {{0}};
 414
 415         rtempl.target = PIPE_TEXTURE_2D;
 416         rtempl.format = PIPE_FORMAT_I8_UNORM;
 417         rtempl.bind = PIPE_BIND_SAMPLER_VIEW;
 418         rtempl.width0 = 1;
 419         rtempl.height0 = 1;
 420         rtempl.depth0 = 1;
 421         tex = screen->resource_create(screen, &rtempl);
 422
 423         u_sampler_view_default_template(&vtempl, tex, tex->format);
 424
 425         r300->texkill_sampler = (struct r300_sampler_view*)
 426             r300->context.create_sampler_view(&r300->context, tex, &vtempl);
 427
 428         pipe_resource_reference(&tex, NULL);
 429     }
 430
 431     return &r300->context;
 432
 433  no_upload_ib:
 434     u_upload_destroy(r300->upload_ib);
 435  no_upload_vb:
 436     FREE(r300);
 437     return NULL;
 438 }
 439
 440 void r300_finish(struct r300_context *r300)
 441 {
 442     struct pipe_framebuffer_state *fb;
 443     unsigned i;
 444
 445     /* This is a preliminary implementation of glFinish.
 446      *
 447      * The ideal implementation should use something like EmitIrqLocked and
 448      * WaitIrq, or better, real fences.
 449      */
 450     if (r300->fb_state.state) {
 451         fb = r300->fb_state.state;
 452
 453         for (i = 0; i < fb->nr_cbufs; i++) {
 454             if (fb->cbufs[i]->texture) {
 455                 r300->rws->buffer_wait(r300->rws,
 456                     r300_texture(fb->cbufs[i]->texture)->buffer);
 457                 return;
 458             }
 459         }
 460         if (fb->zsbuf && fb->zsbuf->texture) {
 461             r300->rws->buffer_wait(r300->rws,
 462                 r300_texture(fb->zsbuf->texture)->buffer);
 463         }
 464     }
 465 }