src/mesa/drivers/dri/i965/brw_draw.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include <sys/errno.h>
  29
  30 #include "main/glheader.h"
  31 #include "main/context.h"
  32 #include "main/condrender.h"
  33 #include "main/samplerobj.h"
  34 #include "main/state.h"
  35 #include "main/enums.h"
  36 #include "tnl/tnl.h"
  37 #include "vbo/vbo_context.h"
  38 #include "swrast/swrast.h"
  39 #include "swrast_setup/swrast_setup.h"
  40
  41 #include "brw_draw.h"
  42 #include "brw_defines.h"
  43 #include "brw_context.h"
  44 #include "brw_state.h"
  45
  46 #include "intel_batchbuffer.h"
  47
  48 #define FILE_DEBUG_FLAG DEBUG_PRIMS
  49
  50 static GLuint prim_to_hw_prim[GL_POLYGON+1] = {
  51    _3DPRIM_POINTLIST,
  52    _3DPRIM_LINELIST,
  53    _3DPRIM_LINELOOP,
  54    _3DPRIM_LINESTRIP,
  55    _3DPRIM_TRILIST,
  56    _3DPRIM_TRISTRIP,
  57    _3DPRIM_TRIFAN,
  58    _3DPRIM_QUADLIST,
  59    _3DPRIM_QUADSTRIP,
  60    _3DPRIM_POLYGON
  61 };
  62
  63
  64 static const GLenum reduced_prim[GL_POLYGON+1] = {
  65    GL_POINTS,
  66    GL_LINES,
  67    GL_LINES,
  68    GL_LINES,
  69    GL_TRIANGLES,
  70    GL_TRIANGLES,
  71    GL_TRIANGLES,
  72    GL_TRIANGLES,
  73    GL_TRIANGLES,
  74    GL_TRIANGLES
  75 };
  76
  77
  78 /* When the primitive changes, set a state bit and re-validate.  Not
  79  * the nicest and would rather deal with this by having all the
  80  * programs be immune to the active primitive (ie. cope with all
  81  * possibilities).  That may not be realistic however.
  82  */
  83 static void brw_set_prim(struct brw_context *brw,
  84                          const struct _mesa_prim *prim)
  85 {
  86    struct gl_context *ctx = &brw->intel.ctx;
  87    uint32_t hw_prim = prim_to_hw_prim[prim->mode];
  88
  89    DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
  90
  91    /* Slight optimization to avoid the GS program when not needed:
  92     */
  93    if (prim->mode == GL_QUAD_STRIP &&
  94        ctx->Light.ShadeModel != GL_FLAT &&
  95        ctx->Polygon.FrontMode == GL_FILL &&
  96        ctx->Polygon.BackMode == GL_FILL)
  97       hw_prim = _3DPRIM_TRISTRIP;
  98
  99    if (prim->mode == GL_QUADS && prim->count == 4 &&
 100        ctx->Light.ShadeModel != GL_FLAT &&
 101        ctx->Polygon.FrontMode == GL_FILL &&
 102        ctx->Polygon.BackMode == GL_FILL) {
 103       hw_prim = _3DPRIM_TRIFAN;
 104    }
 105
 106    if (hw_prim != brw->primitive) {
 107       brw->primitive = hw_prim;
 108       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 109
 110       if (reduced_prim[prim->mode] != brw->intel.reduced_primitive) {
 111          brw->intel.reduced_primitive = reduced_prim[prim->mode];
 112          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 113       }
 114    }
 115 }
 116
 117 static void gen6_set_prim(struct brw_context *brw,
 118                           const struct _mesa_prim *prim)
 119 {
 120    uint32_t hw_prim = prim_to_hw_prim[prim->mode];
 121
 122    DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
 123
 124    if (hw_prim != brw->primitive) {
 125       brw->primitive = hw_prim;
 126       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 127    }
 128 }
 129
 130
 131 static GLuint trim(GLenum prim, GLuint length)
 132 {
 133    if (prim == GL_QUAD_STRIP)
 134       return length > 3 ? (length - length % 2) : 0;
 135    else if (prim == GL_QUADS)
 136       return length - length % 4;
 137    else
 138       return length;
 139 }
 140
 141
 142 static void brw_emit_prim(struct brw_context *brw,
 143                           const struct _mesa_prim *prim,
 144                           uint32_t hw_prim)
 145 {
 146    struct intel_context *intel = &brw->intel;
 147    int verts_per_instance;
 148    int vertex_access_type;
 149    int start_vertex_location;
 150    int base_vertex_location;
 151
 152    DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 153        prim->start, prim->count);
 154
 155    start_vertex_location = prim->start;
 156    base_vertex_location = prim->basevertex;
 157    if (prim->indexed) {
 158       vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
 159       start_vertex_location += brw->ib.start_vertex_offset;
 160       base_vertex_location += brw->vb.start_vertex_bias;
 161    } else {
 162       vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
 163       start_vertex_location += brw->vb.start_vertex_bias;
 164    }
 165
 166    verts_per_instance = trim(prim->mode, prim->count);
 167
 168    /* If nothing to emit, just return. */
 169    if (verts_per_instance == 0)
 170       return;
 171
 172    /* If we're set to always flush, do it before and after the primitive emit.
 173     * We want to catch both missed flushes that hurt instruction/state cache
 174     * and missed flushes of the render cache as it heads to other parts of
 175     * the besides the draw code.
 176     */
 177    if (intel->always_flush_cache) {
 178       intel_batchbuffer_emit_mi_flush(intel);
 179    }
 180
 181    BEGIN_BATCH(6);
 182    OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
 183              hw_prim << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
 184              vertex_access_type);
 185    OUT_BATCH(verts_per_instance);
 186    OUT_BATCH(start_vertex_location);
 187    OUT_BATCH(1); // instance count
 188    OUT_BATCH(0); // start instance location
 189    OUT_BATCH(base_vertex_location);
 190    ADVANCE_BATCH();
 191
 192    intel->batch.need_workaround_flush = true;
 193
 194    if (intel->always_flush_cache) {
 195       intel_batchbuffer_emit_mi_flush(intel);
 196    }
 197 }
 198
 199 static void gen7_emit_prim(struct brw_context *brw,
 200                            const struct _mesa_prim *prim,
 201                            uint32_t hw_prim)
 202 {
 203    struct intel_context *intel = &brw->intel;
 204    int verts_per_instance;
 205    int vertex_access_type;
 206    int start_vertex_location;
 207    int base_vertex_location;
 208
 209    DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 210        prim->start, prim->count);
 211
 212    start_vertex_location = prim->start;
 213    base_vertex_location = prim->basevertex;
 214    if (prim->indexed) {
 215       vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
 216       start_vertex_location += brw->ib.start_vertex_offset;
 217       base_vertex_location += brw->vb.start_vertex_bias;
 218    } else {
 219       vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
 220       start_vertex_location += brw->vb.start_vertex_bias;
 221    }
 222
 223    verts_per_instance = trim(prim->mode, prim->count);
 224
 225    /* If nothing to emit, just return. */
 226    if (verts_per_instance == 0)
 227       return;
 228
 229    /* If we're set to always flush, do it before and after the primitive emit.
 230     * We want to catch both missed flushes that hurt instruction/state cache
 231     * and missed flushes of the render cache as it heads to other parts of
 232     * the besides the draw code.
 233     */
 234    if (intel->always_flush_cache) {
 235       intel_batchbuffer_emit_mi_flush(intel);
 236    }
 237
 238    BEGIN_BATCH(7);
 239    OUT_BATCH(CMD_3D_PRIM << 16 | (7 - 2));
 240    OUT_BATCH(hw_prim | vertex_access_type);
 241    OUT_BATCH(verts_per_instance);
 242    OUT_BATCH(start_vertex_location);
 243    OUT_BATCH(1); // instance count
 244    OUT_BATCH(0); // start instance location
 245    OUT_BATCH(base_vertex_location);
 246    ADVANCE_BATCH();
 247
 248    if (intel->always_flush_cache) {
 249       intel_batchbuffer_emit_mi_flush(intel);
 250    }
 251 }
 252
 253
 254 static void brw_merge_inputs( struct brw_context *brw,
 255                        const struct gl_client_array *arrays[])
 256 {
 257    struct brw_vertex_info old = brw->vb.info;
 258    GLuint i;
 259
 260    for (i = 0; i < brw->vb.nr_buffers; i++) {
 261       drm_intel_bo_unreference(brw->vb.buffers[i].bo);
 262       brw->vb.buffers[i].bo = NULL;
 263    }
 264    brw->vb.nr_buffers = 0;
 265
 266    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 267
 268    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 269       brw->vb.inputs[i].buffer = -1;
 270       brw->vb.inputs[i].glarray = arrays[i];
 271       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 272
 273       if (arrays[i]->StrideB != 0)
 274          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 275             ((i%16) * 2);
 276    }
 277
 278    /* Raise statechanges if input sizes have changed. */
 279    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 280       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 281 }
 282
 283 /* May fail if out of video memory for texture or vbo upload, or on
 284  * fallback conditions.
 285  */
 286 static bool brw_try_draw_prims( struct gl_context *ctx,
 287                                      const struct gl_client_array *arrays[],
 288                                      const struct _mesa_prim *prim,
 289                                      GLuint nr_prims,
 290                                      const struct _mesa_index_buffer *ib,
 291                                      GLuint min_index,
 292                                      GLuint max_index )
 293 {
 294    struct intel_context *intel = intel_context(ctx);
 295    struct brw_context *brw = brw_context(ctx);
 296    bool retval = true;
 297    GLuint i;
 298    bool fail_next = false;
 299
 300    if (ctx->NewState)
 301       _mesa_update_state( ctx );
 302
 303    /* We have to validate the textures *before* checking for fallbacks;
 304     * otherwise, the software fallback won't be able to rely on the
 305     * texture state, the firstLevel and lastLevel fields won't be
 306     * set in the intel texture object (they'll both be 0), and the
 307     * software fallback will segfault if it attempts to access any
 308     * texture level other than level 0.
 309     */
 310    brw_validate_textures( brw );
 311
 312    /* Bind all inputs, derive varying and size information:
 313     */
 314    brw_merge_inputs( brw, arrays );
 315
 316    brw->ib.ib = ib;
 317    brw->state.dirty.brw |= BRW_NEW_INDICES;
 318
 319    brw->vb.min_index = min_index;
 320    brw->vb.max_index = max_index;
 321    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 322
 323    /* Have to validate state quite late.  Will rebuild tnl_program,
 324     * which depends on varying information.
 325     *
 326     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 327     * so can't access it earlier.
 328     */
 329
 330    intel_prepare_render(intel);
 331
 332    for (i = 0; i < nr_prims; i++) {
 333       int estimated_max_prim_size;
 334
 335       estimated_max_prim_size = 512; /* batchbuffer commands */
 336       estimated_max_prim_size += (BRW_MAX_TEX_UNIT *
 337                                   (sizeof(struct brw_sampler_state) +
 338                                    sizeof(struct gen5_sampler_default_color)));
 339       estimated_max_prim_size += 1024; /* gen6 VS push constants */
 340       estimated_max_prim_size += 1024; /* gen6 WM push constants */
 341       estimated_max_prim_size += 512; /* misc. pad */
 342
 343       /* Flush the batch if it's approaching full, so that we don't wrap while
 344        * we've got validated state that needs to be in the same batch as the
 345        * primitives.
 346        */
 347       intel_batchbuffer_require_space(intel, estimated_max_prim_size, false);
 348       intel_batchbuffer_save_state(intel);
 349
 350       if (intel->gen < 6)
 351          brw_set_prim(brw, &prim[i]);
 352       else
 353          gen6_set_prim(brw, &prim[i]);
 354
 355 retry:
 356       /* Note that before the loop, brw->state.dirty.brw was set to != 0, and
 357        * that the state updated in the loop outside of this block is that in
 358        * *_set_prim or intel_batchbuffer_flush(), which only impacts
 359        * brw->state.dirty.brw.
 360        */
 361       if (brw->state.dirty.brw) {
 362          intel->no_batch_wrap = true;
 363          brw_upload_state(brw);
 364
 365          if (unlikely(brw->intel.Fallback)) {
 366             intel->no_batch_wrap = false;
 367             retval = false;
 368             goto out;
 369          }
 370       }
 371
 372       if (intel->gen >= 7)
 373          gen7_emit_prim(brw, &prim[i], brw->primitive);
 374       else
 375          brw_emit_prim(brw, &prim[i], brw->primitive);
 376
 377       intel->no_batch_wrap = false;
 378
 379       if (dri_bufmgr_check_aperture_space(&intel->batch.bo, 1)) {
 380          if (!fail_next) {
 381             intel_batchbuffer_reset_to_saved(intel);
 382             intel_batchbuffer_flush(intel);
 383             fail_next = true;
 384             goto retry;
 385          } else {
 386             if (intel_batchbuffer_flush(intel) == -ENOSPC) {
 387                static bool warned = false;
 388
 389                if (!warned) {
 390                   fprintf(stderr, "i965: Single primitive emit exceeded"
 391                           "available aperture space\n");
 392                   warned = true;
 393                }
 394
 395                retval = false;
 396             }
 397          }
 398       }
 399    }
 400
 401    if (intel->always_flush_batch)
 402       intel_batchbuffer_flush(intel);
 403  out:
 404
 405    brw_state_cache_check_size(brw);
 406
 407    return retval;
 408 }
 409
 410 void brw_draw_prims( struct gl_context *ctx,
 411                      const struct gl_client_array *arrays[],
 412                      const struct _mesa_prim *prim,
 413                      GLuint nr_prims,
 414                      const struct _mesa_index_buffer *ib,
 415                      GLboolean index_bounds_valid,
 416                      GLuint min_index,
 417                      GLuint max_index )
 418 {
 419    bool retval;
 420
 421    if (!_mesa_check_conditional_render(ctx))
 422       return;
 423
 424    if (!vbo_all_varyings_in_vbos(arrays)) {
 425       if (!index_bounds_valid)
 426          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 427
 428       /* Decide if we want to rebase.  If so we end up recursing once
 429        * only into this function.
 430        */
 431       if (min_index != 0 && !vbo_any_varyings_in_vbos(arrays)) {
 432          vbo_rebase_prims(ctx, arrays,
 433                           prim, nr_prims,
 434                           ib, min_index, max_index,
 435                           brw_draw_prims );
 436          return;
 437       }
 438    }
 439
 440    /* Make a first attempt at drawing:
 441     */
 442    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 443
 444    /* Otherwise, we really are out of memory.  Pass the drawing
 445     * command to the software tnl module and which will in turn call
 446     * swrast to do the drawing.
 447     */
 448    if (!retval) {
 449        _swsetup_Wakeup(ctx);
 450        _tnl_wakeup(ctx);
 451       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 452    }
 453
 454 }
 455
 456 void brw_draw_init( struct brw_context *brw )
 457 {
 458    struct gl_context *ctx = &brw->intel.ctx;
 459    struct vbo_context *vbo = vbo_context(ctx);
 460    int i;
 461
 462    /* Register our drawing function:
 463     */
 464    vbo->draw_prims = brw_draw_prims;
 465
 466    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 467       brw->vb.inputs[i].buffer = -1;
 468    brw->vb.nr_buffers = 0;
 469    brw->vb.nr_enabled = 0;
 470 }
 471
 472 void brw_draw_destroy( struct brw_context *brw )
 473 {
 474    int i;
 475
 476    for (i = 0; i < brw->vb.nr_buffers; i++) {
 477       drm_intel_bo_unreference(brw->vb.buffers[i].bo);
 478       brw->vb.buffers[i].bo = NULL;
 479    }
 480    brw->vb.nr_buffers = 0;
 481
 482    for (i = 0; i < brw->vb.nr_enabled; i++) {
 483       brw->vb.enabled[i]->buffer = -1;
 484    }
 485    brw->vb.nr_enabled = 0;
 486
 487    drm_intel_bo_unreference(brw->ib.bo);
 488    brw->ib.bo = NULL;
 489 }