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
  29 #include "main/glheader.h"
  30 #include "main/context.h"
  31 #include "main/condrender.h"
  32 #include "main/samplerobj.h"
  33 #include "main/state.h"
  34 #include "main/enums.h"
  35 #include "tnl/tnl.h"
  36 #include "vbo/vbo_context.h"
  37 #include "swrast/swrast.h"
  38 #include "swrast_setup/swrast_setup.h"
  39
  40 #include "brw_draw.h"
  41 #include "brw_defines.h"
  42 #include "brw_context.h"
  43 #include "brw_state.h"
  44
  45 #include "intel_batchbuffer.h"
  46
  47 #define FILE_DEBUG_FLAG DEBUG_PRIMS
  48
  49 static GLuint prim_to_hw_prim[GL_POLYGON+1] = {
  50    _3DPRIM_POINTLIST,
  51    _3DPRIM_LINELIST,
  52    _3DPRIM_LINELOOP,
  53    _3DPRIM_LINESTRIP,
  54    _3DPRIM_TRILIST,
  55    _3DPRIM_TRISTRIP,
  56    _3DPRIM_TRIFAN,
  57    _3DPRIM_QUADLIST,
  58    _3DPRIM_QUADSTRIP,
  59    _3DPRIM_POLYGON
  60 };
  61
  62
  63 static const GLenum reduced_prim[GL_POLYGON+1] = {
  64    GL_POINTS,
  65    GL_LINES,
  66    GL_LINES,
  67    GL_LINES,
  68    GL_TRIANGLES,
  69    GL_TRIANGLES,
  70    GL_TRIANGLES,
  71    GL_TRIANGLES,
  72    GL_TRIANGLES,
  73    GL_TRIANGLES
  74 };
  75
  76
  77 /* When the primitive changes, set a state bit and re-validate.  Not
  78  * the nicest and would rather deal with this by having all the
  79  * programs be immune to the active primitive (ie. cope with all
  80  * possibilities).  That may not be realistic however.
  81  */
  82 static void brw_set_prim(struct brw_context *brw,
  83                          const struct _mesa_prim *prim)
  84 {
  85    struct gl_context *ctx = &brw->intel.ctx;
  86    uint32_t hw_prim = prim_to_hw_prim[prim->mode];
  87
  88    DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
  89
  90    /* Slight optimization to avoid the GS program when not needed:
  91     */
  92    if (prim->mode == GL_QUAD_STRIP &&
  93        ctx->Light.ShadeModel != GL_FLAT &&
  94        ctx->Polygon.FrontMode == GL_FILL &&
  95        ctx->Polygon.BackMode == GL_FILL)
  96       hw_prim = _3DPRIM_TRISTRIP;
  97
  98    if (prim->mode == GL_QUADS && prim->count == 4 &&
  99        ctx->Light.ShadeModel != GL_FLAT &&
 100        ctx->Polygon.FrontMode == GL_FILL &&
 101        ctx->Polygon.BackMode == GL_FILL) {
 102       hw_prim = _3DPRIM_TRIFAN;
 103    }
 104
 105    if (hw_prim != brw->primitive) {
 106       brw->primitive = hw_prim;
 107       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 108
 109       if (reduced_prim[prim->mode] != brw->intel.reduced_primitive) {
 110          brw->intel.reduced_primitive = reduced_prim[prim->mode];
 111          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 112       }
 113    }
 114 }
 115
 116 static void gen6_set_prim(struct brw_context *brw,
 117                           const struct _mesa_prim *prim)
 118 {
 119    uint32_t hw_prim = prim_to_hw_prim[prim->mode];
 120
 121    DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
 122
 123    if (hw_prim != brw->primitive) {
 124       brw->primitive = hw_prim;
 125       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 126    }
 127 }
 128
 129
 130 static GLuint trim(GLenum prim, GLuint length)
 131 {
 132    if (prim == GL_QUAD_STRIP)
 133       return length > 3 ? (length - length % 2) : 0;
 134    else if (prim == GL_QUADS)
 135       return length - length % 4;
 136    else
 137       return length;
 138 }
 139
 140
 141 static void brw_emit_prim(struct brw_context *brw,
 142                           const struct _mesa_prim *prim,
 143                           uint32_t hw_prim)
 144 {
 145    struct intel_context *intel = &brw->intel;
 146    int verts_per_instance;
 147    int vertex_access_type;
 148    int start_vertex_location;
 149    int base_vertex_location;
 150
 151    DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 152        prim->start, prim->count);
 153
 154    start_vertex_location = prim->start;
 155    base_vertex_location = prim->basevertex;
 156    if (prim->indexed) {
 157       vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
 158       start_vertex_location += brw->ib.start_vertex_offset;
 159       base_vertex_location += brw->vb.start_vertex_bias;
 160    } else {
 161       vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
 162       start_vertex_location += brw->vb.start_vertex_bias;
 163    }
 164
 165    verts_per_instance = trim(prim->mode, prim->count);
 166
 167    /* If nothing to emit, just return. */
 168    if (verts_per_instance == 0)
 169       return;
 170
 171    /* If we're set to always flush, do it before and after the primitive emit.
 172     * We want to catch both missed flushes that hurt instruction/state cache
 173     * and missed flushes of the render cache as it heads to other parts of
 174     * the besides the draw code.
 175     */
 176    if (intel->always_flush_cache) {
 177       intel_batchbuffer_emit_mi_flush(intel);
 178    }
 179
 180    BEGIN_BATCH(6);
 181    OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
 182              hw_prim << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
 183              vertex_access_type);
 184    OUT_BATCH(verts_per_instance);
 185    OUT_BATCH(start_vertex_location);
 186    OUT_BATCH(1); // instance count
 187    OUT_BATCH(0); // start instance location
 188    OUT_BATCH(base_vertex_location);
 189    ADVANCE_BATCH();
 190
 191    intel->batch.need_workaround_flush = true;
 192
 193    if (intel->always_flush_cache) {
 194       intel_batchbuffer_emit_mi_flush(intel);
 195    }
 196 }
 197
 198 static void gen7_emit_prim(struct brw_context *brw,
 199                            const struct _mesa_prim *prim,
 200                            uint32_t hw_prim)
 201 {
 202    struct intel_context *intel = &brw->intel;
 203    int verts_per_instance;
 204    int vertex_access_type;
 205    int start_vertex_location;
 206    int base_vertex_location;
 207
 208    DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 209        prim->start, prim->count);
 210
 211    start_vertex_location = prim->start;
 212    base_vertex_location = prim->basevertex;
 213    if (prim->indexed) {
 214       vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
 215       start_vertex_location += brw->ib.start_vertex_offset;
 216       base_vertex_location += brw->vb.start_vertex_bias;
 217    } else {
 218       vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
 219       start_vertex_location += brw->vb.start_vertex_bias;
 220    }
 221
 222    verts_per_instance = trim(prim->mode, prim->count);
 223
 224    /* If nothing to emit, just return. */
 225    if (verts_per_instance == 0)
 226       return;
 227
 228    /* If we're set to always flush, do it before and after the primitive emit.
 229     * We want to catch both missed flushes that hurt instruction/state cache
 230     * and missed flushes of the render cache as it heads to other parts of
 231     * the besides the draw code.
 232     */
 233    if (intel->always_flush_cache) {
 234       intel_batchbuffer_emit_mi_flush(intel);
 235    }
 236
 237    BEGIN_BATCH(7);
 238    OUT_BATCH(CMD_3D_PRIM << 16 | (7 - 2));
 239    OUT_BATCH(hw_prim | vertex_access_type);
 240    OUT_BATCH(verts_per_instance);
 241    OUT_BATCH(start_vertex_location);
 242    OUT_BATCH(1); // instance count
 243    OUT_BATCH(0); // start instance location
 244    OUT_BATCH(base_vertex_location);
 245    ADVANCE_BATCH();
 246
 247    if (intel->always_flush_cache) {
 248       intel_batchbuffer_emit_mi_flush(intel);
 249    }
 250 }
 251
 252
 253 static void brw_merge_inputs( struct brw_context *brw,
 254                        const struct gl_client_array *arrays[])
 255 {
 256    struct brw_vertex_info old = brw->vb.info;
 257    GLuint i;
 258
 259    for (i = 0; i < brw->vb.nr_buffers; i++) {
 260       drm_intel_bo_unreference(brw->vb.buffers[i].bo);
 261       brw->vb.buffers[i].bo = NULL;
 262    }
 263    brw->vb.nr_buffers = 0;
 264
 265    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 266
 267    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 268       brw->vb.inputs[i].buffer = -1;
 269       brw->vb.inputs[i].glarray = arrays[i];
 270       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 271
 272       if (arrays[i]->StrideB != 0)
 273          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 274             ((i%16) * 2);
 275    }
 276
 277    /* Raise statechanges if input sizes have changed. */
 278    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 279       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 280 }
 281
 282 /* May fail if out of video memory for texture or vbo upload, or on
 283  * fallback conditions.
 284  */
 285 static bool brw_try_draw_prims( struct gl_context *ctx,
 286                                      const struct gl_client_array *arrays[],
 287                                      const struct _mesa_prim *prim,
 288                                      GLuint nr_prims,
 289                                      const struct _mesa_index_buffer *ib,
 290                                      GLuint min_index,
 291                                      GLuint max_index )
 292 {
 293    struct intel_context *intel = intel_context(ctx);
 294    struct brw_context *brw = brw_context(ctx);
 295    bool retval = false;
 296    bool warn = false;
 297    GLuint i;
 298
 299    if (ctx->NewState)
 300       _mesa_update_state( ctx );
 301
 302    /* We have to validate the textures *before* checking for fallbacks;
 303     * otherwise, the software fallback won't be able to rely on the
 304     * texture state, the firstLevel and lastLevel fields won't be
 305     * set in the intel texture object (they'll both be 0), and the
 306     * software fallback will segfault if it attempts to access any
 307     * texture level other than level 0.
 308     */
 309    brw_validate_textures( brw );
 310
 311    /* Bind all inputs, derive varying and size information:
 312     */
 313    brw_merge_inputs( brw, arrays );
 314
 315    brw->ib.ib = ib;
 316    brw->state.dirty.brw |= BRW_NEW_INDICES;
 317
 318    brw->vb.min_index = min_index;
 319    brw->vb.max_index = max_index;
 320    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 321
 322    /* Have to validate state quite late.  Will rebuild tnl_program,
 323     * which depends on varying information.
 324     *
 325     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 326     * so can't access it earlier.
 327     */
 328
 329    intel_prepare_render(intel);
 330
 331    for (i = 0; i < nr_prims; i++) {
 332       int estimated_max_prim_size;
 333
 334       estimated_max_prim_size = 512; /* batchbuffer commands */
 335       estimated_max_prim_size += (BRW_MAX_TEX_UNIT *
 336                                   (sizeof(struct brw_sampler_state) +
 337                                    sizeof(struct gen5_sampler_default_color)));
 338       estimated_max_prim_size += 1024; /* gen6 VS push constants */
 339       estimated_max_prim_size += 1024; /* gen6 WM push constants */
 340       estimated_max_prim_size += 512; /* misc. pad */
 341
 342       /* Flush the batch if it's approaching full, so that we don't wrap while
 343        * we've got validated state that needs to be in the same batch as the
 344        * primitives.
 345        */
 346       intel_batchbuffer_require_space(intel, estimated_max_prim_size, false);
 347
 348       if (intel->gen < 6)
 349          brw_set_prim(brw, &prim[i]);
 350       else
 351          gen6_set_prim(brw, &prim[i]);
 352
 353       /* Note that before the loop, brw->state.dirty.brw was set to != 0, and
 354        * that the state updated in the loop outside of this block is that in
 355        * *_set_prim or intel_batchbuffer_flush(), which only impacts
 356        * brw->state.dirty.brw.
 357        */
 358       if (brw->state.dirty.brw) {
 359          brw_validate_state(brw);
 360
 361          /* Various fallback checks:  */
 362          if (brw->intel.Fallback)
 363             goto out;
 364
 365          /* Check that we can fit our state in with our existing batchbuffer, or
 366           * flush otherwise.
 367           */
 368          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 369                                              brw->state.validated_bo_count)) {
 370             static bool warned;
 371             intel_batchbuffer_flush(intel);
 372
 373             /* Validate the state after we flushed the batch (which would have
 374              * changed the set of dirty state).  If we still fail to
 375              * check_aperture, warn of what's happening, but attempt to continue
 376              * on since it may succeed anyway, and the user would probably rather
 377              * see a failure and a warning than a fallback.
 378              */
 379             brw_validate_state(brw);
 380             if (!warned &&
 381                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 382                                                 brw->state.validated_bo_count)) {
 383                warn = true;
 384                warned = true;
 385             }
 386          }
 387
 388          intel->no_batch_wrap = true;
 389          brw_upload_state(brw);
 390       }
 391
 392       if (intel->gen >= 7)
 393          gen7_emit_prim(brw, &prim[i], brw->primitive);
 394       else
 395          brw_emit_prim(brw, &prim[i], brw->primitive);
 396
 397       intel->no_batch_wrap = false;
 398
 399       retval = true;
 400    }
 401
 402    if (intel->always_flush_batch)
 403       intel_batchbuffer_flush(intel);
 404  out:
 405
 406    brw_state_cache_check_size(brw);
 407
 408    if (warn)
 409       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 410               "available aperture space\n");
 411
 412    if (!retval)
 413       DBG("%s failed\n", __FUNCTION__);
 414
 415    return retval;
 416 }
 417
 418 void brw_draw_prims( struct gl_context *ctx,
 419                      const struct gl_client_array *arrays[],
 420                      const struct _mesa_prim *prim,
 421                      GLuint nr_prims,
 422                      const struct _mesa_index_buffer *ib,
 423                      GLboolean index_bounds_valid,
 424                      GLuint min_index,
 425                      GLuint max_index )
 426 {
 427    bool retval;
 428
 429    if (!_mesa_check_conditional_render(ctx))
 430       return;
 431
 432    if (!vbo_all_varyings_in_vbos(arrays)) {
 433       if (!index_bounds_valid)
 434          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 435
 436       /* Decide if we want to rebase.  If so we end up recursing once
 437        * only into this function.
 438        */
 439       if (min_index != 0 && !vbo_any_varyings_in_vbos(arrays)) {
 440          vbo_rebase_prims(ctx, arrays,
 441                           prim, nr_prims,
 442                           ib, min_index, max_index,
 443                           brw_draw_prims );
 444          return;
 445       }
 446    }
 447
 448    /* Make a first attempt at drawing:
 449     */
 450    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 451
 452    /* Otherwise, we really are out of memory.  Pass the drawing
 453     * command to the software tnl module and which will in turn call
 454     * swrast to do the drawing.
 455     */
 456    if (!retval) {
 457        _swsetup_Wakeup(ctx);
 458        _tnl_wakeup(ctx);
 459       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 460    }
 461
 462 }
 463
 464 void brw_draw_init( struct brw_context *brw )
 465 {
 466    struct gl_context *ctx = &brw->intel.ctx;
 467    struct vbo_context *vbo = vbo_context(ctx);
 468    int i;
 469
 470    /* Register our drawing function:
 471     */
 472    vbo->draw_prims = brw_draw_prims;
 473
 474    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 475       brw->vb.inputs[i].buffer = -1;
 476    brw->vb.nr_buffers = 0;
 477    brw->vb.nr_enabled = 0;
 478 }
 479
 480 void brw_draw_destroy( struct brw_context *brw )
 481 {
 482    int i;
 483
 484    for (i = 0; i < brw->vb.nr_buffers; i++) {
 485       drm_intel_bo_unreference(brw->vb.buffers[i].bo);
 486       brw->vb.buffers[i].bo = NULL;
 487    }
 488    brw->vb.nr_buffers = 0;
 489
 490    for (i = 0; i < brw->vb.nr_enabled; i++) {
 491       brw->vb.enabled[i]->buffer = -1;
 492    }
 493    brw->vb.nr_enabled = 0;
 494
 495    drm_intel_bo_unreference(brw->ib.bo);
 496    brw->ib.bo = NULL;
 497 }