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 GLuint brw_set_prim(struct brw_context *brw,
  83                            const struct _mesa_prim *prim)
  84 {
  85    struct gl_context *ctx = &brw->intel.ctx;
  86    GLenum mode = 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 (mode == GL_QUAD_STRIP &&
  93        ctx->Light.ShadeModel != GL_FLAT &&
  94        ctx->Polygon.FrontMode == GL_FILL &&
  95        ctx->Polygon.BackMode == GL_FILL)
  96       mode = GL_TRIANGLE_STRIP;
  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       mode = GL_TRIANGLE_FAN;
 103    }
 104
 105    if (mode != brw->primitive) {
 106       brw->primitive = mode;
 107       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 108
 109       if (reduced_prim[mode] != brw->intel.reduced_primitive) {
 110          brw->intel.reduced_primitive = reduced_prim[mode];
 111          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 112       }
 113    }
 114
 115    return prim_to_hw_prim[mode];
 116 }
 117
 118
 119 static GLuint trim(GLenum prim, GLuint length)
 120 {
 121    if (prim == GL_QUAD_STRIP)
 122       return length > 3 ? (length - length % 2) : 0;
 123    else if (prim == GL_QUADS)
 124       return length - length % 4;
 125    else
 126       return length;
 127 }
 128
 129
 130 static void brw_emit_prim(struct brw_context *brw,
 131                           const struct _mesa_prim *prim,
 132                           uint32_t hw_prim)
 133 {
 134    struct intel_context *intel = &brw->intel;
 135    int verts_per_instance;
 136    int vertex_access_type;
 137    int start_vertex_location;
 138    int base_vertex_location;
 139
 140    DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 141        prim->start, prim->count);
 142
 143    start_vertex_location = prim->start;
 144    base_vertex_location = prim->basevertex;
 145    if (prim->indexed) {
 146       vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
 147       start_vertex_location += brw->ib.start_vertex_offset;
 148       base_vertex_location += brw->vb.start_vertex_bias;
 149    } else {
 150       vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
 151       start_vertex_location += brw->vb.start_vertex_bias;
 152    }
 153
 154    verts_per_instance = trim(prim->mode, prim->count);
 155
 156    /* If nothing to emit, just return. */
 157    if (verts_per_instance == 0)
 158       return;
 159
 160    /* If we're set to always flush, do it before and after the primitive emit.
 161     * We want to catch both missed flushes that hurt instruction/state cache
 162     * and missed flushes of the render cache as it heads to other parts of
 163     * the besides the draw code.
 164     */
 165    if (intel->always_flush_cache) {
 166       intel_batchbuffer_emit_mi_flush(intel);
 167    }
 168
 169    BEGIN_BATCH(6);
 170    OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
 171              hw_prim << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
 172              vertex_access_type);
 173    OUT_BATCH(verts_per_instance);
 174    OUT_BATCH(start_vertex_location);
 175    OUT_BATCH(1); // instance count
 176    OUT_BATCH(0); // start instance location
 177    OUT_BATCH(base_vertex_location);
 178    ADVANCE_BATCH();
 179
 180    if (intel->always_flush_cache) {
 181       intel_batchbuffer_emit_mi_flush(intel);
 182    }
 183 }
 184
 185 static void gen7_emit_prim(struct brw_context *brw,
 186                            const struct _mesa_prim *prim,
 187                            uint32_t hw_prim)
 188 {
 189    struct intel_context *intel = &brw->intel;
 190    int verts_per_instance;
 191    int vertex_access_type;
 192    int start_vertex_location;
 193    int base_vertex_location;
 194
 195    DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 196        prim->start, prim->count);
 197
 198    start_vertex_location = prim->start;
 199    base_vertex_location = prim->basevertex;
 200    if (prim->indexed) {
 201       vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
 202       start_vertex_location += brw->ib.start_vertex_offset;
 203       base_vertex_location += brw->vb.start_vertex_bias;
 204    } else {
 205       vertex_access_type = GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
 206       start_vertex_location += brw->vb.start_vertex_bias;
 207    }
 208
 209    verts_per_instance = trim(prim->mode, prim->count);
 210
 211    /* If nothing to emit, just return. */
 212    if (verts_per_instance == 0)
 213       return;
 214
 215    /* If we're set to always flush, do it before and after the primitive emit.
 216     * We want to catch both missed flushes that hurt instruction/state cache
 217     * and missed flushes of the render cache as it heads to other parts of
 218     * the besides the draw code.
 219     */
 220    if (intel->always_flush_cache) {
 221       intel_batchbuffer_emit_mi_flush(intel);
 222    }
 223
 224    BEGIN_BATCH(7);
 225    OUT_BATCH(CMD_3D_PRIM << 16 | (7 - 2));
 226    OUT_BATCH(hw_prim | vertex_access_type);
 227    OUT_BATCH(verts_per_instance);
 228    OUT_BATCH(start_vertex_location);
 229    OUT_BATCH(1); // instance count
 230    OUT_BATCH(0); // start instance location
 231    OUT_BATCH(base_vertex_location);
 232    ADVANCE_BATCH();
 233
 234    if (intel->always_flush_cache) {
 235       intel_batchbuffer_emit_mi_flush(intel);
 236    }
 237 }
 238
 239
 240 static void brw_merge_inputs( struct brw_context *brw,
 241                        const struct gl_client_array *arrays[])
 242 {
 243    struct brw_vertex_info old = brw->vb.info;
 244    GLuint i;
 245
 246    for (i = 0; i < brw->vb.nr_buffers; i++) {
 247       drm_intel_bo_unreference(brw->vb.buffers[i].bo);
 248       brw->vb.buffers[i].bo = NULL;
 249    }
 250    brw->vb.nr_buffers = 0;
 251
 252    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 253
 254    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 255       brw->vb.inputs[i].buffer = -1;
 256       brw->vb.inputs[i].glarray = arrays[i];
 257       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 258
 259       if (arrays[i]->StrideB != 0)
 260          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 261             ((i%16) * 2);
 262    }
 263
 264    /* Raise statechanges if input sizes have changed. */
 265    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 266       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 267 }
 268
 269 /* XXX: could split the primitive list to fallback only on the
 270  * non-conformant primitives.
 271  */
 272 static GLboolean check_fallbacks( struct brw_context *brw,
 273                                   const struct _mesa_prim *prim,
 274                                   GLuint nr_prims )
 275 {
 276    struct gl_context *ctx = &brw->intel.ctx;
 277    GLuint i;
 278
 279    /* If we don't require strict OpenGL conformance, never
 280     * use fallbacks.  If we're forcing fallbacks, always
 281     * use fallfacks.
 282     */
 283    if (brw->intel.conformance_mode == 0)
 284       return GL_FALSE;
 285
 286    /* Stipple -- these fallbacks could be resolved with a little
 287     * bit of work?
 288     */
 289    if (ctx->Line.StippleFlag) {
 290       for (i = 0; i < nr_prims; i++) {
 291          /* GS doesn't get enough information to know when to reset
 292           * the stipple counter?!?
 293           */
 294          if (prim[i].mode == GL_LINE_LOOP || prim[i].mode == GL_LINE_STRIP)
 295             return GL_TRUE;
 296
 297          if (prim[i].mode == GL_POLYGON &&
 298              (ctx->Polygon.FrontMode == GL_LINE ||
 299               ctx->Polygon.BackMode == GL_LINE))
 300             return GL_TRUE;
 301       }
 302    }
 303
 304    if (ctx->Point.SmoothFlag) {
 305       for (i = 0; i < nr_prims; i++)
 306          if (prim[i].mode == GL_POINTS)
 307             return GL_TRUE;
 308    }
 309
 310    /* Nothing stopping us from the fast path now */
 311    return GL_FALSE;
 312 }
 313
 314 /* May fail if out of video memory for texture or vbo upload, or on
 315  * fallback conditions.
 316  */
 317 static GLboolean brw_try_draw_prims( struct gl_context *ctx,
 318                                      const struct gl_client_array *arrays[],
 319                                      const struct _mesa_prim *prim,
 320                                      GLuint nr_prims,
 321                                      const struct _mesa_index_buffer *ib,
 322                                      GLuint min_index,
 323                                      GLuint max_index )
 324 {
 325    struct intel_context *intel = intel_context(ctx);
 326    struct brw_context *brw = brw_context(ctx);
 327    GLboolean retval = GL_FALSE;
 328    GLboolean warn = GL_FALSE;
 329    GLuint i;
 330
 331    if (ctx->NewState)
 332       _mesa_update_state( ctx );
 333
 334    /* We have to validate the textures *before* checking for fallbacks;
 335     * otherwise, the software fallback won't be able to rely on the
 336     * texture state, the firstLevel and lastLevel fields won't be
 337     * set in the intel texture object (they'll both be 0), and the
 338     * software fallback will segfault if it attempts to access any
 339     * texture level other than level 0.
 340     */
 341    brw_validate_textures( brw );
 342
 343    if (check_fallbacks(brw, prim, nr_prims))
 344       return GL_FALSE;
 345
 346    /* Bind all inputs, derive varying and size information:
 347     */
 348    brw_merge_inputs( brw, arrays );
 349
 350    brw->ib.ib = ib;
 351    brw->state.dirty.brw |= BRW_NEW_INDICES;
 352
 353    brw->vb.min_index = min_index;
 354    brw->vb.max_index = max_index;
 355    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 356
 357    /* Have to validate state quite late.  Will rebuild tnl_program,
 358     * which depends on varying information.
 359     *
 360     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 361     * so can't access it earlier.
 362     */
 363
 364    intel_prepare_render(intel);
 365
 366    for (i = 0; i < nr_prims; i++) {
 367       uint32_t hw_prim;
 368       int estimated_max_prim_size;
 369
 370       estimated_max_prim_size = 512; /* batchbuffer commands */
 371       estimated_max_prim_size += (BRW_MAX_TEX_UNIT *
 372                                   (sizeof(struct brw_sampler_state) +
 373                                    sizeof(struct gen5_sampler_default_color)));
 374       estimated_max_prim_size += 1024; /* gen6 VS push constants */
 375       estimated_max_prim_size += 1024; /* gen6 WM push constants */
 376       estimated_max_prim_size += 512; /* misc. pad */
 377
 378       /* Flush the batch if it's approaching full, so that we don't wrap while
 379        * we've got validated state that needs to be in the same batch as the
 380        * primitives.
 381        */
 382       intel_batchbuffer_require_space(intel, estimated_max_prim_size, false);
 383
 384       hw_prim = brw_set_prim(brw, &prim[i]);
 385       if (brw->state.dirty.brw) {
 386          brw_validate_state(brw);
 387
 388          /* Various fallback checks:  */
 389          if (brw->intel.Fallback)
 390             goto out;
 391
 392          /* Check that we can fit our state in with our existing batchbuffer, or
 393           * flush otherwise.
 394           */
 395          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 396                                              brw->state.validated_bo_count)) {
 397             static GLboolean warned;
 398             intel_batchbuffer_flush(intel);
 399
 400             /* Validate the state after we flushed the batch (which would have
 401              * changed the set of dirty state).  If we still fail to
 402              * check_aperture, warn of what's happening, but attempt to continue
 403              * on since it may succeed anyway, and the user would probably rather
 404              * see a failure and a warning than a fallback.
 405              */
 406             brw_validate_state(brw);
 407             if (!warned &&
 408                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 409                                                 brw->state.validated_bo_count)) {
 410                warn = GL_TRUE;
 411                warned = GL_TRUE;
 412             }
 413          }
 414
 415          intel->no_batch_wrap = GL_TRUE;
 416          brw_upload_state(brw);
 417       }
 418
 419       if (intel->gen >= 7)
 420          gen7_emit_prim(brw, &prim[i], hw_prim);
 421       else
 422          brw_emit_prim(brw, &prim[i], hw_prim);
 423
 424       intel->no_batch_wrap = GL_FALSE;
 425
 426       retval = GL_TRUE;
 427    }
 428
 429    if (intel->always_flush_batch)
 430       intel_batchbuffer_flush(intel);
 431  out:
 432
 433    brw_state_cache_check_size(brw);
 434
 435    if (warn)
 436       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 437               "available aperture space\n");
 438
 439    if (!retval)
 440       DBG("%s failed\n", __FUNCTION__);
 441
 442    return retval;
 443 }
 444
 445 void brw_draw_prims( struct gl_context *ctx,
 446                      const struct gl_client_array *arrays[],
 447                      const struct _mesa_prim *prim,
 448                      GLuint nr_prims,
 449                      const struct _mesa_index_buffer *ib,
 450                      GLboolean index_bounds_valid,
 451                      GLuint min_index,
 452                      GLuint max_index )
 453 {
 454    GLboolean retval;
 455
 456    if (!_mesa_check_conditional_render(ctx))
 457       return;
 458
 459    if (!vbo_all_varyings_in_vbos(arrays)) {
 460       if (!index_bounds_valid)
 461          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 462
 463       /* Decide if we want to rebase.  If so we end up recursing once
 464        * only into this function.
 465        */
 466       if (min_index != 0 && !vbo_any_varyings_in_vbos(arrays)) {
 467          vbo_rebase_prims(ctx, arrays,
 468                           prim, nr_prims,
 469                           ib, min_index, max_index,
 470                           brw_draw_prims );
 471          return;
 472       }
 473    }
 474
 475    /* Make a first attempt at drawing:
 476     */
 477    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 478
 479    /* Otherwise, we really are out of memory.  Pass the drawing
 480     * command to the software tnl module and which will in turn call
 481     * swrast to do the drawing.
 482     */
 483    if (!retval) {
 484        _swsetup_Wakeup(ctx);
 485       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 486    }
 487
 488 }
 489
 490 void brw_draw_init( struct brw_context *brw )
 491 {
 492    struct gl_context *ctx = &brw->intel.ctx;
 493    struct vbo_context *vbo = vbo_context(ctx);
 494    int i;
 495
 496    /* Register our drawing function:
 497     */
 498    vbo->draw_prims = brw_draw_prims;
 499
 500    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 501       brw->vb.inputs[i].buffer = -1;
 502    brw->vb.nr_buffers = 0;
 503    brw->vb.nr_enabled = 0;
 504 }
 505
 506 void brw_draw_destroy( struct brw_context *brw )
 507 {
 508    int i;
 509
 510    for (i = 0; i < brw->vb.nr_buffers; i++) {
 511       drm_intel_bo_unreference(brw->vb.buffers[i].bo);
 512       brw->vb.buffers[i].bo = NULL;
 513    }
 514    brw->vb.nr_buffers = 0;
 515
 516    for (i = 0; i < brw->vb.nr_enabled; i++) {
 517       brw->vb.enabled[i]->buffer = -1;
 518    }
 519    brw->vb.nr_enabled = 0;
 520
 521    drm_intel_bo_unreference(brw->ib.bo);
 522    brw->ib.bo = NULL;
 523 }