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/state.h"
  32 #include "main/enums.h"
  33 #include "tnl/tnl.h"
  34 #include "vbo/vbo_context.h"
  35 #include "swrast/swrast.h"
  36 #include "swrast_setup/swrast_setup.h"
  37
  38 #include "brw_draw.h"
  39 #include "brw_defines.h"
  40 #include "brw_context.h"
  41 #include "brw_state.h"
  42 #include "brw_fallback.h"
  43
  44 #include "intel_batchbuffer.h"
  45 #include "intel_buffer_objects.h"
  46
  47 #define FILE_DEBUG_FLAG DEBUG_BATCH
  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, GLenum prim)
  83 {
  84    GLcontext *ctx = &brw->intel.ctx;
  85
  86    if (INTEL_DEBUG & DEBUG_PRIMS)
  87       _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
  88
  89    /* Slight optimization to avoid the GS program when not needed:
  90     */
  91    if (prim == GL_QUAD_STRIP &&
  92        ctx->Light.ShadeModel != GL_FLAT &&
  93        ctx->Polygon.FrontMode == GL_FILL &&
  94        ctx->Polygon.BackMode == GL_FILL)
  95       prim = GL_TRIANGLE_STRIP;
  96
  97    if (prim != brw->primitive) {
  98       brw->primitive = prim;
  99       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 100
 101       if (reduced_prim[prim] != brw->intel.reduced_primitive) {
 102          brw->intel.reduced_primitive = reduced_prim[prim];
 103          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 104       }
 105    }
 106
 107    return prim_to_hw_prim[prim];
 108 }
 109
 110
 111 static GLuint trim(GLenum prim, GLuint length)
 112 {
 113    if (prim == GL_QUAD_STRIP)
 114       return length > 3 ? (length - length % 2) : 0;
 115    else if (prim == GL_QUADS)
 116       return length - length % 4;
 117    else
 118       return length;
 119 }
 120
 121
 122 static void brw_emit_prim(struct brw_context *brw,
 123                           const struct _mesa_prim *prim,
 124                           uint32_t hw_prim)
 125 {
 126    struct brw_3d_primitive prim_packet;
 127    struct intel_context *intel = &brw->intel;
 128
 129    if (INTEL_DEBUG & DEBUG_PRIMS)
 130       _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 131                    prim->start, prim->count);
 132
 133    prim_packet.header.opcode = CMD_3D_PRIM;
 134    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 135    prim_packet.header.pad = 0;
 136    prim_packet.header.topology = hw_prim;
 137    prim_packet.header.indexed = prim->indexed;
 138
 139    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 140    prim_packet.start_vert_location = prim->start;
 141    if (prim->indexed)
 142       prim_packet.start_vert_location += brw->ib.start_vertex_offset;
 143    prim_packet.instance_count = 1;
 144    prim_packet.start_instance_location = 0;
 145    prim_packet.base_vert_location = prim->basevertex;
 146
 147    /* Can't wrap here, since we rely on the validated state. */
 148    brw->no_batch_wrap = GL_TRUE;
 149
 150    /* If we're set to always flush, do it before and after the primitive emit.
 151     * We want to catch both missed flushes that hurt instruction/state cache
 152     * and missed flushes of the render cache as it heads to other parts of
 153     * the besides the draw code.
 154     */
 155    if (intel->always_flush_cache) {
 156       BEGIN_BATCH(1, IGNORE_CLIPRECTS);
 157       OUT_BATCH(intel->vtbl.flush_cmd());
 158       ADVANCE_BATCH();
 159    }
 160    if (prim_packet.verts_per_instance) {
 161       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 162                               sizeof(prim_packet), LOOP_CLIPRECTS);
 163    }
 164    if (intel->always_flush_cache) {
 165       BEGIN_BATCH(1, IGNORE_CLIPRECTS);
 166       OUT_BATCH(intel->vtbl.flush_cmd());
 167       ADVANCE_BATCH();
 168    }
 169
 170    brw->no_batch_wrap = GL_FALSE;
 171 }
 172
 173 static void brw_merge_inputs( struct brw_context *brw,
 174                        const struct gl_client_array *arrays[])
 175 {
 176    struct brw_vertex_info old = brw->vb.info;
 177    GLuint i;
 178
 179    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 180       dri_bo_unreference(brw->vb.inputs[i].bo);
 181
 182    memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs));
 183    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 184
 185    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 186       brw->vb.inputs[i].glarray = arrays[i];
 187       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 188
 189       if (arrays[i]->StrideB != 0)
 190          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 191             ((i%16) * 2);
 192    }
 193
 194    /* Raise statechanges if input sizes have changed. */
 195    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 196       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 197 }
 198
 199 /* XXX: could split the primitive list to fallback only on the
 200  * non-conformant primitives.
 201  */
 202 static GLboolean check_fallbacks( struct brw_context *brw,
 203                                   const struct _mesa_prim *prim,
 204                                   GLuint nr_prims )
 205 {
 206    GLcontext *ctx = &brw->intel.ctx;
 207    GLuint i;
 208
 209    /* If we don't require strict OpenGL conformance, never
 210     * use fallbacks.  If we're forcing fallbacks, always
 211     * use fallfacks.
 212     */
 213    if (brw->intel.conformance_mode == 0)
 214       return GL_FALSE;
 215
 216    if (brw->intel.conformance_mode == 2)
 217       return GL_TRUE;
 218
 219    if (ctx->Polygon.SmoothFlag) {
 220       for (i = 0; i < nr_prims; i++)
 221          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 222             return GL_TRUE;
 223    }
 224
 225    /* BRW hardware will do AA lines, but they are non-conformant it
 226     * seems.  TBD whether we keep this fallback:
 227     */
 228    if (ctx->Line.SmoothFlag) {
 229       for (i = 0; i < nr_prims; i++)
 230          if (reduced_prim[prim[i].mode] == GL_LINES)
 231             return GL_TRUE;
 232    }
 233
 234    /* Stipple -- these fallbacks could be resolved with a little
 235     * bit of work?
 236     */
 237    if (ctx->Line.StippleFlag) {
 238       for (i = 0; i < nr_prims; i++) {
 239          /* GS doesn't get enough information to know when to reset
 240           * the stipple counter?!?
 241           */
 242          if (prim[i].mode == GL_LINE_LOOP || prim[i].mode == GL_LINE_STRIP)
 243             return GL_TRUE;
 244
 245          if (prim[i].mode == GL_POLYGON &&
 246              (ctx->Polygon.FrontMode == GL_LINE ||
 247               ctx->Polygon.BackMode == GL_LINE))
 248             return GL_TRUE;
 249       }
 250    }
 251
 252    if (ctx->Point.SmoothFlag) {
 253       for (i = 0; i < nr_prims; i++)
 254          if (prim[i].mode == GL_POINTS)
 255             return GL_TRUE;
 256    }
 257
 258    /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
 259     * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
 260     * as GL_CLAMP_TO_EDGE instead.  If we're using GL_CLAMP, and
 261     * we want strict conformance, force the fallback.
 262     * Right now, we only do this for 2D textures.
 263     */
 264    {
 265       int u;
 266       for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
 267          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[u];
 268          if (texUnit->Enabled) {
 269             if (texUnit->Enabled & TEXTURE_1D_BIT) {
 270                if (texUnit->CurrentTex[TEXTURE_1D_INDEX]->WrapS == GL_CLAMP) {
 271                    return GL_TRUE;
 272                }
 273             }
 274             if (texUnit->Enabled & TEXTURE_2D_BIT) {
 275                if (texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapS == GL_CLAMP ||
 276                    texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapT == GL_CLAMP) {
 277                    return GL_TRUE;
 278                }
 279             }
 280             if (texUnit->Enabled & TEXTURE_3D_BIT) {
 281                if (texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapS == GL_CLAMP ||
 282                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapT == GL_CLAMP ||
 283                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapR == GL_CLAMP) {
 284                    return GL_TRUE;
 285                }
 286             }
 287          }
 288       }
 289    }
 290
 291    /* Nothing stopping us from the fast path now */
 292    return GL_FALSE;
 293 }
 294
 295 /* May fail if out of video memory for texture or vbo upload, or on
 296  * fallback conditions.
 297  */
 298 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 299                                      const struct gl_client_array *arrays[],
 300                                      const struct _mesa_prim *prim,
 301                                      GLuint nr_prims,
 302                                      const struct _mesa_index_buffer *ib,
 303                                      GLuint min_index,
 304                                      GLuint max_index )
 305 {
 306    struct intel_context *intel = intel_context(ctx);
 307    struct brw_context *brw = brw_context(ctx);
 308    GLboolean retval = GL_FALSE;
 309    GLboolean warn = GL_FALSE;
 310    GLboolean first_time = GL_TRUE;
 311    GLuint i;
 312
 313    if (ctx->NewState)
 314       _mesa_update_state( ctx );
 315
 316    /* We have to validate the textures *before* checking for fallbacks;
 317     * otherwise, the software fallback won't be able to rely on the
 318     * texture state, the firstLevel and lastLevel fields won't be
 319     * set in the intel texture object (they'll both be 0), and the
 320     * software fallback will segfault if it attempts to access any
 321     * texture level other than level 0.
 322     */
 323    brw_validate_textures( brw );
 324
 325    if (check_fallbacks(brw, prim, nr_prims))
 326       return GL_FALSE;
 327
 328    /* Bind all inputs, derive varying and size information:
 329     */
 330    brw_merge_inputs( brw, arrays );
 331
 332    brw->ib.ib = ib;
 333    brw->state.dirty.brw |= BRW_NEW_INDICES;
 334
 335    brw->vb.min_index = min_index;
 336    brw->vb.max_index = max_index;
 337    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 338
 339    /* Have to validate state quite late.  Will rebuild tnl_program,
 340     * which depends on varying information.
 341     *
 342     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 343     * so can't access it earlier.
 344     */
 345
 346    LOCK_HARDWARE(intel);
 347
 348    if (!intel->constant_cliprect && intel->driDrawable->numClipRects == 0) {
 349       UNLOCK_HARDWARE(intel);
 350       return GL_TRUE;
 351    }
 352
 353    for (i = 0; i < nr_prims; i++) {
 354       uint32_t hw_prim;
 355
 356       /* Flush the batch if it's approaching full, so that we don't wrap while
 357        * we've got validated state that needs to be in the same batch as the
 358        * primitives.  This fraction is just a guess (minimal full state plus
 359        * a primitive is around 512 bytes), and would be better if we had
 360        * an upper bound of how much we might emit in a single
 361        * brw_try_draw_prims().
 362        */
 363       intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4,
 364                                       LOOP_CLIPRECTS);
 365
 366       hw_prim = brw_set_prim(brw, prim[i].mode);
 367
 368       if (first_time || (brw->state.dirty.brw & BRW_NEW_PRIMITIVE)) {
 369          first_time = GL_FALSE;
 370
 371          brw_validate_state(brw);
 372
 373          /* Various fallback checks:  */
 374          if (brw->intel.Fallback)
 375             goto out;
 376
 377          /* Check that we can fit our state in with our existing batchbuffer, or
 378           * flush otherwise.
 379           */
 380          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 381                                              brw->state.validated_bo_count)) {
 382             static GLboolean warned;
 383             intel_batchbuffer_flush(intel->batch);
 384
 385             /* Validate the state after we flushed the batch (which would have
 386              * changed the set of dirty state).  If we still fail to
 387              * check_aperture, warn of what's happening, but attempt to continue
 388              * on since it may succeed anyway, and the user would probably rather
 389              * see a failure and a warning than a fallback.
 390              */
 391             brw_validate_state(brw);
 392             if (!warned &&
 393                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 394                                                 brw->state.validated_bo_count)) {
 395                warn = GL_TRUE;
 396                warned = GL_TRUE;
 397             }
 398          }
 399
 400          brw_upload_state(brw);
 401       }
 402
 403       brw_emit_prim(brw, &prim[i], hw_prim);
 404
 405       retval = GL_TRUE;
 406    }
 407
 408    if (intel->always_flush_batch)
 409       intel_batchbuffer_flush(intel->batch);
 410  out:
 411    UNLOCK_HARDWARE(intel);
 412
 413    brw_state_cache_check_size(brw);
 414
 415    if (warn)
 416       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 417               "available aperture space\n");
 418
 419    if (!retval)
 420       DBG("%s failed\n", __FUNCTION__);
 421
 422    return retval;
 423 }
 424
 425 void brw_draw_prims( GLcontext *ctx,
 426                      const struct gl_client_array *arrays[],
 427                      const struct _mesa_prim *prim,
 428                      GLuint nr_prims,
 429                      const struct _mesa_index_buffer *ib,
 430                      GLboolean index_bounds_valid,
 431                      GLuint min_index,
 432                      GLuint max_index )
 433 {
 434    GLboolean retval;
 435
 436    if (!vbo_all_varyings_in_vbos(arrays)) {
 437       if (!index_bounds_valid)
 438          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 439
 440       /* Decide if we want to rebase.  If so we end up recursing once
 441        * only into this function.
 442        */
 443       if (min_index != 0) {
 444          vbo_rebase_prims(ctx, arrays,
 445                           prim, nr_prims,
 446                           ib, min_index, max_index,
 447                           brw_draw_prims );
 448          return;
 449       }
 450    }
 451
 452    /* Make a first attempt at drawing:
 453     */
 454    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 455
 456    /* Otherwise, we really are out of memory.  Pass the drawing
 457     * command to the software tnl module and which will in turn call
 458     * swrast to do the drawing.
 459     */
 460    if (!retval) {
 461        _swsetup_Wakeup(ctx);
 462       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 463    }
 464
 465 }
 466
 467 void brw_draw_init( struct brw_context *brw )
 468 {
 469    GLcontext *ctx = &brw->intel.ctx;
 470    struct vbo_context *vbo = vbo_context(ctx);
 471
 472    /* Register our drawing function:
 473     */
 474    vbo->draw_prims = brw_draw_prims;
 475 }
 476
 477 void brw_draw_destroy( struct brw_context *brw )
 478 {
 479    int i;
 480
 481    if (brw->vb.upload.bo != NULL) {
 482       dri_bo_unreference(brw->vb.upload.bo);
 483       brw->vb.upload.bo = NULL;
 484    }
 485
 486    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 487       dri_bo_unreference(brw->vb.inputs[i].bo);
 488       brw->vb.inputs[i].bo = NULL;
 489    }
 490
 491    dri_bo_unreference(brw->ib.bo);
 492    brw->ib.bo = NULL;
 493 }