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
  43 #include "intel_batchbuffer.h"
  44
  45 #define FILE_DEBUG_FLAG DEBUG_BATCH
  46
  47 static GLuint prim_to_hw_prim[GL_POLYGON+1] = {
  48    _3DPRIM_POINTLIST,
  49    _3DPRIM_LINELIST,
  50    _3DPRIM_LINELOOP,
  51    _3DPRIM_LINESTRIP,
  52    _3DPRIM_TRILIST,
  53    _3DPRIM_TRISTRIP,
  54    _3DPRIM_TRIFAN,
  55    _3DPRIM_QUADLIST,
  56    _3DPRIM_QUADSTRIP,
  57    _3DPRIM_POLYGON
  58 };
  59
  60
  61 static const GLenum reduced_prim[GL_POLYGON+1] = {
  62    GL_POINTS,
  63    GL_LINES,
  64    GL_LINES,
  65    GL_LINES,
  66    GL_TRIANGLES,
  67    GL_TRIANGLES,
  68    GL_TRIANGLES,
  69    GL_TRIANGLES,
  70    GL_TRIANGLES,
  71    GL_TRIANGLES
  72 };
  73
  74
  75 /* When the primitive changes, set a state bit and re-validate.  Not
  76  * the nicest and would rather deal with this by having all the
  77  * programs be immune to the active primitive (ie. cope with all
  78  * possibilities).  That may not be realistic however.
  79  */
  80 static GLuint brw_set_prim(struct brw_context *brw,
  81                            const struct _mesa_prim *prim)
  82 {
  83    GLcontext *ctx = &brw->intel.ctx;
  84    GLenum mode = prim->mode;
  85
  86    if (INTEL_DEBUG & DEBUG_PRIMS)
  87       printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
  88
  89    /* Slight optimization to avoid the GS program when not needed:
  90     */
  91    if (mode == GL_QUAD_STRIP &&
  92        ctx->Light.ShadeModel != GL_FLAT &&
  93        ctx->Polygon.FrontMode == GL_FILL &&
  94        ctx->Polygon.BackMode == GL_FILL)
  95       mode = GL_TRIANGLE_STRIP;
  96
  97    if (prim->mode == GL_QUADS && prim->count == 4 &&
  98        ctx->Light.ShadeModel != GL_FLAT &&
  99        ctx->Polygon.FrontMode == GL_FILL &&
 100        ctx->Polygon.BackMode == GL_FILL) {
 101       mode = GL_TRIANGLE_FAN;
 102    }
 103
 104    if (mode != brw->primitive) {
 105       brw->primitive = mode;
 106       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 107
 108       if (reduced_prim[mode] != brw->intel.reduced_primitive) {
 109          brw->intel.reduced_primitive = reduced_prim[mode];
 110          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 111       }
 112    }
 113
 114    return prim_to_hw_prim[mode];
 115 }
 116
 117
 118 static GLuint trim(GLenum prim, GLuint length)
 119 {
 120    if (prim == GL_QUAD_STRIP)
 121       return length > 3 ? (length - length % 2) : 0;
 122    else if (prim == GL_QUADS)
 123       return length - length % 4;
 124    else
 125       return length;
 126 }
 127
 128
 129 static void brw_emit_prim(struct brw_context *brw,
 130                           const struct _mesa_prim *prim,
 131                           uint32_t hw_prim)
 132 {
 133    struct brw_3d_primitive prim_packet;
 134    struct intel_context *intel = &brw->intel;
 135
 136    if (INTEL_DEBUG & DEBUG_PRIMS)
 137       printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 138                    prim->start, prim->count);
 139
 140    prim_packet.header.opcode = CMD_3D_PRIM;
 141    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 142    prim_packet.header.pad = 0;
 143    prim_packet.header.topology = hw_prim;
 144    prim_packet.header.indexed = prim->indexed;
 145
 146    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 147    prim_packet.start_vert_location = prim->start;
 148    if (prim->indexed)
 149       prim_packet.start_vert_location += brw->ib.start_vertex_offset;
 150    prim_packet.instance_count = 1;
 151    prim_packet.start_instance_location = 0;
 152    prim_packet.base_vert_location = prim->basevertex;
 153
 154    /* Can't wrap here, since we rely on the validated state. */
 155    intel->no_batch_wrap = GL_TRUE;
 156
 157    /* If we're set to always flush, do it before and after the primitive emit.
 158     * We want to catch both missed flushes that hurt instruction/state cache
 159     * and missed flushes of the render cache as it heads to other parts of
 160     * the besides the draw code.
 161     */
 162    if (intel->always_flush_cache) {
 163       intel_batchbuffer_emit_mi_flush(intel->batch);
 164    }
 165    if (prim_packet.verts_per_instance) {
 166       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 167                               sizeof(prim_packet));
 168    }
 169    if (intel->always_flush_cache) {
 170       intel_batchbuffer_emit_mi_flush(intel->batch);
 171    }
 172
 173    intel->no_batch_wrap = GL_FALSE;
 174 }
 175
 176 static void brw_merge_inputs( struct brw_context *brw,
 177                        const struct gl_client_array *arrays[])
 178 {
 179    struct brw_vertex_info old = brw->vb.info;
 180    GLuint i;
 181
 182    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 183       dri_bo_unreference(brw->vb.inputs[i].bo);
 184
 185    memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs));
 186    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 187
 188    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 189       brw->vb.inputs[i].glarray = arrays[i];
 190       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 191
 192       if (arrays[i]->StrideB != 0)
 193          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 194             ((i%16) * 2);
 195    }
 196
 197    /* Raise statechanges if input sizes have changed. */
 198    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 199       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 200 }
 201
 202 /* XXX: could split the primitive list to fallback only on the
 203  * non-conformant primitives.
 204  */
 205 static GLboolean check_fallbacks( struct brw_context *brw,
 206                                   const struct _mesa_prim *prim,
 207                                   GLuint nr_prims )
 208 {
 209    GLcontext *ctx = &brw->intel.ctx;
 210    GLuint i;
 211
 212    /* If we don't require strict OpenGL conformance, never
 213     * use fallbacks.  If we're forcing fallbacks, always
 214     * use fallfacks.
 215     */
 216    if (brw->intel.conformance_mode == 0)
 217       return GL_FALSE;
 218
 219    if (brw->intel.conformance_mode == 2)
 220       return GL_TRUE;
 221
 222    if (ctx->Polygon.SmoothFlag) {
 223       for (i = 0; i < nr_prims; i++)
 224          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 225             return GL_TRUE;
 226    }
 227
 228    /* BRW hardware will do AA lines, but they are non-conformant it
 229     * seems.  TBD whether we keep this fallback:
 230     */
 231    if (ctx->Line.SmoothFlag) {
 232       for (i = 0; i < nr_prims; i++)
 233          if (reduced_prim[prim[i].mode] == GL_LINES)
 234             return GL_TRUE;
 235    }
 236
 237    /* Stipple -- these fallbacks could be resolved with a little
 238     * bit of work?
 239     */
 240    if (ctx->Line.StippleFlag) {
 241       for (i = 0; i < nr_prims; i++) {
 242          /* GS doesn't get enough information to know when to reset
 243           * the stipple counter?!?
 244           */
 245          if (prim[i].mode == GL_LINE_LOOP || prim[i].mode == GL_LINE_STRIP)
 246             return GL_TRUE;
 247
 248          if (prim[i].mode == GL_POLYGON &&
 249              (ctx->Polygon.FrontMode == GL_LINE ||
 250               ctx->Polygon.BackMode == GL_LINE))
 251             return GL_TRUE;
 252       }
 253    }
 254
 255    if (ctx->Point.SmoothFlag) {
 256       for (i = 0; i < nr_prims; i++)
 257          if (prim[i].mode == GL_POINTS)
 258             return GL_TRUE;
 259    }
 260
 261    /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
 262     * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
 263     * as GL_CLAMP_TO_EDGE instead.  If we're using GL_CLAMP, and
 264     * we want strict conformance, force the fallback.
 265     * Right now, we only do this for 2D textures.
 266     */
 267    {
 268       int u;
 269       for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
 270          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[u];
 271          if (texUnit->Enabled) {
 272             if (texUnit->Enabled & TEXTURE_1D_BIT) {
 273                if (texUnit->CurrentTex[TEXTURE_1D_INDEX]->WrapS == GL_CLAMP) {
 274                    return GL_TRUE;
 275                }
 276             }
 277             if (texUnit->Enabled & TEXTURE_2D_BIT) {
 278                if (texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapS == GL_CLAMP ||
 279                    texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapT == GL_CLAMP) {
 280                    return GL_TRUE;
 281                }
 282             }
 283             if (texUnit->Enabled & TEXTURE_3D_BIT) {
 284                if (texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapS == GL_CLAMP ||
 285                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapT == GL_CLAMP ||
 286                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapR == GL_CLAMP) {
 287                    return GL_TRUE;
 288                }
 289             }
 290          }
 291       }
 292    }
 293
 294    /* Nothing stopping us from the fast path now */
 295    return GL_FALSE;
 296 }
 297
 298 /* May fail if out of video memory for texture or vbo upload, or on
 299  * fallback conditions.
 300  */
 301 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 302                                      const struct gl_client_array *arrays[],
 303                                      const struct _mesa_prim *prim,
 304                                      GLuint nr_prims,
 305                                      const struct _mesa_index_buffer *ib,
 306                                      GLuint min_index,
 307                                      GLuint max_index )
 308 {
 309    struct intel_context *intel = intel_context(ctx);
 310    struct brw_context *brw = brw_context(ctx);
 311    GLboolean retval = GL_FALSE;
 312    GLboolean warn = GL_FALSE;
 313    GLboolean first_time = GL_TRUE;
 314    GLuint i;
 315
 316    if (ctx->NewState)
 317       _mesa_update_state( ctx );
 318
 319    /* We have to validate the textures *before* checking for fallbacks;
 320     * otherwise, the software fallback won't be able to rely on the
 321     * texture state, the firstLevel and lastLevel fields won't be
 322     * set in the intel texture object (they'll both be 0), and the
 323     * software fallback will segfault if it attempts to access any
 324     * texture level other than level 0.
 325     */
 326    brw_validate_textures( brw );
 327
 328    if (check_fallbacks(brw, prim, nr_prims))
 329       return GL_FALSE;
 330
 331    /* Bind all inputs, derive varying and size information:
 332     */
 333    brw_merge_inputs( brw, arrays );
 334
 335    brw->ib.ib = ib;
 336    brw->state.dirty.brw |= BRW_NEW_INDICES;
 337
 338    brw->vb.min_index = min_index;
 339    brw->vb.max_index = max_index;
 340    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 341
 342    /* Have to validate state quite late.  Will rebuild tnl_program,
 343     * which depends on varying information.
 344     *
 345     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 346     * so can't access it earlier.
 347     */
 348
 349    intel_prepare_render(intel);
 350
 351    for (i = 0; i < nr_prims; i++) {
 352       uint32_t hw_prim;
 353
 354       /* Flush the batch if it's approaching full, so that we don't wrap while
 355        * we've got validated state that needs to be in the same batch as the
 356        * primitives.  This fraction is just a guess (minimal full state plus
 357        * a primitive is around 512 bytes), and would be better if we had
 358        * an upper bound of how much we might emit in a single
 359        * brw_try_draw_prims().
 360        */
 361       intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4);
 362
 363       hw_prim = brw_set_prim(brw, &prim[i]);
 364
 365       if (first_time || (brw->state.dirty.brw & BRW_NEW_PRIMITIVE)) {
 366          first_time = GL_FALSE;
 367
 368          brw_validate_state(brw);
 369
 370          /* Various fallback checks:  */
 371          if (brw->intel.Fallback)
 372             goto out;
 373
 374          /* Check that we can fit our state in with our existing batchbuffer, or
 375           * flush otherwise.
 376           */
 377          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 378                                              brw->state.validated_bo_count)) {
 379             static GLboolean warned;
 380             intel_batchbuffer_flush(intel->batch);
 381
 382             /* Validate the state after we flushed the batch (which would have
 383              * changed the set of dirty state).  If we still fail to
 384              * check_aperture, warn of what's happening, but attempt to continue
 385              * on since it may succeed anyway, and the user would probably rather
 386              * see a failure and a warning than a fallback.
 387              */
 388             brw_validate_state(brw);
 389             if (!warned &&
 390                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 391                                                 brw->state.validated_bo_count)) {
 392                warn = GL_TRUE;
 393                warned = GL_TRUE;
 394             }
 395          }
 396
 397          brw_upload_state(brw);
 398       }
 399
 400       brw_emit_prim(brw, &prim[i], hw_prim);
 401
 402       retval = GL_TRUE;
 403    }
 404
 405    if (intel->always_flush_batch)
 406       intel_batchbuffer_flush(intel->batch);
 407  out:
 408
 409    brw_state_cache_check_size(brw);
 410
 411    if (warn)
 412       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 413               "available aperture space\n");
 414
 415    if (!retval)
 416       DBG("%s failed\n", __FUNCTION__);
 417
 418    return retval;
 419 }
 420
 421 void brw_draw_prims( GLcontext *ctx,
 422                      const struct gl_client_array *arrays[],
 423                      const struct _mesa_prim *prim,
 424                      GLuint nr_prims,
 425                      const struct _mesa_index_buffer *ib,
 426                      GLboolean index_bounds_valid,
 427                      GLuint min_index,
 428                      GLuint max_index )
 429 {
 430    GLboolean retval;
 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) {
 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_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 459    }
 460
 461 }
 462
 463 void brw_draw_init( struct brw_context *brw )
 464 {
 465    GLcontext *ctx = &brw->intel.ctx;
 466    struct vbo_context *vbo = vbo_context(ctx);
 467
 468    /* Register our drawing function:
 469     */
 470    vbo->draw_prims = brw_draw_prims;
 471 }
 472
 473 void brw_draw_destroy( struct brw_context *brw )
 474 {
 475    int i;
 476
 477    if (brw->vb.upload.bo != NULL) {
 478       dri_bo_unreference(brw->vb.upload.bo);
 479       brw->vb.upload.bo = NULL;
 480    }
 481
 482    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 483       dri_bo_unreference(brw->vb.inputs[i].bo);
 484       brw->vb.inputs[i].bo = NULL;
 485    }
 486
 487    dri_bo_unreference(brw->ib.bo);
 488    brw->ib.bo = NULL;
 489 }