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 <stdlib.h>
  29
  30 #include "glheader.h"
  31 #include "context.h"
  32 #include "state.h"
  33 #include "api_validate.h"
  34 #include "enums.h"
  35
  36 #include "brw_draw.h"
  37 #include "brw_defines.h"
  38 #include "brw_context.h"
  39 #include "brw_aub.h"
  40 #include "brw_state.h"
  41 #include "brw_fallback.h"
  42
  43 #include "intel_ioctl.h"
  44 #include "intel_batchbuffer.h"
  45 #include "intel_buffer_objects.h"
  46
  47 #include "tnl/tnl.h"
  48 #include "vbo/vbo_context.h"
  49 #include "swrast/swrast.h"
  50 #include "swrast_setup/swrast_setup.h"
  51
  52
  53
  54 static GLuint hw_prim[GL_POLYGON+1] = {
  55    _3DPRIM_POINTLIST,
  56    _3DPRIM_LINELIST,
  57    _3DPRIM_LINELOOP,
  58    _3DPRIM_LINESTRIP,
  59    _3DPRIM_TRILIST,
  60    _3DPRIM_TRISTRIP,
  61    _3DPRIM_TRIFAN,
  62    _3DPRIM_QUADLIST,
  63    _3DPRIM_QUADSTRIP,
  64    _3DPRIM_POLYGON
  65 };
  66
  67
  68 static const GLenum reduced_prim[GL_POLYGON+1] = {
  69    GL_POINTS,
  70    GL_LINES,
  71    GL_LINES,
  72    GL_LINES,
  73    GL_TRIANGLES,
  74    GL_TRIANGLES,
  75    GL_TRIANGLES,
  76    GL_TRIANGLES,
  77    GL_TRIANGLES,
  78    GL_TRIANGLES
  79 };
  80
  81
  82 /* When the primitive changes, set a state bit and re-validate.  Not
  83  * the nicest and would rather deal with this by having all the
  84  * programs be immune to the active primitive (ie. cope with all
  85  * possibilities).  That may not be realistic however.
  86  */
  87 static GLuint brw_set_prim(struct brw_context *brw, GLenum prim)
  88 {
  89    if (INTEL_DEBUG & DEBUG_PRIMS)
  90       _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
  91
  92    /* Slight optimization to avoid the GS program when not needed:
  93     */
  94    if (prim == GL_QUAD_STRIP &&
  95        brw->attribs.Light->ShadeModel != GL_FLAT &&
  96        brw->attribs.Polygon->FrontMode == GL_FILL &&
  97        brw->attribs.Polygon->BackMode == GL_FILL)
  98       prim = GL_TRIANGLE_STRIP;
  99
 100    if (prim != brw->primitive) {
 101       brw->primitive = prim;
 102       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 103
 104       if (reduced_prim[prim] != brw->intel.reduced_primitive) {
 105          brw->intel.reduced_primitive = reduced_prim[prim];
 106          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 107       }
 108
 109       brw_validate_state(brw);
 110    }
 111
 112    return hw_prim[prim];
 113 }
 114
 115
 116 static GLuint trim(GLenum prim, GLuint length)
 117 {
 118    if (prim == GL_QUAD_STRIP)
 119       return length > 3 ? (length - length % 2) : 0;
 120    else if (prim == GL_QUADS)
 121       return length - length % 4;
 122    else
 123       return length;
 124 }
 125
 126
 127 static void brw_emit_cliprect( struct brw_context *brw,
 128                                const drm_clip_rect_t *rect )
 129 {
 130    struct brw_drawrect bdr;
 131
 132    bdr.header.opcode = CMD_DRAW_RECT;
 133    bdr.header.length = sizeof(bdr)/4 - 2;
 134    bdr.xmin = rect->x1;
 135    bdr.xmax = rect->x2 - 1;
 136    bdr.ymin = rect->y1;
 137    bdr.ymax = rect->y2 - 1;
 138    bdr.xorg = brw->intel.drawX;
 139    bdr.yorg = brw->intel.drawY;
 140
 141    intel_batchbuffer_data( brw->intel.batch, &bdr, sizeof(bdr),
 142                            INTEL_BATCH_NO_CLIPRECTS);
 143 }
 144
 145
 146 static void brw_emit_prim( struct brw_context *brw,
 147                            const struct _mesa_prim *prim )
 148
 149 {
 150    struct brw_3d_primitive prim_packet;
 151
 152    if (INTEL_DEBUG & DEBUG_PRIMS)
 153       _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 154                    prim->start, prim->count);
 155
 156    prim_packet.header.opcode = CMD_3D_PRIM;
 157    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 158    prim_packet.header.pad = 0;
 159    prim_packet.header.topology = brw_set_prim(brw, prim->mode);
 160    prim_packet.header.indexed = prim->indexed;
 161
 162    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 163    prim_packet.start_vert_location = prim->start;
 164    prim_packet.instance_count = 1;
 165    prim_packet.start_instance_location = 0;
 166    prim_packet.base_vert_location = 0;
 167
 168    if (prim_packet.verts_per_instance) {
 169       intel_batchbuffer_data( brw->intel.batch, &prim_packet, sizeof(prim_packet),
 170                               INTEL_BATCH_NO_CLIPRECTS);
 171    }
 172 }
 173
 174 static void brw_merge_inputs( struct brw_context *brw,
 175                        const struct gl_client_array *arrays[])
 176 {
 177    struct brw_vertex_element *inputs = brw->vb.inputs;
 178    struct brw_vertex_info old = brw->vb.info;
 179    GLuint i;
 180
 181    memset(inputs, 0, sizeof(*inputs));
 182    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 183
 184    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 185       brw->vb.inputs[i].glarray = arrays[i];
 186
 187       /* XXX: metaops passes null arrays */
 188       if (arrays[i]) {
 189          if (arrays[i]->StrideB != 0)
 190             brw->vb.info.varying |= 1 << i;
 191
 192          brw->vb.info.sizes[i/16] |= (inputs[i].glarray->Size - 1) << ((i%16) * 2);
 193       }
 194    }
 195
 196    /* Raise statechanges if input sizes and varying have changed:
 197     */
 198    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 199       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 200
 201    if (brw->vb.info.varying != old.varying)
 202       brw->state.dirty.brw |= BRW_NEW_INPUT_VARYING;
 203 }
 204
 205 /* XXX: could split the primitive list to fallback only on the
 206  * non-conformant primitives.
 207  */
 208 static GLboolean check_fallbacks( struct brw_context *brw,
 209                                   const struct _mesa_prim *prim,
 210                                   GLuint nr_prims )
 211 {
 212    GLuint i;
 213
 214    if (!brw->intel.strict_conformance)
 215       return GL_FALSE;
 216
 217    if (brw->attribs.Polygon->SmoothFlag) {
 218       for (i = 0; i < nr_prims; i++)
 219          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 220             return GL_TRUE;
 221    }
 222
 223    /* BRW hardware will do AA lines, but they are non-conformant it
 224     * seems.  TBD whether we keep this fallback:
 225     */
 226    if (brw->attribs.Line->SmoothFlag) {
 227       for (i = 0; i < nr_prims; i++)
 228          if (reduced_prim[prim[i].mode] == GL_LINES)
 229             return GL_TRUE;
 230    }
 231
 232    /* Stipple -- these fallbacks could be resolved with a little
 233     * bit of work?
 234     */
 235    if (brw->attribs.Line->StippleFlag) {
 236       for (i = 0; i < nr_prims; i++) {
 237          /* GS doesn't get enough information to know when to reset
 238           * the stipple counter?!?
 239           */
 240          if (prim[i].mode == GL_LINE_LOOP)
 241             return GL_TRUE;
 242
 243          if (prim[i].mode == GL_POLYGON &&
 244              (brw->attribs.Polygon->FrontMode == GL_LINE ||
 245               brw->attribs.Polygon->BackMode == GL_LINE))
 246             return GL_TRUE;
 247       }
 248    }
 249
 250
 251    if (brw->attribs.Point->SmoothFlag) {
 252       for (i = 0; i < nr_prims; i++)
 253          if (prim[i].mode == GL_POINTS)
 254             return GL_TRUE;
 255    }
 256
 257    return GL_FALSE;
 258 }
 259
 260 /* May fail if out of video memory for texture or vbo upload, or on
 261  * fallback conditions.
 262  */
 263 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 264                                      const struct gl_client_array *arrays[],
 265                                      const struct _mesa_prim *prim,
 266                                      GLuint nr_prims,
 267                                      const struct _mesa_index_buffer *ib,
 268                                      GLuint min_index,
 269                                      GLuint max_index )
 270 {
 271    struct intel_context *intel = intel_context(ctx);
 272    struct brw_context *brw = brw_context(ctx);
 273    GLboolean retval = GL_FALSE;
 274    GLuint i, j;
 275
 276    if (ctx->NewState)
 277       _mesa_update_state( ctx );
 278
 279    /* Bind all inputs, derive varying and size information:
 280     */
 281    brw_merge_inputs( brw, arrays );
 282
 283    /* Have to validate state quite late.  Will rebuild tnl_program,
 284     * which depends on varying information.
 285     *
 286     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 287     * so can't access it earlier.
 288     */
 289
 290    LOCK_HARDWARE(intel);
 291
 292    if (brw->intel.numClipRects == 0) {
 293       assert(intel->batch->ptr == intel->batch->map + intel->batch->offset);
 294       UNLOCK_HARDWARE(intel);
 295       return GL_TRUE;
 296    }
 297
 298    {
 299       /* Set the first primitive early, ahead of validate_state:
 300        */
 301       brw_set_prim(brw, prim[0].mode);
 302
 303       /* XXX:  Need to separate validate and upload of state.
 304        */
 305       brw_validate_state( brw );
 306
 307       /* Various fallback checks:
 308        */
 309       if (brw->intel.Fallback)
 310          goto out;
 311
 312       if (check_fallbacks( brw, prim, nr_prims ))
 313          goto out;
 314
 315       /* Upload index, vertex data:
 316        */
 317       if (ib)
 318          brw_upload_indices( brw, ib );
 319
 320       if (!brw_upload_vertices( brw, min_index, max_index)) {
 321          goto out;
 322       }
 323
 324       /* For single cliprect, state is already emitted:
 325        */
 326       if (brw->intel.numClipRects == 1) {
 327          for (i = 0; i < nr_prims; i++) {
 328             brw_emit_prim(brw, &prim[i]);
 329          }
 330       }
 331       else {
 332          /* Otherwise, explicitly do the cliprects at this point:
 333           */
 334          for (j = 0; j < brw->intel.numClipRects; j++) {
 335             brw_emit_cliprect(brw, &brw->intel.pClipRects[j]);
 336
 337             /* Emit prims to batchbuffer:
 338              */
 339             for (i = 0; i < nr_prims; i++) {
 340                brw_emit_prim(brw, &prim[i]);
 341             }
 342          }
 343       }
 344
 345       intel->need_flush = GL_TRUE;
 346       retval = GL_TRUE;
 347    }
 348
 349  out:
 350
 351    /* Currently have to do this to synchronize with the map/unmap of
 352     * the vertex buffer in brw_exec_api.c.  Not sure if there is any
 353     * way around this, as not every flush is due to a buffer filling
 354     * up.
 355     */
 356    if (!intel_batchbuffer_flush( brw->intel.batch )) {
 357       DBG("%s intel_batchbuffer_flush failed\n", __FUNCTION__);
 358       retval = GL_FALSE;
 359    }
 360
 361    if (retval && intel->thrashing) {
 362       bmSetFence(intel);
 363    }
 364
 365    /* Free any old data so it doesn't clog up texture memory - we
 366     * won't be referencing it again.
 367     */
 368    while (brw->vb.upload.wrap != brw->vb.upload.buf) {
 369       ctx->Driver.BufferData(ctx,
 370                              GL_ARRAY_BUFFER_ARB,
 371                              BRW_UPLOAD_INIT_SIZE,
 372                              NULL,
 373                              GL_DYNAMIC_DRAW_ARB,
 374                              brw->vb.upload.vbo[brw->vb.upload.wrap]);
 375       brw->vb.upload.wrap++;
 376       brw->vb.upload.wrap %= BRW_NR_UPLOAD_BUFS;
 377    }
 378
 379    UNLOCK_HARDWARE(intel);
 380
 381    if (!retval)
 382       DBG("%s failed\n", __FUNCTION__);
 383
 384    return retval;
 385 }
 386
 387 static GLboolean brw_need_rebase( GLcontext *ctx,
 388                                   const struct gl_client_array *arrays[],
 389                                   const struct _mesa_index_buffer *ib,
 390                                   GLuint min_index )
 391 {
 392    if (min_index == 0)
 393       return GL_FALSE;
 394
 395    if (ib) {
 396       if (!vbo_all_varyings_in_vbos(arrays))
 397          return GL_TRUE;
 398       else
 399          return GL_FALSE;
 400    }
 401    else {
 402       /* Hmm.  This isn't quite what I wanted.  BRW can actually
 403        * handle the mixed case well enough that we shouldn't need to
 404        * rebase.  However, it's probably not very common, nor hugely
 405        * expensive to do it this way:
 406        */
 407       if (!vbo_all_varyings_in_vbos(arrays))
 408          return GL_TRUE;
 409       else
 410          return GL_FALSE;
 411    }
 412 }
 413
 414
 415 void brw_draw_prims( GLcontext *ctx,
 416                      const struct gl_client_array *arrays[],
 417                      const struct _mesa_prim *prim,
 418                      GLuint nr_prims,
 419                      const struct _mesa_index_buffer *ib,
 420                      GLuint min_index,
 421                      GLuint max_index )
 422 {
 423    struct intel_context *intel = intel_context(ctx);
 424    GLboolean retval;
 425
 426    /* Decide if we want to rebase.  If so we end up recursing once
 427     * only into this function.
 428     */
 429    if (brw_need_rebase( ctx, arrays, ib, min_index )) {
 430       vbo_rebase_prims( ctx, arrays,
 431                         prim, nr_prims,
 432                         ib, min_index, max_index,
 433                         brw_draw_prims );
 434
 435       return;
 436    }
 437
 438
 439    /* Make a first attempt at drawing:
 440     */
 441    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 442
 443
 444    /* This looks like out-of-memory but potentially we have
 445     * situation where there is enough memory but it has become
 446     * fragmented.  Clear out all heaps and start from scratch by
 447     * faking a contended lock event:  (done elsewhere)
 448     */
 449    if (!retval && !intel->Fallback && bmError(intel)) {
 450       DBG("retrying\n");
 451       /* Then try a second time only to upload textures and draw the
 452        * primitives:
 453        */
 454       retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 455    }
 456
 457    /* Otherwise, we really are out of memory.  Pass the drawing
 458     * command to the software tnl module and which will in turn call
 459     * swrast to do the drawing.
 460     */
 461    if (!retval) {
 462        _swsetup_Wakeup(ctx);
 463       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 464    }
 465
 466    if (intel->aub_file && (INTEL_DEBUG & DEBUG_SYNC)) {
 467       intelFinish( &intel->ctx );
 468       intel->aub_wrap = 1;
 469    }
 470 }
 471
 472
 473 static void brw_invalidate_vbo_cb( struct intel_context *intel, void *ptr )
 474 {
 475    /* nothing to do, we don't rely on the contents being preserved */
 476 }
 477
 478
 479 void brw_draw_init( struct brw_context *brw )
 480 {
 481    GLcontext *ctx = &brw->intel.ctx;
 482    struct vbo_context *vbo = vbo_context(ctx);
 483    GLuint i;
 484
 485    /* Register our drawing function:
 486     */
 487    vbo->draw_prims = brw_draw_prims;
 488
 489    brw->vb.upload.size = BRW_UPLOAD_INIT_SIZE;
 490
 491    for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++) {
 492       brw->vb.upload.vbo[i] = ctx->Driver.NewBufferObject(ctx, 1, GL_ARRAY_BUFFER_ARB);
 493
 494       /* NOTE:  These are set to no-backing-store.
 495        */
 496       bmBufferSetInvalidateCB(&brw->intel,
 497                               intel_bufferobj_buffer(intel_buffer_object(brw->vb.upload.vbo[i])),
 498                               brw_invalidate_vbo_cb,
 499                               &brw->intel,
 500                               GL_TRUE);
 501    }
 502
 503    ctx->Driver.BufferData( ctx,
 504                            GL_ARRAY_BUFFER_ARB,
 505                            BRW_UPLOAD_INIT_SIZE,
 506                            NULL,
 507                            GL_DYNAMIC_DRAW_ARB,
 508                            brw->vb.upload.vbo[0] );
 509 }
 510
 511 void brw_draw_destroy( struct brw_context *brw )
 512 {
 513    GLcontext *ctx = &brw->intel.ctx;
 514    GLuint i;
 515
 516    for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++)
 517       ctx->Driver.DeleteBuffer(ctx, brw->vb.upload.vbo[i]);
 518 }