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 "main/glheader.h"
  31 #include "main/context.h"
  32 #include "main/state.h"
  33 #include "main/api_validate.h"
  34 #include "main/enums.h"
  35
  36 #include "brw_draw.h"
  37 #include "brw_defines.h"
  38 #include "brw_context.h"
  39 #include "brw_state.h"
  40 #include "brw_fallback.h"
  41
  42 #include "intel_batchbuffer.h"
  43 #include "intel_buffer_objects.h"
  44
  45 #include "tnl/tnl.h"
  46 #include "vbo/vbo_context.h"
  47 #include "swrast/swrast.h"
  48 #include "swrast_setup/swrast_setup.h"
  49
  50 #define FILE_DEBUG_FLAG DEBUG_BATCH
  51
  52 static GLuint prim_to_hw_prim[GL_POLYGON+1] = {
  53    _3DPRIM_POINTLIST,
  54    _3DPRIM_LINELIST,
  55    _3DPRIM_LINELOOP,
  56    _3DPRIM_LINESTRIP,
  57    _3DPRIM_TRILIST,
  58    _3DPRIM_TRISTRIP,
  59    _3DPRIM_TRIFAN,
  60    _3DPRIM_QUADLIST,
  61    _3DPRIM_QUADSTRIP,
  62    _3DPRIM_POLYGON
  63 };
  64
  65
  66 static const GLenum reduced_prim[GL_POLYGON+1] = {
  67    GL_POINTS,
  68    GL_LINES,
  69    GL_LINES,
  70    GL_LINES,
  71    GL_TRIANGLES,
  72    GL_TRIANGLES,
  73    GL_TRIANGLES,
  74    GL_TRIANGLES,
  75    GL_TRIANGLES,
  76    GL_TRIANGLES
  77 };
  78
  79
  80 /* When the primitive changes, set a state bit and re-validate.  Not
  81  * the nicest and would rather deal with this by having all the
  82  * programs be immune to the active primitive (ie. cope with all
  83  * possibilities).  That may not be realistic however.
  84  */
  85 static GLuint brw_set_prim(struct brw_context *brw, GLenum prim)
  86 {
  87    if (INTEL_DEBUG & DEBUG_PRIMS)
  88       _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
  89
  90    /* Slight optimization to avoid the GS program when not needed:
  91     */
  92    if (prim == GL_QUAD_STRIP &&
  93        brw->attribs.Light->ShadeModel != GL_FLAT &&
  94        brw->attribs.Polygon->FrontMode == GL_FILL &&
  95        brw->attribs.Polygon->BackMode == GL_FILL)
  96       prim = GL_TRIANGLE_STRIP;
  97
  98    if (prim != brw->primitive) {
  99       brw->primitive = prim;
 100       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 101
 102       if (reduced_prim[prim] != brw->intel.reduced_primitive) {
 103          brw->intel.reduced_primitive = reduced_prim[prim];
 104          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 105       }
 106    }
 107
 108    return prim_to_hw_prim[prim];
 109 }
 110
 111
 112 static GLuint trim(GLenum prim, GLuint length)
 113 {
 114    if (prim == GL_QUAD_STRIP)
 115       return length > 3 ? (length - length % 2) : 0;
 116    else if (prim == GL_QUADS)
 117       return length - length % 4;
 118    else
 119       return length;
 120 }
 121
 122
 123 static void brw_emit_prim(struct brw_context *brw,
 124                           const struct _mesa_prim *prim,
 125                           uint32_t hw_prim)
 126 {
 127    struct brw_3d_primitive prim_packet;
 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    prim_packet.instance_count = 1;
 142    prim_packet.start_instance_location = 0;
 143    prim_packet.base_vert_location = 0;
 144
 145    /* Can't wrap here, since we rely on the validated state. */
 146    brw->no_batch_wrap = GL_TRUE;
 147    if (prim_packet.verts_per_instance) {
 148       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 149                               sizeof(prim_packet), LOOP_CLIPRECTS);
 150    }
 151    brw->no_batch_wrap = GL_FALSE;
 152 }
 153
 154 static void brw_merge_inputs( struct brw_context *brw,
 155                        const struct gl_client_array *arrays[])
 156 {
 157    struct brw_vertex_info old = brw->vb.info;
 158    GLuint i;
 159
 160    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 161       dri_bo_unreference(brw->vb.inputs[i].bo);
 162
 163    memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs));
 164    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 165
 166    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 167       brw->vb.inputs[i].glarray = arrays[i];
 168
 169       if (arrays[i]->StrideB != 0)
 170          brw->vb.info.varying |= 1 << i;
 171
 172          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 173             ((i%16) * 2);
 174    }
 175
 176    /* Raise statechanges if input sizes and varying have changed:
 177     */
 178    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 179       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 180
 181    if (brw->vb.info.varying != old.varying)
 182       brw->state.dirty.brw |= BRW_NEW_INPUT_VARYING;
 183 }
 184
 185 /* XXX: could split the primitive list to fallback only on the
 186  * non-conformant primitives.
 187  */
 188 static GLboolean check_fallbacks( struct brw_context *brw,
 189                                   const struct _mesa_prim *prim,
 190                                   GLuint nr_prims )
 191 {
 192    GLuint i;
 193
 194    if (!brw->intel.strict_conformance)
 195       return GL_FALSE;
 196
 197    if (brw->attribs.Polygon->SmoothFlag) {
 198       for (i = 0; i < nr_prims; i++)
 199          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 200             return GL_TRUE;
 201    }
 202
 203    /* BRW hardware will do AA lines, but they are non-conformant it
 204     * seems.  TBD whether we keep this fallback:
 205     */
 206    if (brw->attribs.Line->SmoothFlag) {
 207       for (i = 0; i < nr_prims; i++)
 208          if (reduced_prim[prim[i].mode] == GL_LINES)
 209             return GL_TRUE;
 210    }
 211
 212    /* Stipple -- these fallbacks could be resolved with a little
 213     * bit of work?
 214     */
 215    if (brw->attribs.Line->StippleFlag) {
 216       for (i = 0; i < nr_prims; i++) {
 217          /* GS doesn't get enough information to know when to reset
 218           * the stipple counter?!?
 219           */
 220          if (prim[i].mode == GL_LINE_LOOP)
 221             return GL_TRUE;
 222
 223          if (prim[i].mode == GL_POLYGON &&
 224              (brw->attribs.Polygon->FrontMode == GL_LINE ||
 225               brw->attribs.Polygon->BackMode == GL_LINE))
 226             return GL_TRUE;
 227       }
 228    }
 229
 230
 231    if (brw->attribs.Point->SmoothFlag) {
 232       for (i = 0; i < nr_prims; i++)
 233          if (prim[i].mode == GL_POINTS)
 234             return GL_TRUE;
 235    }
 236
 237    return GL_FALSE;
 238 }
 239
 240 /* May fail if out of video memory for texture or vbo upload, or on
 241  * fallback conditions.
 242  */
 243 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 244                                      const struct gl_client_array *arrays[],
 245                                      const struct _mesa_prim *prim,
 246                                      GLuint nr_prims,
 247                                      const struct _mesa_index_buffer *ib,
 248                                      GLuint min_index,
 249                                      GLuint max_index )
 250 {
 251    struct intel_context *intel = intel_context(ctx);
 252    struct brw_context *brw = brw_context(ctx);
 253    GLboolean retval = GL_FALSE;
 254    GLboolean warn = GL_FALSE;
 255    GLboolean first_time = GL_TRUE;
 256    GLuint i;
 257
 258    if (ctx->NewState)
 259       _mesa_update_state( ctx );
 260
 261    if (check_fallbacks(brw, prim, nr_prims))
 262       return GL_FALSE;
 263
 264    brw_validate_textures( brw );
 265
 266    /* Bind all inputs, derive varying and size information:
 267     */
 268    brw_merge_inputs( brw, arrays );
 269
 270    brw->ib.ib = ib;
 271    brw->state.dirty.brw |= BRW_NEW_INDICES;
 272
 273    brw->vb.min_index = min_index;
 274    brw->vb.max_index = max_index;
 275    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 276
 277    /* Have to validate state quite late.  Will rebuild tnl_program,
 278     * which depends on varying information.
 279     *
 280     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 281     * so can't access it earlier.
 282     */
 283
 284    LOCK_HARDWARE(intel);
 285
 286    if (!intel->constant_cliprect && intel->driDrawable->numClipRects == 0) {
 287       UNLOCK_HARDWARE(intel);
 288       return GL_TRUE;
 289    }
 290
 291    for (i = 0; i < nr_prims; i++) {
 292       uint32_t hw_prim;
 293
 294       /* Flush the batch if it's approaching full, so that we don't wrap while
 295        * we've got validated state that needs to be in the same batch as the
 296        * primitives.  This fraction is just a guess (minimal full state plus
 297        * a primitive is around 512 bytes), and would be better if we had
 298        * an upper bound of how much we might emit in a single
 299        * brw_try_draw_prims().
 300        */
 301       intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4,
 302                                       LOOP_CLIPRECTS);
 303
 304       hw_prim = brw_set_prim(brw, prim[i].mode);
 305
 306       if (first_time || (brw->state.dirty.brw & BRW_NEW_PRIMITIVE)) {
 307          first_time = GL_FALSE;
 308
 309          brw_validate_state(brw);
 310
 311          /* Various fallback checks:  */
 312          if (brw->intel.Fallback)
 313             goto out;
 314
 315          /* Check that we can fit our state in with our existing batchbuffer, or
 316           * flush otherwise.
 317           */
 318          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 319                                              brw->state.validated_bo_count)) {
 320             static GLboolean warned;
 321             intel_batchbuffer_flush(intel->batch);
 322
 323             /* Validate the state after we flushed the batch (which would have
 324              * changed the set of dirty state).  If we still fail to
 325              * check_aperture, warn of what's happening, but attempt to continue
 326              * on since it may succeed anyway, and the user would probably rather
 327              * see a failure and a warning than a fallback.
 328              */
 329             brw_validate_state(brw);
 330             if (!warned &&
 331                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 332                                                 brw->state.validated_bo_count)) {
 333                warn = GL_TRUE;
 334                warned = GL_TRUE;
 335             }
 336          }
 337
 338          brw_upload_state(brw);
 339       }
 340
 341       brw_emit_prim(brw, &prim[i], hw_prim);
 342
 343       retval = GL_TRUE;
 344    }
 345
 346  out:
 347    UNLOCK_HARDWARE(intel);
 348
 349    if (warn)
 350       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 351               "available aperture space\n");
 352
 353    if (!retval)
 354       DBG("%s failed\n", __FUNCTION__);
 355
 356    return retval;
 357 }
 358
 359 static GLboolean brw_need_rebase( GLcontext *ctx,
 360                                   const struct gl_client_array *arrays[],
 361                                   const struct _mesa_index_buffer *ib,
 362                                   GLuint min_index )
 363 {
 364    if (min_index == 0)
 365       return GL_FALSE;
 366
 367    if (ib) {
 368       if (!vbo_all_varyings_in_vbos(arrays))
 369          return GL_TRUE;
 370       else
 371          return GL_FALSE;
 372    }
 373    else {
 374       /* Hmm.  This isn't quite what I wanted.  BRW can actually
 375        * handle the mixed case well enough that we shouldn't need to
 376        * rebase.  However, it's probably not very common, nor hugely
 377        * expensive to do it this way:
 378        */
 379       if (!vbo_all_varyings_in_vbos(arrays))
 380          return GL_TRUE;
 381       else
 382          return GL_FALSE;
 383    }
 384 }
 385
 386
 387 void brw_draw_prims( GLcontext *ctx,
 388                      const struct gl_client_array *arrays[],
 389                      const struct _mesa_prim *prim,
 390                      GLuint nr_prims,
 391                      const struct _mesa_index_buffer *ib,
 392                      GLuint min_index,
 393                      GLuint max_index )
 394 {
 395    GLboolean retval;
 396
 397    /* Decide if we want to rebase.  If so we end up recursing once
 398     * only into this function.
 399     */
 400    if (brw_need_rebase( ctx, arrays, ib, min_index )) {
 401       vbo_rebase_prims( ctx, arrays,
 402                         prim, nr_prims,
 403                         ib, min_index, max_index,
 404                         brw_draw_prims );
 405
 406       return;
 407    }
 408
 409    /* Make a first attempt at drawing:
 410     */
 411    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 412
 413    /* Otherwise, we really are out of memory.  Pass the drawing
 414     * command to the software tnl module and which will in turn call
 415     * swrast to do the drawing.
 416     */
 417    if (!retval) {
 418        _swsetup_Wakeup(ctx);
 419       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 420    }
 421
 422 }
 423
 424 void brw_draw_init( struct brw_context *brw )
 425 {
 426    GLcontext *ctx = &brw->intel.ctx;
 427    struct vbo_context *vbo = vbo_context(ctx);
 428
 429    /* Register our drawing function:
 430     */
 431    vbo->draw_prims = brw_draw_prims;
 432 }
 433
 434 void brw_draw_destroy( struct brw_context *brw )
 435 {
 436    int i;
 437
 438    if (brw->vb.upload.bo != NULL) {
 439       dri_bo_unreference(brw->vb.upload.bo);
 440       brw->vb.upload.bo = NULL;
 441    }
 442
 443    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 444       dri_bo_unreference(brw->vb.inputs[i].bo);
 445       brw->vb.inputs[i].bo = NULL;
 446    }
 447
 448    dri_bo_unreference(brw->ib.bo);
 449    brw->ib.bo = NULL;
 450 }