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_attrib.h"
  39 #include "brw_context.h"
  40 #include "brw_aub.h"
  41 #include "brw_state.h"
  42 #include "brw_fallback.h"
  43
  44 #include "intel_ioctl.h"
  45 #include "intel_batchbuffer.h"
  46 #include "intel_buffer_objects.h"
  47
  48
  49
  50
  51
  52
  53 static GLuint hw_prim[GL_POLYGON+1] = {
  54    _3DPRIM_POINTLIST,
  55    _3DPRIM_LINELIST,
  56    _3DPRIM_LINELOOP,
  57    _3DPRIM_LINESTRIP,
  58    _3DPRIM_TRILIST,
  59    _3DPRIM_TRISTRIP,
  60    _3DPRIM_TRIFAN,
  61    _3DPRIM_QUADLIST,
  62    _3DPRIM_QUADSTRIP,
  63    _3DPRIM_POLYGON
  64 };
  65
  66
  67 static const GLenum reduced_prim[GL_POLYGON+1] = {
  68    GL_POINTS,
  69    GL_LINES,
  70    GL_LINES,
  71    GL_LINES,
  72    GL_TRIANGLES,
  73    GL_TRIANGLES,
  74    GL_TRIANGLES,
  75    GL_TRIANGLES,
  76    GL_TRIANGLES,
  77    GL_TRIANGLES
  78 };
  79
  80
  81 /* When the primitive changes, set a state bit and re-validate.  Not
  82  * the nicest and would rather deal with this by having all the
  83  * programs be immune to the active primitive (ie. cope with all
  84  * possibilities).  That may not be realistic however.
  85  */
  86 static GLuint brw_set_prim(struct brw_context *brw, GLenum prim)
  87 {
  88    if (INTEL_DEBUG & DEBUG_PRIMS)
  89       _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
  90
  91    /* Slight optimization to avoid the GS program when not needed:
  92     */
  93    if (prim == GL_QUAD_STRIP &&
  94        brw->attribs.Light->ShadeModel != GL_FLAT &&
  95        brw->attribs.Polygon->FrontMode == GL_FILL &&
  96        brw->attribs.Polygon->BackMode == GL_FILL)
  97       prim = GL_TRIANGLE_STRIP;
  98
  99    if (prim != brw->primitive) {
 100       brw->primitive = prim;
 101       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 102
 103       if (reduced_prim[prim] != brw->intel.reduced_primitive) {
 104          brw->intel.reduced_primitive = reduced_prim[prim];
 105          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 106       }
 107
 108       brw_validate_state(brw);
 109    }
 110
 111    return hw_prim[prim];
 112 }
 113
 114
 115 static GLuint trim(GLenum prim, GLuint length)
 116 {
 117    if (prim == GL_QUAD_STRIP)
 118       return length > 3 ? (length - length % 2) : 0;
 119    else if (prim == GL_QUADS)
 120       return length - length % 4;
 121    else
 122       return length;
 123 }
 124
 125
 126 static void brw_emit_cliprect( struct brw_context *brw,
 127                                const drm_clip_rect_t *rect )
 128 {
 129    struct brw_drawrect bdr;
 130
 131    bdr.header.opcode = CMD_DRAW_RECT;
 132    bdr.header.length = sizeof(bdr)/4 - 2;
 133    bdr.xmin = rect->x1;
 134    bdr.xmax = rect->x2 - 1;
 135    bdr.ymin = rect->y1;
 136    bdr.ymax = rect->y2 - 1;
 137    bdr.xorg = brw->intel.drawX;
 138    bdr.yorg = brw->intel.drawY;
 139
 140    intel_batchbuffer_data( brw->intel.batch, &bdr, sizeof(bdr),
 141                            INTEL_BATCH_NO_CLIPRECTS);
 142 }
 143
 144
 145 static void brw_emit_prim( struct brw_context *brw,
 146                            const struct brw_draw_prim *prim )
 147
 148 {
 149    struct brw_3d_primitive prim_packet;
 150
 151    if (INTEL_DEBUG & DEBUG_PRIMS)
 152       _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 153                    prim->start, prim->count);
 154
 155    prim_packet.header.opcode = CMD_3D_PRIM;
 156    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 157    prim_packet.header.pad = 0;
 158    prim_packet.header.topology = brw_set_prim(brw, prim->mode);
 159    prim_packet.header.indexed = prim->indexed;
 160
 161    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 162    prim_packet.start_vert_location = prim->start;
 163    prim_packet.instance_count = 1;
 164    prim_packet.start_instance_location = 0;
 165    prim_packet.base_vert_location = 0;
 166
 167    if (prim_packet.verts_per_instance) {
 168       intel_batchbuffer_data( brw->intel.batch, &prim_packet, sizeof(prim_packet),
 169                               INTEL_BATCH_NO_CLIPRECTS);
 170    }
 171 }
 172
 173
 174
 175 static void update_current_size( struct gl_client_array *array)
 176 {
 177    const GLfloat *ptr = (const GLfloat *)array->Ptr;
 178
 179    assert(array->StrideB == 0);
 180    assert(array->Type == GL_FLOAT || array->Type == GL_UNSIGNED_BYTE);
 181
 182    if (ptr[3] != 1.0)
 183       array->Size = 4;
 184    else if (ptr[2] != 0.0)
 185       array->Size = 3;
 186    else if (ptr[1] != 0.0)
 187       array->Size = 2;
 188    else
 189       array->Size = 1;
 190 }
 191
 192
 193
 194 /* Fill in any gaps in passed arrays with pointers to current
 195  * attributes:
 196  */
 197 static void brw_merge_inputs( struct brw_context *brw,
 198                        const struct gl_client_array *arrays[])
 199 {
 200    struct gl_client_array *current_values = brw->vb.current_values;
 201    struct brw_vertex_element *inputs = brw->vb.inputs;
 202    struct brw_vertex_info old = brw->vb.info;
 203    GLuint i;
 204
 205    memset(inputs, 0, sizeof(*inputs));
 206    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 207
 208    for (i = 0; i < BRW_ATTRIB_MAX; i++) {
 209       if (arrays[i] && arrays[i]->Enabled)
 210       {
 211          brw->vb.inputs[i].glarray = arrays[i];
 212          brw->vb.info.varying |= (GLuint64EXT) 1 << i;
 213       }
 214       else
 215       {
 216          brw->vb.inputs[i].glarray = &current_values[i];
 217          update_current_size(&current_values[i]);
 218       }
 219
 220       brw->vb.info.sizes[i/16] |= (inputs[i].glarray->Size - 1) << ((i%16) * 2);
 221    }
 222
 223    /* Raise statechanges if input sizes and varying have changed:
 224     */
 225    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 226       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 227
 228    if (brw->vb.info.varying != old.varying)
 229       brw->state.dirty.brw |= BRW_NEW_INPUT_VARYING;
 230 }
 231
 232 static GLboolean check_fallbacks( struct brw_context *brw,
 233                                   const struct brw_draw_prim *prim,
 234                                   GLuint nr_prims )
 235 {
 236    GLuint i;
 237
 238    if (!brw->intel.strict_conformance)
 239       return GL_FALSE;
 240
 241    if (brw->attribs.Polygon->SmoothFlag) {
 242       for (i = 0; i < nr_prims; i++)
 243          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 244             return GL_TRUE;
 245    }
 246
 247    /* BRW hardware will do AA lines, but they are non-conformant it
 248     * seems.  TBD whether we keep this fallback:
 249     */
 250    if (brw->attribs.Line->SmoothFlag) {
 251       for (i = 0; i < nr_prims; i++)
 252          if (reduced_prim[prim[i].mode] == GL_LINES)
 253             return GL_TRUE;
 254    }
 255
 256    /* Stipple -- these fallbacks could be resolved with a little
 257     * bit of work?
 258     */
 259    if (brw->attribs.Line->StippleFlag) {
 260       for (i = 0; i < nr_prims; i++) {
 261          /* GS doesn't get enough information to know when to reset
 262           * the stipple counter?!?
 263           */
 264          if (prim[i].mode == GL_LINE_LOOP)
 265             return GL_TRUE;
 266
 267          if (prim[i].mode == GL_POLYGON &&
 268              (brw->attribs.Polygon->FrontMode == GL_LINE ||
 269               brw->attribs.Polygon->BackMode == GL_LINE))
 270             return GL_TRUE;
 271       }
 272    }
 273
 274
 275    if (brw->attribs.Point->SmoothFlag) {
 276       for (i = 0; i < nr_prims; i++)
 277          if (prim[i].mode == GL_POINTS)
 278             return GL_TRUE;
 279    }
 280
 281    return GL_FALSE;
 282 }
 283
 284
 285 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 286                                      const struct gl_client_array *arrays[],
 287                                      const struct brw_draw_prim *prim,
 288                                      GLuint nr_prims,
 289                                      const struct brw_draw_index_buffer *ib,
 290                                      GLuint min_index,
 291                                      GLuint max_index,
 292                                      GLuint flags )
 293 {
 294    struct intel_context *intel = intel_context(ctx);
 295    struct brw_context *brw = brw_context(ctx);
 296    GLboolean retval = GL_FALSE;
 297    GLuint i, j;
 298
 299    if (ctx->NewState)
 300       _mesa_update_state( ctx );
 301
 302    /* Bind all inputs, derive varying and size information:
 303     */
 304    brw_merge_inputs( brw, arrays );
 305
 306    /* Have to validate state quite late.  Will rebuild tnl_program,
 307     * which depends on varying information.
 308     *
 309     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 310     * so can't access it earlier.
 311     */
 312
 313    LOCK_HARDWARE(intel);
 314
 315    if (brw->intel.numClipRects == 0) {
 316       assert(intel->batch->ptr == intel->batch->map + intel->batch->offset);
 317       UNLOCK_HARDWARE(intel);
 318       return GL_TRUE;
 319    }
 320
 321    {
 322       assert(intel->locked);
 323
 324
 325
 326       /* Set the first primitive early, ahead of validate_state:
 327        */
 328       brw_set_prim(brw, prim[0].mode);
 329
 330       /* XXX:  Need to separate validate and upload of state.
 331        */
 332       brw_validate_state( brw );
 333
 334       /* Various fallback checks:
 335        */
 336       if (brw->intel.Fallback)
 337          goto out;
 338
 339       if (check_fallbacks( brw, prim, nr_prims ))
 340          goto out;
 341
 342       /* Upload index, vertex data:
 343        */
 344       if (ib)
 345          brw_upload_indices( brw, ib );
 346
 347       if (!brw_upload_vertices( brw, min_index, max_index)) {
 348          goto out;
 349       }
 350
 351       /* For single cliprect, state is already emitted:
 352        */
 353       if (brw->intel.numClipRects == 1) {
 354          for (i = 0; i < nr_prims; i++) {
 355             brw_emit_prim(brw, &prim[i]);
 356          }
 357       }
 358       else {
 359          /* Otherwise, explicitly do the cliprects at this point:
 360           */
 361          for (j = 0; j < brw->intel.numClipRects; j++) {
 362             brw_emit_cliprect(brw, &brw->intel.pClipRects[j]);
 363
 364             /* Emit prims to batchbuffer:
 365              */
 366             for (i = 0; i < nr_prims; i++) {
 367                brw_emit_prim(brw, &prim[i]);
 368             }
 369          }
 370       }
 371
 372       intel->need_flush = GL_TRUE;
 373       retval = GL_TRUE;
 374    }
 375
 376  out:
 377
 378    /* Currently have to do this to synchronize with the map/unmap of
 379     * the vertex buffer in brw_exec_api.c.  Not sure if there is any
 380     * way around this, as not every flush is due to a buffer filling
 381     * up.
 382     */
 383    if (!intel_batchbuffer_flush( brw->intel.batch )) {
 384       DBG("%s intel_batchbuffer_flush failed\n", __FUNCTION__);
 385       retval = GL_FALSE;
 386    }
 387
 388    if (retval && intel->thrashing) {
 389       bmSetFence(intel);
 390    }
 391
 392    /* Free any old data so it doesn't clog up texture memory - we
 393     * won't be referencing it again.
 394     */
 395    while (brw->vb.upload.wrap != brw->vb.upload.buf) {
 396       ctx->Driver.BufferData(ctx,
 397                              GL_ARRAY_BUFFER_ARB,
 398                              BRW_UPLOAD_INIT_SIZE,
 399                              NULL,
 400                              GL_DYNAMIC_DRAW_ARB,
 401                              brw->vb.upload.vbo[brw->vb.upload.wrap]);
 402       brw->vb.upload.wrap++;
 403       brw->vb.upload.wrap %= BRW_NR_UPLOAD_BUFS;
 404    }
 405
 406    UNLOCK_HARDWARE(intel);
 407
 408    if (!retval)
 409       DBG("%s failed\n", __FUNCTION__);
 410
 411    return retval;
 412 }
 413
 414
 415 GLboolean brw_draw_prims( GLcontext *ctx,
 416                           const struct gl_client_array *arrays[],
 417                           const struct brw_draw_prim *prim,
 418                           GLuint nr_prims,
 419                           const struct brw_draw_index_buffer *ib,
 420                           GLuint min_index,
 421                           GLuint max_index,
 422                           GLuint flags )
 423 {
 424    struct intel_context *intel = intel_context(ctx);
 425    GLboolean retval;
 426
 427    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index, flags);
 428
 429
 430    if (!retval && bmError(intel)) {
 431
 432       DBG("retrying\n");
 433       /* This looks like out-of-memory but potentially we have
 434        * situation where there is enough memory but it has become
 435        * fragmented.  Clear out all heaps and start from scratch by
 436        * faking a contended lock event:  (done elsewhere)
 437        */
 438
 439       /* Then try a second time only to upload textures and draw the
 440        * primitives:
 441        */
 442       retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index, flags);
 443    }
 444
 445    if (intel->aub_file && (INTEL_DEBUG & DEBUG_SYNC)) {
 446       intelFinish( &intel->ctx );
 447       intel->aub_wrap = 1;
 448    }
 449
 450
 451    return retval;
 452 }
 453
 454
 455 static void brw_invalidate_vbo_cb( struct intel_context *intel, void *ptr )
 456 {
 457    /* nothing to do, we don't rely on the contents being preserved */
 458 }
 459
 460
 461 void brw_draw_init( struct brw_context *brw )
 462 {
 463    GLcontext *ctx = &brw->intel.ctx;
 464    GLuint i;
 465
 466    brw->vb.upload.size = BRW_UPLOAD_INIT_SIZE;
 467
 468    for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++) {
 469       brw->vb.upload.vbo[i] = ctx->Driver.NewBufferObject(ctx, 1, GL_ARRAY_BUFFER_ARB);
 470
 471       /* XXX: Set these to no-backing-store
 472        */
 473       bmBufferSetInvalidateCB(&brw->intel,
 474                               intel_bufferobj_buffer(intel_buffer_object(brw->vb.upload.vbo[i])),
 475                               brw_invalidate_vbo_cb,
 476                               &brw->intel,
 477                               GL_TRUE);
 478
 479    }
 480
 481    ctx->Driver.BufferData( ctx,
 482                            GL_ARRAY_BUFFER_ARB,
 483                            BRW_UPLOAD_INIT_SIZE,
 484                            NULL,
 485                            GL_DYNAMIC_DRAW_ARB,
 486                            brw->vb.upload.vbo[0] );
 487
 488
 489    brw_init_current_values(ctx, brw->vb.current_values);
 490 }
 491
 492 void brw_draw_destroy( struct brw_context *brw )
 493 {
 494    GLcontext *ctx = &brw->intel.ctx;
 495    GLuint i;
 496
 497    for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++)
 498       ctx->Driver.DeleteBuffer(ctx, brw->vb.upload.vbo[i]);
 499 }