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    GLcontext *ctx = &brw->intel.ctx;
  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        ctx->Light.ShadeModel != GL_FLAT &&
  96        ctx->Polygon.FrontMode == GL_FILL &&
  97        ctx->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
 110    return prim_to_hw_prim[prim];
 111 }
 112
 113
 114 static GLuint trim(GLenum prim, GLuint length)
 115 {
 116    if (prim == GL_QUAD_STRIP)
 117       return length > 3 ? (length - length % 2) : 0;
 118    else if (prim == GL_QUADS)
 119       return length - length % 4;
 120    else
 121       return length;
 122 }
 123
 124
 125 static void brw_emit_prim(struct brw_context *brw,
 126                           const struct _mesa_prim *prim,
 127                           uint32_t hw_prim)
 128 {
 129    struct brw_3d_primitive prim_packet;
 130    struct intel_context *intel = &brw->intel;
 131
 132    if (INTEL_DEBUG & DEBUG_PRIMS)
 133       _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 134                    prim->start, prim->count);
 135
 136    prim_packet.header.opcode = CMD_3D_PRIM;
 137    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 138    prim_packet.header.pad = 0;
 139    prim_packet.header.topology = hw_prim;
 140    prim_packet.header.indexed = prim->indexed;
 141
 142    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 143    prim_packet.start_vert_location = prim->start;
 144    if (prim->indexed)
 145       prim_packet.start_vert_location += brw->ib.start_vertex_offset;
 146    prim_packet.instance_count = 1;
 147    prim_packet.start_instance_location = 0;
 148    prim_packet.base_vert_location = 0;
 149
 150    /* Can't wrap here, since we rely on the validated state. */
 151    brw->no_batch_wrap = GL_TRUE;
 152
 153    /* If we're set to always flush, do it before and after the primitive emit.
 154     * We want to catch both missed flushes that hurt instruction/state cache
 155     * and missed flushes of the render cache as it heads to other parts of
 156     * the besides the draw code.
 157     */
 158    if (intel->always_flush_cache) {
 159       BEGIN_BATCH(1, IGNORE_CLIPRECTS);
 160       OUT_BATCH(intel->vtbl.flush_cmd());
 161       ADVANCE_BATCH();
 162    }
 163    if (prim_packet.verts_per_instance) {
 164       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 165                               sizeof(prim_packet), LOOP_CLIPRECTS);
 166    }
 167    if (intel->always_flush_cache) {
 168       BEGIN_BATCH(1, IGNORE_CLIPRECTS);
 169       OUT_BATCH(intel->vtbl.flush_cmd());
 170       ADVANCE_BATCH();
 171    }
 172
 173    brw->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
 191       if (arrays[i]->StrideB != 0)
 192          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 193             ((i%16) * 2);
 194    }
 195
 196    /* Raise statechanges if input sizes have changed. */
 197    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 198       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 199 }
 200
 201 /* XXX: could split the primitive list to fallback only on the
 202  * non-conformant primitives.
 203  */
 204 static GLboolean check_fallbacks( struct brw_context *brw,
 205                                   const struct _mesa_prim *prim,
 206                                   GLuint nr_prims )
 207 {
 208    GLcontext *ctx = &brw->intel.ctx;
 209    GLuint i;
 210
 211    /* If we don't require strict OpenGL conformance, never
 212     * use fallbacks.  If we're forcing fallbacks, always
 213     * use fallfacks.
 214     */
 215    if (brw->intel.conformance_mode == 0)
 216       return GL_FALSE;
 217
 218    if (brw->intel.conformance_mode == 2)
 219       return GL_TRUE;
 220
 221    if (ctx->Polygon.SmoothFlag) {
 222       for (i = 0; i < nr_prims; i++)
 223          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 224             return GL_TRUE;
 225    }
 226
 227    /* BRW hardware will do AA lines, but they are non-conformant it
 228     * seems.  TBD whether we keep this fallback:
 229     */
 230    if (ctx->Line.SmoothFlag) {
 231       for (i = 0; i < nr_prims; i++)
 232          if (reduced_prim[prim[i].mode] == GL_LINES)
 233             return GL_TRUE;
 234    }
 235
 236    /* Stipple -- these fallbacks could be resolved with a little
 237     * bit of work?
 238     */
 239    if (ctx->Line.StippleFlag) {
 240       for (i = 0; i < nr_prims; i++) {
 241          /* GS doesn't get enough information to know when to reset
 242           * the stipple counter?!?
 243           */
 244          if (prim[i].mode == GL_LINE_LOOP || prim[i].mode == GL_LINE_STRIP)
 245             return GL_TRUE;
 246
 247          if (prim[i].mode == GL_POLYGON &&
 248              (ctx->Polygon.FrontMode == GL_LINE ||
 249               ctx->Polygon.BackMode == GL_LINE))
 250             return GL_TRUE;
 251       }
 252    }
 253
 254    if (ctx->Point.SmoothFlag) {
 255       for (i = 0; i < nr_prims; i++)
 256          if (prim[i].mode == GL_POINTS)
 257             return GL_TRUE;
 258    }
 259
 260    /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
 261     * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
 262     * as GL_CLAMP_TO_EDGE instead.  If we're using GL_CLAMP, and
 263     * we want strict conformance, force the fallback.
 264     * Right now, we only do this for 2D textures.
 265     */
 266    {
 267       int u;
 268       for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
 269          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[u];
 270          if (texUnit->Enabled) {
 271             if (texUnit->Enabled & TEXTURE_1D_BIT) {
 272                if (texUnit->CurrentTex[TEXTURE_1D_INDEX]->WrapS == GL_CLAMP) {
 273                    return GL_TRUE;
 274                }
 275             }
 276             if (texUnit->Enabled & TEXTURE_2D_BIT) {
 277                if (texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapS == GL_CLAMP ||
 278                    texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapT == GL_CLAMP) {
 279                    return GL_TRUE;
 280                }
 281             }
 282             if (texUnit->Enabled & TEXTURE_3D_BIT) {
 283                if (texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapS == GL_CLAMP ||
 284                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapT == GL_CLAMP ||
 285                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapR == GL_CLAMP) {
 286                    return GL_TRUE;
 287                }
 288             }
 289          }
 290       }
 291    }
 292
 293    /* Nothing stopping us from the fast path now */
 294    return GL_FALSE;
 295 }
 296
 297 /* May fail if out of video memory for texture or vbo upload, or on
 298  * fallback conditions.
 299  */
 300 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 301                                      const struct gl_client_array *arrays[],
 302                                      const struct _mesa_prim *prim,
 303                                      GLuint nr_prims,
 304                                      const struct _mesa_index_buffer *ib,
 305                                      GLuint min_index,
 306                                      GLuint max_index )
 307 {
 308    struct intel_context *intel = intel_context(ctx);
 309    struct brw_context *brw = brw_context(ctx);
 310    GLboolean retval = GL_FALSE;
 311    GLboolean warn = GL_FALSE;
 312    GLboolean first_time = GL_TRUE;
 313    GLuint i;
 314
 315    if (ctx->NewState)
 316       _mesa_update_state( ctx );
 317
 318    /* We have to validate the textures *before* checking for fallbacks;
 319     * otherwise, the software fallback won't be able to rely on the
 320     * texture state, the firstLevel and lastLevel fields won't be
 321     * set in the intel texture object (they'll both be 0), and the
 322     * software fallback will segfault if it attempts to access any
 323     * texture level other than level 0.
 324     */
 325    brw_validate_textures( brw );
 326
 327    if (check_fallbacks(brw, prim, nr_prims))
 328       return GL_FALSE;
 329
 330    /* Bind all inputs, derive varying and size information:
 331     */
 332    brw_merge_inputs( brw, arrays );
 333
 334    brw->ib.ib = ib;
 335    brw->state.dirty.brw |= BRW_NEW_INDICES;
 336
 337    brw->vb.min_index = min_index;
 338    brw->vb.max_index = max_index;
 339    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 340
 341    /* Have to validate state quite late.  Will rebuild tnl_program,
 342     * which depends on varying information.
 343     *
 344     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 345     * so can't access it earlier.
 346     */
 347
 348    LOCK_HARDWARE(intel);
 349
 350    if (!intel->constant_cliprect && intel->driDrawable->numClipRects == 0) {
 351       UNLOCK_HARDWARE(intel);
 352       return GL_TRUE;
 353    }
 354
 355    for (i = 0; i < nr_prims; i++) {
 356       uint32_t hw_prim;
 357
 358       /* Flush the batch if it's approaching full, so that we don't wrap while
 359        * we've got validated state that needs to be in the same batch as the
 360        * primitives.  This fraction is just a guess (minimal full state plus
 361        * a primitive is around 512 bytes), and would be better if we had
 362        * an upper bound of how much we might emit in a single
 363        * brw_try_draw_prims().
 364        */
 365       intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4,
 366                                       LOOP_CLIPRECTS);
 367
 368       hw_prim = brw_set_prim(brw, prim[i].mode);
 369
 370       if (first_time || (brw->state.dirty.brw & BRW_NEW_PRIMITIVE)) {
 371          first_time = GL_FALSE;
 372
 373          brw_validate_state(brw);
 374
 375          /* Various fallback checks:  */
 376          if (brw->intel.Fallback)
 377             goto out;
 378
 379          /* Check that we can fit our state in with our existing batchbuffer, or
 380           * flush otherwise.
 381           */
 382          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 383                                              brw->state.validated_bo_count)) {
 384             static GLboolean warned;
 385             intel_batchbuffer_flush(intel->batch);
 386
 387             /* Validate the state after we flushed the batch (which would have
 388              * changed the set of dirty state).  If we still fail to
 389              * check_aperture, warn of what's happening, but attempt to continue
 390              * on since it may succeed anyway, and the user would probably rather
 391              * see a failure and a warning than a fallback.
 392              */
 393             brw_validate_state(brw);
 394             if (!warned &&
 395                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 396                                                 brw->state.validated_bo_count)) {
 397                warn = GL_TRUE;
 398                warned = GL_TRUE;
 399             }
 400          }
 401
 402          brw_upload_state(brw);
 403       }
 404
 405       brw_emit_prim(brw, &prim[i], hw_prim);
 406
 407       retval = GL_TRUE;
 408    }
 409
 410    if (intel->always_flush_batch)
 411       intel_batchbuffer_flush(intel->batch);
 412  out:
 413    UNLOCK_HARDWARE(intel);
 414
 415    brw_state_cache_check_size(brw);
 416
 417    if (warn)
 418       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 419               "available aperture space\n");
 420
 421    if (!retval)
 422       DBG("%s failed\n", __FUNCTION__);
 423
 424    return retval;
 425 }
 426
 427 void brw_draw_prims( GLcontext *ctx,
 428                      const struct gl_client_array *arrays[],
 429                      const struct _mesa_prim *prim,
 430                      GLuint nr_prims,
 431                      const struct _mesa_index_buffer *ib,
 432                      GLboolean index_bounds_valid,
 433                      GLuint min_index,
 434                      GLuint max_index )
 435 {
 436    GLboolean retval;
 437
 438    if (!vbo_all_varyings_in_vbos(arrays)) {
 439       if (!index_bounds_valid)
 440          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 441
 442       /* Decide if we want to rebase.  If so we end up recursing once
 443        * only into this function.
 444        */
 445       if (min_index != 0) {
 446          vbo_rebase_prims(ctx, arrays,
 447                           prim, nr_prims,
 448                           ib, min_index, max_index,
 449                           brw_draw_prims );
 450          return;
 451       }
 452    }
 453
 454    /* Make a first attempt at drawing:
 455     */
 456    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 457
 458    /* Otherwise, we really are out of memory.  Pass the drawing
 459     * command to the software tnl module and which will in turn call
 460     * swrast to do the drawing.
 461     */
 462    if (!retval) {
 463        _swsetup_Wakeup(ctx);
 464       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 465    }
 466
 467 }
 468
 469 void brw_draw_init( struct brw_context *brw )
 470 {
 471    GLcontext *ctx = &brw->intel.ctx;
 472    struct vbo_context *vbo = vbo_context(ctx);
 473
 474    /* Register our drawing function:
 475     */
 476    vbo->draw_prims = brw_draw_prims;
 477 }
 478
 479 void brw_draw_destroy( struct brw_context *brw )
 480 {
 481    int i;
 482
 483    if (brw->vb.upload.bo != NULL) {
 484       dri_bo_unreference(brw->vb.upload.bo);
 485       brw->vb.upload.bo = NULL;
 486    }
 487
 488    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 489       dri_bo_unreference(brw->vb.inputs[i].bo);
 490       brw->vb.inputs[i].bo = NULL;
 491    }
 492
 493    dri_bo_unreference(brw->ib.bo);
 494    brw->ib.bo = NULL;
 495 }