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    /* If we're set to always flush, do it before and after the primitive emit.
 155     * We want to catch both missed flushes that hurt instruction/state cache
 156     * and missed flushes of the render cache as it heads to other parts of
 157     * the besides the draw code.
 158     */
 159    if (intel->always_flush_cache) {
 160       intel_batchbuffer_emit_mi_flush(intel->batch);
 161    }
 162    if (prim_packet.verts_per_instance) {
 163       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 164                               sizeof(prim_packet));
 165    }
 166    if (intel->always_flush_cache) {
 167       intel_batchbuffer_emit_mi_flush(intel->batch);
 168    }
 169 }
 170
 171 static void brw_merge_inputs( struct brw_context *brw,
 172                        const struct gl_client_array *arrays[])
 173 {
 174    struct brw_vertex_info old = brw->vb.info;
 175    GLuint i;
 176
 177    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 178       drm_intel_bo_unreference(brw->vb.inputs[i].bo);
 179
 180    memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs));
 181    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 182
 183    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 184       brw->vb.inputs[i].glarray = arrays[i];
 185       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 186
 187       if (arrays[i]->StrideB != 0)
 188          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 189             ((i%16) * 2);
 190    }
 191
 192    /* Raise statechanges if input sizes have changed. */
 193    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 194       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 195 }
 196
 197 /* XXX: could split the primitive list to fallback only on the
 198  * non-conformant primitives.
 199  */
 200 static GLboolean check_fallbacks( struct brw_context *brw,
 201                                   const struct _mesa_prim *prim,
 202                                   GLuint nr_prims )
 203 {
 204    GLcontext *ctx = &brw->intel.ctx;
 205    GLuint i;
 206
 207    /* XXX FIXME */
 208    if (brw->intel.gen >= 6) {
 209        for (i = 0; i < nr_prims; i++)
 210            if (prim[i].mode == GL_LINE_LOOP)
 211                return GL_TRUE;
 212    }
 213
 214    /* If we don't require strict OpenGL conformance, never
 215     * use fallbacks.  If we're forcing fallbacks, always
 216     * use fallfacks.
 217     */
 218    if (brw->intel.conformance_mode == 0)
 219       return GL_FALSE;
 220
 221    if (brw->intel.conformance_mode == 2)
 222       return GL_TRUE;
 223
 224    if (ctx->Polygon.SmoothFlag) {
 225       for (i = 0; i < nr_prims; i++)
 226          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 227             return GL_TRUE;
 228    }
 229
 230    /* BRW hardware will do AA lines, but they are non-conformant it
 231     * seems.  TBD whether we keep this fallback:
 232     */
 233    if (ctx->Line.SmoothFlag) {
 234       for (i = 0; i < nr_prims; i++)
 235          if (reduced_prim[prim[i].mode] == GL_LINES)
 236             return GL_TRUE;
 237    }
 238
 239    /* Stipple -- these fallbacks could be resolved with a little
 240     * bit of work?
 241     */
 242    if (ctx->Line.StippleFlag) {
 243       for (i = 0; i < nr_prims; i++) {
 244          /* GS doesn't get enough information to know when to reset
 245           * the stipple counter?!?
 246           */
 247          if (prim[i].mode == GL_LINE_LOOP || prim[i].mode == GL_LINE_STRIP)
 248             return GL_TRUE;
 249
 250          if (prim[i].mode == GL_POLYGON &&
 251              (ctx->Polygon.FrontMode == GL_LINE ||
 252               ctx->Polygon.BackMode == GL_LINE))
 253             return GL_TRUE;
 254       }
 255    }
 256
 257    if (ctx->Point.SmoothFlag) {
 258       for (i = 0; i < nr_prims; i++)
 259          if (prim[i].mode == GL_POINTS)
 260             return GL_TRUE;
 261    }
 262
 263    /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
 264     * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
 265     * as GL_CLAMP_TO_EDGE instead.  If we're using GL_CLAMP, and
 266     * we want strict conformance, force the fallback.
 267     * Right now, we only do this for 2D textures.
 268     */
 269    {
 270       int u;
 271       for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
 272          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[u];
 273          if (texUnit->Enabled) {
 274             if (texUnit->Enabled & TEXTURE_1D_BIT) {
 275                if (texUnit->CurrentTex[TEXTURE_1D_INDEX]->WrapS == GL_CLAMP) {
 276                    return GL_TRUE;
 277                }
 278             }
 279             if (texUnit->Enabled & TEXTURE_2D_BIT) {
 280                if (texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapS == GL_CLAMP ||
 281                    texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapT == GL_CLAMP) {
 282                    return GL_TRUE;
 283                }
 284             }
 285             if (texUnit->Enabled & TEXTURE_3D_BIT) {
 286                if (texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapS == GL_CLAMP ||
 287                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapT == GL_CLAMP ||
 288                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapR == GL_CLAMP) {
 289                    return GL_TRUE;
 290                }
 291             }
 292          }
 293       }
 294    }
 295
 296    /* Nothing stopping us from the fast path now */
 297    return GL_FALSE;
 298 }
 299
 300 /* May fail if out of video memory for texture or vbo upload, or on
 301  * fallback conditions.
 302  */
 303 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 304                                      const struct gl_client_array *arrays[],
 305                                      const struct _mesa_prim *prim,
 306                                      GLuint nr_prims,
 307                                      const struct _mesa_index_buffer *ib,
 308                                      GLuint min_index,
 309                                      GLuint max_index )
 310 {
 311    struct intel_context *intel = intel_context(ctx);
 312    struct brw_context *brw = brw_context(ctx);
 313    GLboolean retval = GL_FALSE;
 314    GLboolean warn = GL_FALSE;
 315    GLboolean first_time = GL_TRUE;
 316    GLuint i;
 317
 318    if (ctx->NewState)
 319       _mesa_update_state( ctx );
 320
 321    /* We have to validate the textures *before* checking for fallbacks;
 322     * otherwise, the software fallback won't be able to rely on the
 323     * texture state, the firstLevel and lastLevel fields won't be
 324     * set in the intel texture object (they'll both be 0), and the
 325     * software fallback will segfault if it attempts to access any
 326     * texture level other than level 0.
 327     */
 328    brw_validate_textures( brw );
 329
 330    if (check_fallbacks(brw, prim, nr_prims))
 331       return GL_FALSE;
 332
 333    /* Bind all inputs, derive varying and size information:
 334     */
 335    brw_merge_inputs( brw, arrays );
 336
 337    brw->ib.ib = ib;
 338    brw->state.dirty.brw |= BRW_NEW_INDICES;
 339
 340    brw->vb.min_index = min_index;
 341    brw->vb.max_index = max_index;
 342    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 343
 344    /* Have to validate state quite late.  Will rebuild tnl_program,
 345     * which depends on varying information.
 346     *
 347     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 348     * so can't access it earlier.
 349     */
 350
 351    intel_prepare_render(intel);
 352
 353    for (i = 0; i < nr_prims; i++) {
 354       uint32_t hw_prim;
 355
 356       /* Flush the batch if it's approaching full, so that we don't wrap while
 357        * we've got validated state that needs to be in the same batch as the
 358        * primitives.  This fraction is just a guess (minimal full state plus
 359        * a primitive is around 512 bytes), and would be better if we had
 360        * an upper bound of how much we might emit in a single
 361        * brw_try_draw_prims().
 362        */
 363       intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4);
 364
 365       hw_prim = brw_set_prim(brw, &prim[i]);
 366
 367       if (first_time || (brw->state.dirty.brw & BRW_NEW_PRIMITIVE)) {
 368          first_time = GL_FALSE;
 369
 370          brw_validate_state(brw);
 371
 372          /* Various fallback checks:  */
 373          if (brw->intel.Fallback)
 374             goto out;
 375
 376          /* Check that we can fit our state in with our existing batchbuffer, or
 377           * flush otherwise.
 378           */
 379          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 380                                              brw->state.validated_bo_count)) {
 381             static GLboolean warned;
 382             intel_batchbuffer_flush(intel->batch);
 383
 384             /* Validate the state after we flushed the batch (which would have
 385              * changed the set of dirty state).  If we still fail to
 386              * check_aperture, warn of what's happening, but attempt to continue
 387              * on since it may succeed anyway, and the user would probably rather
 388              * see a failure and a warning than a fallback.
 389              */
 390             brw_validate_state(brw);
 391             if (!warned &&
 392                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 393                                                 brw->state.validated_bo_count)) {
 394                warn = GL_TRUE;
 395                warned = GL_TRUE;
 396             }
 397          }
 398
 399          intel->no_batch_wrap = GL_TRUE;
 400          brw_upload_state(brw);
 401       }
 402
 403       brw_emit_prim(brw, &prim[i], hw_prim);
 404
 405       intel->no_batch_wrap = GL_FALSE;
 406
 407       retval = GL_TRUE;
 408    }
 409
 410    if (intel->always_flush_batch)
 411       intel_batchbuffer_flush(intel->batch);
 412  out:
 413
 414    brw_state_cache_check_size(brw);
 415
 416    if (warn)
 417       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 418               "available aperture space\n");
 419
 420    if (!retval)
 421       DBG("%s failed\n", __FUNCTION__);
 422
 423    return retval;
 424 }
 425
 426 void brw_draw_prims( GLcontext *ctx,
 427                      const struct gl_client_array *arrays[],
 428                      const struct _mesa_prim *prim,
 429                      GLuint nr_prims,
 430                      const struct _mesa_index_buffer *ib,
 431                      GLboolean index_bounds_valid,
 432                      GLuint min_index,
 433                      GLuint max_index )
 434 {
 435    GLboolean retval;
 436
 437    if (!vbo_all_varyings_in_vbos(arrays)) {
 438       if (!index_bounds_valid)
 439          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 440
 441       /* Decide if we want to rebase.  If so we end up recursing once
 442        * only into this function.
 443        */
 444       if (min_index != 0) {
 445          vbo_rebase_prims(ctx, arrays,
 446                           prim, nr_prims,
 447                           ib, min_index, max_index,
 448                           brw_draw_prims );
 449          return;
 450       }
 451    }
 452
 453    /* Make a first attempt at drawing:
 454     */
 455    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 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 }
 467
 468 void brw_draw_init( struct brw_context *brw )
 469 {
 470    GLcontext *ctx = &brw->intel.ctx;
 471    struct vbo_context *vbo = vbo_context(ctx);
 472
 473    /* Register our drawing function:
 474     */
 475    vbo->draw_prims = brw_draw_prims;
 476 }
 477
 478 void brw_draw_destroy( struct brw_context *brw )
 479 {
 480    int i;
 481
 482    if (brw->vb.upload.bo != NULL) {
 483       drm_intel_bo_unreference(brw->vb.upload.bo);
 484       brw->vb.upload.bo = NULL;
 485    }
 486
 487    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 488       drm_intel_bo_unreference(brw->vb.inputs[i].bo);
 489       brw->vb.inputs[i].bo = NULL;
 490    }
 491
 492    drm_intel_bo_unreference(brw->ib.bo);
 493    brw->ib.bo = NULL;
 494 }