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, GLenum prim)
  81 {
  82    GLcontext *ctx = &brw->intel.ctx;
  83
  84    if (INTEL_DEBUG & DEBUG_PRIMS)
  85       printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
  86
  87    /* Slight optimization to avoid the GS program when not needed:
  88     */
  89    if (prim == GL_QUAD_STRIP &&
  90        ctx->Light.ShadeModel != GL_FLAT &&
  91        ctx->Polygon.FrontMode == GL_FILL &&
  92        ctx->Polygon.BackMode == GL_FILL)
  93       prim = GL_TRIANGLE_STRIP;
  94
  95    if (prim != brw->primitive) {
  96       brw->primitive = prim;
  97       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
  98
  99       if (reduced_prim[prim] != brw->intel.reduced_primitive) {
 100          brw->intel.reduced_primitive = reduced_prim[prim];
 101          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 102       }
 103    }
 104
 105    return prim_to_hw_prim[prim];
 106 }
 107
 108
 109 static GLuint trim(GLenum prim, GLuint length)
 110 {
 111    if (prim == GL_QUAD_STRIP)
 112       return length > 3 ? (length - length % 2) : 0;
 113    else if (prim == GL_QUADS)
 114       return length - length % 4;
 115    else
 116       return length;
 117 }
 118
 119
 120 static void brw_emit_prim(struct brw_context *brw,
 121                           const struct _mesa_prim *prim,
 122                           uint32_t hw_prim)
 123 {
 124    struct brw_3d_primitive prim_packet;
 125    struct intel_context *intel = &brw->intel;
 126
 127    if (INTEL_DEBUG & DEBUG_PRIMS)
 128       printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 129                    prim->start, prim->count);
 130
 131    prim_packet.header.opcode = CMD_3D_PRIM;
 132    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 133    prim_packet.header.pad = 0;
 134    prim_packet.header.topology = hw_prim;
 135    prim_packet.header.indexed = prim->indexed;
 136
 137    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 138    prim_packet.start_vert_location = prim->start;
 139    if (prim->indexed)
 140       prim_packet.start_vert_location += brw->ib.start_vertex_offset;
 141    prim_packet.instance_count = 1;
 142    prim_packet.start_instance_location = 0;
 143    prim_packet.base_vert_location = prim->basevertex;
 144
 145    /* Can't wrap here, since we rely on the validated state. */
 146    intel->no_batch_wrap = GL_TRUE;
 147
 148    /* If we're set to always flush, do it before and after the primitive emit.
 149     * We want to catch both missed flushes that hurt instruction/state cache
 150     * and missed flushes of the render cache as it heads to other parts of
 151     * the besides the draw code.
 152     */
 153    if (intel->always_flush_cache) {
 154       intel_batchbuffer_emit_mi_flush(intel->batch);
 155    }
 156    if (prim_packet.verts_per_instance) {
 157       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 158                               sizeof(prim_packet));
 159    }
 160    if (intel->always_flush_cache) {
 161       intel_batchbuffer_emit_mi_flush(intel->batch);
 162    }
 163
 164    intel->no_batch_wrap = GL_FALSE;
 165 }
 166
 167 static void brw_merge_inputs( struct brw_context *brw,
 168                        const struct gl_client_array *arrays[])
 169 {
 170    struct brw_vertex_info old = brw->vb.info;
 171    GLuint i;
 172
 173    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 174       dri_bo_unreference(brw->vb.inputs[i].bo);
 175
 176    memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs));
 177    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 178
 179    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 180       brw->vb.inputs[i].glarray = arrays[i];
 181       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 182
 183       if (arrays[i]->StrideB != 0)
 184          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 185             ((i%16) * 2);
 186    }
 187
 188    /* Raise statechanges if input sizes have changed. */
 189    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 190       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 191 }
 192
 193 /* XXX: could split the primitive list to fallback only on the
 194  * non-conformant primitives.
 195  */
 196 static GLboolean check_fallbacks( struct brw_context *brw,
 197                                   const struct _mesa_prim *prim,
 198                                   GLuint nr_prims )
 199 {
 200    GLcontext *ctx = &brw->intel.ctx;
 201    GLuint i;
 202
 203    /* If we don't require strict OpenGL conformance, never
 204     * use fallbacks.  If we're forcing fallbacks, always
 205     * use fallfacks.
 206     */
 207    if (brw->intel.conformance_mode == 0)
 208       return GL_FALSE;
 209
 210    if (brw->intel.conformance_mode == 2)
 211       return GL_TRUE;
 212
 213    if (ctx->Polygon.SmoothFlag) {
 214       for (i = 0; i < nr_prims; i++)
 215          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 216             return GL_TRUE;
 217    }
 218
 219    /* BRW hardware will do AA lines, but they are non-conformant it
 220     * seems.  TBD whether we keep this fallback:
 221     */
 222    if (ctx->Line.SmoothFlag) {
 223       for (i = 0; i < nr_prims; i++)
 224          if (reduced_prim[prim[i].mode] == GL_LINES)
 225             return GL_TRUE;
 226    }
 227
 228    /* Stipple -- these fallbacks could be resolved with a little
 229     * bit of work?
 230     */
 231    if (ctx->Line.StippleFlag) {
 232       for (i = 0; i < nr_prims; i++) {
 233          /* GS doesn't get enough information to know when to reset
 234           * the stipple counter?!?
 235           */
 236          if (prim[i].mode == GL_LINE_LOOP || prim[i].mode == GL_LINE_STRIP)
 237             return GL_TRUE;
 238
 239          if (prim[i].mode == GL_POLYGON &&
 240              (ctx->Polygon.FrontMode == GL_LINE ||
 241               ctx->Polygon.BackMode == GL_LINE))
 242             return GL_TRUE;
 243       }
 244    }
 245
 246    if (ctx->Point.SmoothFlag) {
 247       for (i = 0; i < nr_prims; i++)
 248          if (prim[i].mode == GL_POINTS)
 249             return GL_TRUE;
 250    }
 251
 252    /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
 253     * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
 254     * as GL_CLAMP_TO_EDGE instead.  If we're using GL_CLAMP, and
 255     * we want strict conformance, force the fallback.
 256     * Right now, we only do this for 2D textures.
 257     */
 258    {
 259       int u;
 260       for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
 261          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[u];
 262          if (texUnit->Enabled) {
 263             if (texUnit->Enabled & TEXTURE_1D_BIT) {
 264                if (texUnit->CurrentTex[TEXTURE_1D_INDEX]->WrapS == GL_CLAMP) {
 265                    return GL_TRUE;
 266                }
 267             }
 268             if (texUnit->Enabled & TEXTURE_2D_BIT) {
 269                if (texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapS == GL_CLAMP ||
 270                    texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapT == GL_CLAMP) {
 271                    return GL_TRUE;
 272                }
 273             }
 274             if (texUnit->Enabled & TEXTURE_3D_BIT) {
 275                if (texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapS == GL_CLAMP ||
 276                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapT == GL_CLAMP ||
 277                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapR == GL_CLAMP) {
 278                    return GL_TRUE;
 279                }
 280             }
 281          }
 282       }
 283    }
 284
 285    /* Nothing stopping us from the fast path now */
 286    return GL_FALSE;
 287 }
 288
 289 /* May fail if out of video memory for texture or vbo upload, or on
 290  * fallback conditions.
 291  */
 292 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 293                                      const struct gl_client_array *arrays[],
 294                                      const struct _mesa_prim *prim,
 295                                      GLuint nr_prims,
 296                                      const struct _mesa_index_buffer *ib,
 297                                      GLuint min_index,
 298                                      GLuint max_index )
 299 {
 300    struct intel_context *intel = intel_context(ctx);
 301    struct brw_context *brw = brw_context(ctx);
 302    GLboolean retval = GL_FALSE;
 303    GLboolean warn = GL_FALSE;
 304    GLboolean first_time = GL_TRUE;
 305    GLuint i;
 306
 307    if (ctx->NewState)
 308       _mesa_update_state( ctx );
 309
 310    /* We have to validate the textures *before* checking for fallbacks;
 311     * otherwise, the software fallback won't be able to rely on the
 312     * texture state, the firstLevel and lastLevel fields won't be
 313     * set in the intel texture object (they'll both be 0), and the
 314     * software fallback will segfault if it attempts to access any
 315     * texture level other than level 0.
 316     */
 317    brw_validate_textures( brw );
 318
 319    if (check_fallbacks(brw, prim, nr_prims))
 320       return GL_FALSE;
 321
 322    /* Bind all inputs, derive varying and size information:
 323     */
 324    brw_merge_inputs( brw, arrays );
 325
 326    brw->ib.ib = ib;
 327    brw->state.dirty.brw |= BRW_NEW_INDICES;
 328
 329    brw->vb.min_index = min_index;
 330    brw->vb.max_index = max_index;
 331    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 332
 333    /* Have to validate state quite late.  Will rebuild tnl_program,
 334     * which depends on varying information.
 335     *
 336     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 337     * so can't access it earlier.
 338     */
 339
 340    intel_prepare_render(intel);
 341
 342    for (i = 0; i < nr_prims; i++) {
 343       uint32_t hw_prim;
 344
 345       /* Flush the batch if it's approaching full, so that we don't wrap while
 346        * we've got validated state that needs to be in the same batch as the
 347        * primitives.  This fraction is just a guess (minimal full state plus
 348        * a primitive is around 512 bytes), and would be better if we had
 349        * an upper bound of how much we might emit in a single
 350        * brw_try_draw_prims().
 351        */
 352       intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4);
 353
 354       hw_prim = brw_set_prim(brw, prim[i].mode);
 355
 356       if (first_time || (brw->state.dirty.brw & BRW_NEW_PRIMITIVE)) {
 357          first_time = GL_FALSE;
 358
 359          brw_validate_state(brw);
 360
 361          /* Various fallback checks:  */
 362          if (brw->intel.Fallback)
 363             goto out;
 364
 365          /* Check that we can fit our state in with our existing batchbuffer, or
 366           * flush otherwise.
 367           */
 368          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 369                                              brw->state.validated_bo_count)) {
 370             static GLboolean warned;
 371             intel_batchbuffer_flush(intel->batch);
 372
 373             /* Validate the state after we flushed the batch (which would have
 374              * changed the set of dirty state).  If we still fail to
 375              * check_aperture, warn of what's happening, but attempt to continue
 376              * on since it may succeed anyway, and the user would probably rather
 377              * see a failure and a warning than a fallback.
 378              */
 379             brw_validate_state(brw);
 380             if (!warned &&
 381                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 382                                                 brw->state.validated_bo_count)) {
 383                warn = GL_TRUE;
 384                warned = GL_TRUE;
 385             }
 386          }
 387
 388          brw_upload_state(brw);
 389       }
 390
 391       brw_emit_prim(brw, &prim[i], hw_prim);
 392
 393       retval = GL_TRUE;
 394    }
 395
 396    if (intel->always_flush_batch)
 397       intel_batchbuffer_flush(intel->batch);
 398  out:
 399
 400    brw_state_cache_check_size(brw);
 401
 402    if (warn)
 403       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 404               "available aperture space\n");
 405
 406    if (!retval)
 407       DBG("%s failed\n", __FUNCTION__);
 408
 409    return retval;
 410 }
 411
 412 void brw_draw_prims( GLcontext *ctx,
 413                      const struct gl_client_array *arrays[],
 414                      const struct _mesa_prim *prim,
 415                      GLuint nr_prims,
 416                      const struct _mesa_index_buffer *ib,
 417                      GLboolean index_bounds_valid,
 418                      GLuint min_index,
 419                      GLuint max_index )
 420 {
 421    GLboolean retval;
 422
 423    if (!vbo_all_varyings_in_vbos(arrays)) {
 424       if (!index_bounds_valid)
 425          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 426
 427       /* Decide if we want to rebase.  If so we end up recursing once
 428        * only into this function.
 429        */
 430       if (min_index != 0) {
 431          vbo_rebase_prims(ctx, arrays,
 432                           prim, nr_prims,
 433                           ib, min_index, max_index,
 434                           brw_draw_prims );
 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    /* Otherwise, we really are out of memory.  Pass the drawing
 444     * command to the software tnl module and which will in turn call
 445     * swrast to do the drawing.
 446     */
 447    if (!retval) {
 448        _swsetup_Wakeup(ctx);
 449       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 450    }
 451
 452 }
 453
 454 void brw_draw_init( struct brw_context *brw )
 455 {
 456    GLcontext *ctx = &brw->intel.ctx;
 457    struct vbo_context *vbo = vbo_context(ctx);
 458
 459    /* Register our drawing function:
 460     */
 461    vbo->draw_prims = brw_draw_prims;
 462 }
 463
 464 void brw_draw_destroy( struct brw_context *brw )
 465 {
 466    int i;
 467
 468    if (brw->vb.upload.bo != NULL) {
 469       dri_bo_unreference(brw->vb.upload.bo);
 470       brw->vb.upload.bo = NULL;
 471    }
 472
 473    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 474       dri_bo_unreference(brw->vb.inputs[i].bo);
 475       brw->vb.inputs[i].bo = NULL;
 476    }
 477
 478    dri_bo_unreference(brw->ib.bo);
 479    brw->ib.bo = NULL;
 480 }