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_PRIMS
  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    struct gl_context *ctx = &brw->intel.ctx;
  84    GLenum mode = prim->mode;
  85
  86    DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
  87
  88    /* Slight optimization to avoid the GS program when not needed:
  89     */
  90    if (mode == GL_QUAD_STRIP &&
  91        ctx->Light.ShadeModel != GL_FLAT &&
  92        ctx->Polygon.FrontMode == GL_FILL &&
  93        ctx->Polygon.BackMode == GL_FILL)
  94       mode = GL_TRIANGLE_STRIP;
  95
  96    if (prim->mode == GL_QUADS && prim->count == 4 &&
  97        ctx->Light.ShadeModel != GL_FLAT &&
  98        ctx->Polygon.FrontMode == GL_FILL &&
  99        ctx->Polygon.BackMode == GL_FILL) {
 100       mode = GL_TRIANGLE_FAN;
 101    }
 102
 103    if (mode != brw->primitive) {
 104       brw->primitive = mode;
 105       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 106
 107       if (reduced_prim[mode] != brw->intel.reduced_primitive) {
 108          brw->intel.reduced_primitive = reduced_prim[mode];
 109          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 110       }
 111    }
 112
 113    return prim_to_hw_prim[mode];
 114 }
 115
 116
 117 static GLuint trim(GLenum prim, GLuint length)
 118 {
 119    if (prim == GL_QUAD_STRIP)
 120       return length > 3 ? (length - length % 2) : 0;
 121    else if (prim == GL_QUADS)
 122       return length - length % 4;
 123    else
 124       return length;
 125 }
 126
 127
 128 static void brw_emit_prim(struct brw_context *brw,
 129                           const struct _mesa_prim *prim,
 130                           uint32_t hw_prim)
 131 {
 132    struct brw_3d_primitive prim_packet;
 133    struct intel_context *intel = &brw->intel;
 134
 135    DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 136        prim->start, prim->count);
 137
 138    prim_packet.header.opcode = CMD_3D_PRIM;
 139    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 140    prim_packet.header.pad = 0;
 141    prim_packet.header.topology = hw_prim;
 142    prim_packet.header.indexed = prim->indexed;
 143
 144    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 145    prim_packet.start_vert_location = prim->start;
 146    if (prim->indexed)
 147       prim_packet.start_vert_location += brw->ib.start_vertex_offset;
 148    prim_packet.instance_count = 1;
 149    prim_packet.start_instance_location = 0;
 150    prim_packet.base_vert_location = prim->basevertex;
 151
 152    /* If we're set to always flush, do it before and after the primitive emit.
 153     * We want to catch both missed flushes that hurt instruction/state cache
 154     * and missed flushes of the render cache as it heads to other parts of
 155     * the besides the draw code.
 156     */
 157    if (intel->always_flush_cache) {
 158       intel_batchbuffer_emit_mi_flush(intel->batch);
 159    }
 160    if (prim_packet.verts_per_instance) {
 161       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 162                               sizeof(prim_packet));
 163    }
 164    if (intel->always_flush_cache) {
 165       intel_batchbuffer_emit_mi_flush(intel->batch);
 166    }
 167 }
 168
 169 static void brw_merge_inputs( struct brw_context *brw,
 170                        const struct gl_client_array *arrays[])
 171 {
 172    struct brw_vertex_info old = brw->vb.info;
 173    GLuint i;
 174
 175    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 176       drm_intel_bo_unreference(brw->vb.inputs[i].bo);
 177
 178    memset(&brw->vb.inputs, 0, sizeof(brw->vb.inputs));
 179    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 180
 181    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 182       brw->vb.inputs[i].glarray = arrays[i];
 183       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 184
 185       if (arrays[i]->StrideB != 0)
 186          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 187             ((i%16) * 2);
 188    }
 189
 190    /* Raise statechanges if input sizes have changed. */
 191    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 192       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 193 }
 194
 195 /* XXX: could split the primitive list to fallback only on the
 196  * non-conformant primitives.
 197  */
 198 static GLboolean check_fallbacks( struct brw_context *brw,
 199                                   const struct _mesa_prim *prim,
 200                                   GLuint nr_prims )
 201 {
 202    struct gl_context *ctx = &brw->intel.ctx;
 203    GLuint i;
 204
 205    /* If we don't require strict OpenGL conformance, never
 206     * use fallbacks.  If we're forcing fallbacks, always
 207     * use fallfacks.
 208     */
 209    if (brw->intel.conformance_mode == 0)
 210       return GL_FALSE;
 211
 212    if (brw->intel.conformance_mode == 2)
 213       return GL_TRUE;
 214
 215    if (ctx->Polygon.SmoothFlag) {
 216       for (i = 0; i < nr_prims; i++)
 217          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 218             return GL_TRUE;
 219    }
 220
 221    /* BRW hardware will do AA lines, but they are non-conformant it
 222     * seems.  TBD whether we keep this fallback:
 223     */
 224    if (ctx->Line.SmoothFlag) {
 225       for (i = 0; i < nr_prims; i++)
 226          if (reduced_prim[prim[i].mode] == GL_LINES)
 227             return GL_TRUE;
 228    }
 229
 230    /* Stipple -- these fallbacks could be resolved with a little
 231     * bit of work?
 232     */
 233    if (ctx->Line.StippleFlag) {
 234       for (i = 0; i < nr_prims; i++) {
 235          /* GS doesn't get enough information to know when to reset
 236           * the stipple counter?!?
 237           */
 238          if (prim[i].mode == GL_LINE_LOOP || prim[i].mode == GL_LINE_STRIP)
 239             return GL_TRUE;
 240
 241          if (prim[i].mode == GL_POLYGON &&
 242              (ctx->Polygon.FrontMode == GL_LINE ||
 243               ctx->Polygon.BackMode == GL_LINE))
 244             return GL_TRUE;
 245       }
 246    }
 247
 248    if (ctx->Point.SmoothFlag) {
 249       for (i = 0; i < nr_prims; i++)
 250          if (prim[i].mode == GL_POINTS)
 251             return GL_TRUE;
 252    }
 253
 254    /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
 255     * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
 256     * as GL_CLAMP_TO_EDGE instead.  If we're using GL_CLAMP, and
 257     * we want strict conformance, force the fallback.
 258     * Right now, we only do this for 2D textures.
 259     */
 260    {
 261       int u;
 262       for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
 263          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[u];
 264          if (texUnit->Enabled) {
 265             if (texUnit->Enabled & TEXTURE_1D_BIT) {
 266                if (texUnit->CurrentTex[TEXTURE_1D_INDEX]->WrapS == GL_CLAMP) {
 267                    return GL_TRUE;
 268                }
 269             }
 270             if (texUnit->Enabled & TEXTURE_2D_BIT) {
 271                if (texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapS == GL_CLAMP ||
 272                    texUnit->CurrentTex[TEXTURE_2D_INDEX]->WrapT == GL_CLAMP) {
 273                    return GL_TRUE;
 274                }
 275             }
 276             if (texUnit->Enabled & TEXTURE_3D_BIT) {
 277                if (texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapS == GL_CLAMP ||
 278                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapT == GL_CLAMP ||
 279                    texUnit->CurrentTex[TEXTURE_3D_INDEX]->WrapR == GL_CLAMP) {
 280                    return GL_TRUE;
 281                }
 282             }
 283          }
 284       }
 285    }
 286
 287    /* Nothing stopping us from the fast path now */
 288    return GL_FALSE;
 289 }
 290
 291 /* May fail if out of video memory for texture or vbo upload, or on
 292  * fallback conditions.
 293  */
 294 static GLboolean brw_try_draw_prims( struct gl_context *ctx,
 295                                      const struct gl_client_array *arrays[],
 296                                      const struct _mesa_prim *prim,
 297                                      GLuint nr_prims,
 298                                      const struct _mesa_index_buffer *ib,
 299                                      GLuint min_index,
 300                                      GLuint max_index )
 301 {
 302    struct intel_context *intel = intel_context(ctx);
 303    struct brw_context *brw = brw_context(ctx);
 304    GLboolean retval = GL_FALSE;
 305    GLboolean warn = GL_FALSE;
 306    GLboolean first_time = GL_TRUE;
 307    GLuint i;
 308
 309    if (ctx->NewState)
 310       _mesa_update_state( ctx );
 311
 312    /* We have to validate the textures *before* checking for fallbacks;
 313     * otherwise, the software fallback won't be able to rely on the
 314     * texture state, the firstLevel and lastLevel fields won't be
 315     * set in the intel texture object (they'll both be 0), and the
 316     * software fallback will segfault if it attempts to access any
 317     * texture level other than level 0.
 318     */
 319    brw_validate_textures( brw );
 320
 321    if (check_fallbacks(brw, prim, nr_prims))
 322       return GL_FALSE;
 323
 324    /* Bind all inputs, derive varying and size information:
 325     */
 326    brw_merge_inputs( brw, arrays );
 327
 328    brw->ib.ib = ib;
 329    brw->state.dirty.brw |= BRW_NEW_INDICES;
 330
 331    brw->vb.min_index = min_index;
 332    brw->vb.max_index = max_index;
 333    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 334
 335    /* Have to validate state quite late.  Will rebuild tnl_program,
 336     * which depends on varying information.
 337     *
 338     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 339     * so can't access it earlier.
 340     */
 341
 342    intel_prepare_render(intel);
 343
 344    for (i = 0; i < nr_prims; i++) {
 345       uint32_t hw_prim;
 346
 347       /* Flush the batch if it's approaching full, so that we don't wrap while
 348        * we've got validated state that needs to be in the same batch as the
 349        * primitives.  This fraction is just a guess (minimal full state plus
 350        * a primitive is around 512 bytes), and would be better if we had
 351        * an upper bound of how much we might emit in a single
 352        * brw_try_draw_prims().
 353        */
 354       intel_batchbuffer_require_space(intel->batch, intel->batch->size / 4);
 355
 356       hw_prim = brw_set_prim(brw, &prim[i]);
 357
 358       if (first_time || (brw->state.dirty.brw & BRW_NEW_PRIMITIVE)) {
 359          first_time = GL_FALSE;
 360
 361          brw_validate_state(brw);
 362
 363          /* Various fallback checks:  */
 364          if (brw->intel.Fallback)
 365             goto out;
 366
 367          /* Check that we can fit our state in with our existing batchbuffer, or
 368           * flush otherwise.
 369           */
 370          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 371                                              brw->state.validated_bo_count)) {
 372             static GLboolean warned;
 373             intel_batchbuffer_flush(intel->batch);
 374
 375             /* Validate the state after we flushed the batch (which would have
 376              * changed the set of dirty state).  If we still fail to
 377              * check_aperture, warn of what's happening, but attempt to continue
 378              * on since it may succeed anyway, and the user would probably rather
 379              * see a failure and a warning than a fallback.
 380              */
 381             brw_validate_state(brw);
 382             if (!warned &&
 383                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 384                                                 brw->state.validated_bo_count)) {
 385                warn = GL_TRUE;
 386                warned = GL_TRUE;
 387             }
 388          }
 389
 390          intel->no_batch_wrap = GL_TRUE;
 391          brw_upload_state(brw);
 392       }
 393
 394       brw_emit_prim(brw, &prim[i], hw_prim);
 395
 396       intel->no_batch_wrap = GL_FALSE;
 397
 398       retval = GL_TRUE;
 399    }
 400
 401    if (intel->always_flush_batch)
 402       intel_batchbuffer_flush(intel->batch);
 403  out:
 404
 405    brw_state_cache_check_size(brw);
 406
 407    if (warn)
 408       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 409               "available aperture space\n");
 410
 411    if (!retval)
 412       DBG("%s failed\n", __FUNCTION__);
 413
 414    return retval;
 415 }
 416
 417 void brw_draw_prims( struct gl_context *ctx,
 418                      const struct gl_client_array *arrays[],
 419                      const struct _mesa_prim *prim,
 420                      GLuint nr_prims,
 421                      const struct _mesa_index_buffer *ib,
 422                      GLboolean index_bounds_valid,
 423                      GLuint min_index,
 424                      GLuint max_index )
 425 {
 426    GLboolean retval;
 427
 428    if (!vbo_all_varyings_in_vbos(arrays)) {
 429       if (!index_bounds_valid)
 430          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 431
 432       /* Decide if we want to rebase.  If so we end up recursing once
 433        * only into this function.
 434        */
 435       if (min_index != 0 && !vbo_any_varyings_in_vbos(arrays)) {
 436          vbo_rebase_prims(ctx, arrays,
 437                           prim, nr_prims,
 438                           ib, min_index, max_index,
 439                           brw_draw_prims );
 440          return;
 441       }
 442    }
 443
 444    /* Make a first attempt at drawing:
 445     */
 446    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 447
 448    /* Otherwise, we really are out of memory.  Pass the drawing
 449     * command to the software tnl module and which will in turn call
 450     * swrast to do the drawing.
 451     */
 452    if (!retval) {
 453        _swsetup_Wakeup(ctx);
 454       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 455    }
 456
 457 }
 458
 459 void brw_draw_init( struct brw_context *brw )
 460 {
 461    struct gl_context *ctx = &brw->intel.ctx;
 462    struct vbo_context *vbo = vbo_context(ctx);
 463
 464    /* Register our drawing function:
 465     */
 466    vbo->draw_prims = brw_draw_prims;
 467 }
 468
 469 void brw_draw_destroy( struct brw_context *brw )
 470 {
 471    int i;
 472
 473    if (brw->vb.upload.bo != NULL) {
 474       drm_intel_bo_unreference(brw->vb.upload.bo);
 475       brw->vb.upload.bo = NULL;
 476    }
 477
 478    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 479       drm_intel_bo_unreference(brw->vb.inputs[i].bo);
 480       brw->vb.inputs[i].bo = NULL;
 481    }
 482
 483    drm_intel_bo_unreference(brw->ib.bo);
 484    brw->ib.bo = NULL;
 485 }