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/condrender.h"
  32 #include "main/samplerobj.h"
  33 #include "main/state.h"
  34 #include "main/enums.h"
  35 #include "tnl/tnl.h"
  36 #include "vbo/vbo_context.h"
  37 #include "swrast/swrast.h"
  38 #include "swrast_setup/swrast_setup.h"
  39
  40 #include "brw_draw.h"
  41 #include "brw_defines.h"
  42 #include "brw_context.h"
  43 #include "brw_state.h"
  44
  45 #include "intel_batchbuffer.h"
  46
  47 #define FILE_DEBUG_FLAG DEBUG_PRIMS
  48
  49 static GLuint prim_to_hw_prim[GL_POLYGON+1] = {
  50    _3DPRIM_POINTLIST,
  51    _3DPRIM_LINELIST,
  52    _3DPRIM_LINELOOP,
  53    _3DPRIM_LINESTRIP,
  54    _3DPRIM_TRILIST,
  55    _3DPRIM_TRISTRIP,
  56    _3DPRIM_TRIFAN,
  57    _3DPRIM_QUADLIST,
  58    _3DPRIM_QUADSTRIP,
  59    _3DPRIM_POLYGON
  60 };
  61
  62
  63 static const GLenum reduced_prim[GL_POLYGON+1] = {
  64    GL_POINTS,
  65    GL_LINES,
  66    GL_LINES,
  67    GL_LINES,
  68    GL_TRIANGLES,
  69    GL_TRIANGLES,
  70    GL_TRIANGLES,
  71    GL_TRIANGLES,
  72    GL_TRIANGLES,
  73    GL_TRIANGLES
  74 };
  75
  76
  77 /* When the primitive changes, set a state bit and re-validate.  Not
  78  * the nicest and would rather deal with this by having all the
  79  * programs be immune to the active primitive (ie. cope with all
  80  * possibilities).  That may not be realistic however.
  81  */
  82 static GLuint brw_set_prim(struct brw_context *brw,
  83                            const struct _mesa_prim *prim)
  84 {
  85    struct gl_context *ctx = &brw->intel.ctx;
  86    GLenum mode = prim->mode;
  87
  88    DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim->mode));
  89
  90    /* Slight optimization to avoid the GS program when not needed:
  91     */
  92    if (mode == GL_QUAD_STRIP &&
  93        ctx->Light.ShadeModel != GL_FLAT &&
  94        ctx->Polygon.FrontMode == GL_FILL &&
  95        ctx->Polygon.BackMode == GL_FILL)
  96       mode = GL_TRIANGLE_STRIP;
  97
  98    if (prim->mode == GL_QUADS && prim->count == 4 &&
  99        ctx->Light.ShadeModel != GL_FLAT &&
 100        ctx->Polygon.FrontMode == GL_FILL &&
 101        ctx->Polygon.BackMode == GL_FILL) {
 102       mode = GL_TRIANGLE_FAN;
 103    }
 104
 105    if (mode != brw->primitive) {
 106       brw->primitive = mode;
 107       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 108
 109       if (reduced_prim[mode] != brw->intel.reduced_primitive) {
 110          brw->intel.reduced_primitive = reduced_prim[mode];
 111          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 112       }
 113    }
 114
 115    return prim_to_hw_prim[mode];
 116 }
 117
 118
 119 static GLuint trim(GLenum prim, GLuint length)
 120 {
 121    if (prim == GL_QUAD_STRIP)
 122       return length > 3 ? (length - length % 2) : 0;
 123    else if (prim == GL_QUADS)
 124       return length - length % 4;
 125    else
 126       return length;
 127 }
 128
 129
 130 static void brw_emit_prim(struct brw_context *brw,
 131                           const struct _mesa_prim *prim,
 132                           uint32_t hw_prim)
 133 {
 134    struct intel_context *intel = &brw->intel;
 135    int verts_per_instance;
 136    int vertex_access_type;
 137    int start_vertex_location;
 138    int base_vertex_location;
 139
 140    DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 141        prim->start, prim->count);
 142
 143    start_vertex_location = prim->start;
 144    base_vertex_location = prim->basevertex;
 145    if (prim->indexed) {
 146       vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
 147       start_vertex_location += brw->ib.start_vertex_offset;
 148       base_vertex_location += brw->vb.start_vertex_bias;
 149    } else {
 150       vertex_access_type = GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
 151       start_vertex_location += brw->vb.start_vertex_bias;
 152    }
 153
 154    verts_per_instance = trim(prim->mode, prim->count);
 155
 156    /* If nothing to emit, just return. */
 157    if (verts_per_instance == 0)
 158       return;
 159
 160    /* If we're set to always flush, do it before and after the primitive emit.
 161     * We want to catch both missed flushes that hurt instruction/state cache
 162     * and missed flushes of the render cache as it heads to other parts of
 163     * the besides the draw code.
 164     */
 165    if (intel->always_flush_cache) {
 166       intel_batchbuffer_emit_mi_flush(intel);
 167    }
 168
 169    BEGIN_BATCH(6);
 170    OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
 171              hw_prim << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
 172              vertex_access_type);
 173    OUT_BATCH(verts_per_instance);
 174    OUT_BATCH(start_vertex_location);
 175    OUT_BATCH(1); // instance count
 176    OUT_BATCH(0); // start instance location
 177    OUT_BATCH(base_vertex_location);
 178    ADVANCE_BATCH();
 179
 180    if (intel->always_flush_cache) {
 181       intel_batchbuffer_emit_mi_flush(intel);
 182    }
 183 }
 184
 185 static void brw_merge_inputs( struct brw_context *brw,
 186                        const struct gl_client_array *arrays[])
 187 {
 188    struct brw_vertex_info old = brw->vb.info;
 189    GLuint i;
 190
 191    for (i = 0; i < brw->vb.nr_buffers; i++) {
 192       drm_intel_bo_unreference(brw->vb.buffers[i].bo);
 193       brw->vb.buffers[i].bo = NULL;
 194    }
 195    brw->vb.nr_buffers = 0;
 196
 197    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 198
 199    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 200       brw->vb.inputs[i].buffer = -1;
 201       brw->vb.inputs[i].glarray = arrays[i];
 202       brw->vb.inputs[i].attrib = (gl_vert_attrib) i;
 203
 204       if (arrays[i]->StrideB != 0)
 205          brw->vb.info.sizes[i/16] |= (brw->vb.inputs[i].glarray->Size - 1) <<
 206             ((i%16) * 2);
 207    }
 208
 209    /* Raise statechanges if input sizes have changed. */
 210    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 211       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 212 }
 213
 214 /* XXX: could split the primitive list to fallback only on the
 215  * non-conformant primitives.
 216  */
 217 static GLboolean check_fallbacks( struct brw_context *brw,
 218                                   const struct _mesa_prim *prim,
 219                                   GLuint nr_prims )
 220 {
 221    struct gl_context *ctx = &brw->intel.ctx;
 222    GLuint i;
 223
 224    /* If we don't require strict OpenGL conformance, never
 225     * use fallbacks.  If we're forcing fallbacks, always
 226     * use fallfacks.
 227     */
 228    if (brw->intel.conformance_mode == 0)
 229       return GL_FALSE;
 230
 231    if (brw->intel.conformance_mode == 2)
 232       return GL_TRUE;
 233
 234    if (ctx->Polygon.SmoothFlag) {
 235       for (i = 0; i < nr_prims; i++)
 236          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 237             return GL_TRUE;
 238    }
 239
 240    /* BRW hardware will do AA lines, but they are non-conformant it
 241     * seems.  TBD whether we keep this fallback:
 242     */
 243    if (ctx->Line.SmoothFlag) {
 244       for (i = 0; i < nr_prims; i++)
 245          if (reduced_prim[prim[i].mode] == GL_LINES)
 246             return GL_TRUE;
 247    }
 248
 249    /* Stipple -- these fallbacks could be resolved with a little
 250     * bit of work?
 251     */
 252    if (ctx->Line.StippleFlag) {
 253       for (i = 0; i < nr_prims; i++) {
 254          /* GS doesn't get enough information to know when to reset
 255           * the stipple counter?!?
 256           */
 257          if (prim[i].mode == GL_LINE_LOOP || prim[i].mode == GL_LINE_STRIP)
 258             return GL_TRUE;
 259
 260          if (prim[i].mode == GL_POLYGON &&
 261              (ctx->Polygon.FrontMode == GL_LINE ||
 262               ctx->Polygon.BackMode == GL_LINE))
 263             return GL_TRUE;
 264       }
 265    }
 266
 267    if (ctx->Point.SmoothFlag) {
 268       for (i = 0; i < nr_prims; i++)
 269          if (prim[i].mode == GL_POINTS)
 270             return GL_TRUE;
 271    }
 272
 273    /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
 274     * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
 275     * as GL_CLAMP_TO_EDGE instead.  If we're using GL_CLAMP, and
 276     * we want strict conformance, force the fallback.
 277     * Right now, we only do this for 2D textures.
 278     */
 279    {
 280       int u;
 281       for (u = 0; u < ctx->Const.MaxTextureCoordUnits; u++) {
 282          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[u];
 283
 284          if (texUnit->Enabled) {
 285             struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, u);
 286
 287             if (texUnit->Enabled & TEXTURE_1D_BIT) {
 288                if (sampler->WrapS == GL_CLAMP) {
 289                    return GL_TRUE;
 290                }
 291             }
 292             if (texUnit->Enabled & TEXTURE_2D_BIT) {
 293                if (sampler->WrapS == GL_CLAMP ||
 294                    sampler->WrapT == GL_CLAMP) {
 295                    return GL_TRUE;
 296                }
 297             }
 298             if (texUnit->Enabled & TEXTURE_3D_BIT) {
 299                if (sampler->WrapS == GL_CLAMP ||
 300                    sampler->WrapT == GL_CLAMP ||
 301                    sampler->WrapR == GL_CLAMP) {
 302                    return GL_TRUE;
 303                }
 304             }
 305          }
 306       }
 307    }
 308
 309    /* Nothing stopping us from the fast path now */
 310    return GL_FALSE;
 311 }
 312
 313 /* May fail if out of video memory for texture or vbo upload, or on
 314  * fallback conditions.
 315  */
 316 static GLboolean brw_try_draw_prims( struct gl_context *ctx,
 317                                      const struct gl_client_array *arrays[],
 318                                      const struct _mesa_prim *prim,
 319                                      GLuint nr_prims,
 320                                      const struct _mesa_index_buffer *ib,
 321                                      GLuint min_index,
 322                                      GLuint max_index )
 323 {
 324    struct intel_context *intel = intel_context(ctx);
 325    struct brw_context *brw = brw_context(ctx);
 326    GLboolean retval = GL_FALSE;
 327    GLboolean warn = GL_FALSE;
 328    GLuint i;
 329
 330    if (ctx->NewState)
 331       _mesa_update_state( ctx );
 332
 333    /* We have to validate the textures *before* checking for fallbacks;
 334     * otherwise, the software fallback won't be able to rely on the
 335     * texture state, the firstLevel and lastLevel fields won't be
 336     * set in the intel texture object (they'll both be 0), and the
 337     * software fallback will segfault if it attempts to access any
 338     * texture level other than level 0.
 339     */
 340    brw_validate_textures( brw );
 341
 342    if (check_fallbacks(brw, prim, nr_prims))
 343       return GL_FALSE;
 344
 345    /* Bind all inputs, derive varying and size information:
 346     */
 347    brw_merge_inputs( brw, arrays );
 348
 349    brw->ib.ib = ib;
 350    brw->state.dirty.brw |= BRW_NEW_INDICES;
 351
 352    brw->vb.min_index = min_index;
 353    brw->vb.max_index = max_index;
 354    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 355
 356    /* Have to validate state quite late.  Will rebuild tnl_program,
 357     * which depends on varying information.
 358     *
 359     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 360     * so can't access it earlier.
 361     */
 362
 363    intel_prepare_render(intel);
 364
 365    for (i = 0; i < nr_prims; i++) {
 366       uint32_t hw_prim;
 367
 368       /* Flush the batch if it's approaching full, so that we don't wrap while
 369        * we've got validated state that needs to be in the same batch as the
 370        * primitives.  This fraction is just a guess (minimal full state plus
 371        * a primitive is around 512 bytes), and would be better if we had
 372        * an upper bound of how much we might emit in a single
 373        * brw_try_draw_prims().
 374        */
 375       intel_batchbuffer_require_space(intel, 1024, false);
 376
 377       hw_prim = brw_set_prim(brw, &prim[i]);
 378       if (brw->state.dirty.brw) {
 379          brw_validate_state(brw);
 380
 381          /* Various fallback checks:  */
 382          if (brw->intel.Fallback)
 383             goto out;
 384
 385          /* Check that we can fit our state in with our existing batchbuffer, or
 386           * flush otherwise.
 387           */
 388          if (dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 389                                              brw->state.validated_bo_count)) {
 390             static GLboolean warned;
 391             intel_batchbuffer_flush(intel);
 392
 393             /* Validate the state after we flushed the batch (which would have
 394              * changed the set of dirty state).  If we still fail to
 395              * check_aperture, warn of what's happening, but attempt to continue
 396              * on since it may succeed anyway, and the user would probably rather
 397              * see a failure and a warning than a fallback.
 398              */
 399             brw_validate_state(brw);
 400             if (!warned &&
 401                 dri_bufmgr_check_aperture_space(brw->state.validated_bos,
 402                                                 brw->state.validated_bo_count)) {
 403                warn = GL_TRUE;
 404                warned = GL_TRUE;
 405             }
 406          }
 407
 408          intel->no_batch_wrap = GL_TRUE;
 409          brw_upload_state(brw);
 410       }
 411
 412       brw_emit_prim(brw, &prim[i], hw_prim);
 413
 414       intel->no_batch_wrap = GL_FALSE;
 415
 416       retval = GL_TRUE;
 417    }
 418
 419    if (intel->always_flush_batch)
 420       intel_batchbuffer_flush(intel);
 421  out:
 422
 423    brw_state_cache_check_size(brw);
 424
 425    if (warn)
 426       fprintf(stderr, "i965: Single primitive emit potentially exceeded "
 427               "available aperture space\n");
 428
 429    if (!retval)
 430       DBG("%s failed\n", __FUNCTION__);
 431
 432    return retval;
 433 }
 434
 435 void brw_draw_prims( struct gl_context *ctx,
 436                      const struct gl_client_array *arrays[],
 437                      const struct _mesa_prim *prim,
 438                      GLuint nr_prims,
 439                      const struct _mesa_index_buffer *ib,
 440                      GLboolean index_bounds_valid,
 441                      GLuint min_index,
 442                      GLuint max_index )
 443 {
 444    GLboolean retval;
 445
 446    if (!_mesa_check_conditional_render(ctx))
 447       return;
 448
 449    if (!vbo_all_varyings_in_vbos(arrays)) {
 450       if (!index_bounds_valid)
 451          vbo_get_minmax_index(ctx, prim, ib, &min_index, &max_index);
 452
 453       /* Decide if we want to rebase.  If so we end up recursing once
 454        * only into this function.
 455        */
 456       if (min_index != 0 && !vbo_any_varyings_in_vbos(arrays)) {
 457          vbo_rebase_prims(ctx, arrays,
 458                           prim, nr_prims,
 459                           ib, min_index, max_index,
 460                           brw_draw_prims );
 461          return;
 462       }
 463    }
 464
 465    /* Make a first attempt at drawing:
 466     */
 467    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 468
 469    /* Otherwise, we really are out of memory.  Pass the drawing
 470     * command to the software tnl module and which will in turn call
 471     * swrast to do the drawing.
 472     */
 473    if (!retval) {
 474        _swsetup_Wakeup(ctx);
 475       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 476    }
 477
 478 }
 479
 480 void brw_draw_init( struct brw_context *brw )
 481 {
 482    struct gl_context *ctx = &brw->intel.ctx;
 483    struct vbo_context *vbo = vbo_context(ctx);
 484    int i;
 485
 486    /* Register our drawing function:
 487     */
 488    vbo->draw_prims = brw_draw_prims;
 489
 490    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 491       brw->vb.inputs[i].buffer = -1;
 492    brw->vb.nr_buffers = 0;
 493    brw->vb.nr_enabled = 0;
 494 }
 495
 496 void brw_draw_destroy( struct brw_context *brw )
 497 {
 498    int i;
 499
 500    for (i = 0; i < brw->vb.nr_buffers; i++) {
 501       drm_intel_bo_unreference(brw->vb.buffers[i].bo);
 502       brw->vb.buffers[i].bo = NULL;
 503    }
 504    brw->vb.nr_buffers = 0;
 505
 506    for (i = 0; i < brw->vb.nr_enabled; i++) {
 507       brw->vb.enabled[i]->buffer = -1;
 508    }
 509    brw->vb.nr_enabled = 0;
 510
 511    drm_intel_bo_unreference(brw->ib.bo);
 512    brw->ib.bo = NULL;
 513 }