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 "glheader.h"
  31 #include "context.h"
  32 #include "state.h"
  33 #include "api_validate.h"
  34 #include "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 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    if (INTEL_DEBUG & DEBUG_PRIMS)
  88       _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
  89
  90    /* Slight optimization to avoid the GS program when not needed:
  91     */
  92    if (prim == GL_QUAD_STRIP &&
  93        brw->attribs.Light->ShadeModel != GL_FLAT &&
  94        brw->attribs.Polygon->FrontMode == GL_FILL &&
  95        brw->attribs.Polygon->BackMode == GL_FILL)
  96       prim = GL_TRIANGLE_STRIP;
  97
  98    if (prim != brw->primitive) {
  99       brw->primitive = prim;
 100       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 101
 102       if (reduced_prim[prim] != brw->intel.reduced_primitive) {
 103          brw->intel.reduced_primitive = reduced_prim[prim];
 104          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 105       }
 106
 107       brw_validate_state(brw);
 108    }
 109
 110    return 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
 128 {
 129    struct brw_3d_primitive prim_packet;
 130
 131    if (INTEL_DEBUG & DEBUG_PRIMS)
 132       _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 133                    prim->start, prim->count);
 134
 135    prim_packet.header.opcode = CMD_3D_PRIM;
 136    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 137    prim_packet.header.pad = 0;
 138    prim_packet.header.topology = brw_set_prim(brw, prim->mode);
 139    prim_packet.header.indexed = prim->indexed;
 140
 141    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 142    prim_packet.start_vert_location = prim->start;
 143    prim_packet.instance_count = 1;
 144    prim_packet.start_instance_location = 0;
 145    prim_packet.base_vert_location = 0;
 146
 147    /* Can't wrap here, since we rely on the validated state. */
 148    brw->no_batch_wrap = GL_TRUE;
 149    if (prim_packet.verts_per_instance) {
 150       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 151                               sizeof(prim_packet), LOOP_CLIPRECTS);
 152    }
 153    brw->no_batch_wrap = GL_FALSE;
 154 }
 155
 156 static void brw_merge_inputs( struct brw_context *brw,
 157                        const struct gl_client_array *arrays[])
 158 {
 159    struct brw_vertex_element *inputs = brw->vb.inputs;
 160    struct brw_vertex_info old = brw->vb.info;
 161    GLuint i;
 162
 163    memset(inputs, 0, sizeof(*inputs));
 164    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 165
 166    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 167       brw->vb.inputs[i].glarray = arrays[i];
 168
 169       /* XXX: metaops passes null arrays */
 170       if (arrays[i]) {
 171          if (arrays[i]->StrideB != 0)
 172             brw->vb.info.varying |= 1 << i;
 173
 174          brw->vb.info.sizes[i/16] |= (inputs[i].glarray->Size - 1) << ((i%16) * 2);
 175       }
 176    }
 177
 178    /* Raise statechanges if input sizes and varying have changed:
 179     */
 180    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 181       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 182
 183    if (brw->vb.info.varying != old.varying)
 184       brw->state.dirty.brw |= BRW_NEW_INPUT_VARYING;
 185 }
 186
 187 /* XXX: could split the primitive list to fallback only on the
 188  * non-conformant primitives.
 189  */
 190 static GLboolean check_fallbacks( struct brw_context *brw,
 191                                   const struct _mesa_prim *prim,
 192                                   GLuint nr_prims )
 193 {
 194    GLuint i;
 195
 196    if (!brw->intel.strict_conformance)
 197       return GL_FALSE;
 198
 199    if (brw->attribs.Polygon->SmoothFlag) {
 200       for (i = 0; i < nr_prims; i++)
 201          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 202             return GL_TRUE;
 203    }
 204
 205    /* BRW hardware will do AA lines, but they are non-conformant it
 206     * seems.  TBD whether we keep this fallback:
 207     */
 208    if (brw->attribs.Line->SmoothFlag) {
 209       for (i = 0; i < nr_prims; i++)
 210          if (reduced_prim[prim[i].mode] == GL_LINES)
 211             return GL_TRUE;
 212    }
 213
 214    /* Stipple -- these fallbacks could be resolved with a little
 215     * bit of work?
 216     */
 217    if (brw->attribs.Line->StippleFlag) {
 218       for (i = 0; i < nr_prims; i++) {
 219          /* GS doesn't get enough information to know when to reset
 220           * the stipple counter?!?
 221           */
 222          if (prim[i].mode == GL_LINE_LOOP)
 223             return GL_TRUE;
 224
 225          if (prim[i].mode == GL_POLYGON &&
 226              (brw->attribs.Polygon->FrontMode == GL_LINE ||
 227               brw->attribs.Polygon->BackMode == GL_LINE))
 228             return GL_TRUE;
 229       }
 230    }
 231
 232
 233    if (brw->attribs.Point->SmoothFlag) {
 234       for (i = 0; i < nr_prims; i++)
 235          if (prim[i].mode == GL_POINTS)
 236             return GL_TRUE;
 237    }
 238
 239    return GL_FALSE;
 240 }
 241
 242 /* May fail if out of video memory for texture or vbo upload, or on
 243  * fallback conditions.
 244  */
 245 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 246                                      const struct gl_client_array *arrays[],
 247                                      const struct _mesa_prim *prim,
 248                                      GLuint nr_prims,
 249                                      const struct _mesa_index_buffer *ib,
 250                                      GLuint min_index,
 251                                      GLuint max_index )
 252 {
 253    struct intel_context *intel = intel_context(ctx);
 254    struct brw_context *brw = brw_context(ctx);
 255    GLboolean retval = GL_FALSE;
 256    GLuint i;
 257
 258    if (ctx->NewState)
 259       _mesa_update_state( ctx );
 260
 261    brw_validate_textures( brw );
 262
 263    /* Bind all inputs, derive varying and size information:
 264     */
 265    brw_merge_inputs( brw, arrays );
 266
 267    brw->ib.ib = ib;
 268    brw->state.dirty.brw |= BRW_NEW_INDICES;
 269
 270    brw->vb.min_index = min_index;
 271    brw->vb.max_index = max_index;
 272    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 273    /* Have to validate state quite late.  Will rebuild tnl_program,
 274     * which depends on varying information.
 275     *
 276     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 277     * so can't access it earlier.
 278     */
 279
 280    LOCK_HARDWARE(intel);
 281
 282    if (brw->intel.numClipRects == 0) {
 283       UNLOCK_HARDWARE(intel);
 284       return GL_TRUE;
 285    }
 286
 287    {
 288       /* Flush the batch if it's approaching full, so that we don't wrap while
 289        * we've got validated state that needs to be in the same batch as the
 290        * primitives.  This fraction is just a guess (minimal full state plus
 291        * a primitive is around 512 bytes), and would be better if we had
 292        * an upper bound of how much we might emit in a single
 293        * brw_try_draw_prims().
 294        */
 295       if (intel->batch->ptr - intel->batch->map > intel->batch->size * 3 / 4
 296         /* brw_emit_prim may change the cliprect_mode to LOOP_CLIPRECTS */
 297           || intel->batch->cliprect_mode != LOOP_CLIPRECTS)
 298               intel_batchbuffer_flush(intel->batch);
 299
 300       /* Set the first primitive early, ahead of validate_state:
 301        */
 302       brw_set_prim(brw, prim[0].mode);
 303
 304       /* XXX:  Need to separate validate and upload of state.
 305        */
 306       brw_validate_state( brw );
 307
 308       /* Various fallback checks:
 309        */
 310       if (brw->intel.Fallback)
 311          goto out;
 312
 313       if (check_fallbacks( brw, prim, nr_prims ))
 314          goto out;
 315
 316       for (i = 0; i < nr_prims; i++) {
 317          brw_emit_prim(brw, &prim[i]);
 318       }
 319
 320       retval = GL_TRUE;
 321    }
 322
 323  out:
 324    UNLOCK_HARDWARE(intel);
 325
 326    if (!retval)
 327       DBG("%s failed\n", __FUNCTION__);
 328
 329    return retval;
 330 }
 331
 332 static GLboolean brw_need_rebase( GLcontext *ctx,
 333                                   const struct gl_client_array *arrays[],
 334                                   const struct _mesa_index_buffer *ib,
 335                                   GLuint min_index )
 336 {
 337    if (min_index == 0)
 338       return GL_FALSE;
 339
 340    if (ib) {
 341       if (!vbo_all_varyings_in_vbos(arrays))
 342          return GL_TRUE;
 343       else
 344          return GL_FALSE;
 345    }
 346    else {
 347       /* Hmm.  This isn't quite what I wanted.  BRW can actually
 348        * handle the mixed case well enough that we shouldn't need to
 349        * rebase.  However, it's probably not very common, nor hugely
 350        * expensive to do it this way:
 351        */
 352       if (!vbo_all_varyings_in_vbos(arrays))
 353          return GL_TRUE;
 354       else
 355          return GL_FALSE;
 356    }
 357 }
 358
 359
 360 void brw_draw_prims( GLcontext *ctx,
 361                      const struct gl_client_array *arrays[],
 362                      const struct _mesa_prim *prim,
 363                      GLuint nr_prims,
 364                      const struct _mesa_index_buffer *ib,
 365                      GLuint min_index,
 366                      GLuint max_index )
 367 {
 368    GLboolean retval;
 369
 370    /* Decide if we want to rebase.  If so we end up recursing once
 371     * only into this function.
 372     */
 373    if (brw_need_rebase( ctx, arrays, ib, min_index )) {
 374       vbo_rebase_prims( ctx, arrays,
 375                         prim, nr_prims,
 376                         ib, min_index, max_index,
 377                         brw_draw_prims );
 378
 379       return;
 380    }
 381
 382
 383    /* Make a first attempt at drawing:
 384     */
 385    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 386
 387    /* Otherwise, we really are out of memory.  Pass the drawing
 388     * command to the software tnl module and which will in turn call
 389     * swrast to do the drawing.
 390     */
 391    if (!retval) {
 392        _swsetup_Wakeup(ctx);
 393       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 394    }
 395 }
 396
 397 void brw_draw_init( struct brw_context *brw )
 398 {
 399    GLcontext *ctx = &brw->intel.ctx;
 400    struct vbo_context *vbo = vbo_context(ctx);
 401
 402    /* Register our drawing function:
 403     */
 404    vbo->draw_prims = brw_draw_prims;
 405 }
 406
 407 void brw_draw_destroy( struct brw_context *brw )
 408 {
 409    if (brw->vb.upload.bo != NULL) {
 410       dri_bo_unreference(brw->vb.upload.bo);
 411       brw->vb.upload.bo = NULL;
 412    }
 413 }