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_ioctl.h"
  43 #include "intel_batchbuffer.h"
  44 #include "intel_buffer_objects.h"
  45
  46 #include "tnl/tnl.h"
  47 #include "vbo/vbo_context.h"
  48 #include "swrast/swrast.h"
  49 #include "swrast_setup/swrast_setup.h"
  50
  51 #define FILE_DEBUG_FLAG DEBUG_BATCH
  52
  53 static GLuint hw_prim[GL_POLYGON+1] = {
  54    _3DPRIM_POINTLIST,
  55    _3DPRIM_LINELIST,
  56    _3DPRIM_LINELOOP,
  57    _3DPRIM_LINESTRIP,
  58    _3DPRIM_TRILIST,
  59    _3DPRIM_TRISTRIP,
  60    _3DPRIM_TRIFAN,
  61    _3DPRIM_QUADLIST,
  62    _3DPRIM_QUADSTRIP,
  63    _3DPRIM_POLYGON
  64 };
  65
  66
  67 static const GLenum reduced_prim[GL_POLYGON+1] = {
  68    GL_POINTS,
  69    GL_LINES,
  70    GL_LINES,
  71    GL_LINES,
  72    GL_TRIANGLES,
  73    GL_TRIANGLES,
  74    GL_TRIANGLES,
  75    GL_TRIANGLES,
  76    GL_TRIANGLES,
  77    GL_TRIANGLES
  78 };
  79
  80
  81 /* When the primitive changes, set a state bit and re-validate.  Not
  82  * the nicest and would rather deal with this by having all the
  83  * programs be immune to the active primitive (ie. cope with all
  84  * possibilities).  That may not be realistic however.
  85  */
  86 static GLuint brw_set_prim(struct brw_context *brw, GLenum prim)
  87 {
  88    if (INTEL_DEBUG & DEBUG_PRIMS)
  89       _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim));
  90
  91    /* Slight optimization to avoid the GS program when not needed:
  92     */
  93    if (prim == GL_QUAD_STRIP &&
  94        brw->attribs.Light->ShadeModel != GL_FLAT &&
  95        brw->attribs.Polygon->FrontMode == GL_FILL &&
  96        brw->attribs.Polygon->BackMode == GL_FILL)
  97       prim = GL_TRIANGLE_STRIP;
  98
  99    if (prim != brw->primitive) {
 100       brw->primitive = prim;
 101       brw->state.dirty.brw |= BRW_NEW_PRIMITIVE;
 102
 103       if (reduced_prim[prim] != brw->intel.reduced_primitive) {
 104          brw->intel.reduced_primitive = reduced_prim[prim];
 105          brw->state.dirty.brw |= BRW_NEW_REDUCED_PRIMITIVE;
 106       }
 107
 108       brw_validate_state(brw);
 109    }
 110
 111    return hw_prim[prim];
 112 }
 113
 114
 115 static GLuint trim(GLenum prim, GLuint length)
 116 {
 117    if (prim == GL_QUAD_STRIP)
 118       return length > 3 ? (length - length % 2) : 0;
 119    else if (prim == GL_QUADS)
 120       return length - length % 4;
 121    else
 122       return length;
 123 }
 124
 125
 126 static void brw_emit_prim( struct brw_context *brw,
 127                            const struct _mesa_prim *prim )
 128
 129 {
 130    struct brw_3d_primitive prim_packet;
 131
 132    if (INTEL_DEBUG & DEBUG_PRIMS)
 133       _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim->mode),
 134                    prim->start, prim->count);
 135
 136    prim_packet.header.opcode = CMD_3D_PRIM;
 137    prim_packet.header.length = sizeof(prim_packet)/4 - 2;
 138    prim_packet.header.pad = 0;
 139    prim_packet.header.topology = brw_set_prim(brw, prim->mode);
 140    prim_packet.header.indexed = prim->indexed;
 141
 142    prim_packet.verts_per_instance = trim(prim->mode, prim->count);
 143    prim_packet.start_vert_location = prim->start;
 144    prim_packet.instance_count = 1;
 145    prim_packet.start_instance_location = 0;
 146    prim_packet.base_vert_location = 0;
 147
 148    /* Can't wrap here, since we rely on the validated state. */
 149    brw->no_batch_wrap = GL_TRUE;
 150    if (prim_packet.verts_per_instance) {
 151       intel_batchbuffer_data( brw->intel.batch, &prim_packet,
 152                               sizeof(prim_packet), LOOP_CLIPRECTS);
 153    }
 154    brw->no_batch_wrap = GL_FALSE;
 155 }
 156
 157 static void brw_merge_inputs( struct brw_context *brw,
 158                        const struct gl_client_array *arrays[])
 159 {
 160    struct brw_vertex_element *inputs = brw->vb.inputs;
 161    struct brw_vertex_info old = brw->vb.info;
 162    GLuint i;
 163
 164    memset(inputs, 0, sizeof(*inputs));
 165    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 166
 167    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 168       brw->vb.inputs[i].glarray = arrays[i];
 169
 170       /* XXX: metaops passes null arrays */
 171       if (arrays[i]) {
 172          if (arrays[i]->StrideB != 0)
 173             brw->vb.info.varying |= 1 << i;
 174
 175          brw->vb.info.sizes[i/16] |= (inputs[i].glarray->Size - 1) << ((i%16) * 2);
 176       }
 177    }
 178
 179    /* Raise statechanges if input sizes and varying have changed:
 180     */
 181    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 182       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 183
 184    if (brw->vb.info.varying != old.varying)
 185       brw->state.dirty.brw |= BRW_NEW_INPUT_VARYING;
 186 }
 187
 188 /* XXX: could split the primitive list to fallback only on the
 189  * non-conformant primitives.
 190  */
 191 static GLboolean check_fallbacks( struct brw_context *brw,
 192                                   const struct _mesa_prim *prim,
 193                                   GLuint nr_prims )
 194 {
 195    GLuint i;
 196
 197    if (!brw->intel.strict_conformance)
 198       return GL_FALSE;
 199
 200    if (brw->attribs.Polygon->SmoothFlag) {
 201       for (i = 0; i < nr_prims; i++)
 202          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 203             return GL_TRUE;
 204    }
 205
 206    /* BRW hardware will do AA lines, but they are non-conformant it
 207     * seems.  TBD whether we keep this fallback:
 208     */
 209    if (brw->attribs.Line->SmoothFlag) {
 210       for (i = 0; i < nr_prims; i++)
 211          if (reduced_prim[prim[i].mode] == GL_LINES)
 212             return GL_TRUE;
 213    }
 214
 215    /* Stipple -- these fallbacks could be resolved with a little
 216     * bit of work?
 217     */
 218    if (brw->attribs.Line->StippleFlag) {
 219       for (i = 0; i < nr_prims; i++) {
 220          /* GS doesn't get enough information to know when to reset
 221           * the stipple counter?!?
 222           */
 223          if (prim[i].mode == GL_LINE_LOOP)
 224             return GL_TRUE;
 225
 226          if (prim[i].mode == GL_POLYGON &&
 227              (brw->attribs.Polygon->FrontMode == GL_LINE ||
 228               brw->attribs.Polygon->BackMode == GL_LINE))
 229             return GL_TRUE;
 230       }
 231    }
 232
 233
 234    if (brw->attribs.Point->SmoothFlag) {
 235       for (i = 0; i < nr_prims; i++)
 236          if (prim[i].mode == GL_POINTS)
 237             return GL_TRUE;
 238    }
 239
 240    return GL_FALSE;
 241 }
 242
 243 /* May fail if out of video memory for texture or vbo upload, or on
 244  * fallback conditions.
 245  */
 246 static GLboolean brw_try_draw_prims( GLcontext *ctx,
 247                                      const struct gl_client_array *arrays[],
 248                                      const struct _mesa_prim *prim,
 249                                      GLuint nr_prims,
 250                                      const struct _mesa_index_buffer *ib,
 251                                      GLuint min_index,
 252                                      GLuint max_index )
 253 {
 254    struct intel_context *intel = intel_context(ctx);
 255    struct brw_context *brw = brw_context(ctx);
 256    GLboolean retval = GL_FALSE;
 257    GLuint i;
 258
 259    if (ctx->NewState)
 260       _mesa_update_state( ctx );
 261
 262    brw_validate_textures( brw );
 263
 264    /* Bind all inputs, derive varying and size information:
 265     */
 266    brw_merge_inputs( brw, arrays );
 267
 268    brw->ib.ib = ib;
 269    brw->state.dirty.brw |= BRW_NEW_INDICES;
 270
 271    brw->vb.min_index = min_index;
 272    brw->vb.max_index = max_index;
 273    brw->state.dirty.brw |= BRW_NEW_VERTICES;
 274    /* Have to validate state quite late.  Will rebuild tnl_program,
 275     * which depends on varying information.
 276     *
 277     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 278     * so can't access it earlier.
 279     */
 280
 281    LOCK_HARDWARE(intel);
 282
 283    if (brw->intel.numClipRects == 0) {
 284       UNLOCK_HARDWARE(intel);
 285       return GL_TRUE;
 286    }
 287
 288    {
 289       /* Flush the batch if it's approaching full, so that we don't wrap while
 290        * we've got validated state that needs to be in the same batch as the
 291        * primitives.  This fraction is just a guess (minimal full state plus
 292        * a primitive is around 512 bytes), and would be better if we had
 293        * an upper bound of how much we might emit in a single
 294        * brw_try_draw_prims().
 295        */
 296       if (intel->batch->ptr - intel->batch->map > intel->batch->size * 3 / 4
 297         /* brw_emit_prim may change the cliprect_mode to LOOP_CLIPRECTS */
 298           || intel->batch->cliprect_mode != LOOP_CLIPRECTS)
 299               intel_batchbuffer_flush(intel->batch);
 300
 301       /* Set the first primitive early, ahead of validate_state:
 302        */
 303       brw_set_prim(brw, prim[0].mode);
 304
 305       /* XXX:  Need to separate validate and upload of state.
 306        */
 307       brw_validate_state( brw );
 308
 309       /* Various fallback checks:
 310        */
 311       if (brw->intel.Fallback)
 312          goto out;
 313
 314       if (check_fallbacks( brw, prim, nr_prims ))
 315          goto out;
 316
 317       for (i = 0; i < nr_prims; i++) {
 318          brw_emit_prim(brw, &prim[i]);
 319       }
 320
 321       retval = GL_TRUE;
 322    }
 323
 324  out:
 325    UNLOCK_HARDWARE(intel);
 326
 327    if (!retval)
 328       DBG("%s failed\n", __FUNCTION__);
 329
 330    return retval;
 331 }
 332
 333 static GLboolean brw_need_rebase( GLcontext *ctx,
 334                                   const struct gl_client_array *arrays[],
 335                                   const struct _mesa_index_buffer *ib,
 336                                   GLuint min_index )
 337 {
 338    if (min_index == 0)
 339       return GL_FALSE;
 340
 341    if (ib) {
 342       if (!vbo_all_varyings_in_vbos(arrays))
 343          return GL_TRUE;
 344       else
 345          return GL_FALSE;
 346    }
 347    else {
 348       /* Hmm.  This isn't quite what I wanted.  BRW can actually
 349        * handle the mixed case well enough that we shouldn't need to
 350        * rebase.  However, it's probably not very common, nor hugely
 351        * expensive to do it this way:
 352        */
 353       if (!vbo_all_varyings_in_vbos(arrays))
 354          return GL_TRUE;
 355       else
 356          return GL_FALSE;
 357    }
 358 }
 359
 360
 361 void brw_draw_prims( GLcontext *ctx,
 362                      const struct gl_client_array *arrays[],
 363                      const struct _mesa_prim *prim,
 364                      GLuint nr_prims,
 365                      const struct _mesa_index_buffer *ib,
 366                      GLuint min_index,
 367                      GLuint max_index )
 368 {
 369    GLboolean retval;
 370
 371    /* Decide if we want to rebase.  If so we end up recursing once
 372     * only into this function.
 373     */
 374    if (brw_need_rebase( ctx, arrays, ib, min_index )) {
 375       vbo_rebase_prims( ctx, arrays,
 376                         prim, nr_prims,
 377                         ib, min_index, max_index,
 378                         brw_draw_prims );
 379
 380       return;
 381    }
 382
 383
 384    /* Make a first attempt at drawing:
 385     */
 386    retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 387
 388    /* Otherwise, we really are out of memory.  Pass the drawing
 389     * command to the software tnl module and which will in turn call
 390     * swrast to do the drawing.
 391     */
 392    if (!retval) {
 393        _swsetup_Wakeup(ctx);
 394       _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index);
 395    }
 396 }
 397
 398 void brw_draw_init( struct brw_context *brw )
 399 {
 400    GLcontext *ctx = &brw->intel.ctx;
 401    struct vbo_context *vbo = vbo_context(ctx);
 402
 403    /* Register our drawing function:
 404     */
 405    vbo->draw_prims = brw_draw_prims;
 406 }
 407
 408 void brw_draw_destroy( struct brw_context *brw )
 409 {
 410    if (brw->vb.upload.bo != NULL) {
 411       dri_bo_unreference(brw->vb.upload.bo);
 412       brw->vb.upload.bo = NULL;
 413    }
 414 }