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_attrib.h"
  39 #include "brw_context.h"
  40 #include "brw_aub.h"
  41 #include "brw_state.h"
  42 #include "brw_fallback.h"
  43
  44 #include "intel_ioctl.h"
  45 #include "intel_batchbuffer.h"
  46 #include "intel_buffer_objects.h"
  47
  48
  49
  50
  51
  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
 127
 128 static void brw_emit_prim( struct brw_context *brw,
 129                            const struct brw_draw_prim *prim )
 130
 131 {
 132    struct brw_3d_primitive prim_packet;
 133
 134    if (INTEL_DEBUG & DEBUG_PRIMS)
 135       _mesa_printf("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 = brw_set_prim(brw, prim->mode);
 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    prim_packet.instance_count = 1;
 147    prim_packet.start_instance_location = 0;
 148    prim_packet.base_vert_location = 0;
 149
 150    if (prim_packet.verts_per_instance) {
 151       intel_batchbuffer_data( brw->intel.batch, &prim_packet, sizeof(prim_packet),
 152                               INTEL_BATCH_CLIPRECTS);
 153    }
 154 }
 155
 156
 157
 158 static void update_current_size( struct gl_client_array *array)
 159 {
 160    const GLfloat *ptr = (const GLfloat *)array->Ptr;
 161
 162    assert(array->StrideB == 0);
 163    assert(array->Type == GL_FLOAT);
 164
 165    if (ptr[3] != 1.0)
 166       array->Size = 4;
 167    else if (ptr[2] != 0.0)
 168       array->Size = 3;
 169    else if (ptr[1] != 0.0)
 170       array->Size = 2;
 171    else
 172       array->Size = 1;
 173 }
 174
 175
 176
 177 /* Fill in any gaps in passed arrays with pointers to current
 178  * attributes:
 179  */
 180 static void brw_merge_inputs( struct brw_context *brw,
 181                        const struct gl_client_array *arrays[])
 182 {
 183    struct gl_client_array *current_values = brw->vb.current_values;
 184    struct brw_vertex_element *inputs = brw->vb.inputs;
 185    struct brw_vertex_info old = brw->vb.info;
 186    GLuint i;
 187
 188    memset(inputs, 0, sizeof(*inputs));
 189    memset(&brw->vb.info, 0, sizeof(brw->vb.info));
 190
 191    for (i = 0; i < BRW_ATTRIB_MAX; i++) {
 192       if (arrays[i] && arrays[i]->Enabled)
 193       {
 194          brw->vb.inputs[i].glarray = arrays[i];
 195          brw->vb.info.varying[i/32] |= 1 << (i%32);
 196       }
 197       else
 198       {
 199          brw->vb.inputs[i].glarray = &current_values[i];
 200
 201          /* XXX: This will change on the Mesa trunk as Brian has moved
 202           * edgeflag, index into this range:
 203           */
 204          if (i < VERT_ATTRIB_GENERIC0)
 205             update_current_size(&current_values[i]);
 206       }
 207
 208       brw->vb.info.sizes[i/16] |= (inputs[i].glarray->Size - 1) << ((i%16) * 2);
 209    }
 210
 211    /* Raise statechanges if input sizes and varying have changed:
 212     */
 213    if (memcmp(brw->vb.info.sizes, old.sizes, sizeof(old.sizes)) != 0)
 214       brw->state.dirty.brw |= BRW_NEW_INPUT_DIMENSIONS;
 215
 216    if (memcmp(brw->vb.info.varying, old.varying, sizeof(old.varying)) != 0)
 217       brw->state.dirty.brw |= BRW_NEW_INPUT_VARYING;
 218 }
 219
 220 static GLboolean check_fallbacks( struct brw_context *brw,
 221                                   const struct brw_draw_prim *prim,
 222                                   GLuint nr_prims )
 223 {
 224    GLuint i;
 225
 226    if (!brw->intel.strict_conformance)
 227       return GL_FALSE;
 228
 229    if (brw->attribs.Polygon->SmoothFlag) {
 230       for (i = 0; i < nr_prims; i++)
 231          if (reduced_prim[prim[i].mode] == GL_TRIANGLES)
 232             return GL_TRUE;
 233    }
 234
 235    /* BRW hardware will do AA lines, but they are non-conformant it
 236     * seems.  TBD whether we keep this fallback:
 237     */
 238    if (brw->attribs.Line->SmoothFlag) {
 239       for (i = 0; i < nr_prims; i++)
 240          if (reduced_prim[prim[i].mode] == GL_LINES)
 241             return GL_TRUE;
 242    }
 243
 244    /* Stipple -- these fallbacks could be resolved with a little
 245     * bit of work?
 246     */
 247    if (brw->attribs.Line->StippleFlag) {
 248       for (i = 0; i < nr_prims; i++) {
 249          /* GS doesn't get enough information to know when to reset
 250           * the stipple counter?!?
 251           */
 252          if (prim[i].mode == GL_LINE_LOOP)
 253             return GL_TRUE;
 254
 255          if (prim[i].mode == GL_POLYGON &&
 256              (brw->attribs.Polygon->FrontMode == GL_LINE ||
 257               brw->attribs.Polygon->BackMode == GL_LINE))
 258             return GL_TRUE;
 259       }
 260    }
 261
 262
 263    if (brw->attribs.Point->SmoothFlag) {
 264       for (i = 0; i < nr_prims; i++)
 265          if (prim[i].mode == GL_POINTS)
 266             return GL_TRUE;
 267    }
 268
 269    return GL_FALSE;
 270 }
 271
 272
 273 GLboolean brw_draw_prims( GLcontext *ctx,
 274                           const struct gl_client_array *arrays[],
 275                           const struct brw_draw_prim *prim,
 276                           GLuint nr_prims,
 277                           const struct brw_draw_index_buffer *ib,
 278                           GLuint min_index,
 279                           GLuint max_index,
 280                           GLuint flags )
 281 {
 282    struct intel_context *intel = intel_context(ctx);
 283    struct brw_context *brw = brw_context(ctx);
 284    GLboolean retval = GL_FALSE;
 285    GLuint i;
 286
 287    if (ctx->NewState)
 288       _mesa_update_state( ctx );
 289
 290    /* Bind all inputs, derive varying and size information:
 291     */
 292    brw_merge_inputs( brw, arrays );
 293
 294    /* Have to validate state quite late.  Will rebuild tnl_program,
 295     * which depends on varying information.
 296     *
 297     * Note this is where brw->vs->prog_data.inputs_read is calculated,
 298     * so can't access it earlier.
 299     */
 300
 301    LOCK_HARDWARE(intel);
 302    {
 303       assert(intel->locked);
 304
 305       /* Set the first primitive early, ahead of validate_state:
 306        */
 307       brw_set_prim(brw, prim[0].mode);
 308
 309       /* XXX:  Need to separate validate and upload of state.
 310        */
 311       brw_validate_state( brw );
 312
 313       /* Various fallback checks:
 314        */
 315       if (brw->intel.Fallback)
 316          goto out;
 317
 318       if (check_fallbacks( brw, prim, nr_prims ))
 319          goto out;
 320
 321       /* Upload index, vertex data:
 322        */
 323       if (ib)
 324          brw_upload_indices( brw, ib );
 325
 326       if (!brw_upload_vertices( brw, min_index, max_index)) {
 327          goto out;
 328       }
 329
 330       /* Emit prims to batchbuffer:
 331        */
 332       for (i = 0; i < nr_prims; i++) {
 333          brw_emit_prim(brw, &prim[i]);
 334       }
 335
 336       retval = GL_TRUE;
 337    }
 338
 339  out:
 340
 341    /* Currently have to do this to synchronize with the map/unmap of
 342     * the vertex buffer in brw_exec_api.c.  Not sure if there is any
 343     * way around this, as not every flush is due to a buffer filling
 344     * up.
 345     */
 346    intel_batchbuffer_flush( brw->intel.batch );
 347
 348    if (intel->thrashing) {
 349       bmSetFence(intel);
 350    }
 351
 352    /* Free any old data so it doesn't clog up texture memory - we
 353     * won't be referencing it again.
 354     */
 355    while (brw->vb.upload.wrap != brw->vb.upload.buf) {
 356       ctx->Driver.BufferData(ctx,
 357                              GL_ARRAY_BUFFER_ARB,
 358                              BRW_UPLOAD_INIT_SIZE,
 359                              NULL,
 360                              GL_DYNAMIC_DRAW_ARB,
 361                              brw->vb.upload.vbo[brw->vb.upload.wrap]);
 362       brw->vb.upload.wrap++;
 363       brw->vb.upload.wrap %= BRW_NR_UPLOAD_BUFS;
 364    }
 365
 366    UNLOCK_HARDWARE(intel);
 367    return retval;
 368 }
 369
 370 static void brw_invalidate_vbo_cb( struct intel_context *intel, void *ptr )
 371 {
 372    /* nothing to do, we don't rely on the contents being preserved */
 373 }
 374
 375
 376 void brw_draw_init( struct brw_context *brw )
 377 {
 378    GLcontext *ctx = &brw->intel.ctx;
 379    GLuint i;
 380
 381    brw->vb.upload.size = BRW_UPLOAD_INIT_SIZE;
 382
 383    for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++) {
 384       brw->vb.upload.vbo[i] = ctx->Driver.NewBufferObject(ctx, 1, GL_ARRAY_BUFFER_ARB);
 385
 386       /* XXX: Set these to no-backing-store
 387        */
 388       bmBufferSetInvalidateCB(&brw->intel,
 389                               intel_bufferobj_buffer(intel_buffer_object(brw->vb.upload.vbo[i])),
 390                               brw_invalidate_vbo_cb,
 391                               &brw->intel,
 392                               GL_TRUE);
 393
 394    }
 395
 396    ctx->Driver.BufferData( ctx,
 397                            GL_ARRAY_BUFFER_ARB,
 398                            BRW_UPLOAD_INIT_SIZE,
 399                            NULL,
 400                            GL_DYNAMIC_DRAW_ARB,
 401                            brw->vb.upload.vbo[0] );
 402
 403
 404    brw_init_current_values(ctx, brw->vb.current_values);
 405 }
 406
 407 void brw_draw_destroy( struct brw_context *brw )
 408 {
 409    GLcontext *ctx = &brw->intel.ctx;
 410    GLuint i;
 411
 412    for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++)
 413       ctx->Driver.DeleteBuffer(ctx, brw->vb.upload.vbo[i]);
 414 }