src/gallium/drivers/i965/brw_draw_upload.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 "pipe/p_context.h"
  29 #include "util/u_inlines.h"
  30
  31 #include "util/u_upload_mgr.h"
  32 #include "util/u_math.h"
  33
  34 #include "brw_draw.h"
  35 #include "brw_defines.h"
  36 #include "brw_context.h"
  37 #include "brw_state.h"
  38 #include "brw_screen.h"
  39 #include "brw_batchbuffer.h"
  40 #include "brw_debug.h"
  41 #include "brw_resource.h"
  42
  43
  44
  45
  46 static unsigned get_index_type(int type)
  47 {
  48    switch (type) {
  49    case 1: return BRW_INDEX_BYTE;
  50    case 2: return BRW_INDEX_WORD;
  51    case 4: return BRW_INDEX_DWORD;
  52    default: assert(0); return 0;
  53    }
  54 }
  55
  56
  57 static int brw_prepare_vertices(struct brw_context *brw)
  58 {
  59    unsigned int min_index = brw->curr.min_index;
  60    unsigned int max_index = brw->curr.max_index;
  61    GLuint i;
  62    int ret;
  63
  64    if (BRW_DEBUG & DEBUG_VERTS)
  65       debug_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
  66
  67
  68    for (i = 0; i < brw->curr.num_vertex_buffers; i++) {
  69       struct pipe_vertex_buffer *vb = &brw->curr.vertex_buffer[i];
  70       struct brw_winsys_buffer *bo;
  71       struct pipe_resource *upload_buf = NULL;
  72       unsigned offset;
  73
  74       if (BRW_DEBUG & DEBUG_VERTS)
  75          debug_printf("%s vb[%d] user:%d offset:0x%x sz:0x%x stride:0x%x\n",
  76                       __FUNCTION__, i,
  77                       brw_buffer_is_user_buffer(vb->buffer),
  78                       vb->buffer_offset,
  79                       vb->buffer->width0,
  80                       vb->stride);
  81
  82       if (brw_buffer_is_user_buffer(vb->buffer)) {
  83
  84          /* XXX: simplify this.  Stop the state trackers from generating
  85           * zero-stride buffers & have them use additional constants (or
  86           * add support for >1 constant buffer) instead.
  87           */
  88          unsigned size = (vb->stride == 0 ?
  89                           vb->buffer->width0 - vb->buffer_offset :
  90                           MAX2(vb->buffer->width0 - vb->buffer_offset,
  91                                vb->stride * (max_index + 1 - min_index)));
  92          boolean flushed;
  93
  94          ret = u_upload_buffer( brw->vb.upload_vertex,
  95                                 0,
  96                                 vb->buffer_offset + min_index * vb->stride,
  97                                 size,
  98                                 vb->buffer,
  99                                 &offset,
 100                                 &upload_buf,
 101                                 &flushed );
 102          if (ret)
 103             return ret;
 104
 105          bo = brw_buffer(upload_buf)->bo;
 106
 107          assert(offset + size <= bo->size);
 108       }
 109       else
 110       {
 111          offset = vb->buffer_offset;
 112          bo = brw_buffer(vb->buffer)->bo;
 113       }
 114
 115       assert(offset < bo->size);
 116
 117       /* Set up post-upload info about this vertex buffer:
 118        */
 119       brw->vb.vb[i].offset = offset;
 120       brw->vb.vb[i].stride = vb->stride;
 121       brw->vb.vb[i].vertex_count = (vb->stride == 0 ?
 122                                     1 :
 123                                     (bo->size - offset) / vb->stride);
 124
 125       bo_reference( &brw->vb.vb[i].bo,  bo );
 126
 127       /* Don't need to retain this reference.  We have a reference on
 128        * the underlying winsys buffer:
 129        */
 130       pipe_resource_reference( &upload_buf, NULL );
 131    }
 132
 133    brw->vb.nr_vb = i;
 134    brw_prepare_query_begin(brw);
 135
 136    for (i = 0; i < brw->vb.nr_vb; i++) {
 137       brw_add_validated_bo(brw, brw->vb.vb[i].bo);
 138    }
 139
 140    return 0;
 141 }
 142
 143 static int brw_emit_vertex_buffers( struct brw_context *brw )
 144 {
 145    int i;
 146
 147    /* If the VS doesn't read any inputs (calculating vertex position from
 148     * a state variable for some reason, for example), just bail.
 149     *
 150     * The stale VB state stays in place, but they don't do anything unless
 151     * a VE loads from them.
 152     */
 153    if (brw->vb.nr_vb == 0) {
 154       if (BRW_DEBUG & DEBUG_VERTS)
 155          debug_printf("%s: no active vertex buffers\n", __FUNCTION__);
 156
 157       return 0;
 158    }
 159
 160    /* Emit VB state packets.
 161     */
 162    BEGIN_BATCH(1 + brw->vb.nr_vb * 4, IGNORE_CLIPRECTS);
 163    OUT_BATCH((CMD_VERTEX_BUFFER << 16) |
 164              ((1 + brw->vb.nr_vb * 4) - 2));
 165
 166    for (i = 0; i < brw->vb.nr_vb; i++) {
 167       OUT_BATCH((i << BRW_VB0_INDEX_SHIFT) |
 168                 BRW_VB0_ACCESS_VERTEXDATA |
 169                 (brw->vb.vb[i].stride << BRW_VB0_PITCH_SHIFT));
 170       OUT_RELOC(brw->vb.vb[i].bo,
 171                 BRW_USAGE_VERTEX,
 172                 brw->vb.vb[i].offset);
 173       if (brw->gen == 5) {
 174          OUT_RELOC(brw->vb.vb[i].bo,
 175                    BRW_USAGE_VERTEX,
 176                    brw->vb.vb[i].bo->size - 1);
 177       } else
 178          OUT_BATCH(brw->vb.vb[i].stride ? brw->vb.vb[i].vertex_count : 0);
 179       OUT_BATCH(0); /* Instance data step rate */
 180    }
 181    ADVANCE_BATCH();
 182    return 0;
 183 }
 184
 185
 186
 187 static int brw_emit_vertex_elements(struct brw_context *brw)
 188 {
 189    const struct brw_vertex_element_packet *brw_velems = brw->curr.velems;
 190    unsigned size = brw_velems->header.length + 2;
 191
 192    /* why is this here */
 193    brw_emit_query_begin(brw);
 194
 195    brw_batchbuffer_data(brw->batch, brw_velems, size * 4, IGNORE_CLIPRECTS);
 196
 197    return 0;
 198 }
 199
 200
 201 static int brw_emit_vertices( struct brw_context *brw )
 202 {
 203    int ret;
 204
 205    ret = brw_emit_vertex_buffers( brw );
 206    if (ret)
 207       return ret;
 208
 209    /* XXX should separate this? */
 210    ret = brw_emit_vertex_elements( brw );
 211    if (ret)
 212       return ret;
 213
 214    return 0;
 215 }
 216
 217
 218 const struct brw_tracked_state brw_vertices = {
 219    .dirty = {
 220       .mesa = (PIPE_NEW_INDEX_RANGE |
 221                PIPE_NEW_VERTEX_BUFFER |
 222                PIPE_NEW_VERTEX_ELEMENT),
 223       .brw = BRW_NEW_BATCH,
 224       .cache = 0,
 225    },
 226    .prepare = brw_prepare_vertices,
 227    .emit = brw_emit_vertices,
 228 };
 229
 230
 231 static int brw_prepare_indices(struct brw_context *brw)
 232 {
 233    struct pipe_resource *index_buffer = brw->curr.index_buffer;
 234    struct pipe_resource *upload_buf = NULL;
 235    struct brw_winsys_buffer *bo = NULL;
 236    GLuint offset;
 237    GLuint index_size, index_offset;
 238    GLuint ib_size;
 239    int ret;
 240
 241    if (index_buffer == NULL)
 242       return 0;
 243
 244    if (BRW_DEBUG & DEBUG_VERTS)
 245       debug_printf("%s: index_size:%d index_buffer->size:%d\n",
 246                    __FUNCTION__,
 247                    brw->curr.index_size,
 248                    brw->curr.index_buffer->width0);
 249
 250    ib_size = index_buffer->width0;
 251    index_size = brw->curr.index_size;
 252    index_offset = brw->curr.index_offset;
 253
 254    /* Turn userbuffer into a proper hardware buffer?
 255     */
 256    if (brw_buffer_is_user_buffer(index_buffer)) {
 257       boolean flushed;
 258
 259       ret = u_upload_buffer( brw->vb.upload_index,
 260                              0,
 261                              index_offset,
 262                              ib_size,
 263                              index_buffer,
 264                              &offset,
 265                              &upload_buf,
 266                              &flushed );
 267       if (ret)
 268          return ret;
 269
 270       bo = brw_buffer(upload_buf)->bo;
 271
 272       /* XXX: annotate the userbuffer with the upload information so
 273        * that successive calls don't get re-uploaded.
 274        */
 275    }
 276    else {
 277       bo = brw_buffer(index_buffer)->bo;
 278       ib_size = bo->size;
 279       offset = index_offset;
 280    }
 281
 282    /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading the
 283     * index buffer state when we're just moving the start index of our
 284     * drawing.
 285     *
 286     * In gallium this will happen in the case where successive draw
 287     * calls are made with (distinct?) userbuffers, but the upload_mgr
 288     * places the data into a single winsys buffer.
 289     *
 290     * This statechange doesn't raise any state flags and is always
 291     * just merged into the final draw packet:
 292     */
 293    if (1) {
 294       assert((offset & (index_size - 1)) == 0);
 295       brw->ib.start_vertex_offset = offset / index_size;
 296    }
 297
 298    /* These statechanges trigger a new CMD_INDEX_BUFFER packet:
 299     */
 300    if (brw->ib.bo != bo ||
 301        brw->ib.size != ib_size)
 302    {
 303       bo_reference(&brw->ib.bo, bo);
 304       brw->ib.size = ib_size;
 305       brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
 306    }
 307
 308    pipe_resource_reference( &upload_buf, NULL );
 309    brw_add_validated_bo(brw, brw->ib.bo);
 310    return 0;
 311 }
 312
 313 const struct brw_tracked_state brw_indices = {
 314    .dirty = {
 315       .mesa = PIPE_NEW_INDEX_BUFFER,
 316       .brw = 0,
 317       .cache = 0,
 318    },
 319    .prepare = brw_prepare_indices,
 320 };
 321
 322 static int brw_emit_index_buffer(struct brw_context *brw)
 323 {
 324    /* Emit the indexbuffer packet:
 325     */
 326    if (brw->ib.bo)
 327    {
 328       struct brw_indexbuffer ib;
 329
 330       memset(&ib, 0, sizeof(ib));
 331
 332       ib.header.bits.opcode = CMD_INDEX_BUFFER;
 333       ib.header.bits.length = sizeof(ib)/4 - 2;
 334       ib.header.bits.index_format = get_index_type(brw->ib.size);
 335       ib.header.bits.cut_index_enable = 0;
 336
 337       BEGIN_BATCH(4, IGNORE_CLIPRECTS);
 338       OUT_BATCH( ib.header.dword );
 339       OUT_RELOC(brw->ib.bo,
 340                 BRW_USAGE_VERTEX,
 341                 brw->ib.offset);
 342       OUT_RELOC(brw->ib.bo,
 343                 BRW_USAGE_VERTEX,
 344                 brw->ib.offset + brw->ib.size - 1);
 345       OUT_BATCH( 0 );
 346       ADVANCE_BATCH();
 347    }
 348
 349    return 0;
 350 }
 351
 352 const struct brw_tracked_state brw_index_buffer = {
 353    .dirty = {
 354       .mesa = 0,
 355       .brw = BRW_NEW_BATCH | BRW_NEW_INDEX_BUFFER,
 356       .cache = 0,
 357    },
 358    .emit = brw_emit_index_buffer,
 359 };