src/gallium/drivers/svga/svga_pipe_draw.c

   1 /**********************************************************
   2  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   3  *
   4  * Permission is hereby granted, free of charge, to any person
   5  * obtaining a copy of this software and associated documentation
   6  * files (the "Software"), to deal in the Software without
   7  * restriction, including without limitation the rights to use, copy,
   8  * modify, merge, publish, distribute, sublicense, and/or sell copies
   9  * of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  **********************************************************/
  25
  26 #include "svga_cmd.h"
  27
  28 #include "util/u_format.h"
  29 #include "util/u_inlines.h"
  30 #include "util/u_prim.h"
  31 #include "util/u_time.h"
  32 #include "indices/u_indices.h"
  33
  34 #include "svga_hw_reg.h"
  35 #include "svga_context.h"
  36 #include "svga_screen.h"
  37 #include "svga_draw.h"
  38 #include "svga_state.h"
  39 #include "svga_swtnl.h"
  40 #include "svga_debug.h"
  41 #include "svga_resource_buffer.h"
  42 #include "util/u_upload_mgr.h"
  43
  44 /**
  45  * Determine the ranges to upload for the user-buffers referenced
  46  * by the next draw command.
  47  *
  48  * TODO: It might be beneficial to support multiple ranges. In that case,
  49  * the struct svga_buffer::uploaded member should be made an array or a
  50  * list, since we need to account for the possibility that different ranges
  51  * may be uploaded to different hardware buffers chosen by the utility
  52  * upload manager.
  53  */
  54
  55 static void
  56 svga_user_buffer_range(struct svga_context *svga,
  57                        unsigned start,
  58                        unsigned count,
  59                        unsigned instance_count)
  60 {
  61    const struct pipe_vertex_element *ve = svga->curr.velems->velem;
  62    int i;
  63
  64    /*
  65     * Release old uploaded range (if not done already) and
  66     * initialize new ranges.
  67     */
  68
  69    for (i=0; i < svga->curr.velems->count; i++) {
  70       struct pipe_vertex_buffer *vb =
  71          &svga->curr.vb[ve[i].vertex_buffer_index];
  72
  73       if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
  74          struct svga_buffer *buffer = svga_buffer(vb->buffer);
  75
  76          pipe_resource_reference(&buffer->uploaded.buffer, NULL);
  77          buffer->uploaded.start = ~0;
  78          buffer->uploaded.end = 0;
  79       }
  80    }
  81
  82    for (i=0; i < svga->curr.velems->count; i++) {
  83       struct pipe_vertex_buffer *vb =
  84          &svga->curr.vb[ve[i].vertex_buffer_index];
  85
  86       if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
  87          struct svga_buffer *buffer = svga_buffer(vb->buffer);
  88          unsigned first, size;
  89          unsigned instance_div = ve[i].instance_divisor;
  90          unsigned elemSize = util_format_get_blocksize(ve[i].src_format);
  91
  92          svga->dirty |= SVGA_NEW_VBUFFER;
  93
  94          if (instance_div) {
  95             first = ve[i].src_offset;
  96             count = (instance_count + instance_div - 1) / instance_div;
  97             size = vb->stride * (count - 1) + elemSize;
  98          } else if (vb->stride) {
  99             first = vb->stride * start + ve[i].src_offset;
 100             size = vb->stride * (count - 1) + elemSize;
 101          } else {
 102             /* Only a single vertex!
 103              * Upload with the largest vertex size the hw supports,
 104              * if possible.
 105              */
 106             first = ve[i].src_offset;
 107             size = MIN2(16, vb->buffer->width0);
 108          }
 109
 110          buffer->uploaded.start = MIN2(buffer->uploaded.start, first);
 111          buffer->uploaded.end = MAX2(buffer->uploaded.end, first + size);
 112       }
 113    }
 114 }
 115
 116 /**
 117  * svga_upload_user_buffers - upload parts of user buffers
 118  *
 119  * This function streams a part of a user buffer to hw and fills
 120  * svga_buffer::uploaded with information on the upload.
 121  */
 122
 123 static int
 124 svga_upload_user_buffers(struct svga_context *svga,
 125                          unsigned start,
 126                          unsigned count,
 127                          unsigned instance_count)
 128 {
 129    const struct pipe_vertex_element *ve = svga->curr.velems->velem;
 130    unsigned i;
 131    int ret;
 132
 133    svga_user_buffer_range(svga, start, count, instance_count);
 134
 135    for (i=0; i < svga->curr.velems->count; i++) {
 136       struct pipe_vertex_buffer *vb =
 137          &svga->curr.vb[ve[i].vertex_buffer_index];
 138
 139       if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
 140          struct svga_buffer *buffer = svga_buffer(vb->buffer);
 141
 142          /*
 143           * Check if already uploaded. Otherwise go ahead and upload.
 144           */
 145
 146          if (buffer->uploaded.buffer)
 147             continue;
 148
 149          ret = u_upload_buffer( svga->upload_vb,
 150                                 0,
 151                                 buffer->uploaded.start,
 152                                 buffer->uploaded.end - buffer->uploaded.start,
 153                                 &buffer->b.b,
 154                                 &buffer->uploaded.offset,
 155                                 &buffer->uploaded.buffer);
 156
 157          if (ret)
 158             return ret;
 159
 160          if (0)
 161             debug_printf("%s: %d: orig buf %p upl buf %p ofs %d sofs %d"
 162                          " sz %d\n",
 163                          __FUNCTION__,
 164                          i,
 165                          buffer,
 166                          buffer->uploaded.buffer,
 167                          buffer->uploaded.offset,
 168                          buffer->uploaded.start,
 169                          buffer->uploaded.end - buffer->uploaded.start);
 170
 171          vb->buffer_offset = buffer->uploaded.offset;
 172       }
 173    }
 174
 175    return PIPE_OK;
 176 }
 177
 178 /**
 179  * svga_release_user_upl_buffers - release uploaded parts of user buffers
 180  *
 181  * This function releases the hw copy of the uploaded fraction of the
 182  * user-buffer. It's important to do this as soon as all draw calls
 183  * affecting the uploaded fraction are issued, as this allows for
 184  * efficient reuse of the hardware surface backing the uploaded fraction.
 185  *
 186  * svga_buffer::source_offset is set to 0, and svga_buffer::uploaded::buffer
 187  * is set to 0.
 188  */
 189
 190 static void
 191 svga_release_user_upl_buffers(struct svga_context *svga)
 192 {
 193    unsigned i;
 194    unsigned nr;
 195
 196    nr = svga->curr.num_vertex_buffers;
 197
 198    for (i = 0; i < nr; ++i) {
 199       struct pipe_vertex_buffer *vb = &svga->curr.vb[i];
 200
 201       if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
 202          struct svga_buffer *buffer = svga_buffer(vb->buffer);
 203
 204          /* The buffer_offset is relative to the uploaded buffer.
 205           * Since we're discarding that buffer we need to reset this offset
 206           * so it's not inadvertantly applied to a subsequent draw.
 207           *
 208           * XXX a root problem here is that the svga->curr.vb[] information
 209           * is getting set both by gallium API calls and by code in
 210           * svga_upload_user_buffers().  We should instead have two copies
 211           * of the vertex buffer information and choose between as needed.
 212           */
 213          vb->buffer_offset = 0;
 214
 215          buffer->uploaded.start = ~0;
 216          buffer->uploaded.end = 0;
 217          if (buffer->uploaded.buffer)
 218             pipe_resource_reference(&buffer->uploaded.buffer, NULL);
 219       }
 220    }
 221 }
 222
 223
 224
 225 static enum pipe_error
 226 retry_draw_range_elements( struct svga_context *svga,
 227                            struct pipe_resource *index_buffer,
 228                            unsigned index_size,
 229                            int index_bias,
 230                            unsigned min_index,
 231                            unsigned max_index,
 232                            unsigned prim,
 233                            unsigned start,
 234                            unsigned count,
 235                            unsigned instance_count,
 236                            boolean do_retry )
 237 {
 238    enum pipe_error ret = PIPE_OK;
 239
 240    svga_hwtnl_set_unfilled( svga->hwtnl,
 241                             svga->curr.rast->hw_unfilled );
 242
 243    svga_hwtnl_set_flatshade( svga->hwtnl,
 244                              svga->curr.rast->templ.flatshade,
 245                              svga->curr.rast->templ.flatshade_first );
 246
 247    ret = svga_upload_user_buffers( svga, min_index + index_bias,
 248                                    max_index - min_index + 1, instance_count );
 249    if (ret != PIPE_OK)
 250       goto retry;
 251
 252    ret = svga_update_state( svga, SVGA_STATE_HW_DRAW );
 253    if (ret != PIPE_OK)
 254       goto retry;
 255
 256    ret = svga_hwtnl_draw_range_elements( svga->hwtnl,
 257                                          index_buffer, index_size, index_bias,
 258                                          min_index, max_index,
 259                                          prim, start, count );
 260    if (ret != PIPE_OK)
 261       goto retry;
 262
 263    return PIPE_OK;
 264
 265 retry:
 266    svga_context_flush( svga, NULL );
 267
 268    if (do_retry)
 269    {
 270       return retry_draw_range_elements( svga,
 271                                         index_buffer, index_size, index_bias,
 272                                         min_index, max_index,
 273                                         prim, start, count,
 274                                         instance_count, FALSE );
 275    }
 276
 277    return ret;
 278 }
 279
 280
 281 static enum pipe_error
 282 retry_draw_arrays( struct svga_context *svga,
 283                    unsigned prim,
 284                    unsigned start,
 285                    unsigned count,
 286                    unsigned instance_count,
 287                    boolean do_retry )
 288 {
 289    enum pipe_error ret;
 290
 291    svga_hwtnl_set_unfilled( svga->hwtnl,
 292                             svga->curr.rast->hw_unfilled );
 293
 294    svga_hwtnl_set_flatshade( svga->hwtnl,
 295                              svga->curr.rast->templ.flatshade,
 296                              svga->curr.rast->templ.flatshade_first );
 297
 298    ret = svga_upload_user_buffers( svga, start, count, instance_count );
 299
 300    if (ret != PIPE_OK)
 301       goto retry;
 302
 303    ret = svga_update_state( svga, SVGA_STATE_HW_DRAW );
 304    if (ret != PIPE_OK)
 305       goto retry;
 306
 307    ret = svga_hwtnl_draw_arrays( svga->hwtnl, prim,
 308                                  start, count );
 309    if (ret != PIPE_OK)
 310       goto retry;
 311
 312    return PIPE_OK;
 313
 314 retry:
 315    if (ret == PIPE_ERROR_OUT_OF_MEMORY && do_retry)
 316    {
 317       svga_context_flush( svga, NULL );
 318
 319       return retry_draw_arrays( svga,
 320                                 prim,
 321                                 start,
 322                                 count,
 323                                 instance_count,
 324                                 FALSE );
 325    }
 326
 327    return ret;
 328 }
 329
 330
 331 static void
 332 svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 333 {
 334    struct svga_context *svga = svga_context( pipe );
 335    unsigned reduced_prim = u_reduced_prim( info->mode );
 336    unsigned count = info->count;
 337    enum pipe_error ret = 0;
 338    boolean needed_swtnl;
 339
 340    if (!u_trim_pipe_prim( info->mode, &count ))
 341       return;
 342
 343    /*
 344     * Mark currently bound target surfaces as dirty
 345     * doesn't really matter if it is done before drawing.
 346     *
 347     * TODO If we ever normaly return something other then
 348     * true we should not mark it as dirty then.
 349     */
 350    svga_mark_surfaces_dirty(svga_context(pipe));
 351
 352    if (svga->curr.reduced_prim != reduced_prim) {
 353       svga->curr.reduced_prim = reduced_prim;
 354       svga->dirty |= SVGA_NEW_REDUCED_PRIMITIVE;
 355    }
 356
 357    needed_swtnl = svga->state.sw.need_swtnl;
 358
 359    svga_update_state_retry( svga, SVGA_STATE_NEED_SWTNL );
 360
 361 #ifdef DEBUG
 362    if (svga->curr.vs->base.id == svga->debug.disable_shader ||
 363        svga->curr.fs->base.id == svga->debug.disable_shader)
 364       return;
 365 #endif
 366
 367    if (svga->state.sw.need_swtnl) {
 368       if (!needed_swtnl) {
 369          /*
 370           * We're switching from HW to SW TNL.  SW TNL will require mapping all
 371           * currently bound vertex buffers, some of which may already be
 372           * referenced in the current command buffer as result of previous HW
 373           * TNL. So flush now, to prevent the context to flush while a referred
 374           * vertex buffer is mapped.
 375           */
 376
 377          svga_context_flush(svga, NULL);
 378       }
 379
 380       /* Avoid leaking the previous hwtnl bias to swtnl */
 381       svga_hwtnl_set_index_bias( svga->hwtnl, 0 );
 382       ret = svga_swtnl_draw_vbo( svga, info );
 383    }
 384    else {
 385       if (info->indexed && svga->curr.ib.buffer) {
 386          unsigned offset;
 387
 388          assert(svga->curr.ib.offset % svga->curr.ib.index_size == 0);
 389          offset = svga->curr.ib.offset / svga->curr.ib.index_size;
 390
 391          ret = retry_draw_range_elements( svga,
 392                                           svga->curr.ib.buffer,
 393                                           svga->curr.ib.index_size,
 394                                           info->index_bias,
 395                                           info->min_index,
 396                                           info->max_index,
 397                                           info->mode,
 398                                           info->start + offset,
 399                                           info->count,
 400                                           info->instance_count,
 401                                           TRUE );
 402       }
 403       else {
 404          ret = retry_draw_arrays( svga,
 405                                   info->mode,
 406                                   info->start,
 407                                   info->count,
 408                                   info->instance_count,
 409                                   TRUE );
 410       }
 411    }
 412
 413    /* XXX: Silence warnings, do something sensible here? */
 414    (void)ret;
 415
 416    svga_release_user_upl_buffers( svga );
 417
 418    if (SVGA_DEBUG & DEBUG_FLUSH) {
 419       svga_hwtnl_flush_retry( svga );
 420       svga_context_flush(svga, NULL);
 421    }
 422 }
 423
 424
 425 void svga_init_draw_functions( struct svga_context *svga )
 426 {
 427    svga->pipe.draw_vbo = svga_draw_vbo;
 428 }