src/gallium/drivers/i965/brw_clip_state.c

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a 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, sublicense, 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
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32 #include "util/u_math.h"
  33
  34 #include "brw_context.h"
  35 #include "brw_clip.h"
  36 #include "brw_state.h"
  37 #include "brw_defines.h"
  38 #include "brw_debug.h"
  39
  40 struct brw_clip_unit_key {
  41    unsigned int total_grf;
  42    unsigned int urb_entry_read_length;
  43    unsigned int curb_entry_read_length;
  44    unsigned int clip_mode;
  45
  46    unsigned int curbe_offset;
  47
  48    unsigned int nr_urb_entries, urb_size;
  49
  50    GLboolean depth_clamp;
  51 };
  52
  53 static void
  54 clip_unit_populate_key(struct brw_context *brw, struct brw_clip_unit_key *key)
  55 {
  56    memset(key, 0, sizeof(*key));
  57
  58    /* CACHE_NEW_CLIP_PROG */
  59    key->total_grf = brw->clip.prog_data->total_grf;
  60    key->urb_entry_read_length = brw->clip.prog_data->urb_read_length;
  61    key->curb_entry_read_length = brw->clip.prog_data->curb_read_length;
  62    key->clip_mode = brw->clip.prog_data->clip_mode;
  63
  64    /* BRW_NEW_CURBE_OFFSETS */
  65    key->curbe_offset = brw->curbe.clip_start;
  66
  67    /* BRW_NEW_URB_FENCE */
  68    key->nr_urb_entries = brw->urb.nr_clip_entries;
  69    key->urb_size = brw->urb.vsize;
  70
  71    /*  */
  72    key->depth_clamp = 0; /* XXX: add this to gallium: ctx->Transform.DepthClamp; */
  73 }
  74
  75 static enum pipe_error
  76 clip_unit_create_from_key(struct brw_context *brw,
  77                           struct brw_clip_unit_key *key,
  78                           struct brw_winsys_reloc *reloc,
  79                           struct brw_winsys_buffer **bo_out)
  80 {
  81    struct brw_clip_unit_state clip;
  82    enum pipe_error ret;
  83
  84    memset(&clip, 0, sizeof(clip));
  85
  86    clip.thread0.grf_reg_count = align(key->total_grf, 16) / 16 - 1;
  87    /* reloc */
  88    clip.thread0.kernel_start_pointer = 0;
  89
  90    clip.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
  91    clip.thread1.single_program_flow = 1;
  92
  93    clip.thread3.urb_entry_read_length = key->urb_entry_read_length;
  94    clip.thread3.const_urb_entry_read_length = key->curb_entry_read_length;
  95    clip.thread3.const_urb_entry_read_offset = key->curbe_offset * 2;
  96    clip.thread3.dispatch_grf_start_reg = 1;
  97    clip.thread3.urb_entry_read_offset = 0;
  98
  99    clip.thread4.nr_urb_entries = key->nr_urb_entries;
 100    clip.thread4.urb_entry_allocation_size = key->urb_size - 1;
 101    /* If we have enough clip URB entries to run two threads, do so.
 102     */
 103    if (key->nr_urb_entries >= 10) {
 104       /* Half of the URB entries go to each thread, and it has to be an
 105        * even number.
 106        */
 107       assert(key->nr_urb_entries % 2 == 0);
 108
 109       /* Although up to 16 concurrent Clip threads are allowed on IGDNG,
 110        * only 2 threads can output VUEs at a time.
 111        */
 112       if (BRW_IS_IGDNG(brw))
 113          clip.thread4.max_threads = 16 - 1;
 114       else
 115          clip.thread4.max_threads = 2 - 1;
 116    } else {
 117       assert(key->nr_urb_entries >= 5);
 118       clip.thread4.max_threads = 1 - 1;
 119    }
 120
 121    if (BRW_DEBUG & DEBUG_SINGLE_THREAD)
 122       clip.thread4.max_threads = 0;
 123
 124    if (BRW_DEBUG & DEBUG_STATS)
 125       clip.thread4.stats_enable = 1;
 126
 127    clip.clip5.userclip_enable_flags = 0x7f;
 128    clip.clip5.userclip_must_clip = 1;
 129    clip.clip5.guard_band_enable = 0;
 130    if (!key->depth_clamp)
 131       clip.clip5.viewport_z_clip_enable = 1;
 132    clip.clip5.viewport_xy_clip_enable = 1;
 133    clip.clip5.vertex_position_space = BRW_CLIP_NDCSPACE;
 134    clip.clip5.api_mode = BRW_CLIP_API_OGL;
 135    clip.clip5.clip_mode = key->clip_mode;
 136
 137    if (BRW_IS_G4X(brw))
 138       clip.clip5.negative_w_clip_test = 1;
 139
 140    clip.clip6.clipper_viewport_state_ptr = 0;
 141    clip.viewport_xmin = -1;
 142    clip.viewport_xmax = 1;
 143    clip.viewport_ymin = -1;
 144    clip.viewport_ymax = 1;
 145
 146    ret = brw_upload_cache(&brw->cache, BRW_CLIP_UNIT,
 147                           key, sizeof(*key),
 148                           reloc, 1,
 149                           &clip, sizeof(clip),
 150                           NULL, NULL,
 151                           bo_out);
 152    if (ret)
 153       return ret;
 154
 155    return PIPE_OK;
 156 }
 157
 158 static int upload_clip_unit( struct brw_context *brw )
 159 {
 160    struct brw_clip_unit_key key;
 161    struct brw_winsys_reloc reloc[1];
 162    unsigned grf_reg_count;
 163    enum pipe_error ret;
 164
 165    clip_unit_populate_key(brw, &key);
 166
 167    grf_reg_count = align(key.total_grf, 16) / 16 - 1;
 168
 169    /* clip program relocation
 170     *
 171     * XXX: these reloc structs are long lived and only need to be
 172     * updated when the bound BO changes.  Hopefully the stuff mixed in
 173     * in the delta's is non-orthogonal.
 174     */
 175    assert(brw->clip.prog_bo);
 176    make_reloc(&reloc[0],
 177               BRW_USAGE_STATE,
 178               grf_reg_count << 1,
 179               offsetof(struct brw_clip_unit_state, thread0),
 180               brw->clip.prog_bo);
 181
 182
 183    if (brw_search_cache(&brw->cache, BRW_CLIP_UNIT,
 184                         &key, sizeof(key),
 185                         reloc, 1,
 186                         NULL,
 187                         &brw->clip.state_bo))
 188       return PIPE_OK;
 189
 190    /* Create new:
 191     */
 192    ret = clip_unit_create_from_key(brw, &key,
 193                                    reloc,
 194                                    &brw->clip.state_bo);
 195    if (ret)
 196       return ret;
 197
 198    return PIPE_OK;
 199 }
 200
 201 const struct brw_tracked_state brw_clip_unit = {
 202    .dirty = {
 203       .mesa  = 0,
 204       .brw   = (BRW_NEW_CURBE_OFFSETS |
 205                 BRW_NEW_URB_FENCE),
 206       .cache = CACHE_NEW_CLIP_PROG
 207    },
 208    .prepare = upload_clip_unit,
 209 };