src/gallium/drivers/i965/brw_sf_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 "pipe/p_state.h"
  35
  36 #include "brw_context.h"
  37 #include "brw_state.h"
  38 #include "brw_defines.h"
  39 #include "brw_debug.h"
  40 #include "brw_pipe_rast.h"
  41
  42 static enum pipe_error upload_sf_vp(struct brw_context *brw)
  43 {
  44    const struct pipe_viewport_state *vp = &brw->curr.viewport;
  45    const struct pipe_scissor_state *scissor = &brw->curr.scissor;
  46    struct brw_sf_viewport sfv;
  47    enum pipe_error ret;
  48
  49    memset(&sfv, 0, sizeof(sfv));
  50
  51    /* PIPE_NEW_VIEWPORT, PIPE_NEW_SCISSOR */
  52
  53    sfv.viewport.m00 = vp->scale[0];
  54    sfv.viewport.m11 = vp->scale[1];
  55    sfv.viewport.m22 = vp->scale[2];
  56    sfv.viewport.m30 = vp->translate[0];
  57    sfv.viewport.m31 = vp->translate[1];
  58    sfv.viewport.m32 = vp->translate[2];
  59
  60    sfv.scissor.xmin = scissor->minx;
  61    sfv.scissor.xmax = scissor->maxx - 1; /* ? */
  62    sfv.scissor.ymin = scissor->miny;
  63    sfv.scissor.ymax = scissor->maxy - 1; /* ? */
  64
  65    ret = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0,
  66                          &brw->sf.vp_bo );
  67    if (ret)
  68       return ret;
  69
  70    return PIPE_OK;
  71 }
  72
  73 const struct brw_tracked_state brw_sf_vp = {
  74    .dirty = {
  75       .mesa  = (PIPE_NEW_VIEWPORT |
  76                 PIPE_NEW_SCISSOR),
  77       .brw   = 0,
  78       .cache = 0
  79    },
  80    .prepare = upload_sf_vp
  81 };
  82
  83 struct brw_sf_unit_key {
  84    unsigned int total_grf;
  85    unsigned int urb_entry_read_length;
  86    unsigned int nr_urb_entries, urb_size, sfsize;
  87
  88    unsigned scissor:1;
  89    unsigned line_smooth:1;
  90    unsigned point_sprite:1;
  91    unsigned point_attenuated:1;
  92    unsigned front_face:2;
  93    unsigned cull_mode:2;
  94    unsigned flatshade_first:1;
  95    unsigned gl_rasterization_rules:1;
  96    unsigned line_last_pixel_enable:1;
  97    float line_width;
  98    float point_size;
  99 };
 100
 101 static void
 102 sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key)
 103 {
 104    const struct pipe_rasterizer_state *rast = &brw->curr.rast->templ;
 105    memset(key, 0, sizeof(*key));
 106
 107    /* CACHE_NEW_SF_PROG */
 108    key->total_grf = brw->sf.prog_data->total_grf;
 109    key->urb_entry_read_length = brw->sf.prog_data->urb_read_length;
 110
 111    /* BRW_NEW_URB_FENCE */
 112    key->nr_urb_entries = brw->urb.nr_sf_entries;
 113    key->urb_size = brw->urb.vsize;
 114    key->sfsize = brw->urb.sfsize;
 115
 116    /* PIPE_NEW_RAST */
 117    key->scissor = rast->scissor;
 118    key->front_face = rast->front_winding;
 119    key->cull_mode = rast->cull_mode;
 120    key->line_smooth = rast->line_smooth;
 121    key->line_width = rast->line_width;
 122    key->flatshade_first = rast->flatshade_first;
 123    key->line_last_pixel_enable = rast->line_last_pixel;
 124    key->gl_rasterization_rules = rast->gl_rasterization_rules;
 125
 126    key->point_sprite = rast->point_sprite;
 127    key->point_attenuated = rast->point_size_per_vertex;
 128
 129    key->point_size = CLAMP(rast->point_size,
 130                            rast->point_size_min,
 131                            rast->point_size_max);
 132 }
 133
 134 static enum pipe_error
 135 sf_unit_create_from_key(struct brw_context *brw,
 136                         struct brw_sf_unit_key *key,
 137                         struct brw_winsys_reloc *reloc,
 138                         struct brw_winsys_buffer **bo_out)
 139 {
 140    struct brw_sf_unit_state sf;
 141    enum pipe_error ret;
 142    int chipset_max_threads;
 143    memset(&sf, 0, sizeof(sf));
 144
 145    sf.thread0.grf_reg_count = align(key->total_grf, 16) / 16 - 1;
 146    /* reloc */
 147    sf.thread0.kernel_start_pointer = 0;
 148
 149    sf.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
 150
 151    sf.thread3.dispatch_grf_start_reg = 3;
 152
 153    if (BRW_IS_IGDNG(brw))
 154        sf.thread3.urb_entry_read_offset = 3;
 155    else
 156        sf.thread3.urb_entry_read_offset = 1;
 157
 158    sf.thread3.urb_entry_read_length = key->urb_entry_read_length;
 159
 160    sf.thread4.nr_urb_entries = key->nr_urb_entries;
 161    sf.thread4.urb_entry_allocation_size = key->sfsize - 1;
 162
 163    /* Each SF thread produces 1 PUE, and there can be up to 24(Pre-IGDNG) or
 164     * 48(IGDNG) threads
 165     */
 166    if (BRW_IS_IGDNG(brw))
 167       chipset_max_threads = 48;
 168    else
 169       chipset_max_threads = 24;
 170
 171    sf.thread4.max_threads = MIN2(chipset_max_threads, key->nr_urb_entries) - 1;
 172
 173    if (BRW_DEBUG & DEBUG_SINGLE_THREAD)
 174       sf.thread4.max_threads = 0;
 175
 176    if (BRW_DEBUG & DEBUG_STATS)
 177       sf.thread4.stats_enable = 1;
 178
 179    /* CACHE_NEW_SF_VP */
 180    /* reloc */
 181    sf.sf5.sf_viewport_state_offset = 0;
 182
 183    sf.sf5.viewport_transform = 1;
 184
 185    if (key->scissor)
 186       sf.sf6.scissor = 1;
 187
 188    if (key->front_face == PIPE_WINDING_CCW)
 189       sf.sf5.front_winding = BRW_FRONTWINDING_CCW;
 190    else
 191       sf.sf5.front_winding = BRW_FRONTWINDING_CW;
 192
 193    switch (key->cull_mode) {
 194    case PIPE_WINDING_CCW:
 195    case PIPE_WINDING_CW:
 196       sf.sf6.cull_mode = (key->front_face == key->cull_mode ?
 197                           BRW_CULLMODE_FRONT :
 198                           BRW_CULLMODE_BACK);
 199       break;
 200    case PIPE_WINDING_BOTH:
 201       sf.sf6.cull_mode = BRW_CULLMODE_BOTH;
 202       break;
 203    case PIPE_WINDING_NONE:
 204       sf.sf6.cull_mode = BRW_CULLMODE_NONE;
 205       break;
 206    default:
 207       assert(0);
 208       sf.sf6.cull_mode = BRW_CULLMODE_NONE;
 209       break;
 210    }
 211
 212    /* _NEW_LINE */
 213    /* XXX use ctx->Const.Min/MaxLineWidth here */
 214    sf.sf6.line_width = CLAMP(key->line_width, 1.0, 5.0) * (1<<1);
 215
 216    sf.sf6.line_endcap_aa_region_width = 1;
 217    if (key->line_smooth)
 218       sf.sf6.aa_enable = 1;
 219    else if (sf.sf6.line_width <= 0x2)
 220        sf.sf6.line_width = 0;
 221
 222    /* XXX: gl_rasterization_rules?  something else?
 223     */
 224    sf.sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT;
 225    sf.sf6.point_rast_rule = BRW_RASTRULE_LOWER_RIGHT;
 226    sf.sf6.point_rast_rule = 1;
 227
 228    /* XXX clamp max depends on AA vs. non-AA */
 229
 230    /* _NEW_POINT */
 231    sf.sf7.sprite_point = key->point_sprite;
 232    sf.sf7.point_size = CLAMP(rint(key->point_size), 1, 255) * (1<<3);
 233    sf.sf7.use_point_size_state = !key->point_attenuated;
 234    sf.sf7.aa_line_distance_mode = 0;
 235
 236    /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
 237     */
 238    if (!key->flatshade_first) {
 239       sf.sf7.trifan_pv = 2;
 240       sf.sf7.linestrip_pv = 1;
 241       sf.sf7.tristrip_pv = 2;
 242    } else {
 243       sf.sf7.trifan_pv = 1;
 244       sf.sf7.linestrip_pv = 0;
 245       sf.sf7.tristrip_pv = 0;
 246    }
 247
 248    sf.sf7.line_last_pixel_enable = key->line_last_pixel_enable;
 249
 250    /* Set bias for OpenGL rasterization rules:
 251     */
 252    if (key->gl_rasterization_rules) {
 253       sf.sf6.dest_org_vbias = 0x8;
 254       sf.sf6.dest_org_hbias = 0x8;
 255    }
 256    else {
 257       sf.sf6.dest_org_vbias = 0x0;
 258       sf.sf6.dest_org_hbias = 0x0;
 259    }
 260
 261    ret = brw_upload_cache(&brw->cache, BRW_SF_UNIT,
 262                           key, sizeof(*key),
 263                           reloc, 2,
 264                           &sf, sizeof(sf),
 265                           NULL, NULL,
 266                           bo_out);
 267    if (ret)
 268       return ret;
 269
 270
 271    return PIPE_OK;
 272 }
 273
 274 static enum pipe_error upload_sf_unit( struct brw_context *brw )
 275 {
 276    struct brw_sf_unit_key key;
 277    struct brw_winsys_reloc reloc[2];
 278    unsigned total_grf;
 279    unsigned viewport_transform;
 280    unsigned front_winding;
 281    enum pipe_error ret;
 282
 283    sf_unit_populate_key(brw, &key);
 284
 285    /* XXX: cut this crap and pre calculate the key:
 286     */
 287    total_grf = (align(key.total_grf, 16) / 16 - 1);
 288    viewport_transform = 1;
 289    front_winding = (key.front_face == PIPE_WINDING_CCW ?
 290                     BRW_FRONTWINDING_CCW :
 291                     BRW_FRONTWINDING_CW);
 292
 293    /* Emit SF program relocation */
 294    make_reloc(&reloc[0],
 295               BRW_USAGE_STATE,
 296               total_grf << 1,
 297               offsetof(struct brw_sf_unit_state, thread0),
 298               brw->sf.prog_bo);
 299
 300    /* Emit SF viewport relocation */
 301    make_reloc(&reloc[1],
 302               BRW_USAGE_STATE,
 303               front_winding | (viewport_transform << 1),
 304               offsetof(struct brw_sf_unit_state, sf5),
 305               brw->sf.vp_bo);
 306
 307
 308    if (brw_search_cache(&brw->cache, BRW_SF_UNIT,
 309                         &key, sizeof(key),
 310                         reloc, 2,
 311                         NULL,
 312                         &brw->sf.state_bo))
 313       return PIPE_OK;
 314
 315
 316    ret = sf_unit_create_from_key(brw, &key,
 317                                  reloc,
 318                                  &brw->sf.state_bo);
 319    if (ret)
 320       return ret;
 321
 322    return PIPE_OK;
 323 }
 324
 325 const struct brw_tracked_state brw_sf_unit = {
 326    .dirty = {
 327       .mesa  = (PIPE_NEW_RAST),
 328       .brw   = BRW_NEW_URB_FENCE,
 329       .cache = (CACHE_NEW_SF_VP |
 330                 CACHE_NEW_SF_PROG)
 331    },
 332    .prepare = upload_sf_unit,
 333 };