src/gallium/drivers/i965/brw_wm_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_state.h"
  36 #include "brw_defines.h"
  37 #include "brw_wm.h"
  38 #include "brw_debug.h"
  39 #include "brw_pipe_rast.h"
  40
  41 /***********************************************************************
  42  * WM unit - fragment programs and rasterization
  43  */
  44
  45 struct brw_wm_unit_key {
  46    unsigned int total_grf, total_scratch;
  47    unsigned int urb_entry_read_length;
  48    unsigned int curb_entry_read_length;
  49    unsigned int dispatch_grf_start_reg;
  50
  51    unsigned int curbe_offset;
  52    unsigned int urb_size;
  53
  54    unsigned int max_threads;
  55
  56    unsigned int nr_surfaces, sampler_count;
  57    GLboolean uses_depth, computes_depth, uses_kill, has_flow_control;
  58    GLboolean polygon_stipple, stats_wm, line_stipple, offset_enable;
  59    GLfloat offset_units, offset_factor;
  60 };
  61
  62 static void
  63 wm_unit_populate_key(struct brw_context *brw, struct brw_wm_unit_key *key)
  64 {
  65    const struct brw_fragment_shader *fp = brw->curr.fragment_shader;
  66
  67    memset(key, 0, sizeof(*key));
  68
  69    if (BRW_DEBUG & DEBUG_SINGLE_THREAD)
  70       key->max_threads = 1;
  71    else {
  72       /* WM maximum threads is number of EUs times number of threads per EU. */
  73       if (BRW_IS_IGDNG(brw))
  74          key->max_threads = 12 * 6;
  75       else if (BRW_IS_G4X(brw))
  76          key->max_threads = 10 * 5;
  77       else
  78          key->max_threads = 8 * 4;
  79    }
  80
  81    /* CACHE_NEW_WM_PROG */
  82    key->total_grf = brw->wm.prog_data->total_grf;
  83    key->urb_entry_read_length = brw->wm.prog_data->urb_read_length;
  84    key->curb_entry_read_length = brw->wm.prog_data->curb_read_length;
  85    key->dispatch_grf_start_reg = brw->wm.prog_data->first_curbe_grf;
  86    key->total_scratch = align(brw->wm.prog_data->total_scratch, 1024);
  87
  88    /* BRW_NEW_URB_FENCE */
  89    key->urb_size = brw->urb.vsize;
  90
  91    /* BRW_NEW_CURBE_OFFSETS */
  92    key->curbe_offset = brw->curbe.wm_start;
  93
  94    /* BRW_NEW_NR_SURFACEs */
  95    key->nr_surfaces = brw->wm.nr_surfaces;
  96
  97    /* CACHE_NEW_SAMPLER */
  98    key->sampler_count = brw->wm.sampler_count;
  99
 100    /* PIPE_NEW_RAST */
 101    key->polygon_stipple = brw->curr.rast->templ.poly_stipple_enable;
 102
 103    /* PIPE_NEW_FRAGMENT_PROGRAM */
 104    key->uses_depth = fp->uses_depth;
 105    key->computes_depth = fp->info.writes_z;
 106
 107    /* PIPE_NEW_DEPTH_BUFFER
 108     *
 109     * Override for NULL depthbuffer case, required by the Pixel Shader Computed
 110     * Depth field.
 111     */
 112    if (brw->curr.fb.zsbuf == NULL)
 113       key->computes_depth = 0;
 114
 115    /* PIPE_NEW_DEPTH_STENCIL_ALPHA */
 116    key->uses_kill = (fp->info.uses_kill ||
 117                      brw->curr.zstencil->cc3.alpha_test);
 118
 119    key->has_flow_control = fp->has_flow_control;
 120
 121    /* temporary sanity check assertion */
 122    assert(fp->has_flow_control == 0);
 123
 124    /* PIPE_NEW_QUERY */
 125    key->stats_wm = (brw->query.stats_wm != 0);
 126
 127    /* PIPE_NEW_RAST */
 128    key->line_stipple = brw->curr.rast->templ.line_stipple_enable;
 129
 130
 131    key->offset_enable = (brw->curr.rast->templ.offset_cw ||
 132                          brw->curr.rast->templ.offset_ccw);
 133
 134    key->offset_units = brw->curr.rast->templ.offset_units;
 135    key->offset_factor = brw->curr.rast->templ.offset_scale;
 136 }
 137
 138 /**
 139  * Setup wm hardware state.  See page 225 of Volume 2
 140  */
 141 static enum pipe_error
 142 wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key,
 143                         struct brw_winsys_reloc *reloc,
 144                         unsigned nr_reloc,
 145                         struct brw_winsys_buffer **bo_out)
 146 {
 147    struct brw_wm_unit_state wm;
 148    enum pipe_error ret;
 149
 150    memset(&wm, 0, sizeof(wm));
 151
 152    wm.thread0.grf_reg_count = align(key->total_grf, 16) / 16 - 1;
 153    wm.thread0.kernel_start_pointer = 0; /* reloc */
 154    wm.thread1.depth_coef_urb_read_offset = 1;
 155    wm.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
 156
 157    if (BRW_IS_IGDNG(brw))
 158       wm.thread1.binding_table_entry_count = 0; /* hardware requirement */
 159    else
 160       wm.thread1.binding_table_entry_count = key->nr_surfaces;
 161
 162    if (key->total_scratch != 0) {
 163       wm.thread2.scratch_space_base_pointer = 0; /* reloc */
 164       wm.thread2.per_thread_scratch_space = key->total_scratch / 1024 - 1;
 165    } else {
 166       wm.thread2.scratch_space_base_pointer = 0;
 167       wm.thread2.per_thread_scratch_space = 0;
 168    }
 169
 170    wm.thread3.dispatch_grf_start_reg = key->dispatch_grf_start_reg;
 171    wm.thread3.urb_entry_read_length = key->urb_entry_read_length;
 172    wm.thread3.urb_entry_read_offset = 0;
 173    wm.thread3.const_urb_entry_read_length = key->curb_entry_read_length;
 174    wm.thread3.const_urb_entry_read_offset = key->curbe_offset * 2;
 175
 176    if (BRW_IS_IGDNG(brw))
 177       wm.wm4.sampler_count = 0; /* hardware requirement */
 178    else
 179       wm.wm4.sampler_count = (key->sampler_count + 1) / 4;
 180
 181    /* reloc */
 182    wm.wm4.sampler_state_pointer = 0;
 183
 184    wm.wm5.program_uses_depth = key->uses_depth;
 185    wm.wm5.program_computes_depth = key->computes_depth;
 186    wm.wm5.program_uses_killpixel = key->uses_kill;
 187
 188    if (key->has_flow_control)
 189       wm.wm5.enable_8_pix = 1;
 190    else
 191       wm.wm5.enable_16_pix = 1;
 192
 193    wm.wm5.max_threads = key->max_threads - 1;
 194    wm.wm5.thread_dispatch_enable = 1;   /* AKA: color_write */
 195    wm.wm5.legacy_line_rast = 0;
 196    wm.wm5.legacy_global_depth_bias = 0;
 197    wm.wm5.early_depth_test = 1;         /* never need to disable */
 198    wm.wm5.line_aa_region_width = 0;
 199    wm.wm5.line_endcap_aa_region_width = 1;
 200
 201    wm.wm5.polygon_stipple = key->polygon_stipple;
 202
 203    if (key->offset_enable) {
 204       wm.wm5.depth_offset = 1;
 205       /* Something wierd going on with legacy_global_depth_bias,
 206        * offset_constant, scaling and MRD.  This value passes glean
 207        * but gives some odd results elsewere (eg. the
 208        * quad-offset-units test).
 209        */
 210       wm.global_depth_offset_constant = key->offset_units * 2;
 211
 212       /* This is the only value that passes glean:
 213        */
 214       wm.global_depth_offset_scale = key->offset_factor;
 215    }
 216
 217    wm.wm5.line_stipple = key->line_stipple;
 218
 219    if ((BRW_DEBUG & DEBUG_STATS) || key->stats_wm)
 220       wm.wm4.stats_enable = 1;
 221
 222    ret = brw_upload_cache(&brw->cache, BRW_WM_UNIT,
 223                           key, sizeof(*key),
 224                           reloc, nr_reloc,
 225                           &wm, sizeof(wm),
 226                           NULL, NULL,
 227                           bo_out);
 228    if (ret)
 229       return ret;
 230
 231    return PIPE_OK;
 232 }
 233
 234
 235 static enum pipe_error upload_wm_unit( struct brw_context *brw )
 236 {
 237    struct brw_wm_unit_key key;
 238    struct brw_winsys_reloc reloc[3];
 239    unsigned nr_reloc = 0;
 240    enum pipe_error ret;
 241    unsigned grf_reg_count;
 242    unsigned per_thread_scratch_space;
 243    unsigned stats_enable;
 244    unsigned sampler_count;
 245
 246    wm_unit_populate_key(brw, &key);
 247
 248
 249    /* Allocate the necessary scratch space if we haven't already.  Don't
 250     * bother reducing the allocation later, since we use scratch so
 251     * rarely.
 252     */
 253    assert(key.total_scratch <= 12 * 1024);
 254    if (key.total_scratch) {
 255       GLuint total = key.total_scratch * key.max_threads;
 256
 257       /* Do we need a new buffer:
 258        */
 259       if (brw->wm.scratch_bo && total > brw->wm.scratch_bo->size)
 260          bo_reference(&brw->wm.scratch_bo, NULL);
 261
 262       if (brw->wm.scratch_bo == NULL) {
 263          ret = brw->sws->bo_alloc(brw->sws,
 264                                   BRW_BUFFER_TYPE_SHADER_SCRATCH,
 265                                   total,
 266                                   4096,
 267                                   &brw->wm.scratch_bo);
 268          if (ret)
 269             return ret;
 270       }
 271    }
 272
 273
 274    /* XXX: temporary:
 275     */
 276    grf_reg_count = (align(key.total_grf, 16) / 16 - 1);
 277    per_thread_scratch_space = key.total_scratch / 1024 - 1;
 278    stats_enable = (BRW_DEBUG & DEBUG_STATS) || key.stats_wm;
 279    sampler_count = BRW_IS_IGDNG(brw) ? 0 :(key.sampler_count + 1) / 4;
 280
 281    /* Emit WM program relocation */
 282    make_reloc(&reloc[nr_reloc++],
 283               BRW_USAGE_STATE,
 284               grf_reg_count << 1,
 285               offsetof(struct brw_wm_unit_state, thread0),
 286               brw->wm.prog_bo);
 287
 288    /* Emit scratch space relocation */
 289    if (key.total_scratch != 0) {
 290       make_reloc(&reloc[nr_reloc++],
 291                  BRW_USAGE_SCRATCH,
 292                  per_thread_scratch_space,
 293                  offsetof(struct brw_wm_unit_state, thread2),
 294                  brw->wm.scratch_bo);
 295    }
 296
 297    /* Emit sampler state relocation */
 298    if (key.sampler_count != 0) {
 299       make_reloc(&reloc[nr_reloc++],
 300                  BRW_USAGE_STATE,
 301                  stats_enable | (sampler_count << 2),
 302                  offsetof(struct brw_wm_unit_state, wm4),
 303                  brw->wm.sampler_bo);
 304    }
 305
 306
 307    if (brw_search_cache(&brw->cache, BRW_WM_UNIT,
 308                         &key, sizeof(key),
 309                         reloc, nr_reloc,
 310                         NULL,
 311                         &brw->wm.state_bo))
 312       return PIPE_OK;
 313
 314    ret = wm_unit_create_from_key(brw, &key,
 315                                  reloc, nr_reloc,
 316                                  &brw->wm.state_bo);
 317    if (ret)
 318       return ret;
 319
 320    return PIPE_OK;
 321 }
 322
 323 const struct brw_tracked_state brw_wm_unit = {
 324    .dirty = {
 325       .mesa = (PIPE_NEW_FRAGMENT_SHADER |
 326                PIPE_NEW_DEPTH_BUFFER |
 327                PIPE_NEW_RAST |
 328                PIPE_NEW_DEPTH_STENCIL_ALPHA |
 329                PIPE_NEW_QUERY),
 330
 331       .brw = (BRW_NEW_CURBE_OFFSETS |
 332               BRW_NEW_NR_WM_SURFACES),
 333
 334       .cache = (CACHE_NEW_WM_PROG |
 335                 CACHE_NEW_SAMPLER)
 336    },
 337    .prepare = upload_wm_unit,
 338 };
 339