src/mesa/drivers/dri/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
  33
  34 #include "brw_context.h"
  35 #include "brw_state.h"
  36 #include "brw_defines.h"
  37 #include "brw_wm.h"
  38
  39 /***********************************************************************
  40  * WM unit - fragment programs and rasterization
  41  */
  42
  43 struct brw_wm_unit_key {
  44    unsigned int total_grf, total_scratch;
  45    unsigned int urb_entry_read_length;
  46    unsigned int curb_entry_read_length;
  47    unsigned int dispatch_grf_start_reg;
  48
  49    unsigned int curbe_offset;
  50    unsigned int urb_size;
  51
  52    unsigned int max_threads;
  53
  54    unsigned int nr_surfaces, sampler_count;
  55    GLboolean uses_depth, computes_depth, uses_kill, is_glsl;
  56    GLboolean polygon_stipple, stats_wm, line_stipple, offset_enable;
  57    GLfloat offset_units, offset_factor;
  58 };
  59
  60 static void
  61 wm_unit_populate_key(struct brw_context *brw, struct brw_wm_unit_key *key)
  62 {
  63    const struct gl_fragment_program *fp = brw->fragment_program;
  64    struct intel_context *intel = &brw->intel;
  65
  66    memset(key, 0, sizeof(*key));
  67
  68    if (INTEL_DEBUG & DEBUG_SINGLE_THREAD)
  69       key->max_threads = 1;
  70    else {
  71       /* WM maximum threads is number of EUs times number of threads per EU. */
  72       if (BRW_IS_G4X(brw))
  73          key->max_threads = 10 * 5;
  74       else
  75          key->max_threads = 8 * 4;
  76    }
  77
  78    /* CACHE_NEW_WM_PROG */
  79    key->total_grf = brw->wm.prog_data->total_grf;
  80    key->urb_entry_read_length = brw->wm.prog_data->urb_read_length;
  81    key->curb_entry_read_length = brw->wm.prog_data->curb_read_length;
  82    key->dispatch_grf_start_reg = brw->wm.prog_data->first_curbe_grf;
  83    key->total_scratch = ALIGN(brw->wm.prog_data->total_scratch, 1024);
  84
  85    /* BRW_NEW_URB_FENCE */
  86    key->urb_size = brw->urb.vsize;
  87
  88    /* BRW_NEW_CURBE_OFFSETS */
  89    key->curbe_offset = brw->curbe.wm_start;
  90
  91    /* CACHE_NEW_SURFACE */
  92    key->nr_surfaces = brw->wm.nr_surfaces;
  93
  94    /* CACHE_NEW_SAMPLER */
  95    key->sampler_count = brw->wm.sampler_count;
  96
  97    /* _NEW_POLYGONSTIPPLE */
  98    key->polygon_stipple = brw->attribs.Polygon->StippleFlag;
  99
 100    /* BRW_NEW_FRAGMENT_PROGRAM */
 101    key->uses_depth = (fp->Base.InputsRead & (1 << FRAG_ATTRIB_WPOS)) != 0;
 102
 103    /* as far as we can tell */
 104    key->computes_depth =
 105       (fp->Base.OutputsWritten & (1 << FRAG_RESULT_DEPR)) != 0;
 106
 107    /* _NEW_COLOR */
 108    key->uses_kill = fp->UsesKill || brw->attribs.Color->AlphaEnabled;
 109    key->is_glsl = brw_wm_is_glsl(fp);
 110
 111    /* XXX: This needs a flag to indicate when it changes. */
 112    key->stats_wm = intel->stats_wm;
 113
 114    /* _NEW_LINE */
 115    key->line_stipple = brw->attribs.Line->StippleFlag;
 116
 117    /* _NEW_POLYGON */
 118    key->offset_enable = brw->attribs.Polygon->OffsetFill;
 119    key->offset_units = brw->attribs.Polygon->OffsetUnits;
 120    key->offset_factor = brw->attribs.Polygon->OffsetFactor;
 121 }
 122
 123 static dri_bo *
 124 wm_unit_create_from_key(struct brw_context *brw, struct brw_wm_unit_key *key,
 125                         dri_bo **reloc_bufs)
 126 {
 127    struct brw_wm_unit_state wm;
 128    dri_bo *bo;
 129
 130    memset(&wm, 0, sizeof(wm));
 131
 132    wm.thread0.grf_reg_count = ALIGN(key->total_grf, 16) / 16 - 1;
 133    wm.thread0.kernel_start_pointer = brw->wm.prog_bo->offset >> 6; /* reloc */
 134    wm.thread1.depth_coef_urb_read_offset = 1;
 135    wm.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
 136    wm.thread1.binding_table_entry_count = key->nr_surfaces;
 137
 138    if (key->total_scratch != 0) {
 139       wm.thread2.scratch_space_base_pointer =
 140          brw->wm.scratch_buffer->offset >> 10; /* reloc */
 141       wm.thread2.per_thread_scratch_space = key->total_scratch / 1024 - 1;
 142    } else {
 143       wm.thread2.scratch_space_base_pointer = 0;
 144       wm.thread2.per_thread_scratch_space = 0;
 145    }
 146
 147    wm.thread3.dispatch_grf_start_reg = key->dispatch_grf_start_reg;
 148    wm.thread3.urb_entry_read_length = key->urb_entry_read_length;
 149    wm.thread3.const_urb_entry_read_length = key->curb_entry_read_length;
 150    wm.thread3.const_urb_entry_read_offset = key->curbe_offset * 2;
 151    wm.thread3.urb_entry_read_offset = 0;
 152
 153    wm.wm4.sampler_count = (key->sampler_count + 1) / 4;
 154    if (brw->wm.sampler_bo != NULL) {
 155       /* reloc */
 156       wm.wm4.sampler_state_pointer = brw->wm.sampler_bo->offset >> 5;
 157    } else {
 158       wm.wm4.sampler_state_pointer = 0;
 159    }
 160
 161    wm.wm5.program_uses_depth = key->uses_depth;
 162    wm.wm5.program_computes_depth = key->computes_depth;
 163    wm.wm5.program_uses_killpixel = key->uses_kill;
 164
 165    if (key->is_glsl)
 166       wm.wm5.enable_8_pix = 1;
 167    else
 168       wm.wm5.enable_16_pix = 1;
 169
 170    wm.wm5.max_threads = key->max_threads - 1;
 171    wm.wm5.thread_dispatch_enable = 1;   /* AKA: color_write */
 172    wm.wm5.legacy_line_rast = 0;
 173    wm.wm5.legacy_global_depth_bias = 0;
 174    wm.wm5.early_depth_test = 1;         /* never need to disable */
 175    wm.wm5.line_aa_region_width = 0;
 176    wm.wm5.line_endcap_aa_region_width = 1;
 177
 178    wm.wm5.polygon_stipple = key->polygon_stipple;
 179
 180    if (key->offset_enable) {
 181       wm.wm5.depth_offset = 1;
 182       /* Something wierd going on with legacy_global_depth_bias,
 183        * offset_constant, scaling and MRD.  This value passes glean
 184        * but gives some odd results elsewere (eg. the
 185        * quad-offset-units test).
 186        */
 187       wm.global_depth_offset_constant = key->offset_units * 2;
 188
 189       /* This is the only value that passes glean:
 190        */
 191       wm.global_depth_offset_scale = key->offset_factor;
 192    }
 193
 194    wm.wm5.line_stipple = key->line_stipple;
 195
 196    if (INTEL_DEBUG & DEBUG_STATS || key->stats_wm)
 197       wm.wm4.stats_enable = 1;
 198
 199    bo = brw_upload_cache(&brw->cache, BRW_WM_UNIT,
 200                          key, sizeof(*key),
 201                          reloc_bufs, 3,
 202                          &wm, sizeof(wm),
 203                          NULL, NULL);
 204
 205    /* Emit WM program relocation */
 206    dri_bo_emit_reloc(bo,
 207                      I915_GEM_DOMAIN_INSTRUCTION, 0,
 208                      wm.thread0.grf_reg_count << 1,
 209                      offsetof(struct brw_wm_unit_state, thread0),
 210                      brw->wm.prog_bo);
 211
 212    /* Emit scratch space relocation */
 213    if (key->total_scratch != 0) {
 214       dri_bo_emit_reloc(bo,
 215                         0, 0,
 216                         wm.thread2.per_thread_scratch_space,
 217                         offsetof(struct brw_wm_unit_state, thread2),
 218                         brw->wm.scratch_buffer);
 219    }
 220
 221    /* Emit sampler state relocation */
 222    if (key->sampler_count != 0) {
 223       dri_bo_emit_reloc(bo,
 224                         I915_GEM_DOMAIN_INSTRUCTION, 0,
 225                         wm.wm4.stats_enable | (wm.wm4.sampler_count << 2),
 226                         offsetof(struct brw_wm_unit_state, wm4),
 227                         brw->wm.sampler_bo);
 228    }
 229
 230    return bo;
 231 }
 232
 233
 234 static void upload_wm_unit( struct brw_context *brw )
 235 {
 236    struct intel_context *intel = &brw->intel;
 237    struct brw_wm_unit_key key;
 238    dri_bo *reloc_bufs[3];
 239    wm_unit_populate_key(brw, &key);
 240
 241    /* Allocate the necessary scratch space if we haven't already.  Don't
 242     * bother reducing the allocation later, since we use scratch so
 243     * rarely.
 244     */
 245    assert(key.total_scratch <= 12 * 1024);
 246    if (key.total_scratch) {
 247       GLuint total = key.total_scratch * key.max_threads;
 248
 249       if (brw->wm.scratch_buffer && total > brw->wm.scratch_buffer->size) {
 250          dri_bo_unreference(brw->wm.scratch_buffer);
 251          brw->wm.scratch_buffer = NULL;
 252       }
 253       if (brw->wm.scratch_buffer == NULL) {
 254          brw->wm.scratch_buffer = dri_bo_alloc(intel->bufmgr,
 255                                                "wm scratch",
 256                                                total,
 257                                                4096);
 258       }
 259    }
 260
 261    reloc_bufs[0] = brw->wm.prog_bo;
 262    reloc_bufs[1] = brw->wm.scratch_buffer;
 263    reloc_bufs[2] = brw->wm.sampler_bo;
 264
 265    dri_bo_unreference(brw->wm.state_bo);
 266    brw->wm.state_bo = brw_search_cache(&brw->cache, BRW_WM_UNIT,
 267                                        &key, sizeof(key),
 268                                        reloc_bufs, 3,
 269                                        NULL);
 270    if (brw->wm.state_bo == NULL) {
 271       brw->wm.state_bo = wm_unit_create_from_key(brw, &key, reloc_bufs);
 272    }
 273 }
 274
 275 const struct brw_tracked_state brw_wm_unit = {
 276    .dirty = {
 277       .mesa = (_NEW_POLYGON |
 278                _NEW_POLYGONSTIPPLE |
 279                _NEW_LINE |
 280                _NEW_COLOR),
 281
 282       .brw = (BRW_NEW_FRAGMENT_PROGRAM |
 283               BRW_NEW_CURBE_OFFSETS |
 284               BRW_NEW_LOCK),
 285
 286       .cache = (CACHE_NEW_SURFACE |
 287                 CACHE_NEW_WM_PROG |
 288                 CACHE_NEW_SAMPLER)
 289    },
 290    .prepare = upload_wm_unit,
 291 };
 292