src/mesa/drivers/dri/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
  33
  34 #include "brw_context.h"
  35 #include "brw_state.h"
  36 #include "brw_defines.h"
  37 #include "main/macros.h"
  38 #include "intel_fbo.h"
  39
  40 static void upload_sf_vp(struct brw_context *brw)
  41 {
  42    GLcontext *ctx = &brw->intel.ctx;
  43    const GLfloat depth_scale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
  44    struct brw_sf_viewport sfv;
  45    GLfloat y_scale, y_bias;
  46    const GLboolean render_to_fbo = (ctx->DrawBuffer->Name != 0);
  47    const GLfloat *v = ctx->Viewport._WindowMap.m;
  48
  49    memset(&sfv, 0, sizeof(sfv));
  50
  51    if (render_to_fbo) {
  52       y_scale = 1.0;
  53       y_bias = 0;
  54    }
  55    else {
  56       y_scale = -1.0;
  57       y_bias = ctx->DrawBuffer->Height;
  58    }
  59
  60    /* _NEW_VIEWPORT */
  61
  62    sfv.viewport.m00 = v[MAT_SX];
  63    sfv.viewport.m11 = v[MAT_SY] * y_scale;
  64    sfv.viewport.m22 = v[MAT_SZ] * depth_scale;
  65    sfv.viewport.m30 = v[MAT_TX];
  66    sfv.viewport.m31 = v[MAT_TY] * y_scale + y_bias;
  67    sfv.viewport.m32 = v[MAT_TZ] * depth_scale;
  68
  69    /* _NEW_SCISSOR | _NEW_BUFFERS | _NEW_VIEWPORT
  70     * for DrawBuffer->_[XY]{min,max}
  71     */
  72
  73    /* The scissor only needs to handle the intersection of drawable and
  74     * scissor rect.  Clipping to the boundaries of static shared buffers
  75     * for front/back/depth is covered by looping over cliprects in brw_draw.c.
  76     *
  77     * Note that the hardware's coordinates are inclusive, while Mesa's min is
  78     * inclusive but max is exclusive.
  79     */
  80    if (render_to_fbo) {
  81       /* texmemory: Y=0=bottom */
  82       sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
  83       sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
  84       sfv.scissor.ymin = ctx->DrawBuffer->_Ymin;
  85       sfv.scissor.ymax = ctx->DrawBuffer->_Ymax - 1;
  86    }
  87    else {
  88       /* memory: Y=0=top */
  89       sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
  90       sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
  91       sfv.scissor.ymin = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymax;
  92       sfv.scissor.ymax = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymin - 1;
  93    }
  94
  95    dri_bo_unreference(brw->sf.vp_bo);
  96    brw->sf.vp_bo = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0 );
  97 }
  98
  99 const struct brw_tracked_state brw_sf_vp = {
 100    .dirty = {
 101       .mesa  = (_NEW_VIEWPORT |
 102                 _NEW_SCISSOR |
 103                 _NEW_BUFFERS),
 104       .brw   = 0,
 105       .cache = 0
 106    },
 107    .prepare = upload_sf_vp
 108 };
 109
 110 struct brw_sf_unit_key {
 111    unsigned int total_grf;
 112    unsigned int urb_entry_read_length;
 113
 114    unsigned int nr_urb_entries, urb_size, sfsize;
 115
 116    GLenum front_face, cull_face, provoking_vertex;
 117    unsigned scissor:1;
 118    unsigned line_smooth:1;
 119    unsigned point_sprite:1;
 120    unsigned point_attenuated:1;
 121    unsigned render_to_fbo:1;
 122    float line_width;
 123    float point_size;
 124 };
 125
 126 static void
 127 sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key)
 128 {
 129    GLcontext *ctx = &brw->intel.ctx;
 130    memset(key, 0, sizeof(*key));
 131
 132    /* CACHE_NEW_SF_PROG */
 133    key->total_grf = brw->sf.prog_data->total_grf;
 134    key->urb_entry_read_length = brw->sf.prog_data->urb_read_length;
 135
 136    /* BRW_NEW_URB_FENCE */
 137    key->nr_urb_entries = brw->urb.nr_sf_entries;
 138    key->urb_size = brw->urb.vsize;
 139    key->sfsize = brw->urb.sfsize;
 140
 141    key->scissor = ctx->Scissor.Enabled;
 142    key->front_face = ctx->Polygon.FrontFace;
 143
 144    if (ctx->Polygon.CullFlag)
 145       key->cull_face = ctx->Polygon.CullFaceMode;
 146    else
 147       key->cull_face = GL_NONE;
 148
 149    key->line_width = ctx->Line.Width;
 150    key->line_smooth = ctx->Line.SmoothFlag;
 151
 152    key->point_sprite = ctx->Point.PointSprite;
 153    key->point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 154    key->point_attenuated = ctx->Point._Attenuated;
 155
 156    /* _NEW_LIGHT */
 157    key->provoking_vertex = ctx->Light.ProvokingVertex;
 158
 159    key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 160 }
 161
 162 static dri_bo *
 163 sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
 164                         dri_bo **reloc_bufs)
 165 {
 166    struct brw_sf_unit_state sf;
 167    dri_bo *bo;
 168    int chipset_max_threads;
 169    memset(&sf, 0, sizeof(sf));
 170
 171    sf.thread0.grf_reg_count = ALIGN(key->total_grf, 16) / 16 - 1;
 172    sf.thread0.kernel_start_pointer = brw->sf.prog_bo->offset >> 6; /* reloc */
 173
 174    sf.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
 175
 176    sf.thread3.dispatch_grf_start_reg = 3;
 177
 178    if (BRW_IS_IGDNG(brw))
 179        sf.thread3.urb_entry_read_offset = 3;
 180    else
 181        sf.thread3.urb_entry_read_offset = 1;
 182
 183    sf.thread3.urb_entry_read_length = key->urb_entry_read_length;
 184
 185    sf.thread4.nr_urb_entries = key->nr_urb_entries;
 186    sf.thread4.urb_entry_allocation_size = key->sfsize - 1;
 187
 188    /* Each SF thread produces 1 PUE, and there can be up to 24(Pre-IGDNG) or
 189     * 48(IGDNG) threads
 190     */
 191    if (BRW_IS_IGDNG(brw))
 192       chipset_max_threads = 48;
 193    else
 194       chipset_max_threads = 24;
 195
 196    sf.thread4.max_threads = MIN2(chipset_max_threads, key->nr_urb_entries) - 1;
 197
 198    if (INTEL_DEBUG & DEBUG_SINGLE_THREAD)
 199       sf.thread4.max_threads = 0;
 200
 201    if (INTEL_DEBUG & DEBUG_STATS)
 202       sf.thread4.stats_enable = 1;
 203
 204    /* CACHE_NEW_SF_VP */
 205    sf.sf5.sf_viewport_state_offset = brw->sf.vp_bo->offset >> 5; /* reloc */
 206
 207    sf.sf5.viewport_transform = 1;
 208
 209    /* _NEW_SCISSOR */
 210    if (key->scissor)
 211       sf.sf6.scissor = 1;
 212
 213    /* _NEW_POLYGON */
 214    if (key->front_face == GL_CCW)
 215       sf.sf5.front_winding = BRW_FRONTWINDING_CCW;
 216    else
 217       sf.sf5.front_winding = BRW_FRONTWINDING_CW;
 218
 219    /* The viewport is inverted for rendering to a FBO, and that inverts
 220     * polygon front/back orientation.
 221     */
 222    sf.sf5.front_winding ^= key->render_to_fbo;
 223
 224    switch (key->cull_face) {
 225    case GL_FRONT:
 226       sf.sf6.cull_mode = BRW_CULLMODE_FRONT;
 227       break;
 228    case GL_BACK:
 229       sf.sf6.cull_mode = BRW_CULLMODE_BACK;
 230       break;
 231    case GL_FRONT_AND_BACK:
 232       sf.sf6.cull_mode = BRW_CULLMODE_BOTH;
 233       break;
 234    case GL_NONE:
 235       sf.sf6.cull_mode = BRW_CULLMODE_NONE;
 236       break;
 237    default:
 238       assert(0);
 239       break;
 240    }
 241
 242    /* _NEW_LINE */
 243    /* XXX use ctx->Const.Min/MaxLineWidth here */
 244    sf.sf6.line_width = CLAMP(key->line_width, 1.0, 5.0) * (1<<1);
 245
 246    sf.sf6.line_endcap_aa_region_width = 1;
 247    if (key->line_smooth)
 248       sf.sf6.aa_enable = 1;
 249    else if (sf.sf6.line_width <= 0x2)
 250        sf.sf6.line_width = 0;
 251
 252    /* _NEW_BUFFERS */
 253    key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 254    if (!key->render_to_fbo) {
 255       /* Rendering to an OpenGL window */
 256       sf.sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT;
 257    }
 258    else {
 259       /* If rendering to an FBO, the pixel coordinate system is
 260        * inverted with respect to the normal OpenGL coordinate
 261        * system, so BRW_RASTRULE_LOWER_RIGHT is correct.
 262        * But this value is listed as "Reserved, but not seen as useful"
 263        * in Intel documentation (page 212, "Point Rasterization Rule",
 264        * section 7.4 "SF Pipeline State Summary", of document
 265        * "Intel® 965 Express Chipset Family and Intel® G35 Express
 266        * Chipset Graphics Controller Programmer's Reference Manual,
 267        * Volume 2: 3D/Media", Revision 1.0b as of January 2008,
 268        * available at
 269        *     http://intellinuxgraphics.org/documentation.html
 270        * at the time of this writing).
 271        *
 272        * It does work on at least some devices, if not all;
 273        * if devices that don't support it can be identified,
 274        * the likely failure case is that points are rasterized
 275        * incorrectly, which is no worse than occurs without
 276        * the value, so we're using it here.
 277        */
 278       sf.sf6.point_rast_rule = BRW_RASTRULE_LOWER_RIGHT;
 279    }
 280    /* XXX clamp max depends on AA vs. non-AA */
 281
 282    /* _NEW_POINT */
 283    sf.sf7.sprite_point = key->point_sprite;
 284    sf.sf7.point_size = CLAMP(rint(key->point_size), 1, 255) * (1<<3);
 285    sf.sf7.use_point_size_state = !key->point_attenuated;
 286    sf.sf7.aa_line_distance_mode = 0;
 287
 288    /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
 289     */
 290    if (key->provoking_vertex == GL_LAST_VERTEX_CONVENTION) {
 291       sf.sf7.trifan_pv = 2;
 292       sf.sf7.linestrip_pv = 1;
 293       sf.sf7.tristrip_pv = 2;
 294    } else {
 295       sf.sf7.trifan_pv = 1;
 296       sf.sf7.linestrip_pv = 0;
 297       sf.sf7.tristrip_pv = 0;
 298    }
 299    sf.sf7.line_last_pixel_enable = 0;
 300
 301    /* Set bias for OpenGL rasterization rules:
 302     */
 303    sf.sf6.dest_org_vbias = 0x8;
 304    sf.sf6.dest_org_hbias = 0x8;
 305
 306    bo = brw_upload_cache(&brw->cache, BRW_SF_UNIT,
 307                          key, sizeof(*key),
 308                          reloc_bufs, 2,
 309                          &sf, sizeof(sf),
 310                          NULL, NULL);
 311
 312    /* STATE_PREFETCH command description describes this state as being
 313     * something loaded through the GPE (L2 ISC), so it's INSTRUCTION domain.
 314     */
 315    /* Emit SF program relocation */
 316    dri_bo_emit_reloc(bo,
 317                      I915_GEM_DOMAIN_INSTRUCTION, 0,
 318                      sf.thread0.grf_reg_count << 1,
 319                      offsetof(struct brw_sf_unit_state, thread0),
 320                      brw->sf.prog_bo);
 321
 322    /* Emit SF viewport relocation */
 323    dri_bo_emit_reloc(bo,
 324                      I915_GEM_DOMAIN_INSTRUCTION, 0,
 325                      sf.sf5.front_winding | (sf.sf5.viewport_transform << 1),
 326                      offsetof(struct brw_sf_unit_state, sf5),
 327                      brw->sf.vp_bo);
 328
 329    return bo;
 330 }
 331
 332 static void upload_sf_unit( struct brw_context *brw )
 333 {
 334    struct brw_sf_unit_key key;
 335    dri_bo *reloc_bufs[2];
 336
 337    sf_unit_populate_key(brw, &key);
 338
 339    reloc_bufs[0] = brw->sf.prog_bo;
 340    reloc_bufs[1] = brw->sf.vp_bo;
 341
 342    dri_bo_unreference(brw->sf.state_bo);
 343    brw->sf.state_bo = brw_search_cache(&brw->cache, BRW_SF_UNIT,
 344                                        &key, sizeof(key),
 345                                        reloc_bufs, 2,
 346                                        NULL);
 347    if (brw->sf.state_bo == NULL) {
 348       brw->sf.state_bo = sf_unit_create_from_key(brw, &key, reloc_bufs);
 349    }
 350 }
 351
 352 const struct brw_tracked_state brw_sf_unit = {
 353    .dirty = {
 354       .mesa  = (_NEW_POLYGON |
 355                 _NEW_LIGHT |
 356                 _NEW_LINE |
 357                 _NEW_POINT |
 358                 _NEW_SCISSOR |
 359                 _NEW_BUFFERS),
 360       .brw   = BRW_NEW_URB_FENCE,
 361       .cache = (CACHE_NEW_SF_VP |
 362                 CACHE_NEW_SF_PROG)
 363    },
 364    .prepare = upload_sf_unit,
 365 };