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, sizeof(sfv),
  97                                   NULL, 0);
  98 }
  99
 100 const struct brw_tracked_state brw_sf_vp = {
 101    .dirty = {
 102       .mesa  = (_NEW_VIEWPORT |
 103                 _NEW_SCISSOR |
 104                 _NEW_BUFFERS),
 105       .brw   = 0,
 106       .cache = 0
 107    },
 108    .prepare = upload_sf_vp
 109 };
 110
 111 struct brw_sf_unit_key {
 112    unsigned int total_grf;
 113    unsigned int urb_entry_read_length;
 114
 115    unsigned int nr_urb_entries, urb_size, sfsize;
 116
 117    GLenum front_face, cull_face;
 118    unsigned pv_first:1;
 119    unsigned scissor:1;
 120    unsigned line_smooth:1;
 121    unsigned point_sprite:1;
 122    unsigned point_attenuated:1;
 123    unsigned render_to_fbo:1;
 124    float line_width;
 125    float point_size;
 126 };
 127
 128 static void
 129 sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key)
 130 {
 131    GLcontext *ctx = &brw->intel.ctx;
 132    memset(key, 0, sizeof(*key));
 133
 134    /* CACHE_NEW_SF_PROG */
 135    key->total_grf = brw->sf.prog_data->total_grf;
 136    key->urb_entry_read_length = brw->sf.prog_data->urb_read_length;
 137
 138    /* BRW_NEW_URB_FENCE */
 139    key->nr_urb_entries = brw->urb.nr_sf_entries;
 140    key->urb_size = brw->urb.vsize;
 141    key->sfsize = brw->urb.sfsize;
 142
 143    key->scissor = ctx->Scissor.Enabled;
 144    key->front_face = ctx->Polygon.FrontFace;
 145
 146    if (ctx->Polygon.CullFlag)
 147       key->cull_face = ctx->Polygon.CullFaceMode;
 148    else
 149       key->cull_face = GL_NONE;
 150
 151    key->line_width = ctx->Line.Width;
 152    key->line_smooth = ctx->Line.SmoothFlag;
 153
 154    key->point_sprite = ctx->Point.PointSprite;
 155    key->point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 156    key->point_attenuated = ctx->Point._Attenuated;
 157
 158    /* _NEW_LIGHT */
 159    key->pv_first = (ctx->Light.ProvokingVertex == GL_FIRST_VERTEX_CONVENTION);
 160
 161    key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 162 }
 163
 164 static dri_bo *
 165 sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
 166                         dri_bo **reloc_bufs)
 167 {
 168    struct intel_context *intel = &brw->intel;
 169    struct brw_sf_unit_state sf;
 170    dri_bo *bo;
 171    int chipset_max_threads;
 172    memset(&sf, 0, sizeof(sf));
 173
 174    sf.thread0.grf_reg_count = ALIGN(key->total_grf, 16) / 16 - 1;
 175    sf.thread0.kernel_start_pointer = brw->sf.prog_bo->offset >> 6; /* reloc */
 176
 177    sf.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
 178
 179    sf.thread3.dispatch_grf_start_reg = 3;
 180
 181    if (intel->is_ironlake)
 182        sf.thread3.urb_entry_read_offset = 3;
 183    else
 184        sf.thread3.urb_entry_read_offset = 1;
 185
 186    sf.thread3.urb_entry_read_length = key->urb_entry_read_length;
 187
 188    sf.thread4.nr_urb_entries = key->nr_urb_entries;
 189    sf.thread4.urb_entry_allocation_size = key->sfsize - 1;
 190
 191    /* Each SF thread produces 1 PUE, and there can be up to 24 (Pre-Ironlake) or
 192     * 48 (Ironlake) threads.
 193     */
 194    if (intel->is_ironlake)
 195       chipset_max_threads = 48;
 196    else
 197       chipset_max_threads = 24;
 198
 199    sf.thread4.max_threads = MIN2(chipset_max_threads, key->nr_urb_entries) - 1;
 200
 201    if (INTEL_DEBUG & DEBUG_SINGLE_THREAD)
 202       sf.thread4.max_threads = 0;
 203
 204    if (INTEL_DEBUG & DEBUG_STATS)
 205       sf.thread4.stats_enable = 1;
 206
 207    /* CACHE_NEW_SF_VP */
 208    sf.sf5.sf_viewport_state_offset = brw->sf.vp_bo->offset >> 5; /* reloc */
 209
 210    sf.sf5.viewport_transform = 1;
 211
 212    /* _NEW_SCISSOR */
 213    if (key->scissor)
 214       sf.sf6.scissor = 1;
 215
 216    /* _NEW_POLYGON */
 217    if (key->front_face == GL_CCW)
 218       sf.sf5.front_winding = BRW_FRONTWINDING_CCW;
 219    else
 220       sf.sf5.front_winding = BRW_FRONTWINDING_CW;
 221
 222    /* The viewport is inverted for rendering to a FBO, and that inverts
 223     * polygon front/back orientation.
 224     */
 225    sf.sf5.front_winding ^= key->render_to_fbo;
 226
 227    switch (key->cull_face) {
 228    case GL_FRONT:
 229       sf.sf6.cull_mode = BRW_CULLMODE_FRONT;
 230       break;
 231    case GL_BACK:
 232       sf.sf6.cull_mode = BRW_CULLMODE_BACK;
 233       break;
 234    case GL_FRONT_AND_BACK:
 235       sf.sf6.cull_mode = BRW_CULLMODE_BOTH;
 236       break;
 237    case GL_NONE:
 238       sf.sf6.cull_mode = BRW_CULLMODE_NONE;
 239       break;
 240    default:
 241       assert(0);
 242       break;
 243    }
 244
 245    /* _NEW_LINE */
 246    /* XXX use ctx->Const.Min/MaxLineWidth here */
 247    sf.sf6.line_width = CLAMP(key->line_width, 1.0, 5.0) * (1<<1);
 248
 249    sf.sf6.line_endcap_aa_region_width = 1;
 250    if (key->line_smooth)
 251       sf.sf6.aa_enable = 1;
 252    else if (sf.sf6.line_width <= 0x2)
 253        sf.sf6.line_width = 0;
 254
 255    /* _NEW_BUFFERS */
 256    key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 257    if (!key->render_to_fbo) {
 258       /* Rendering to an OpenGL window */
 259       sf.sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT;
 260    }
 261    else {
 262       /* If rendering to an FBO, the pixel coordinate system is
 263        * inverted with respect to the normal OpenGL coordinate
 264        * system, so BRW_RASTRULE_LOWER_RIGHT is correct.
 265        * But this value is listed as "Reserved, but not seen as useful"
 266        * in Intel documentation (page 212, "Point Rasterization Rule",
 267        * section 7.4 "SF Pipeline State Summary", of document
 268        * "Intel® 965 Express Chipset Family and Intel® G35 Express
 269        * Chipset Graphics Controller Programmer's Reference Manual,
 270        * Volume 2: 3D/Media", Revision 1.0b as of January 2008,
 271        * available at
 272        *     http://intellinuxgraphics.org/documentation.html
 273        * at the time of this writing).
 274        *
 275        * It does work on at least some devices, if not all;
 276        * if devices that don't support it can be identified,
 277        * the likely failure case is that points are rasterized
 278        * incorrectly, which is no worse than occurs without
 279        * the value, so we're using it here.
 280        */
 281       sf.sf6.point_rast_rule = BRW_RASTRULE_LOWER_RIGHT;
 282    }
 283    /* XXX clamp max depends on AA vs. non-AA */
 284
 285    /* _NEW_POINT */
 286    sf.sf7.sprite_point = key->point_sprite;
 287    sf.sf7.point_size = CLAMP(rint(key->point_size), 1, 255) * (1<<3);
 288    sf.sf7.use_point_size_state = !key->point_attenuated;
 289    sf.sf7.aa_line_distance_mode = 0;
 290
 291    /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
 292     */
 293    if (!key->pv_first) {
 294       sf.sf7.trifan_pv = 2;
 295       sf.sf7.linestrip_pv = 1;
 296       sf.sf7.tristrip_pv = 2;
 297    } else {
 298       sf.sf7.trifan_pv = 1;
 299       sf.sf7.linestrip_pv = 0;
 300       sf.sf7.tristrip_pv = 0;
 301    }
 302    sf.sf7.line_last_pixel_enable = 0;
 303
 304    /* Set bias for OpenGL rasterization rules:
 305     */
 306    sf.sf6.dest_org_vbias = 0x8;
 307    sf.sf6.dest_org_hbias = 0x8;
 308
 309    bo = brw_upload_cache(&brw->cache, BRW_SF_UNIT,
 310                          key, sizeof(*key),
 311                          reloc_bufs, 2,
 312                          &sf, sizeof(sf));
 313
 314    /* STATE_PREFETCH command description describes this state as being
 315     * something loaded through the GPE (L2 ISC), so it's INSTRUCTION domain.
 316     */
 317    /* Emit SF program relocation */
 318    dri_bo_emit_reloc(bo,
 319                      I915_GEM_DOMAIN_INSTRUCTION, 0,
 320                      sf.thread0.grf_reg_count << 1,
 321                      offsetof(struct brw_sf_unit_state, thread0),
 322                      brw->sf.prog_bo);
 323
 324    /* Emit SF viewport relocation */
 325    dri_bo_emit_reloc(bo,
 326                      I915_GEM_DOMAIN_INSTRUCTION, 0,
 327                      sf.sf5.front_winding | (sf.sf5.viewport_transform << 1),
 328                      offsetof(struct brw_sf_unit_state, sf5),
 329                      brw->sf.vp_bo);
 330
 331    return bo;
 332 }
 333
 334 static void upload_sf_unit( struct brw_context *brw )
 335 {
 336    struct brw_sf_unit_key key;
 337    dri_bo *reloc_bufs[2];
 338
 339    sf_unit_populate_key(brw, &key);
 340
 341    reloc_bufs[0] = brw->sf.prog_bo;
 342    reloc_bufs[1] = brw->sf.vp_bo;
 343
 344    dri_bo_unreference(brw->sf.state_bo);
 345    brw->sf.state_bo = brw_search_cache(&brw->cache, BRW_SF_UNIT,
 346                                        &key, sizeof(key),
 347                                        reloc_bufs, 2,
 348                                        NULL);
 349    if (brw->sf.state_bo == NULL) {
 350       brw->sf.state_bo = sf_unit_create_from_key(brw, &key, reloc_bufs);
 351    }
 352 }
 353
 354 const struct brw_tracked_state brw_sf_unit = {
 355    .dirty = {
 356       .mesa  = (_NEW_POLYGON |
 357                 _NEW_LIGHT |
 358                 _NEW_LINE |
 359                 _NEW_POINT |
 360                 _NEW_SCISSOR |
 361                 _NEW_BUFFERS),
 362       .brw   = BRW_NEW_URB_FENCE,
 363       .cache = (CACHE_NEW_SF_VP |
 364                 CACHE_NEW_SF_PROG)
 365    },
 366    .prepare = upload_sf_unit,
 367 };