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