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