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 */
  70
  71    /* The scissor only needs to handle the intersection of drawable and
  72     * scissor rect.  Clipping to the boundaries of static shared buffers
  73     * for front/back/depth is covered by looping over cliprects in brw_draw.c.
  74     *
  75     * Note that the hardware's coordinates are inclusive, while Mesa's min is
  76     * inclusive but max is exclusive.
  77     */
  78    if (render_to_fbo) {
  79       /* texmemory: Y=0=bottom */
  80       sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
  81       sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
  82       sfv.scissor.ymin = ctx->DrawBuffer->_Ymin;
  83       sfv.scissor.ymax = ctx->DrawBuffer->_Ymax - 1;
  84    }
  85    else {
  86       /* memory: Y=0=top */
  87       sfv.scissor.xmin = ctx->DrawBuffer->_Xmin;
  88       sfv.scissor.xmax = ctx->DrawBuffer->_Xmax - 1;
  89       sfv.scissor.ymin = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymax;
  90       sfv.scissor.ymax = ctx->DrawBuffer->Height - ctx->DrawBuffer->_Ymin - 1;
  91    }
  92
  93    dri_bo_unreference(brw->sf.vp_bo);
  94    brw->sf.vp_bo = brw_cache_data( &brw->cache, BRW_SF_VP, &sfv, NULL, 0 );
  95 }
  96
  97 const struct brw_tracked_state brw_sf_vp = {
  98    .dirty = {
  99       .mesa  = (_NEW_VIEWPORT |
 100                 _NEW_SCISSOR),
 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;
 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    GLcontext *ctx = &brw->intel.ctx;
 127    memset(key, 0, sizeof(*key));
 128
 129    /* CACHE_NEW_SF_PROG */
 130    key->total_grf = brw->sf.prog_data->total_grf;
 131    key->urb_entry_read_length = brw->sf.prog_data->urb_read_length;
 132
 133    /* BRW_NEW_URB_FENCE */
 134    key->nr_urb_entries = brw->urb.nr_sf_entries;
 135    key->urb_size = brw->urb.vsize;
 136    key->sfsize = brw->urb.sfsize;
 137
 138    key->scissor = ctx->Scissor.Enabled;
 139    key->front_face = ctx->Polygon.FrontFace;
 140
 141    if (ctx->Polygon.CullFlag)
 142       key->cull_face = ctx->Polygon.CullFaceMode;
 143    else
 144       key->cull_face = GL_NONE;
 145
 146    key->line_width = ctx->Line.Width;
 147    key->line_smooth = ctx->Line.SmoothFlag;
 148
 149    key->point_sprite = ctx->Point.PointSprite;
 150    key->point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 151    key->point_attenuated = ctx->Point._Attenuated;
 152
 153    key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 154 }
 155
 156 static dri_bo *
 157 sf_unit_create_from_key(struct brw_context *brw, struct brw_sf_unit_key *key,
 158                         dri_bo **reloc_bufs)
 159 {
 160    struct brw_sf_unit_state sf;
 161    dri_bo *bo;
 162
 163    memset(&sf, 0, sizeof(sf));
 164
 165    sf.thread0.grf_reg_count = ALIGN(key->total_grf, 16) / 16 - 1;
 166    sf.thread0.kernel_start_pointer = brw->sf.prog_bo->offset >> 6; /* reloc */
 167
 168    sf.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
 169
 170    sf.thread3.dispatch_grf_start_reg = 3;
 171    sf.thread3.urb_entry_read_offset = 1;
 172    sf.thread3.urb_entry_read_length = key->urb_entry_read_length;
 173
 174    sf.thread4.nr_urb_entries = key->nr_urb_entries;
 175    sf.thread4.urb_entry_allocation_size = key->sfsize - 1;
 176    /* Each SF thread produces 1 PUE, and there can be up to 24 threads */
 177    sf.thread4.max_threads = MIN2(24, key->nr_urb_entries) - 1;
 178
 179    if (INTEL_DEBUG & DEBUG_SINGLE_THREAD)
 180       sf.thread4.max_threads = 0;
 181
 182    if (INTEL_DEBUG & DEBUG_STATS)
 183       sf.thread4.stats_enable = 1;
 184
 185    /* CACHE_NEW_SF_VP */
 186    sf.sf5.sf_viewport_state_offset = brw->sf.vp_bo->offset >> 5; /* reloc */
 187
 188    sf.sf5.viewport_transform = 1;
 189
 190    /* _NEW_SCISSOR */
 191    if (key->scissor)
 192       sf.sf6.scissor = 1;
 193
 194    /* _NEW_POLYGON */
 195    if (key->front_face == GL_CCW)
 196       sf.sf5.front_winding = BRW_FRONTWINDING_CCW;
 197    else
 198       sf.sf5.front_winding = BRW_FRONTWINDING_CW;
 199
 200    /* The viewport is inverted for rendering to a FBO, and that inverts
 201     * polygon front/back orientation.
 202     */
 203    sf.sf5.front_winding ^= key->render_to_fbo;
 204
 205    switch (key->cull_face) {
 206    case GL_FRONT:
 207       sf.sf6.cull_mode = BRW_CULLMODE_FRONT;
 208       break;
 209    case GL_BACK:
 210       sf.sf6.cull_mode = BRW_CULLMODE_BACK;
 211       break;
 212    case GL_FRONT_AND_BACK:
 213       sf.sf6.cull_mode = BRW_CULLMODE_BOTH;
 214       break;
 215    case GL_NONE:
 216       sf.sf6.cull_mode = BRW_CULLMODE_NONE;
 217       break;
 218    default:
 219       assert(0);
 220       break;
 221    }
 222
 223    /* _NEW_LINE */
 224    /* XXX use ctx->Const.Min/MaxLineWidth here */
 225    sf.sf6.line_width = CLAMP(key->line_width, 1.0, 5.0) * (1<<1);
 226
 227    sf.sf6.line_endcap_aa_region_width = 1;
 228    if (key->line_smooth)
 229       sf.sf6.aa_enable = 1;
 230    else if (sf.sf6.line_width <= 0x2)
 231        sf.sf6.line_width = 0;
 232
 233    /* _NEW_POINT */
 234    key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 235    if (!key->render_to_fbo) {
 236       /* Rendering to an OpenGL window */
 237       sf.sf6.point_rast_rule = BRW_RASTRULE_UPPER_RIGHT;
 238    }
 239    else {
 240       /* If rendering to an FBO, the pixel coordinate system is
 241        * inverted with respect to the normal OpenGL coordinate
 242        * system, so BRW_RASTRULE_LOWER_RIGHT is correct.
 243        * But this value is listed as "Reserved, but not seen as useful"
 244        * in Intel documentation (page 212, "Point Rasterization Rule",
 245        * section 7.4 "SF Pipeline State Summary", of document
 246        * "Intel® 965 Express Chipset Family and Intel® G35 Express
 247        * Chipset Graphics Controller Programmer's Reference Manual,
 248        * Volume 2: 3D/Media", Revision 1.0b as of January 2008,
 249        * available at
 250        *     http://intellinuxgraphics.org/documentation.html
 251        * at the time of this writing).
 252        *
 253        * It does work on at least some devices, if not all;
 254        * if devices that don't support it can be identified,
 255        * the likely failure case is that points are rasterized
 256        * incorrectly, which is no worse than occurs without
 257        * the value, so we're using it here.
 258        */
 259       sf.sf6.point_rast_rule = BRW_RASTRULE_LOWER_RIGHT;
 260    }
 261    /* XXX clamp max depends on AA vs. non-AA */
 262
 263    sf.sf7.sprite_point = key->point_sprite;
 264    sf.sf7.point_size = CLAMP(rint(key->point_size), 1, 255) * (1<<3);
 265    sf.sf7.use_point_size_state = !key->point_attenuated;
 266    sf.sf7.aa_line_distance_mode = 0;
 267
 268    /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
 269     */
 270    sf.sf7.trifan_pv = 2;
 271    sf.sf7.linestrip_pv = 1;
 272    sf.sf7.tristrip_pv = 2;
 273    sf.sf7.line_last_pixel_enable = 0;
 274
 275    /* Set bias for OpenGL rasterization rules:
 276     */
 277    sf.sf6.dest_org_vbias = 0x8;
 278    sf.sf6.dest_org_hbias = 0x8;
 279
 280    bo = brw_upload_cache(&brw->cache, BRW_SF_UNIT,
 281                          key, sizeof(*key),
 282                          reloc_bufs, 2,
 283                          &sf, sizeof(sf),
 284                          NULL, NULL);
 285
 286    /* Emit SF program relocation */
 287    dri_bo_emit_reloc(bo,
 288                      I915_GEM_DOMAIN_INSTRUCTION, 0,
 289                      sf.thread0.grf_reg_count << 1,
 290                      offsetof(struct brw_sf_unit_state, thread0),
 291                      brw->sf.prog_bo);
 292
 293    /* Emit SF viewport relocation */
 294    dri_bo_emit_reloc(bo,
 295                      I915_GEM_DOMAIN_INSTRUCTION, 0,
 296                      sf.sf5.front_winding | (sf.sf5.viewport_transform << 1),
 297                      offsetof(struct brw_sf_unit_state, sf5),
 298                      brw->sf.vp_bo);
 299
 300    return bo;
 301 }
 302
 303 static void upload_sf_unit( struct brw_context *brw )
 304 {
 305    struct brw_sf_unit_key key;
 306    dri_bo *reloc_bufs[2];
 307
 308    sf_unit_populate_key(brw, &key);
 309
 310    reloc_bufs[0] = brw->sf.prog_bo;
 311    reloc_bufs[1] = brw->sf.vp_bo;
 312
 313    dri_bo_unreference(brw->sf.state_bo);
 314    brw->sf.state_bo = brw_search_cache(&brw->cache, BRW_SF_UNIT,
 315                                        &key, sizeof(key),
 316                                        reloc_bufs, 2,
 317                                        NULL);
 318    if (brw->sf.state_bo == NULL) {
 319       brw->sf.state_bo = sf_unit_create_from_key(brw, &key, reloc_bufs);
 320    }
 321 }
 322
 323 const struct brw_tracked_state brw_sf_unit = {
 324    .dirty = {
 325       .mesa  = (_NEW_POLYGON |
 326                 _NEW_LINE |
 327                 _NEW_POINT |
 328                 _NEW_SCISSOR),
 329       .brw   = BRW_NEW_URB_FENCE,
 330       .cache = (CACHE_NEW_SF_VP |
 331                 CACHE_NEW_SF_PROG)
 332    },
 333    .prepare = upload_sf_unit,
 334 };