src/mesa/drivers/dri/i965/brw_wm_sampler_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 #include "brw_context.h"
  34 #include "brw_state.h"
  35 #include "brw_defines.h"
  36
  37 #include "main/macros.h"
  38
  39
  40
  41 /* Samplers aren't strictly wm state from the hardware's perspective,
  42  * but that is the only situation in which we use them in this driver.
  43  */
  44
  45
  46
  47 /* The brw (and related graphics cores) do not support GL_CLAMP.  The
  48  * Intel drivers for "other operating systems" implement GL_CLAMP as
  49  * GL_CLAMP_TO_EDGE, so the same is done here.
  50  */
  51 static GLuint translate_wrap_mode( GLenum wrap )
  52 {
  53    switch( wrap ) {
  54    case GL_REPEAT:
  55       return BRW_TEXCOORDMODE_WRAP;
  56    case GL_CLAMP:
  57       return BRW_TEXCOORDMODE_CLAMP;
  58    case GL_CLAMP_TO_EDGE:
  59       return BRW_TEXCOORDMODE_CLAMP; /* conform likes it this way */
  60    case GL_CLAMP_TO_BORDER:
  61       return BRW_TEXCOORDMODE_CLAMP_BORDER;
  62    case GL_MIRRORED_REPEAT:
  63       return BRW_TEXCOORDMODE_MIRROR;
  64    default:
  65       return BRW_TEXCOORDMODE_WRAP;
  66    }
  67 }
  68
  69 static drm_intel_bo *upload_default_color( struct brw_context *brw,
  70                                      const GLfloat *color )
  71 {
  72    struct intel_context *intel = &brw->intel;
  73
  74    if (intel->gen >= 5) {
  75       struct gen5_sampler_default_color sdc;
  76
  77       memset(&sdc, 0, sizeof(sdc));
  78
  79       UNCLAMPED_FLOAT_TO_UBYTE(sdc.ub[0], color[0]);
  80       UNCLAMPED_FLOAT_TO_UBYTE(sdc.ub[1], color[1]);
  81       UNCLAMPED_FLOAT_TO_UBYTE(sdc.ub[2], color[2]);
  82       UNCLAMPED_FLOAT_TO_UBYTE(sdc.ub[3], color[3]);
  83
  84       UNCLAMPED_FLOAT_TO_USHORT(sdc.us[0], color[0]);
  85       UNCLAMPED_FLOAT_TO_USHORT(sdc.us[1], color[1]);
  86       UNCLAMPED_FLOAT_TO_USHORT(sdc.us[2], color[2]);
  87       UNCLAMPED_FLOAT_TO_USHORT(sdc.us[3], color[3]);
  88
  89       UNCLAMPED_FLOAT_TO_SHORT(sdc.s[0], color[0]);
  90       UNCLAMPED_FLOAT_TO_SHORT(sdc.s[1], color[1]);
  91       UNCLAMPED_FLOAT_TO_SHORT(sdc.s[2], color[2]);
  92       UNCLAMPED_FLOAT_TO_SHORT(sdc.s[3], color[3]);
  93
  94       /* XXX: Fill in half floats */
  95       /* XXX: Fill in signed bytes */
  96
  97       COPY_4V(sdc.f, color);
  98
  99       return brw_cache_data(&brw->cache, BRW_SAMPLER_DEFAULT_COLOR,
 100                             &sdc, sizeof(sdc));
 101    } else {
 102       struct brw_sampler_default_color sdc;
 103
 104       COPY_4V(sdc.color, color);
 105
 106       return brw_cache_data(&brw->cache, BRW_SAMPLER_DEFAULT_COLOR,
 107                             &sdc, sizeof(sdc));
 108    }
 109 }
 110
 111
 112 struct wm_sampler_key {
 113    int sampler_count;
 114
 115    struct wm_sampler_entry {
 116       GLenum tex_target;
 117       GLenum wrap_r, wrap_s, wrap_t;
 118       float maxlod, minlod;
 119       float lod_bias;
 120       float max_aniso;
 121       GLenum minfilter, magfilter;
 122       GLenum comparemode, comparefunc;
 123
 124       /** If target is cubemap, take context setting.
 125        */
 126       GLboolean seamless_cube_map;
 127    } sampler[BRW_MAX_TEX_UNIT];
 128 };
 129
 130 /**
 131  * Sets the sampler state for a single unit based off of the sampler key
 132  * entry.
 133  */
 134 static void brw_update_sampler_state(struct brw_context *brw,
 135                                      struct wm_sampler_entry *key,
 136                                      drm_intel_bo *sdc_bo,
 137                                      struct brw_sampler_state *sampler)
 138 {
 139    struct intel_context *intel = &brw->intel;
 140
 141    memset(sampler, 0, sizeof(*sampler));
 142
 143    switch (key->minfilter) {
 144    case GL_NEAREST:
 145       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 146       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
 147       break;
 148    case GL_LINEAR:
 149       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 150       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
 151       break;
 152    case GL_NEAREST_MIPMAP_NEAREST:
 153       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 154       sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
 155       break;
 156    case GL_LINEAR_MIPMAP_NEAREST:
 157       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 158       sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
 159       break;
 160    case GL_NEAREST_MIPMAP_LINEAR:
 161       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 162       sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
 163       break;
 164    case GL_LINEAR_MIPMAP_LINEAR:
 165       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 166       sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
 167       break;
 168    default:
 169       break;
 170    }
 171
 172    /* Set Anisotropy:
 173     */
 174    if (key->max_aniso > 1.0) {
 175       sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC;
 176       sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
 177
 178       if (key->max_aniso > 2.0) {
 179          sampler->ss3.max_aniso = MIN2((key->max_aniso - 2) / 2,
 180                                        BRW_ANISORATIO_16);
 181       }
 182    }
 183    else {
 184       switch (key->magfilter) {
 185       case GL_NEAREST:
 186          sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
 187          break;
 188       case GL_LINEAR:
 189          sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
 190          break;
 191       default:
 192          break;
 193       }
 194    }
 195
 196    sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r);
 197    sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s);
 198    sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t);
 199
 200    if (intel->gen >= 6 &&
 201        sampler->ss0.min_filter != sampler->ss0.mag_filter)
 202         sampler->ss0.min_mag_neq = 1;
 203
 204    /* Cube-maps on 965 and later must use the same wrap mode for all 3
 205     * coordinate dimensions.  Futher, only CUBE and CLAMP are valid.
 206     */
 207    if (key->tex_target == GL_TEXTURE_CUBE_MAP) {
 208       if (key->seamless_cube_map &&
 209           (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) {
 210          sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 211          sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 212          sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 213       } else {
 214          sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 215          sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 216          sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 217       }
 218    } else if (key->tex_target == GL_TEXTURE_1D) {
 219       /* There's a bug in 1D texture sampling - it actually pays
 220        * attention to the wrap_t value, though it should not.
 221        * Override the wrap_t value here to GL_REPEAT to keep
 222        * any nonexistent border pixels from floating in.
 223        */
 224       sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP;
 225    }
 226
 227
 228    /* Set shadow function:
 229     */
 230    if (key->comparemode == GL_COMPARE_R_TO_TEXTURE_ARB) {
 231       /* Shadowing is "enabled" by emitting a particular sampler
 232        * message (sample_c).  So need to recompile WM program when
 233        * shadow comparison is enabled on each/any texture unit.
 234        */
 235       sampler->ss0.shadow_function =
 236          intel_translate_shadow_compare_func(key->comparefunc);
 237    }
 238
 239    /* Set LOD bias:
 240     */
 241    sampler->ss0.lod_bias = S_FIXED(CLAMP(key->lod_bias, -16, 15), 6);
 242
 243    sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
 244    sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
 245
 246    /* Set BaseMipLevel, MaxLOD, MinLOD:
 247     *
 248     * XXX: I don't think that using firstLevel, lastLevel works,
 249     * because we always setup the surface state as if firstLevel ==
 250     * level zero.  Probably have to subtract firstLevel from each of
 251     * these:
 252     */
 253    sampler->ss0.base_level = U_FIXED(0, 1);
 254
 255    sampler->ss1.max_lod = U_FIXED(CLAMP(key->maxlod, 0, 13), 6);
 256    sampler->ss1.min_lod = U_FIXED(CLAMP(key->minlod, 0, 13), 6);
 257
 258    sampler->ss2.default_color_pointer = sdc_bo->offset >> 5; /* reloc */
 259 }
 260
 261
 262 /** Sets up the cache key for sampler state for all texture units */
 263 static void
 264 brw_wm_sampler_populate_key(struct brw_context *brw,
 265                             struct wm_sampler_key *key)
 266 {
 267    struct gl_context *ctx = &brw->intel.ctx;
 268    int unit;
 269    char *last_entry_end = ((char*)&key->sampler_count) +
 270       sizeof(key->sampler_count);
 271
 272    key->sampler_count = 0;
 273
 274    for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) {
 275       if (ctx->Texture.Unit[unit]._ReallyEnabled) {
 276          struct wm_sampler_entry *entry = &key->sampler[unit];
 277          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
 278          struct gl_texture_object *texObj = texUnit->_Current;
 279          struct gl_texture_image *firstImage =
 280             texObj->Image[0][texObj->BaseLevel];
 281
 282          memset(last_entry_end, 0,
 283                 (char*)entry - last_entry_end + sizeof(*entry));
 284          last_entry_end = ((char*)entry) + sizeof(*entry);
 285
 286          entry->tex_target = texObj->Target;
 287
 288          entry->seamless_cube_map = (texObj->Target == GL_TEXTURE_CUBE_MAP)
 289             ? ctx->Texture.CubeMapSeamless : GL_FALSE;
 290
 291          entry->wrap_r = texObj->WrapR;
 292          entry->wrap_s = texObj->WrapS;
 293          entry->wrap_t = texObj->WrapT;
 294
 295          entry->maxlod = texObj->MaxLod;
 296          entry->minlod = texObj->MinLod;
 297          entry->lod_bias = texUnit->LodBias + texObj->LodBias;
 298          entry->max_aniso = texObj->MaxAnisotropy;
 299          entry->minfilter = texObj->MinFilter;
 300          entry->magfilter = texObj->MagFilter;
 301          entry->comparemode = texObj->CompareMode;
 302          entry->comparefunc = texObj->CompareFunc;
 303
 304          drm_intel_bo_unreference(brw->wm.sdc_bo[unit]);
 305          if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) {
 306             float bordercolor[4] = {
 307                texObj->BorderColor.f[0],
 308                texObj->BorderColor.f[0],
 309                texObj->BorderColor.f[0],
 310                texObj->BorderColor.f[0]
 311             };
 312             /* GL specs that border color for depth textures is taken from the
 313              * R channel, while the hardware uses A.  Spam R into all the
 314              * channels for safety.
 315              */
 316             brw->wm.sdc_bo[unit] = upload_default_color(brw, bordercolor);
 317          } else {
 318             brw->wm.sdc_bo[unit] = upload_default_color(brw,
 319                                                         texObj->BorderColor.f);
 320          }
 321          key->sampler_count = unit + 1;
 322       }
 323    }
 324    struct wm_sampler_entry *entry = &key->sampler[key->sampler_count];
 325    memset(last_entry_end, 0, (char*)entry - last_entry_end);
 326 }
 327
 328 /* All samplers must be uploaded in a single contiguous array, which
 329  * complicates various things.  However, this is still too confusing -
 330  * FIXME: simplify all the different new texture state flags.
 331  */
 332 static void upload_wm_samplers( struct brw_context *brw )
 333 {
 334    struct gl_context *ctx = &brw->intel.ctx;
 335    struct wm_sampler_key key;
 336    int i, sampler_key_size;
 337
 338    brw_wm_sampler_populate_key(brw, &key);
 339
 340    if (brw->wm.sampler_count != key.sampler_count) {
 341       brw->wm.sampler_count = key.sampler_count;
 342       brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
 343    }
 344
 345    drm_intel_bo_unreference(brw->wm.sampler_bo);
 346    brw->wm.sampler_bo = NULL;
 347    if (brw->wm.sampler_count == 0)
 348       return;
 349
 350    /* Only include the populated portion of the key in the search. */
 351    sampler_key_size = offsetof(struct wm_sampler_key,
 352                                sampler[key.sampler_count]);
 353    brw->wm.sampler_bo = brw_search_cache(&brw->cache, BRW_SAMPLER,
 354                                          &key, sampler_key_size,
 355                                          brw->wm.sdc_bo, key.sampler_count,
 356                                          NULL);
 357
 358    /* If we didnt find it in the cache, compute the state and put it in the
 359     * cache.
 360     */
 361    if (brw->wm.sampler_bo == NULL) {
 362       struct brw_sampler_state sampler[BRW_MAX_TEX_UNIT];
 363
 364       memset(sampler, 0, sizeof(sampler));
 365       for (i = 0; i < key.sampler_count; i++) {
 366          if (brw->wm.sdc_bo[i] == NULL)
 367             continue;
 368
 369          brw_update_sampler_state(brw, &key.sampler[i], brw->wm.sdc_bo[i],
 370                                   &sampler[i]);
 371       }
 372
 373       brw->wm.sampler_bo = brw_upload_cache(&brw->cache, BRW_SAMPLER,
 374                                             &key, sampler_key_size,
 375                                             brw->wm.sdc_bo, key.sampler_count,
 376                                             &sampler, sizeof(sampler));
 377
 378       /* Emit SDC relocations */
 379       for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
 380          if (!ctx->Texture.Unit[i]._ReallyEnabled)
 381             continue;
 382
 383          drm_intel_bo_emit_reloc(brw->wm.sampler_bo,
 384                                  i * sizeof(struct brw_sampler_state) +
 385                                  offsetof(struct brw_sampler_state, ss2),
 386                                  brw->wm.sdc_bo[i], 0,
 387                                  I915_GEM_DOMAIN_SAMPLER, 0);
 388       }
 389    }
 390 }
 391
 392 const struct brw_tracked_state brw_wm_samplers = {
 393    .dirty = {
 394       .mesa = _NEW_TEXTURE,
 395       .brw = 0,
 396       .cache = 0
 397    },
 398    .prepare = upload_wm_samplers,
 399 };
 400
 401