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 dri_bo *upload_default_color( struct brw_context *brw,
  70                                      const GLfloat *color )
  71 {
  72    struct brw_sampler_default_color sdc;
  73
  74    COPY_4V(sdc.color, color);
  75
  76    return brw_cache_data(&brw->cache, BRW_SAMPLER_DEFAULT_COLOR,
  77                          &sdc, sizeof(sdc), NULL, 0);
  78 }
  79
  80
  81 struct wm_sampler_key {
  82    int sampler_count;
  83
  84    struct wm_sampler_entry {
  85       GLenum tex_target;
  86       GLenum wrap_r, wrap_s, wrap_t;
  87       float maxlod, minlod;
  88       float lod_bias;
  89       float max_aniso;
  90       GLenum minfilter, magfilter;
  91       GLenum comparemode, comparefunc;
  92       dri_bo *sdc_bo;
  93
  94       /** If target is cubemap, take context setting.
  95        */
  96       GLboolean seamless_cube_map;
  97    } sampler[BRW_MAX_TEX_UNIT];
  98 };
  99
 100 /**
 101  * Sets the sampler state for a single unit based off of the sampler key
 102  * entry.
 103  */
 104 static void brw_update_sampler_state(struct wm_sampler_entry *key,
 105                                      dri_bo *sdc_bo,
 106                                      struct brw_sampler_state *sampler)
 107 {
 108    _mesa_memset(sampler, 0, sizeof(*sampler));
 109
 110    switch (key->minfilter) {
 111    case GL_NEAREST:
 112       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 113       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
 114       break;
 115    case GL_LINEAR:
 116       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 117       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
 118       break;
 119    case GL_NEAREST_MIPMAP_NEAREST:
 120       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 121       sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
 122       break;
 123    case GL_LINEAR_MIPMAP_NEAREST:
 124       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 125       sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
 126       break;
 127    case GL_NEAREST_MIPMAP_LINEAR:
 128       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 129       sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
 130       break;
 131    case GL_LINEAR_MIPMAP_LINEAR:
 132       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 133       sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
 134       break;
 135    default:
 136       break;
 137    }
 138
 139    /* Set Anisotropy:
 140     */
 141    if (key->max_aniso > 1.0) {
 142       sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC;
 143       sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
 144
 145       if (key->max_aniso > 2.0) {
 146          sampler->ss3.max_aniso = MIN2((key->max_aniso - 2) / 2,
 147                                        BRW_ANISORATIO_16);
 148       }
 149    }
 150    else {
 151       switch (key->magfilter) {
 152       case GL_NEAREST:
 153          sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
 154          break;
 155       case GL_LINEAR:
 156          sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
 157          break;
 158       default:
 159          break;
 160       }
 161    }
 162
 163    sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r);
 164    sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s);
 165    sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t);
 166
 167    /* Cube-maps on 965 and later must use the same wrap mode for all 3
 168     * coordinate dimensions.  Futher, only CUBE and CLAMP are valid.
 169     */
 170    if (key->tex_target == GL_TEXTURE_CUBE_MAP) {
 171       if (key->seamless_cube_map &&
 172           (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) {
 173          sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 174          sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 175          sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 176       } else {
 177          sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 178          sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 179          sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 180       }
 181    } else if (key->tex_target == GL_TEXTURE_1D) {
 182       /* There's a bug in 1D texture sampling - it actually pays
 183        * attention to the wrap_t value, though it should not.
 184        * Override the wrap_t value here to GL_REPEAT to keep
 185        * any nonexistent border pixels from floating in.
 186        */
 187       sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP;
 188    }
 189
 190
 191    /* Set shadow function:
 192     */
 193    if (key->comparemode == GL_COMPARE_R_TO_TEXTURE_ARB) {
 194       /* Shadowing is "enabled" by emitting a particular sampler
 195        * message (sample_c).  So need to recompile WM program when
 196        * shadow comparison is enabled on each/any texture unit.
 197        */
 198       sampler->ss0.shadow_function =
 199          intel_translate_shadow_compare_func(key->comparefunc);
 200    }
 201
 202    /* Set LOD bias:
 203     */
 204    sampler->ss0.lod_bias = S_FIXED(CLAMP(key->lod_bias, -16, 15), 6);
 205
 206    sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
 207    sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
 208
 209    /* Set BaseMipLevel, MaxLOD, MinLOD:
 210     *
 211     * XXX: I don't think that using firstLevel, lastLevel works,
 212     * because we always setup the surface state as if firstLevel ==
 213     * level zero.  Probably have to subtract firstLevel from each of
 214     * these:
 215     */
 216    sampler->ss0.base_level = U_FIXED(0, 1);
 217
 218    sampler->ss1.max_lod = U_FIXED(CLAMP(key->maxlod, 0, 13), 6);
 219    sampler->ss1.min_lod = U_FIXED(CLAMP(key->minlod, 0, 13), 6);
 220
 221    sampler->ss2.default_color_pointer = sdc_bo->offset >> 5; /* reloc */
 222 }
 223
 224
 225 /** Sets up the cache key for sampler state for all texture units */
 226 static void
 227 brw_wm_sampler_populate_key(struct brw_context *brw,
 228                             struct wm_sampler_key *key)
 229 {
 230    GLcontext *ctx = &brw->intel.ctx;
 231    int unit;
 232
 233    memset(key, 0, sizeof(*key));
 234
 235    for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) {
 236       if (ctx->Texture.Unit[unit]._ReallyEnabled) {
 237          struct wm_sampler_entry *entry = &key->sampler[unit];
 238          struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
 239          struct gl_texture_object *texObj = texUnit->_Current;
 240          struct intel_texture_object *intelObj = intel_texture_object(texObj);
 241          struct gl_texture_image *firstImage =
 242             texObj->Image[0][intelObj->firstLevel];
 243
 244          entry->tex_target = texObj->Target;
 245
 246          entry->seamless_cube_map = (texObj->Target == GL_TEXTURE_CUBE_MAP)
 247             ? ctx->Texture.CubeMapSeamless : GL_FALSE;
 248
 249          entry->wrap_r = texObj->WrapR;
 250          entry->wrap_s = texObj->WrapS;
 251          entry->wrap_t = texObj->WrapT;
 252
 253          entry->maxlod = texObj->MaxLod;
 254          entry->minlod = texObj->MinLod;
 255          entry->lod_bias = texUnit->LodBias + texObj->LodBias;
 256          entry->max_aniso = texObj->MaxAnisotropy;
 257          entry->minfilter = texObj->MinFilter;
 258          entry->magfilter = texObj->MagFilter;
 259          entry->comparemode = texObj->CompareMode;
 260          entry->comparefunc = texObj->CompareFunc;
 261
 262          dri_bo_unreference(brw->wm.sdc_bo[unit]);
 263          if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) {
 264             float bordercolor[4] = {
 265                texObj->BorderColor.f[0],
 266                texObj->BorderColor.f[0],
 267                texObj->BorderColor.f[0],
 268                texObj->BorderColor.f[0]
 269             };
 270             /* GL specs that border color for depth textures is taken from the
 271              * R channel, while the hardware uses A.  Spam R into all the
 272              * channels for safety.
 273              */
 274             brw->wm.sdc_bo[unit] = upload_default_color(brw, bordercolor);
 275          } else {
 276             brw->wm.sdc_bo[unit] = upload_default_color(brw,
 277                                                         texObj->BorderColor.f);
 278          }
 279          key->sampler_count = unit + 1;
 280       }
 281    }
 282 }
 283
 284 /* All samplers must be uploaded in a single contiguous array, which
 285  * complicates various things.  However, this is still too confusing -
 286  * FIXME: simplify all the different new texture state flags.
 287  */
 288 static void upload_wm_samplers( struct brw_context *brw )
 289 {
 290    GLcontext *ctx = &brw->intel.ctx;
 291    struct wm_sampler_key key;
 292    int i;
 293
 294    brw_wm_sampler_populate_key(brw, &key);
 295
 296    if (brw->wm.sampler_count != key.sampler_count) {
 297       brw->wm.sampler_count = key.sampler_count;
 298       brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
 299    }
 300
 301    dri_bo_unreference(brw->wm.sampler_bo);
 302    brw->wm.sampler_bo = NULL;
 303    if (brw->wm.sampler_count == 0)
 304       return;
 305
 306    brw->wm.sampler_bo = brw_search_cache(&brw->cache, BRW_SAMPLER,
 307                                          &key, sizeof(key),
 308                                          brw->wm.sdc_bo, key.sampler_count,
 309                                          NULL);
 310
 311    /* If we didnt find it in the cache, compute the state and put it in the
 312     * cache.
 313     */
 314    if (brw->wm.sampler_bo == NULL) {
 315       struct brw_sampler_state sampler[BRW_MAX_TEX_UNIT];
 316
 317       memset(sampler, 0, sizeof(sampler));
 318       for (i = 0; i < key.sampler_count; i++) {
 319          if (brw->wm.sdc_bo[i] == NULL)
 320             continue;
 321
 322          brw_update_sampler_state(&key.sampler[i], brw->wm.sdc_bo[i],
 323                                   &sampler[i]);
 324       }
 325
 326       brw->wm.sampler_bo = brw_upload_cache(&brw->cache, BRW_SAMPLER,
 327                                             &key, sizeof(key),
 328                                             brw->wm.sdc_bo, key.sampler_count,
 329                                             &sampler, sizeof(sampler));
 330
 331       /* Emit SDC relocations */
 332       for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
 333          if (!ctx->Texture.Unit[i]._ReallyEnabled)
 334             continue;
 335
 336          dri_bo_emit_reloc(brw->wm.sampler_bo,
 337                            I915_GEM_DOMAIN_SAMPLER, 0,
 338                            0,
 339                            i * sizeof(struct brw_sampler_state) +
 340                            offsetof(struct brw_sampler_state, ss2),
 341                            brw->wm.sdc_bo[i]);
 342       }
 343    }
 344 }
 345
 346 const struct brw_tracked_state brw_wm_samplers = {
 347    .dirty = {
 348       .mesa = _NEW_TEXTURE,
 349       .brw = 0,
 350       .cache = 0
 351    },
 352    .prepare = upload_wm_samplers,
 353 };
 354
 355