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 #include "main/samplerobj.h"
  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 uint32_t
  48 translate_wrap_mode(GLenum wrap, bool using_nearest)
  49 {
  50    switch( wrap ) {
  51    case GL_REPEAT:
  52       return BRW_TEXCOORDMODE_WRAP;
  53    case GL_CLAMP:
  54       /* GL_CLAMP is the weird mode where coordinates are clamped to
  55        * [0.0, 1.0], so linear filtering of coordinates outside of
  56        * [0.0, 1.0] give you half edge texel value and half border
  57        * color.  The fragment shader will clamp the coordinates, and
  58        * we set clamp_border here, which gets the result desired.  We
  59        * just use clamp(_to_edge) for nearest, because for nearest
  60        * clamping to 1.0 gives border color instead of the desired
  61        * edge texels.
  62        */
  63       if (using_nearest)
  64          return BRW_TEXCOORDMODE_CLAMP;
  65       else
  66          return BRW_TEXCOORDMODE_CLAMP_BORDER;
  67    case GL_CLAMP_TO_EDGE:
  68       return BRW_TEXCOORDMODE_CLAMP;
  69    case GL_CLAMP_TO_BORDER:
  70       return BRW_TEXCOORDMODE_CLAMP_BORDER;
  71    case GL_MIRRORED_REPEAT:
  72       return BRW_TEXCOORDMODE_MIRROR;
  73    default:
  74       return BRW_TEXCOORDMODE_WRAP;
  75    }
  76 }
  77
  78 /**
  79  * Upload SAMPLER_BORDER_COLOR_STATE.
  80  */
  81 void
  82 upload_default_color(struct brw_context *brw, struct gl_sampler_object *sampler,
  83                      int unit)
  84 {
  85    struct intel_context *intel = &brw->intel;
  86    struct gl_context *ctx = &intel->ctx;
  87    struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
  88    struct gl_texture_object *texObj = texUnit->_Current;
  89    struct gl_texture_image *firstImage = texObj->Image[0][texObj->BaseLevel];
  90    float color[4];
  91
  92    if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) {
  93       /* GL specs that border color for depth textures is taken from the
  94        * R channel, while the hardware uses A.  Spam R into all the
  95        * channels for safety.
  96        */
  97       color[0] = sampler->BorderColor.f[0];
  98       color[1] = sampler->BorderColor.f[0];
  99       color[2] = sampler->BorderColor.f[0];
 100       color[3] = sampler->BorderColor.f[0];
 101    } else {
 102       color[0] = sampler->BorderColor.f[0];
 103       color[1] = sampler->BorderColor.f[1];
 104       color[2] = sampler->BorderColor.f[2];
 105       color[3] = sampler->BorderColor.f[3];
 106    }
 107
 108    if (intel->gen == 5 || intel->gen == 6) {
 109       struct gen5_sampler_default_color *sdc;
 110
 111       sdc = brw_state_batch(brw, AUB_TRACE_SAMPLER_DEFAULT_COLOR,
 112                             sizeof(*sdc), 32, &brw->wm.sdc_offset[unit]);
 113
 114       memset(sdc, 0, sizeof(*sdc));
 115
 116       UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[0], color[0]);
 117       UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[1], color[1]);
 118       UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[2], color[2]);
 119       UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[3], color[3]);
 120
 121       UNCLAMPED_FLOAT_TO_USHORT(sdc->us[0], color[0]);
 122       UNCLAMPED_FLOAT_TO_USHORT(sdc->us[1], color[1]);
 123       UNCLAMPED_FLOAT_TO_USHORT(sdc->us[2], color[2]);
 124       UNCLAMPED_FLOAT_TO_USHORT(sdc->us[3], color[3]);
 125
 126       UNCLAMPED_FLOAT_TO_SHORT(sdc->s[0], color[0]);
 127       UNCLAMPED_FLOAT_TO_SHORT(sdc->s[1], color[1]);
 128       UNCLAMPED_FLOAT_TO_SHORT(sdc->s[2], color[2]);
 129       UNCLAMPED_FLOAT_TO_SHORT(sdc->s[3], color[3]);
 130
 131       sdc->hf[0] = _mesa_float_to_half(color[0]);
 132       sdc->hf[1] = _mesa_float_to_half(color[1]);
 133       sdc->hf[2] = _mesa_float_to_half(color[2]);
 134       sdc->hf[3] = _mesa_float_to_half(color[3]);
 135
 136       sdc->b[0] = sdc->s[0] >> 8;
 137       sdc->b[1] = sdc->s[1] >> 8;
 138       sdc->b[2] = sdc->s[2] >> 8;
 139       sdc->b[3] = sdc->s[3] >> 8;
 140
 141       sdc->f[0] = color[0];
 142       sdc->f[1] = color[1];
 143       sdc->f[2] = color[2];
 144       sdc->f[3] = color[3];
 145    } else {
 146       struct brw_sampler_default_color *sdc;
 147
 148       sdc = brw_state_batch(brw, AUB_TRACE_SAMPLER_DEFAULT_COLOR,
 149                             sizeof(*sdc), 32, &brw->wm.sdc_offset[unit]);
 150
 151       COPY_4V(sdc->color, color);
 152    }
 153 }
 154
 155 /**
 156  * Sets the sampler state for a single unit based off of the sampler key
 157  * entry.
 158  */
 159 static void brw_update_sampler_state(struct brw_context *brw,
 160                                      int unit,
 161                                      struct brw_sampler_state *sampler)
 162 {
 163    struct intel_context *intel = &brw->intel;
 164    struct gl_context *ctx = &intel->ctx;
 165    struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
 166    struct gl_texture_object *texObj = texUnit->_Current;
 167    struct gl_sampler_object *gl_sampler = _mesa_get_samplerobj(ctx, unit);
 168    bool using_nearest = false;
 169
 170    switch (gl_sampler->MinFilter) {
 171    case GL_NEAREST:
 172       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 173       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
 174       using_nearest = true;
 175       break;
 176    case GL_LINEAR:
 177       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 178       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
 179       break;
 180    case GL_NEAREST_MIPMAP_NEAREST:
 181       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 182       sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
 183       break;
 184    case GL_LINEAR_MIPMAP_NEAREST:
 185       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 186       sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
 187       break;
 188    case GL_NEAREST_MIPMAP_LINEAR:
 189       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 190       sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
 191       break;
 192    case GL_LINEAR_MIPMAP_LINEAR:
 193       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 194       sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
 195       break;
 196    default:
 197       break;
 198    }
 199
 200    /* Set Anisotropy:
 201     */
 202    if (gl_sampler->MaxAnisotropy > 1.0) {
 203       sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC;
 204       sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
 205
 206       if (gl_sampler->MaxAnisotropy > 2.0) {
 207          sampler->ss3.max_aniso = MIN2((gl_sampler->MaxAnisotropy - 2) / 2,
 208                                        BRW_ANISORATIO_16);
 209       }
 210    }
 211    else {
 212       switch (gl_sampler->MagFilter) {
 213       case GL_NEAREST:
 214          sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
 215          using_nearest = true;
 216          break;
 217       case GL_LINEAR:
 218          sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
 219          break;
 220       default:
 221          break;
 222       }
 223    }
 224
 225    sampler->ss1.r_wrap_mode = translate_wrap_mode(gl_sampler->WrapR,
 226                                                   using_nearest);
 227    sampler->ss1.s_wrap_mode = translate_wrap_mode(gl_sampler->WrapS,
 228                                                   using_nearest);
 229    sampler->ss1.t_wrap_mode = translate_wrap_mode(gl_sampler->WrapT,
 230                                                   using_nearest);
 231
 232    if (intel->gen >= 6 &&
 233        sampler->ss0.min_filter != sampler->ss0.mag_filter)
 234         sampler->ss0.min_mag_neq = 1;
 235
 236    /* Cube-maps on 965 and later must use the same wrap mode for all 3
 237     * coordinate dimensions.  Futher, only CUBE and CLAMP are valid.
 238     */
 239    if (texObj->Target == GL_TEXTURE_CUBE_MAP) {
 240       if (ctx->Texture.CubeMapSeamless &&
 241           (gl_sampler->MinFilter != GL_NEAREST ||
 242            gl_sampler->MagFilter != GL_NEAREST)) {
 243          sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 244          sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 245          sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 246       } else {
 247          sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 248          sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 249          sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 250       }
 251    } else if (texObj->Target == GL_TEXTURE_1D) {
 252       /* There's a bug in 1D texture sampling - it actually pays
 253        * attention to the wrap_t value, though it should not.
 254        * Override the wrap_t value here to GL_REPEAT to keep
 255        * any nonexistent border pixels from floating in.
 256        */
 257       sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP;
 258    }
 259
 260
 261    /* Set shadow function:
 262     */
 263    if (gl_sampler->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB) {
 264       /* Shadowing is "enabled" by emitting a particular sampler
 265        * message (sample_c).  So need to recompile WM program when
 266        * shadow comparison is enabled on each/any texture unit.
 267        */
 268       sampler->ss0.shadow_function =
 269          intel_translate_shadow_compare_func(gl_sampler->CompareFunc);
 270    }
 271
 272    /* Set LOD bias:
 273     */
 274    sampler->ss0.lod_bias = S_FIXED(CLAMP(texUnit->LodBias +
 275                                          gl_sampler->LodBias, -16, 15), 6);
 276
 277    sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
 278    sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
 279
 280    /* Set BaseMipLevel, MaxLOD, MinLOD:
 281     *
 282     * XXX: I don't think that using firstLevel, lastLevel works,
 283     * because we always setup the surface state as if firstLevel ==
 284     * level zero.  Probably have to subtract firstLevel from each of
 285     * these:
 286     */
 287    sampler->ss0.base_level = U_FIXED(0, 1);
 288
 289    sampler->ss1.max_lod = U_FIXED(CLAMP(gl_sampler->MaxLod, 0, 13), 6);
 290    sampler->ss1.min_lod = U_FIXED(CLAMP(gl_sampler->MinLod, 0, 13), 6);
 291
 292    /* On Gen6+, the sampler can handle non-normalized texture
 293     * rectangle coordinates natively
 294     */
 295    if (intel->gen >= 6 && texObj->Target == GL_TEXTURE_RECTANGLE) {
 296       sampler->ss3.non_normalized_coord = 1;
 297    }
 298
 299    upload_default_color(brw, gl_sampler, unit);
 300
 301    if (intel->gen >= 6) {
 302       sampler->ss2.default_color_pointer = brw->wm.sdc_offset[unit] >> 5;
 303    } else {
 304       /* reloc */
 305       sampler->ss2.default_color_pointer = (intel->batch.bo->offset +
 306                                             brw->wm.sdc_offset[unit]) >> 5;
 307
 308       drm_intel_bo_emit_reloc(intel->batch.bo,
 309                               brw->sampler.offset +
 310                               unit * sizeof(struct brw_sampler_state) +
 311                               offsetof(struct brw_sampler_state, ss2),
 312                               intel->batch.bo, brw->wm.sdc_offset[unit],
 313                               I915_GEM_DOMAIN_SAMPLER, 0);
 314    }
 315
 316    if (sampler->ss0.min_filter != BRW_MAPFILTER_NEAREST)
 317       sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
 318                                     BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
 319                                     BRW_ADDRESS_ROUNDING_ENABLE_R_MIN;
 320    if (sampler->ss0.mag_filter != BRW_MAPFILTER_NEAREST)
 321       sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
 322                                     BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
 323                                     BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
 324 }
 325
 326
 327 /* All samplers must be uploaded in a single contiguous array, which
 328  * complicates various things.  However, this is still too confusing -
 329  * FIXME: simplify all the different new texture state flags.
 330  */
 331 static void
 332 brw_upload_samplers(struct brw_context *brw)
 333 {
 334    struct gl_context *ctx = &brw->intel.ctx;
 335    struct brw_sampler_state *samplers;
 336    int i;
 337
 338    brw->sampler.count = 0;
 339    for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
 340       if (ctx->Texture.Unit[i]._ReallyEnabled)
 341          brw->sampler.count = i + 1;
 342    }
 343
 344    if (brw->sampler.count == 0)
 345       return;
 346
 347    samplers = brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE,
 348                               brw->sampler.count * sizeof(*samplers),
 349                               32, &brw->sampler.offset);
 350    memset(samplers, 0, brw->sampler.count * sizeof(*samplers));
 351
 352    for (i = 0; i < brw->sampler.count; i++) {
 353       if (ctx->Texture.Unit[i]._ReallyEnabled)
 354          brw_update_sampler_state(brw, i, &samplers[i]);
 355    }
 356
 357    brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
 358 }
 359
 360 const struct brw_tracked_state brw_samplers = {
 361    .dirty = {
 362       .mesa = _NEW_TEXTURE,
 363       .brw = BRW_NEW_BATCH,
 364       .cache = 0
 365    },
 366    .emit = brw_upload_samplers,
 367 };
 368
 369