src/mesa/drivers/dri/i965/brw_sampler_state.c

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics 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 <keithw@vmware.com>
  30   */
  31
  32 /**
  33  * @file brw_sampler_state.c
  34  *
  35  * This file contains code for emitting SAMPLER_STATE structures, which
  36  * specifies filter modes, wrap modes, border color, and so on.
  37  */
  38
  39 #include "brw_context.h"
  40 #include "brw_state.h"
  41 #include "brw_defines.h"
  42 #include "intel_mipmap_tree.h"
  43
  44 #include "main/macros.h"
  45 #include "main/samplerobj.h"
  46
  47 uint32_t
  48 translate_wrap_mode(struct brw_context *brw, 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.
  58        *
  59        * Gen8+ supports this natively.
  60        */
  61       if (brw->gen >= 8)
  62          return GEN8_TEXCOORDMODE_HALF_BORDER;
  63
  64       /* On Gen4-7.5, we clamp the coordinates in the fragment shader
  65        * and set clamp_border here, which gets the result desired.
  66        * We just use clamp(_to_edge) for nearest, because for nearest
  67        * clamping to 1.0 gives border color instead of the desired
  68        * edge texels.
  69        */
  70       if (using_nearest)
  71          return BRW_TEXCOORDMODE_CLAMP;
  72       else
  73          return BRW_TEXCOORDMODE_CLAMP_BORDER;
  74    case GL_CLAMP_TO_EDGE:
  75       return BRW_TEXCOORDMODE_CLAMP;
  76    case GL_CLAMP_TO_BORDER:
  77       return BRW_TEXCOORDMODE_CLAMP_BORDER;
  78    case GL_MIRRORED_REPEAT:
  79       return BRW_TEXCOORDMODE_MIRROR;
  80    case GL_MIRROR_CLAMP_TO_EDGE:
  81       return BRW_TEXCOORDMODE_MIRROR_ONCE;
  82    default:
  83       return BRW_TEXCOORDMODE_WRAP;
  84    }
  85 }
  86
  87 /**
  88  * Upload SAMPLER_BORDER_COLOR_STATE.
  89  */
  90 void
  91 upload_default_color(struct brw_context *brw,
  92                      struct gl_sampler_object *sampler,
  93                      int unit,
  94                      uint32_t *sdc_offset)
  95 {
  96    struct gl_context *ctx = &brw->ctx;
  97    struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
  98    struct gl_texture_object *texObj = texUnit->_Current;
  99    struct gl_texture_image *firstImage = texObj->Image[0][texObj->BaseLevel];
 100    float color[4];
 101
 102    switch (firstImage->_BaseFormat) {
 103    case GL_DEPTH_COMPONENT:
 104       /* GL specs that border color for depth textures is taken from the
 105        * R channel, while the hardware uses A.  Spam R into all the
 106        * channels for safety.
 107        */
 108       color[0] = sampler->BorderColor.f[0];
 109       color[1] = sampler->BorderColor.f[0];
 110       color[2] = sampler->BorderColor.f[0];
 111       color[3] = sampler->BorderColor.f[0];
 112       break;
 113    case GL_ALPHA:
 114       color[0] = 0.0;
 115       color[1] = 0.0;
 116       color[2] = 0.0;
 117       color[3] = sampler->BorderColor.f[3];
 118       break;
 119    case GL_INTENSITY:
 120       color[0] = sampler->BorderColor.f[0];
 121       color[1] = sampler->BorderColor.f[0];
 122       color[2] = sampler->BorderColor.f[0];
 123       color[3] = sampler->BorderColor.f[0];
 124       break;
 125    case GL_LUMINANCE:
 126       color[0] = sampler->BorderColor.f[0];
 127       color[1] = sampler->BorderColor.f[0];
 128       color[2] = sampler->BorderColor.f[0];
 129       color[3] = 1.0;
 130       break;
 131    case GL_LUMINANCE_ALPHA:
 132       color[0] = sampler->BorderColor.f[0];
 133       color[1] = sampler->BorderColor.f[0];
 134       color[2] = sampler->BorderColor.f[0];
 135       color[3] = sampler->BorderColor.f[3];
 136       break;
 137    default:
 138       color[0] = sampler->BorderColor.f[0];
 139       color[1] = sampler->BorderColor.f[1];
 140       color[2] = sampler->BorderColor.f[2];
 141       color[3] = sampler->BorderColor.f[3];
 142       break;
 143    }
 144
 145    /* In some cases we use an RGBA surface format for GL RGB textures,
 146     * where we've initialized the A channel to 1.0.  We also have to set
 147     * the border color alpha to 1.0 in that case.
 148     */
 149    if (firstImage->_BaseFormat == GL_RGB)
 150       color[3] = 1.0;
 151
 152    if (brw->gen >= 8) {
 153       /* On Broadwell, the border color is represented as four 32-bit floats,
 154        * integers, or unsigned values, interpreted according to the surface
 155        * format.  This matches the sampler->BorderColor union exactly.  Since
 156        * we use floats both here and in the above reswizzling code, we preserve
 157        * the original bit pattern.  So we actually handle all three formats.
 158        */
 159       float *sdc = brw_state_batch(brw, AUB_TRACE_SAMPLER_DEFAULT_COLOR,
 160                                    4 * 4, 64, sdc_offset);
 161       COPY_4FV(sdc, color);
 162    } else if (brw->gen == 5 || brw->gen == 6) {
 163       struct gen5_sampler_default_color *sdc;
 164
 165       sdc = brw_state_batch(brw, AUB_TRACE_SAMPLER_DEFAULT_COLOR,
 166                             sizeof(*sdc), 32, sdc_offset);
 167
 168       memset(sdc, 0, sizeof(*sdc));
 169
 170       UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[0], color[0]);
 171       UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[1], color[1]);
 172       UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[2], color[2]);
 173       UNCLAMPED_FLOAT_TO_UBYTE(sdc->ub[3], color[3]);
 174
 175       UNCLAMPED_FLOAT_TO_USHORT(sdc->us[0], color[0]);
 176       UNCLAMPED_FLOAT_TO_USHORT(sdc->us[1], color[1]);
 177       UNCLAMPED_FLOAT_TO_USHORT(sdc->us[2], color[2]);
 178       UNCLAMPED_FLOAT_TO_USHORT(sdc->us[3], color[3]);
 179
 180       UNCLAMPED_FLOAT_TO_SHORT(sdc->s[0], color[0]);
 181       UNCLAMPED_FLOAT_TO_SHORT(sdc->s[1], color[1]);
 182       UNCLAMPED_FLOAT_TO_SHORT(sdc->s[2], color[2]);
 183       UNCLAMPED_FLOAT_TO_SHORT(sdc->s[3], color[3]);
 184
 185       sdc->hf[0] = _mesa_float_to_half(color[0]);
 186       sdc->hf[1] = _mesa_float_to_half(color[1]);
 187       sdc->hf[2] = _mesa_float_to_half(color[2]);
 188       sdc->hf[3] = _mesa_float_to_half(color[3]);
 189
 190       sdc->b[0] = sdc->s[0] >> 8;
 191       sdc->b[1] = sdc->s[1] >> 8;
 192       sdc->b[2] = sdc->s[2] >> 8;
 193       sdc->b[3] = sdc->s[3] >> 8;
 194
 195       sdc->f[0] = color[0];
 196       sdc->f[1] = color[1];
 197       sdc->f[2] = color[2];
 198       sdc->f[3] = color[3];
 199    } else {
 200       float *sdc = brw_state_batch(brw, AUB_TRACE_SAMPLER_DEFAULT_COLOR,
 201                                    4 * 4, 32, sdc_offset);
 202       memcpy(sdc, color, 4 * 4);
 203    }
 204 }
 205
 206 /**
 207  * Sets the sampler state for a single unit based off of the sampler key
 208  * entry.
 209  */
 210 static void
 211 brw_update_sampler_state(struct brw_context *brw,
 212                          int unit,
 213                          struct brw_sampler_state *sampler,
 214                          uint32_t batch_offset_for_sampler_state)
 215 {
 216    struct gl_context *ctx = &brw->ctx;
 217    struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
 218    struct gl_texture_object *texObj = texUnit->_Current;
 219    struct gl_sampler_object *gl_sampler = _mesa_get_samplerobj(ctx, unit);
 220    bool using_nearest = false;
 221
 222    /* These don't use samplers at all. */
 223    if (texObj->Target == GL_TEXTURE_BUFFER)
 224       return;
 225
 226    switch (gl_sampler->MinFilter) {
 227    case GL_NEAREST:
 228       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 229       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
 230       using_nearest = true;
 231       break;
 232    case GL_LINEAR:
 233       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 234       sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
 235       break;
 236    case GL_NEAREST_MIPMAP_NEAREST:
 237       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 238       sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
 239       break;
 240    case GL_LINEAR_MIPMAP_NEAREST:
 241       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 242       sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
 243       break;
 244    case GL_NEAREST_MIPMAP_LINEAR:
 245       sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
 246       sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
 247       break;
 248    case GL_LINEAR_MIPMAP_LINEAR:
 249       sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
 250       sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
 251       break;
 252    default:
 253       break;
 254    }
 255
 256    /* Set Anisotropy:
 257     */
 258    if (gl_sampler->MaxAnisotropy > 1.0) {
 259       sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC;
 260       sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
 261
 262       if (gl_sampler->MaxAnisotropy > 2.0) {
 263          sampler->ss3.max_aniso = MIN2((gl_sampler->MaxAnisotropy - 2) / 2,
 264                                        BRW_ANISORATIO_16);
 265       }
 266    }
 267    else {
 268       switch (gl_sampler->MagFilter) {
 269       case GL_NEAREST:
 270          sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
 271          using_nearest = true;
 272          break;
 273       case GL_LINEAR:
 274          sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
 275          break;
 276       default:
 277          break;
 278       }
 279    }
 280
 281    sampler->ss1.r_wrap_mode = translate_wrap_mode(brw, gl_sampler->WrapR,
 282                                                   using_nearest);
 283    sampler->ss1.s_wrap_mode = translate_wrap_mode(brw, gl_sampler->WrapS,
 284                                                   using_nearest);
 285    sampler->ss1.t_wrap_mode = translate_wrap_mode(brw, gl_sampler->WrapT,
 286                                                   using_nearest);
 287
 288    if (brw->gen >= 6 &&
 289        sampler->ss0.min_filter != sampler->ss0.mag_filter)
 290         sampler->ss0.min_mag_neq = 1;
 291
 292    /* Cube-maps on 965 and later must use the same wrap mode for all 3
 293     * coordinate dimensions.  Futher, only CUBE and CLAMP are valid.
 294     */
 295    if (texObj->Target == GL_TEXTURE_CUBE_MAP ||
 296        texObj->Target == GL_TEXTURE_CUBE_MAP_ARRAY) {
 297       if ((ctx->Texture.CubeMapSeamless || gl_sampler->CubeMapSeamless) &&
 298           (gl_sampler->MinFilter != GL_NEAREST ||
 299            gl_sampler->MagFilter != GL_NEAREST)) {
 300          sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 301          sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 302          sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
 303       } else {
 304          sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 305          sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 306          sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
 307       }
 308    } else if (texObj->Target == GL_TEXTURE_1D) {
 309       /* There's a bug in 1D texture sampling - it actually pays
 310        * attention to the wrap_t value, though it should not.
 311        * Override the wrap_t value here to GL_REPEAT to keep
 312        * any nonexistent border pixels from floating in.
 313        */
 314       sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP;
 315    }
 316
 317
 318    /* Set shadow function:
 319     */
 320    if (gl_sampler->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB) {
 321       /* Shadowing is "enabled" by emitting a particular sampler
 322        * message (sample_c).  So need to recompile WM program when
 323        * shadow comparison is enabled on each/any texture unit.
 324        */
 325       sampler->ss0.shadow_function =
 326          intel_translate_shadow_compare_func(gl_sampler->CompareFunc);
 327    }
 328
 329    /* Set LOD bias:
 330     */
 331    sampler->ss0.lod_bias = S_FIXED(CLAMP(texUnit->LodBias +
 332                                          gl_sampler->LodBias, -16, 15), 6);
 333
 334    sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
 335    sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
 336
 337    sampler->ss0.base_level = U_FIXED(0, 1);
 338
 339    sampler->ss1.max_lod = U_FIXED(CLAMP(gl_sampler->MaxLod, 0, 13), 6);
 340    sampler->ss1.min_lod = U_FIXED(CLAMP(gl_sampler->MinLod, 0, 13), 6);
 341
 342    /* On Gen6+, the sampler can handle non-normalized texture
 343     * rectangle coordinates natively
 344     */
 345    if (brw->gen >= 6 && texObj->Target == GL_TEXTURE_RECTANGLE) {
 346       sampler->ss3.non_normalized_coord = 1;
 347    }
 348
 349    uint32_t sdc_offset;
 350    upload_default_color(brw, gl_sampler, unit, &sdc_offset);
 351
 352    if (brw->gen >= 6) {
 353       sampler->ss2.default_color_pointer = sdc_offset >> 5;
 354    } else {
 355       /* reloc */
 356       sampler->ss2.default_color_pointer =
 357          (brw->batch.bo->offset64 + sdc_offset) >> 5;
 358
 359       drm_intel_bo_emit_reloc(brw->batch.bo,
 360                               batch_offset_for_sampler_state +
 361                               offsetof(struct brw_sampler_state, ss2),
 362                               brw->batch.bo, sdc_offset,
 363                               I915_GEM_DOMAIN_SAMPLER, 0);
 364    }
 365
 366    if (sampler->ss0.min_filter != BRW_MAPFILTER_NEAREST)
 367       sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
 368                                     BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
 369                                     BRW_ADDRESS_ROUNDING_ENABLE_R_MIN;
 370    if (sampler->ss0.mag_filter != BRW_MAPFILTER_NEAREST)
 371       sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
 372                                     BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
 373                                     BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
 374 }
 375
 376
 377 static void
 378 brw_upload_sampler_state_table(struct brw_context *brw,
 379                                struct gl_program *prog,
 380                                struct brw_stage_state *stage_state)
 381 {
 382    struct gl_context *ctx = &brw->ctx;
 383    uint32_t sampler_count = stage_state->sampler_count;
 384
 385    GLbitfield SamplersUsed = prog->SamplersUsed;
 386
 387    if (sampler_count == 0)
 388       return;
 389
 390    /* SAMPLER_STATE is 4 DWords on all platforms. */
 391    const int dwords = 4;
 392    const int size_in_bytes = dwords * sizeof(uint32_t);
 393
 394    uint32_t *sampler_state = brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE,
 395                                              sampler_count * size_in_bytes,
 396                                              32, &stage_state->sampler_offset);
 397    memset(sampler_state, 0, sampler_count * size_in_bytes);
 398
 399    uint32_t batch_offset_for_sampler_state = stage_state->sampler_offset;
 400
 401    for (unsigned s = 0; s < sampler_count; s++) {
 402       if (SamplersUsed & (1 << s)) {
 403          const unsigned unit = prog->SamplerUnits[s];
 404          if (ctx->Texture.Unit[unit]._Current) {
 405             brw_update_sampler_state(brw, unit,
 406                                      (struct brw_sampler_state *) sampler_state,
 407                                      batch_offset_for_sampler_state);
 408          }
 409       }
 410
 411       sampler_state += dwords;
 412       batch_offset_for_sampler_state += size_in_bytes;
 413    }
 414
 415    brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
 416 }
 417
 418 static void
 419 brw_upload_fs_samplers(struct brw_context *brw)
 420 {
 421    /* BRW_NEW_FRAGMENT_PROGRAM */
 422    struct gl_program *fs = (struct gl_program *) brw->fragment_program;
 423    brw->vtbl.upload_sampler_state_table(brw, fs, &brw->wm.base);
 424 }
 425
 426 const struct brw_tracked_state brw_fs_samplers = {
 427    .dirty = {
 428       .mesa = _NEW_TEXTURE,
 429       .brw = BRW_NEW_BATCH |
 430              BRW_NEW_FRAGMENT_PROGRAM,
 431       .cache = 0
 432    },
 433    .emit = brw_upload_fs_samplers,
 434 };
 435
 436 static void
 437 brw_upload_vs_samplers(struct brw_context *brw)
 438 {
 439    /* BRW_NEW_VERTEX_PROGRAM */
 440    struct gl_program *vs = (struct gl_program *) brw->vertex_program;
 441    brw->vtbl.upload_sampler_state_table(brw, vs, &brw->vs.base);
 442 }
 443
 444
 445 const struct brw_tracked_state brw_vs_samplers = {
 446    .dirty = {
 447       .mesa = _NEW_TEXTURE,
 448       .brw = BRW_NEW_BATCH |
 449              BRW_NEW_VERTEX_PROGRAM,
 450       .cache = 0
 451    },
 452    .emit = brw_upload_vs_samplers,
 453 };
 454
 455
 456 static void
 457 brw_upload_gs_samplers(struct brw_context *brw)
 458 {
 459    /* BRW_NEW_GEOMETRY_PROGRAM */
 460    struct gl_program *gs = (struct gl_program *) brw->geometry_program;
 461    if (!gs)
 462       return;
 463
 464    brw->vtbl.upload_sampler_state_table(brw, gs, &brw->gs.base);
 465 }
 466
 467
 468 const struct brw_tracked_state brw_gs_samplers = {
 469    .dirty = {
 470       .mesa = _NEW_TEXTURE,
 471       .brw = BRW_NEW_BATCH |
 472              BRW_NEW_GEOMETRY_PROGRAM,
 473       .cache = 0
 474    },
 475    .emit = brw_upload_gs_samplers,
 476 };
 477
 478
 479 void
 480 gen4_init_vtable_sampler_functions(struct brw_context *brw)
 481 {
 482    brw->vtbl.upload_sampler_state_table = brw_upload_sampler_state_table;
 483 }