src/gallium/drivers/svga/svga_pipe_sampler.c

   1 /**********************************************************
   2  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   3  *
   4  * Permission is hereby granted, free of charge, to any person
   5  * obtaining a copy of this software and associated documentation
   6  * files (the "Software"), to deal in the Software without
   7  * restriction, including without limitation the rights to use, copy,
   8  * modify, merge, publish, distribute, sublicense, and/or sell copies
   9  * of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  **********************************************************/
  25
  26 #include "pipe/p_defines.h"
  27 #include "util/u_bitmask.h"
  28 #include "util/u_format.h"
  29 #include "util/u_inlines.h"
  30 #include "util/u_math.h"
  31 #include "util/u_memory.h"
  32 #include "tgsi/tgsi_parse.h"
  33
  34 #include "svga_context.h"
  35 #include "svga_cmd.h"
  36 #include "svga_debug.h"
  37 #include "svga_resource_texture.h"
  38 #include "svga_surface.h"
  39 #include "svga_sampler_view.h"
  40
  41
  42 static inline unsigned
  43 translate_wrap_mode(unsigned wrap)
  44 {
  45    switch (wrap) {
  46    case PIPE_TEX_WRAP_REPEAT:
  47       return SVGA3D_TEX_ADDRESS_WRAP;
  48    case PIPE_TEX_WRAP_CLAMP:
  49       return SVGA3D_TEX_ADDRESS_CLAMP;
  50    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
  51       /* Unfortunately SVGA3D_TEX_ADDRESS_EDGE not respected by
  52        * hardware.
  53        */
  54       return SVGA3D_TEX_ADDRESS_CLAMP;
  55    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
  56       return SVGA3D_TEX_ADDRESS_BORDER;
  57    case PIPE_TEX_WRAP_MIRROR_REPEAT:
  58       return SVGA3D_TEX_ADDRESS_MIRROR;
  59    case PIPE_TEX_WRAP_MIRROR_CLAMP:
  60    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
  61    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
  62       return SVGA3D_TEX_ADDRESS_MIRRORONCE;
  63    default:
  64       assert(0);
  65       return SVGA3D_TEX_ADDRESS_WRAP;
  66    }
  67 }
  68
  69
  70 static inline unsigned
  71 translate_img_filter(unsigned filter)
  72 {
  73    switch (filter) {
  74    case PIPE_TEX_FILTER_NEAREST:
  75       return SVGA3D_TEX_FILTER_NEAREST;
  76    case PIPE_TEX_FILTER_LINEAR:
  77       return SVGA3D_TEX_FILTER_LINEAR;
  78    default:
  79       assert(0);
  80       return SVGA3D_TEX_FILTER_NEAREST;
  81    }
  82 }
  83
  84
  85 static inline unsigned
  86 translate_mip_filter(unsigned filter)
  87 {
  88    switch (filter) {
  89    case PIPE_TEX_MIPFILTER_NONE:
  90       return SVGA3D_TEX_FILTER_NONE;
  91    case PIPE_TEX_MIPFILTER_NEAREST:
  92       return SVGA3D_TEX_FILTER_NEAREST;
  93    case PIPE_TEX_MIPFILTER_LINEAR:
  94       return SVGA3D_TEX_FILTER_LINEAR;
  95    default:
  96       assert(0);
  97       return SVGA3D_TEX_FILTER_NONE;
  98    }
  99 }
 100
 101
 102 static uint8
 103 translate_comparison_func(unsigned func)
 104 {
 105    switch (func) {
 106    case PIPE_FUNC_NEVER:
 107       return SVGA3D_COMPARISON_NEVER;
 108    case PIPE_FUNC_LESS:
 109       return SVGA3D_COMPARISON_LESS;
 110    case PIPE_FUNC_EQUAL:
 111       return SVGA3D_COMPARISON_EQUAL;
 112    case PIPE_FUNC_LEQUAL:
 113       return SVGA3D_COMPARISON_LESS_EQUAL;
 114    case PIPE_FUNC_GREATER:
 115       return SVGA3D_COMPARISON_GREATER;
 116    case PIPE_FUNC_NOTEQUAL:
 117       return SVGA3D_COMPARISON_NOT_EQUAL;
 118    case PIPE_FUNC_GEQUAL:
 119       return SVGA3D_COMPARISON_GREATER_EQUAL;
 120    case PIPE_FUNC_ALWAYS:
 121       return SVGA3D_COMPARISON_ALWAYS;
 122    default:
 123       assert(!"Invalid comparison function");
 124       return SVGA3D_COMPARISON_ALWAYS;
 125    }
 126 }
 127
 128
 129 /**
 130  * Translate filtering state to vgpu10 format.
 131  */
 132 static SVGA3dFilter
 133 translate_filter_mode(unsigned img_filter,
 134                       unsigned min_filter,
 135                       unsigned mag_filter,
 136                       boolean anisotropic,
 137                       boolean compare)
 138 {
 139    SVGA3dFilter mode = 0;
 140
 141    if (img_filter == PIPE_TEX_FILTER_LINEAR)
 142       mode |= SVGA3D_FILTER_MIP_LINEAR;
 143    if (min_filter == PIPE_TEX_FILTER_LINEAR)
 144       mode |= SVGA3D_FILTER_MIN_LINEAR;
 145    if (mag_filter == PIPE_TEX_FILTER_LINEAR)
 146       mode |= SVGA3D_FILTER_MAG_LINEAR;
 147    if (anisotropic)
 148       mode |= SVGA3D_FILTER_ANISOTROPIC;
 149    if (compare)
 150       mode |= SVGA3D_FILTER_COMPARE;
 151
 152    return mode;
 153 }
 154
 155
 156 /**
 157  * Define a vgpu10 sampler state.
 158  */
 159 static void
 160 define_sampler_state_object(struct svga_context *svga,
 161                             struct svga_sampler_state *ss,
 162                             const struct pipe_sampler_state *ps)
 163 {
 164    uint8_t max_aniso = (uint8_t) 255; /* XXX fix me */
 165    boolean anisotropic;
 166    uint8 compare_func;
 167    SVGA3dFilter filter;
 168    SVGA3dRGBAFloat bcolor;
 169    unsigned try;
 170    float min_lod, max_lod;
 171
 172    assert(svga_have_vgpu10(svga));
 173
 174    anisotropic = ss->aniso_level > 1.0f;
 175
 176    filter = translate_filter_mode(ps->min_mip_filter,
 177                                   ps->min_img_filter,
 178                                   ps->mag_img_filter,
 179                                   anisotropic,
 180                                   ss->compare_mode);
 181
 182    compare_func = translate_comparison_func(ss->compare_func);
 183
 184    COPY_4V(bcolor.value, ps->border_color.f);
 185
 186    ss->id = util_bitmask_add(svga->sampler_object_id_bm);
 187
 188    assert(ps->min_lod <= ps->max_lod);
 189
 190    if (ps->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
 191       /* just use the base level image */
 192       min_lod = max_lod = 0.0f;
 193    }
 194    else {
 195       min_lod = ps->min_lod;
 196       max_lod = ps->max_lod;
 197    }
 198
 199    /* Loop in case command buffer is full and we need to flush and retry */
 200    for (try = 0; try < 2; try++) {
 201       enum pipe_error ret =
 202          SVGA3D_vgpu10_DefineSamplerState(svga->swc,
 203                                           ss->id,
 204                                           filter,
 205                                           ss->addressu,
 206                                           ss->addressv,
 207                                           ss->addressw,
 208                                           ss->lod_bias, /* float */
 209                                           max_aniso,
 210                                           compare_func,
 211                                           bcolor,
 212                                           min_lod,       /* float */
 213                                           max_lod);      /* float */
 214       if (ret == PIPE_OK)
 215          return;
 216       svga_context_flush(svga, NULL);
 217    }
 218 }
 219
 220
 221 static void *
 222 svga_create_sampler_state(struct pipe_context *pipe,
 223                           const struct pipe_sampler_state *sampler)
 224 {
 225    struct svga_context *svga = svga_context(pipe);
 226    struct svga_sampler_state *cso = CALLOC_STRUCT( svga_sampler_state );
 227
 228    if (!cso)
 229       return NULL;
 230
 231    cso->mipfilter = translate_mip_filter(sampler->min_mip_filter);
 232    cso->magfilter = translate_img_filter( sampler->mag_img_filter );
 233    cso->minfilter = translate_img_filter( sampler->min_img_filter );
 234    cso->aniso_level = MAX2( sampler->max_anisotropy, 1 );
 235    if (sampler->max_anisotropy)
 236       cso->magfilter = cso->minfilter = SVGA3D_TEX_FILTER_ANISOTROPIC;
 237    cso->lod_bias = sampler->lod_bias;
 238    cso->addressu = translate_wrap_mode(sampler->wrap_s);
 239    cso->addressv = translate_wrap_mode(sampler->wrap_t);
 240    cso->addressw = translate_wrap_mode(sampler->wrap_r);
 241    cso->normalized_coords = sampler->normalized_coords;
 242    cso->compare_mode = sampler->compare_mode;
 243    cso->compare_func = sampler->compare_func;
 244
 245    {
 246       uint32 r = float_to_ubyte(sampler->border_color.f[0]);
 247       uint32 g = float_to_ubyte(sampler->border_color.f[1]);
 248       uint32 b = float_to_ubyte(sampler->border_color.f[2]);
 249       uint32 a = float_to_ubyte(sampler->border_color.f[3]);
 250
 251       cso->bordercolor = (a << 24) | (r << 16) | (g << 8) | b;
 252    }
 253
 254    /* No SVGA3D support for:
 255     *    - min/max LOD clamping
 256     */
 257    cso->min_lod = 0;
 258    cso->view_min_lod = MAX2((int) (sampler->min_lod + 0.5), 0);
 259    cso->view_max_lod = MAX2((int) (sampler->max_lod + 0.5), 0);
 260
 261    /* Use min_mipmap */
 262    if (svga->debug.use_min_mipmap) {
 263       if (cso->view_min_lod == cso->view_max_lod) {
 264          cso->min_lod = cso->view_min_lod;
 265          cso->view_min_lod = 0;
 266          cso->view_max_lod = 1000; /* Just a high number */
 267          cso->mipfilter = SVGA3D_TEX_FILTER_NONE;
 268       }
 269    }
 270
 271    if (svga_have_vgpu10(svga)) {
 272       define_sampler_state_object(svga, cso, sampler);
 273    }
 274
 275    SVGA_DBG(DEBUG_VIEWS, "min %u, view(min %u, max %u) lod, mipfilter %s\n",
 276             cso->min_lod, cso->view_min_lod, cso->view_max_lod,
 277             cso->mipfilter == SVGA3D_TEX_FILTER_NONE ? "SVGA3D_TEX_FILTER_NONE" : "SOMETHING");
 278
 279    svga->hud.num_sampler_objects++;
 280
 281    return cso;
 282 }
 283
 284
 285 static void
 286 svga_bind_sampler_states(struct pipe_context *pipe,
 287                          unsigned shader,
 288                          unsigned start,
 289                          unsigned num,
 290                          void **samplers)
 291 {
 292    struct svga_context *svga = svga_context(pipe);
 293    unsigned i;
 294    boolean any_change = FALSE;
 295
 296    assert(shader < PIPE_SHADER_TYPES);
 297    assert(start + num <= PIPE_MAX_SAMPLERS);
 298
 299    /* Pre-VGPU10 only supports FS textures */
 300    if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
 301       return;
 302
 303    for (i = 0; i < num; i++) {
 304       if (svga->curr.sampler[shader][start + i] != samplers[i])
 305          any_change = TRUE;
 306       svga->curr.sampler[shader][start + i] = samplers[i];
 307    }
 308
 309    if (!any_change) {
 310       return;
 311    }
 312
 313    /* find highest non-null sampler[] entry */
 314    {
 315       unsigned j = MAX2(svga->curr.num_samplers[shader], start + num);
 316       while (j > 0 && svga->curr.sampler[shader][j - 1] == NULL)
 317          j--;
 318       svga->curr.num_samplers[shader] = j;
 319    }
 320
 321    svga->dirty |= SVGA_NEW_SAMPLER;
 322 }
 323
 324
 325 static void
 326 svga_delete_sampler_state(struct pipe_context *pipe, void *sampler)
 327 {
 328    struct svga_sampler_state *ss = (struct svga_sampler_state *) sampler;
 329    struct svga_context *svga = svga_context(pipe);
 330
 331    if (svga_have_vgpu10(svga)) {
 332       enum pipe_error ret;
 333
 334       svga_hwtnl_flush_retry(svga);
 335
 336       ret = SVGA3D_vgpu10_DestroySamplerState(svga->swc, ss->id);
 337       if (ret != PIPE_OK) {
 338          svga_context_flush(svga, NULL);
 339          ret = SVGA3D_vgpu10_DestroySamplerState(svga->swc, ss->id);
 340       }
 341       util_bitmask_clear(svga->sampler_object_id_bm, ss->id);
 342    }
 343
 344    FREE(sampler);
 345    svga->hud.num_sampler_objects--;
 346 }
 347
 348
 349 static struct pipe_sampler_view *
 350 svga_create_sampler_view(struct pipe_context *pipe,
 351                          struct pipe_resource *texture,
 352                          const struct pipe_sampler_view *templ)
 353 {
 354    struct svga_context *svga = svga_context(pipe);
 355    struct svga_pipe_sampler_view *sv = CALLOC_STRUCT(svga_pipe_sampler_view);
 356
 357    if (!sv) {
 358       return NULL;
 359    }
 360
 361    sv->base = *templ;
 362    sv->base.reference.count = 1;
 363    sv->base.texture = NULL;
 364    pipe_resource_reference(&sv->base.texture, texture);
 365
 366    sv->base.context = pipe;
 367    sv->id = SVGA3D_INVALID_ID;
 368
 369    svga->hud.num_samplerview_objects++;
 370
 371    return &sv->base;
 372 }
 373
 374
 375 static void
 376 svga_sampler_view_destroy(struct pipe_context *pipe,
 377                           struct pipe_sampler_view *view)
 378 {
 379    struct svga_context *svga = svga_context(pipe);
 380    struct svga_pipe_sampler_view *sv = svga_pipe_sampler_view(view);
 381
 382    if (svga_have_vgpu10(svga) && sv->id != SVGA3D_INVALID_ID) {
 383       if (view->context != pipe) {
 384          /* The SVGA3D device will generate an error (and on Linux, cause
 385           * us to abort) if we try to destroy a shader resource view from
 386           * a context other than the one it was created with.  Skip the
 387           * SVGA3D_vgpu10_DestroyShaderResourceView() and leak the sampler
 388           * view for now.  This should only sometimes happen when a shared
 389           * texture is deleted.
 390           */
 391          _debug_printf("context mismatch in %s\n", __func__);
 392       }
 393       else {
 394          enum pipe_error ret;
 395
 396          svga_hwtnl_flush_retry(svga); /* XXX is this needed? */
 397
 398          ret = SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc, sv->id);
 399          if (ret != PIPE_OK) {
 400             svga_context_flush(svga, NULL);
 401             ret = SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc, sv->id);
 402          }
 403          util_bitmask_clear(svga->sampler_view_id_bm, sv->id);
 404       }
 405    }
 406
 407    pipe_resource_reference(&sv->base.texture, NULL);
 408
 409    FREE(sv);
 410    svga->hud.num_samplerview_objects--;
 411 }
 412
 413
 414 static void
 415 svga_set_sampler_views(struct pipe_context *pipe,
 416                        unsigned shader,
 417                        unsigned start,
 418                        unsigned num,
 419                        struct pipe_sampler_view **views)
 420 {
 421    struct svga_context *svga = svga_context(pipe);
 422    unsigned flag_1d = 0;
 423    unsigned flag_srgb = 0;
 424    unsigned flag_rect = 0;
 425    unsigned flag_buf = 0;
 426    uint i;
 427    boolean any_change = FALSE;
 428
 429    assert(shader < PIPE_SHADER_TYPES);
 430    assert(start + num <= ARRAY_SIZE(svga->curr.sampler_views[shader]));
 431
 432    /* Pre-VGPU10 only supports FS textures */
 433    if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
 434       return;
 435
 436    /* This bit of code works around a quirk in the CSO module.
 437     * If start=num=0 it means all sampler views should be released.
 438     * Note that the CSO module treats sampler views for fragment shaders
 439     * differently than other shader types.
 440     */
 441    if (start == 0 && num == 0 && svga->curr.num_sampler_views[shader] > 0) {
 442       for (i = 0; i < svga->curr.num_sampler_views[shader]; i++) {
 443          pipe_sampler_view_release(pipe, &svga->curr.sampler_views[shader][i]);
 444       }
 445       any_change = TRUE;
 446    }
 447
 448    for (i = 0; i < num; i++) {
 449       enum pipe_texture_target target;
 450
 451       if (svga->curr.sampler_views[shader][start + i] != views[i]) {
 452          /* Note: we're using pipe_sampler_view_release() here to work around
 453           * a possible crash when the old view belongs to another context that
 454           * was already destroyed.
 455           */
 456          pipe_sampler_view_release(pipe, &svga->curr.sampler_views[shader][start + i]);
 457          pipe_sampler_view_reference(&svga->curr.sampler_views[shader][start + i],
 458                                      views[i]);
 459          any_change = TRUE;
 460       }
 461
 462       if (!views[i])
 463          continue;
 464
 465       if (util_format_is_srgb(views[i]->format))
 466          flag_srgb |= 1 << (start + i);
 467
 468       target = views[i]->texture->target;
 469       if (target == PIPE_TEXTURE_1D)
 470          flag_1d |= 1 << (start + i);
 471       else if (target == PIPE_TEXTURE_RECT)
 472          flag_rect |= 1 << (start + i);
 473       else if (target == PIPE_BUFFER)
 474          flag_buf |= 1 << (start + i);
 475    }
 476
 477    if (!any_change) {
 478       return;
 479    }
 480
 481    /* find highest non-null sampler_views[] entry */
 482    {
 483       unsigned j = MAX2(svga->curr.num_sampler_views[shader], start + num);
 484       while (j > 0 && svga->curr.sampler_views[shader][j - 1] == NULL)
 485          j--;
 486       svga->curr.num_sampler_views[shader] = j;
 487    }
 488
 489    svga->dirty |= SVGA_NEW_TEXTURE_BINDING;
 490
 491    if (flag_srgb != svga->curr.tex_flags.flag_srgb ||
 492        flag_1d != svga->curr.tex_flags.flag_1d) {
 493       svga->dirty |= SVGA_NEW_TEXTURE_FLAGS;
 494       svga->curr.tex_flags.flag_1d = flag_1d;
 495       svga->curr.tex_flags.flag_srgb = flag_srgb;
 496    }
 497
 498    if (flag_rect != svga->curr.tex_flags.flag_rect ||
 499        flag_buf != svga->curr.tex_flags.flag_buf)
 500    {
 501       /* Need to re-emit texture constants */
 502       svga->dirty |= SVGA_NEW_TEXTURE_CONSTS;
 503       svga->curr.tex_flags.flag_rect = flag_rect;
 504       svga->curr.tex_flags.flag_buf = flag_buf;
 505    }
 506
 507    /* Check if any of the sampler view resources collide with the framebuffer
 508     * color buffers or depth stencil resource. If so, enable the NEW_FRAME_BUFFER
 509     * dirty bit so that emit_framebuffer can be invoked to create backed view
 510     * for the conflicted surface view.
 511     */
 512    for (i = 0; i < svga->curr.framebuffer.nr_cbufs; i++) {
 513       if (svga->curr.framebuffer.cbufs[i]) {
 514          struct svga_surface *s = svga_surface(svga->curr.framebuffer.cbufs[i]);
 515          if (svga_check_sampler_view_resource_collision(svga, s->handle, shader)) {
 516             svga->dirty |= SVGA_NEW_FRAME_BUFFER;
 517             break;
 518          }
 519       }
 520    }
 521
 522    if (svga->curr.framebuffer.zsbuf) {
 523       struct svga_surface *s = svga_surface(svga->curr.framebuffer.zsbuf);
 524       if (s) {
 525          if (svga_check_sampler_view_resource_collision(svga, s->handle, shader)) {
 526             svga->dirty |= SVGA_NEW_FRAME_BUFFER;
 527          }
 528       }
 529    }
 530 }
 531
 532
 533 void
 534 svga_init_sampler_functions( struct svga_context *svga )
 535 {
 536    svga->pipe.create_sampler_state = svga_create_sampler_state;
 537    svga->pipe.bind_sampler_states = svga_bind_sampler_states;
 538    svga->pipe.delete_sampler_state = svga_delete_sampler_state;
 539    svga->pipe.set_sampler_views = svga_set_sampler_views;
 540    svga->pipe.create_sampler_view = svga_create_sampler_view;
 541    svga->pipe.sampler_view_destroy = svga_sampler_view_destroy;
 542 }