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
  49    case PIPE_TEX_WRAP_CLAMP:
  50       return SVGA3D_TEX_ADDRESS_CLAMP;
  51
  52    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
  53       /* Unfortunately SVGA3D_TEX_ADDRESS_EDGE not respected by
  54        * hardware.
  55        */
  56       return SVGA3D_TEX_ADDRESS_CLAMP;
  57
  58    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
  59       return SVGA3D_TEX_ADDRESS_BORDER;
  60
  61    case PIPE_TEX_WRAP_MIRROR_REPEAT:
  62       return SVGA3D_TEX_ADDRESS_MIRROR;
  63
  64    case PIPE_TEX_WRAP_MIRROR_CLAMP:
  65    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
  66    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
  67       return SVGA3D_TEX_ADDRESS_MIRRORONCE;
  68
  69    default:
  70       assert(0);
  71       return SVGA3D_TEX_ADDRESS_WRAP;
  72    }
  73 }
  74
  75 static inline unsigned translate_img_filter( unsigned filter )
  76 {
  77    switch (filter) {
  78    case PIPE_TEX_FILTER_NEAREST: return SVGA3D_TEX_FILTER_NEAREST;
  79    case PIPE_TEX_FILTER_LINEAR:  return SVGA3D_TEX_FILTER_LINEAR;
  80    default:
  81       assert(0);
  82       return SVGA3D_TEX_FILTER_NEAREST;
  83    }
  84 }
  85
  86 static inline unsigned translate_mip_filter( unsigned filter )
  87 {
  88    switch (filter) {
  89    case PIPE_TEX_MIPFILTER_NONE:    return SVGA3D_TEX_FILTER_NONE;
  90    case PIPE_TEX_MIPFILTER_NEAREST: return SVGA3D_TEX_FILTER_NEAREST;
  91    case PIPE_TEX_MIPFILTER_LINEAR:  return SVGA3D_TEX_FILTER_LINEAR;
  92    default:
  93       assert(0);
  94       return SVGA3D_TEX_FILTER_NONE;
  95    }
  96 }
  97
  98
  99 static uint8
 100 translate_comparison_func(unsigned func)
 101 {
 102    switch (func) {
 103    case PIPE_FUNC_NEVER:
 104       return SVGA3D_COMPARISON_NEVER;
 105    case PIPE_FUNC_LESS:
 106       return SVGA3D_COMPARISON_LESS;
 107    case PIPE_FUNC_EQUAL:
 108       return SVGA3D_COMPARISON_EQUAL;
 109    case PIPE_FUNC_LEQUAL:
 110       return SVGA3D_COMPARISON_LESS_EQUAL;
 111    case PIPE_FUNC_GREATER:
 112       return SVGA3D_COMPARISON_GREATER;
 113    case PIPE_FUNC_NOTEQUAL:
 114       return SVGA3D_COMPARISON_NOT_EQUAL;
 115    case PIPE_FUNC_GEQUAL:
 116       return SVGA3D_COMPARISON_GREATER_EQUAL;
 117    case PIPE_FUNC_ALWAYS:
 118       return SVGA3D_COMPARISON_ALWAYS;
 119    default:
 120       assert(!"Invalid comparison function");
 121       return SVGA3D_COMPARISON_ALWAYS;
 122    }
 123 }
 124
 125
 126 /**
 127  * Translate filtering state to vgpu10 format.
 128  */
 129 static SVGA3dFilter
 130 translate_filter_mode(unsigned img_filter,
 131                       unsigned min_filter,
 132                       unsigned mag_filter,
 133                       boolean anisotropic,
 134                       boolean compare)
 135 {
 136    SVGA3dFilter mode = 0;
 137
 138    if (img_filter == PIPE_TEX_FILTER_LINEAR)
 139       mode |= SVGA3D_FILTER_MIP_LINEAR;
 140    if (min_filter == PIPE_TEX_FILTER_LINEAR)
 141       mode |= SVGA3D_FILTER_MIN_LINEAR;
 142    if (mag_filter == PIPE_TEX_FILTER_LINEAR)
 143       mode |= SVGA3D_FILTER_MAG_LINEAR;
 144    if (anisotropic)
 145       mode |= SVGA3D_FILTER_ANISOTROPIC;
 146    if (compare)
 147       mode |= SVGA3D_FILTER_COMPARE;
 148
 149    return mode;
 150 }
 151
 152
 153 /**
 154  * Define a vgpu10 sampler state.
 155  */
 156 static void
 157 define_sampler_state_object(struct svga_context *svga,
 158                             struct svga_sampler_state *ss,
 159                             const struct pipe_sampler_state *ps)
 160 {
 161    uint8_t max_aniso = (uint8_t) 255; /* XXX fix me */
 162    boolean anisotropic;
 163    uint8 compare_func;
 164    SVGA3dFilter filter;
 165    SVGA3dRGBAFloat bcolor;
 166    unsigned try;
 167    float min_lod, max_lod;
 168
 169    assert(svga_have_vgpu10(svga));
 170
 171    anisotropic = ss->aniso_level > 1.0f;
 172
 173    filter = translate_filter_mode(ps->min_mip_filter,
 174                                   ps->min_img_filter,
 175                                   ps->mag_img_filter,
 176                                   anisotropic,
 177                                   ss->compare_mode);
 178
 179    compare_func = translate_comparison_func(ss->compare_func);
 180
 181    COPY_4V(bcolor.value, ps->border_color.f);
 182
 183    ss->id = util_bitmask_add(svga->sampler_object_id_bm);
 184
 185    assert(ps->min_lod <= ps->max_lod);
 186
 187    if (ps->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
 188       /* just use the base level image */
 189       min_lod = max_lod = 0.0f;
 190    }
 191    else {
 192       min_lod = ps->min_lod;
 193       max_lod = ps->max_lod;
 194    }
 195
 196    /* Loop in case command buffer is full and we need to flush and retry */
 197    for (try = 0; try < 2; try++) {
 198       enum pipe_error ret =
 199          SVGA3D_vgpu10_DefineSamplerState(svga->swc,
 200                                           ss->id,
 201                                           filter,
 202                                           ss->addressu,
 203                                           ss->addressv,
 204                                           ss->addressw,
 205                                           ss->lod_bias, /* float */
 206                                           max_aniso,
 207                                           compare_func,
 208                                           bcolor,
 209                                           min_lod,       /* float */
 210                                           max_lod);      /* float */
 211       if (ret == PIPE_OK)
 212          return;
 213       svga_context_flush(svga, NULL);
 214    }
 215 }
 216
 217
 218 static void *
 219 svga_create_sampler_state(struct pipe_context *pipe,
 220                           const struct pipe_sampler_state *sampler)
 221 {
 222    struct svga_context *svga = svga_context(pipe);
 223    struct svga_sampler_state *cso = CALLOC_STRUCT( svga_sampler_state );
 224
 225    if (!cso)
 226       return NULL;
 227
 228    cso->mipfilter = translate_mip_filter(sampler->min_mip_filter);
 229    cso->magfilter = translate_img_filter( sampler->mag_img_filter );
 230    cso->minfilter = translate_img_filter( sampler->min_img_filter );
 231    cso->aniso_level = MAX2( sampler->max_anisotropy, 1 );
 232    if(sampler->max_anisotropy)
 233       cso->magfilter = cso->minfilter = SVGA3D_TEX_FILTER_ANISOTROPIC;
 234    cso->lod_bias = sampler->lod_bias;
 235    cso->addressu = translate_wrap_mode(sampler->wrap_s);
 236    cso->addressv = translate_wrap_mode(sampler->wrap_t);
 237    cso->addressw = translate_wrap_mode(sampler->wrap_r);
 238    cso->normalized_coords = sampler->normalized_coords;
 239    cso->compare_mode = sampler->compare_mode;
 240    cso->compare_func = sampler->compare_func;
 241
 242    {
 243       uint32 r = float_to_ubyte(sampler->border_color.f[0]);
 244       uint32 g = float_to_ubyte(sampler->border_color.f[1]);
 245       uint32 b = float_to_ubyte(sampler->border_color.f[2]);
 246       uint32 a = float_to_ubyte(sampler->border_color.f[3]);
 247
 248       cso->bordercolor = (a << 24) | (r << 16) | (g << 8) | b;
 249    }
 250
 251    /* No SVGA3D support for:
 252     *    - min/max LOD clamping
 253     */
 254    cso->min_lod = 0;
 255    cso->view_min_lod = MAX2((int) (sampler->min_lod + 0.5), 0);
 256    cso->view_max_lod = MAX2((int) (sampler->max_lod + 0.5), 0);
 257
 258    /* Use min_mipmap */
 259    if (svga->debug.use_min_mipmap) {
 260       if (cso->view_min_lod == cso->view_max_lod) {
 261          cso->min_lod = cso->view_min_lod;
 262          cso->view_min_lod = 0;
 263          cso->view_max_lod = 1000; /* Just a high number */
 264          cso->mipfilter = SVGA3D_TEX_FILTER_NONE;
 265       }
 266    }
 267
 268    if (svga_have_vgpu10(svga)) {
 269       define_sampler_state_object(svga, cso, sampler);
 270    }
 271
 272    SVGA_DBG(DEBUG_VIEWS, "min %u, view(min %u, max %u) lod, mipfilter %s\n",
 273             cso->min_lod, cso->view_min_lod, cso->view_max_lod,
 274             cso->mipfilter == SVGA3D_TEX_FILTER_NONE ? "SVGA3D_TEX_FILTER_NONE" : "SOMETHING");
 275
 276    return cso;
 277 }
 278
 279 static void
 280 svga_bind_sampler_states(struct pipe_context *pipe,
 281                          unsigned shader,
 282                          unsigned start,
 283                          unsigned num,
 284                          void **samplers)
 285 {
 286    struct svga_context *svga = svga_context(pipe);
 287    unsigned i;
 288
 289    assert(shader < PIPE_SHADER_TYPES);
 290    assert(start + num <= PIPE_MAX_SAMPLERS);
 291
 292    /* Pre-VGPU10 only supports FS textures */
 293    if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
 294       return;
 295
 296    for (i = 0; i < num; i++)
 297       svga->curr.sampler[shader][start + i] = samplers[i];
 298
 299    /* find highest non-null sampler[] entry */
 300    {
 301       unsigned j = MAX2(svga->curr.num_samplers[shader], start + num);
 302       while (j > 0 && svga->curr.sampler[shader][j - 1] == NULL)
 303          j--;
 304       svga->curr.num_samplers[shader] = j;
 305    }
 306
 307    svga->dirty |= SVGA_NEW_SAMPLER;
 308 }
 309
 310
 311 static void svga_delete_sampler_state(struct pipe_context *pipe,
 312                                       void *sampler)
 313 {
 314    struct svga_sampler_state *ss = (struct svga_sampler_state *) sampler;
 315    struct svga_context *svga = svga_context(pipe);
 316
 317    if (svga_have_vgpu10(svga)) {
 318       enum pipe_error ret;
 319
 320       svga_hwtnl_flush_retry(svga);
 321
 322       ret = SVGA3D_vgpu10_DestroySamplerState(svga->swc, ss->id);
 323       if (ret != PIPE_OK) {
 324          svga_context_flush(svga, NULL);
 325          ret = SVGA3D_vgpu10_DestroySamplerState(svga->swc, ss->id);
 326       }
 327       util_bitmask_clear(svga->sampler_object_id_bm, ss->id);
 328    }
 329
 330    FREE(sampler);
 331 }
 332
 333
 334 static struct pipe_sampler_view *
 335 svga_create_sampler_view(struct pipe_context *pipe,
 336                          struct pipe_resource *texture,
 337                          const struct pipe_sampler_view *templ)
 338 {
 339    struct svga_pipe_sampler_view *sv = CALLOC_STRUCT(svga_pipe_sampler_view);
 340
 341    if (!sv) {
 342       return NULL;
 343    }
 344
 345    sv->base = *templ;
 346    sv->base.reference.count = 1;
 347    sv->base.texture = NULL;
 348    pipe_resource_reference(&sv->base.texture, texture);
 349
 350    sv->base.context = pipe;
 351    sv->id = SVGA3D_INVALID_ID;
 352
 353    return &sv->base;
 354 }
 355
 356
 357 static void
 358 svga_sampler_view_destroy(struct pipe_context *pipe,
 359                           struct pipe_sampler_view *view)
 360 {
 361    struct svga_context *svga = svga_context(pipe);
 362    struct svga_pipe_sampler_view *sv = svga_pipe_sampler_view(view);
 363
 364    if (svga_have_vgpu10(svga) && sv->id != SVGA3D_INVALID_ID) {
 365       if (view->context != pipe) {
 366          /* The SVGA3D device will generate an error (and on Linux, cause
 367           * us to abort) if we try to destroy a shader resource view from
 368           * a context other than the one it was created with.  Skip the
 369           * SVGA3D_vgpu10_DestroyShaderResourceView() and leak the sampler
 370           * view for now.  This should only sometimes happen when a shared
 371           * texture is deleted.
 372           */
 373          _debug_printf("context mismatch in %s\n", __func__);
 374       }
 375       else {
 376          enum pipe_error ret;
 377
 378          svga_hwtnl_flush_retry(svga); /* XXX is this needed? */
 379
 380          ret = SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc, sv->id);
 381          if (ret != PIPE_OK) {
 382             svga_context_flush(svga, NULL);
 383             ret = SVGA3D_vgpu10_DestroyShaderResourceView(svga->swc, sv->id);
 384          }
 385          util_bitmask_clear(svga->sampler_view_id_bm, sv->id);
 386       }
 387    }
 388
 389    pipe_resource_reference(&sv->base.texture, NULL);
 390
 391    FREE(sv);
 392 }
 393
 394 static void
 395 svga_set_sampler_views(struct pipe_context *pipe,
 396                        unsigned shader,
 397                        unsigned start,
 398                        unsigned num,
 399                        struct pipe_sampler_view **views)
 400 {
 401    struct svga_context *svga = svga_context(pipe);
 402    unsigned flag_1d = 0;
 403    unsigned flag_srgb = 0;
 404    uint i;
 405
 406    assert(shader < PIPE_SHADER_TYPES);
 407    assert(start + num <= Elements(svga->curr.sampler_views[shader]));
 408
 409    /* Pre-VGPU10 only supports FS textures */
 410    if (!svga_have_vgpu10(svga) && shader != PIPE_SHADER_FRAGMENT)
 411       return;
 412
 413    for (i = 0; i < num; i++) {
 414       if (svga->curr.sampler_views[shader][start + i] != views[i]) {
 415          /* Note: we're using pipe_sampler_view_release() here to work around
 416           * a possible crash when the old view belongs to another context that
 417           * was already destroyed.
 418           */
 419          pipe_sampler_view_release(pipe, &svga->curr.sampler_views[shader][start + i]);
 420          pipe_sampler_view_reference(&svga->curr.sampler_views[shader][start + i],
 421                                      views[i]);
 422       }
 423
 424       if (!views[i])
 425          continue;
 426
 427       if (util_format_is_srgb(views[i]->format))
 428          flag_srgb |= 1 << (start + i);
 429
 430       if (views[i]->texture->target == PIPE_TEXTURE_1D)
 431          flag_1d |= 1 << (start + i);
 432    }
 433
 434    /* find highest non-null sampler_views[] entry */
 435    {
 436       unsigned j = MAX2(svga->curr.num_sampler_views[shader], start + num);
 437       while (j > 0 && svga->curr.sampler_views[shader][j - 1] == NULL)
 438          j--;
 439       svga->curr.num_sampler_views[shader] = j;
 440    }
 441
 442    svga->dirty |= SVGA_NEW_TEXTURE_BINDING;
 443
 444    if (flag_srgb != svga->curr.tex_flags.flag_srgb ||
 445        flag_1d != svga->curr.tex_flags.flag_1d)
 446    {
 447       svga->dirty |= SVGA_NEW_TEXTURE_FLAGS;
 448       svga->curr.tex_flags.flag_1d = flag_1d;
 449       svga->curr.tex_flags.flag_srgb = flag_srgb;
 450    }
 451
 452    /* Check if any of the sampler view resources collide with the framebuffer
 453     * color buffers or depth stencil resource. If so, enable the NEW_FRAME_BUFFER
 454     * dirty bit so that emit_framebuffer can be invoked to create backed view
 455     * for the conflicted surface view.
 456     */
 457    for (i = 0; i < svga->curr.framebuffer.nr_cbufs; i++) {
 458       if (svga->curr.framebuffer.cbufs[i]) {
 459          struct svga_surface *s = svga_surface(svga->curr.framebuffer.cbufs[i]);
 460          if (svga_check_sampler_view_resource_collision(svga, s->handle, shader)) {
 461             svga->dirty |= SVGA_NEW_FRAME_BUFFER;
 462             break;
 463          }
 464       }
 465    }
 466
 467    if (svga->curr.framebuffer.zsbuf) {
 468       struct svga_surface *s = svga_surface(svga->curr.framebuffer.zsbuf);
 469       if (s) {
 470          if (svga_check_sampler_view_resource_collision(svga, s->handle, shader)) {
 471             svga->dirty |= SVGA_NEW_FRAME_BUFFER;
 472          }
 473       }
 474    }
 475 }
 476
 477
 478 void svga_init_sampler_functions( struct svga_context *svga )
 479 {
 480    svga->pipe.create_sampler_state = svga_create_sampler_state;
 481    svga->pipe.bind_sampler_states = svga_bind_sampler_states;
 482    svga->pipe.delete_sampler_state = svga_delete_sampler_state;
 483    svga->pipe.set_sampler_views = svga_set_sampler_views;
 484    svga->pipe.create_sampler_view = svga_create_sampler_view;
 485    svga->pipe.sampler_view_destroy = svga_sampler_view_destroy;
 486 }
 487
 488
 489