src/gallium/drivers/freedreno/a6xx/fd6_texture.c

   1 /*
   2  * Copyright (C) 2016 Rob Clark <robclark@freedesktop.org>
   3  * Copyright © 2018 Google, Inc.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  * Authors:
  25  *    Rob Clark <robclark@freedesktop.org>
  26  */
  27
  28 #include "pipe/p_state.h"
  29 #include "util/u_string.h"
  30 #include "util/u_memory.h"
  31 #include "util/u_inlines.h"
  32 #include "util/format/u_format.h"
  33 #include "util/hash_table.h"
  34
  35 #include "fd6_texture.h"
  36 #include "fd6_resource.h"
  37 #include "fd6_format.h"
  38 #include "fd6_emit.h"
  39
  40 static void fd6_texture_state_destroy(struct fd6_texture_state *state);
  41
  42 static enum a6xx_tex_clamp
  43 tex_clamp(unsigned wrap, bool clamp_to_edge, bool *needs_border)
  44 {
  45         /* Hardware does not support _CLAMP, but we emulate it: */
  46         if (wrap == PIPE_TEX_WRAP_CLAMP) {
  47                 wrap = (clamp_to_edge) ?
  48                         PIPE_TEX_WRAP_CLAMP_TO_EDGE : PIPE_TEX_WRAP_CLAMP_TO_BORDER;
  49         }
  50
  51         switch (wrap) {
  52         case PIPE_TEX_WRAP_REPEAT:
  53                 return A6XX_TEX_REPEAT;
  54         case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
  55                 return A6XX_TEX_CLAMP_TO_EDGE;
  56         case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
  57                 *needs_border = true;
  58                 return A6XX_TEX_CLAMP_TO_BORDER;
  59         case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
  60                 /* only works for PoT.. need to emulate otherwise! */
  61                 return A6XX_TEX_MIRROR_CLAMP;
  62         case PIPE_TEX_WRAP_MIRROR_REPEAT:
  63                 return A6XX_TEX_MIRROR_REPEAT;
  64         case PIPE_TEX_WRAP_MIRROR_CLAMP:
  65         case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
  66                 /* these two we could perhaps emulate, but we currently
  67                  * just don't advertise PIPE_CAP_TEXTURE_MIRROR_CLAMP
  68                  */
  69         default:
  70                 DBG("invalid wrap: %u", wrap);
  71                 return 0;
  72         }
  73 }
  74
  75 static enum a6xx_tex_filter
  76 tex_filter(unsigned filter, bool aniso)
  77 {
  78         switch (filter) {
  79         case PIPE_TEX_FILTER_NEAREST:
  80                 return A6XX_TEX_NEAREST;
  81         case PIPE_TEX_FILTER_LINEAR:
  82                 return aniso ? A6XX_TEX_ANISO : A6XX_TEX_LINEAR;
  83         default:
  84                 DBG("invalid filter: %u", filter);
  85                 return 0;
  86         }
  87 }
  88
  89 static void *
  90 fd6_sampler_state_create(struct pipe_context *pctx,
  91                 const struct pipe_sampler_state *cso)
  92 {
  93         struct fd6_sampler_stateobj *so = CALLOC_STRUCT(fd6_sampler_stateobj);
  94         unsigned aniso = util_last_bit(MIN2(cso->max_anisotropy >> 1, 8));
  95         bool miplinear = false;
  96         bool clamp_to_edge;
  97
  98         if (!so)
  99                 return NULL;
 100
 101         so->base = *cso;
 102         so->seqno = ++fd6_context(fd_context(pctx))->tex_seqno;
 103
 104         if (cso->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR)
 105                 miplinear = true;
 106
 107         /*
 108          * For nearest filtering, _CLAMP means _CLAMP_TO_EDGE;  for linear
 109          * filtering, _CLAMP means _CLAMP_TO_BORDER while additionally
 110          * clamping the texture coordinates to [0.0, 1.0].
 111          *
 112          * The clamping will be taken care of in the shaders.  There are two
 113          * filters here, but let the minification one has a say.
 114          */
 115         clamp_to_edge = (cso->min_img_filter == PIPE_TEX_FILTER_NEAREST);
 116         if (!clamp_to_edge) {
 117                 so->saturate_s = (cso->wrap_s == PIPE_TEX_WRAP_CLAMP);
 118                 so->saturate_t = (cso->wrap_t == PIPE_TEX_WRAP_CLAMP);
 119                 so->saturate_r = (cso->wrap_r == PIPE_TEX_WRAP_CLAMP);
 120         }
 121
 122         so->needs_border = false;
 123         so->texsamp0 =
 124                 COND(miplinear, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR) |
 125                 A6XX_TEX_SAMP_0_XY_MAG(tex_filter(cso->mag_img_filter, aniso)) |
 126                 A6XX_TEX_SAMP_0_XY_MIN(tex_filter(cso->min_img_filter, aniso)) |
 127                 A6XX_TEX_SAMP_0_ANISO(aniso) |
 128                 A6XX_TEX_SAMP_0_WRAP_S(tex_clamp(cso->wrap_s, clamp_to_edge, &so->needs_border)) |
 129                 A6XX_TEX_SAMP_0_WRAP_T(tex_clamp(cso->wrap_t, clamp_to_edge, &so->needs_border)) |
 130                 A6XX_TEX_SAMP_0_WRAP_R(tex_clamp(cso->wrap_r, clamp_to_edge, &so->needs_border));
 131
 132         so->texsamp1 =
 133                 COND(cso->min_mip_filter == PIPE_TEX_MIPFILTER_NONE,
 134                                 A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR) |
 135                 COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) |
 136                 COND(!cso->normalized_coords, A6XX_TEX_SAMP_1_UNNORM_COORDS);
 137
 138         so->texsamp0 |= A6XX_TEX_SAMP_0_LOD_BIAS(cso->lod_bias);
 139         so->texsamp1 |=
 140                 A6XX_TEX_SAMP_1_MIN_LOD(cso->min_lod) |
 141                 A6XX_TEX_SAMP_1_MAX_LOD(cso->max_lod);
 142
 143         if (cso->compare_mode)
 144                 so->texsamp1 |= A6XX_TEX_SAMP_1_COMPARE_FUNC(cso->compare_func); /* maps 1:1 */
 145
 146         return so;
 147 }
 148
 149 static void
 150 fd6_sampler_state_delete(struct pipe_context *pctx, void *hwcso)
 151 {
 152         struct fd6_context *fd6_ctx = fd6_context(fd_context(pctx));
 153         struct fd6_sampler_stateobj *samp = hwcso;
 154
 155         hash_table_foreach(fd6_ctx->tex_cache, entry) {
 156                 struct fd6_texture_state *state = entry->data;
 157
 158                 for (unsigned i = 0; i < ARRAY_SIZE(state->key.samp); i++) {
 159                         if (samp->seqno == state->key.samp[i].seqno) {
 160                                 fd6_texture_state_destroy(entry->data);
 161                                 _mesa_hash_table_remove(fd6_ctx->tex_cache, entry);
 162                                 break;
 163                         }
 164                 }
 165         }
 166
 167         free(hwcso);
 168 }
 169
 170 static void
 171 fd6_sampler_states_bind(struct pipe_context *pctx,
 172                 enum pipe_shader_type shader, unsigned start,
 173                 unsigned nr, void **hwcso)
 174 {
 175         struct fd_context *ctx = fd_context(pctx);
 176         struct fd6_context *fd6_ctx = fd6_context(ctx);
 177         uint16_t saturate_s = 0, saturate_t = 0, saturate_r = 0;
 178         unsigned i;
 179
 180         if (!hwcso)
 181                 nr = 0;
 182
 183         for (i = 0; i < nr; i++) {
 184                 if (hwcso[i]) {
 185                         struct fd6_sampler_stateobj *sampler =
 186                                         fd6_sampler_stateobj(hwcso[i]);
 187                         if (sampler->saturate_s)
 188                                 saturate_s |= (1 << i);
 189                         if (sampler->saturate_t)
 190                                 saturate_t |= (1 << i);
 191                         if (sampler->saturate_r)
 192                                 saturate_r |= (1 << i);
 193                 }
 194         }
 195
 196         fd_sampler_states_bind(pctx, shader, start, nr, hwcso);
 197
 198         if (shader == PIPE_SHADER_FRAGMENT) {
 199                 fd6_ctx->fsaturate =
 200                         (saturate_s != 0) ||
 201                         (saturate_t != 0) ||
 202                         (saturate_r != 0);
 203                 fd6_ctx->fsaturate_s = saturate_s;
 204                 fd6_ctx->fsaturate_t = saturate_t;
 205                 fd6_ctx->fsaturate_r = saturate_r;
 206         } else if (shader == PIPE_SHADER_VERTEX) {
 207                 fd6_ctx->vsaturate =
 208                         (saturate_s != 0) ||
 209                         (saturate_t != 0) ||
 210                         (saturate_r != 0);
 211                 fd6_ctx->vsaturate_s = saturate_s;
 212                 fd6_ctx->vsaturate_t = saturate_t;
 213                 fd6_ctx->vsaturate_r = saturate_r;
 214         }
 215 }
 216
 217 static struct pipe_sampler_view *
 218 fd6_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 219                 const struct pipe_sampler_view *cso)
 220 {
 221         struct fd6_pipe_sampler_view *so = CALLOC_STRUCT(fd6_pipe_sampler_view);
 222         struct fd_resource *rsc = fd_resource(prsc);
 223         enum pipe_format format = cso->format;
 224         unsigned lvl, layers = 0;
 225
 226         if (!so)
 227                 return NULL;
 228
 229         fd6_validate_format(fd_context(pctx), rsc, format);
 230
 231         if (format == PIPE_FORMAT_X32_S8X24_UINT) {
 232                 rsc = rsc->stencil;
 233                 format = rsc->base.format;
 234         }
 235
 236         so->base = *cso;
 237         pipe_reference(NULL, &prsc->reference);
 238         so->base.texture = prsc;
 239         so->base.reference.count = 1;
 240         so->base.context = pctx;
 241         so->seqno = ++fd6_context(fd_context(pctx))->tex_seqno;
 242
 243         if (cso->target == PIPE_BUFFER) {
 244                 unsigned elements = cso->u.buf.size / util_format_get_blocksize(format);
 245
 246                 lvl = 0;
 247                 so->texconst1 =
 248                         A6XX_TEX_CONST_1_WIDTH(elements & MASK(15)) |
 249                         A6XX_TEX_CONST_1_HEIGHT(elements >> 15);
 250                 so->texconst2 =
 251                         A6XX_TEX_CONST_2_UNK4 |
 252                         A6XX_TEX_CONST_2_UNK31;
 253                 so->offset = cso->u.buf.offset;
 254         } else {
 255                 unsigned miplevels;
 256
 257                 lvl = fd_sampler_first_level(cso);
 258                 miplevels = fd_sampler_last_level(cso) - lvl;
 259                 layers = cso->u.tex.last_layer - cso->u.tex.first_layer + 1;
 260
 261                 so->texconst0 |= A6XX_TEX_CONST_0_MIPLVLS(miplevels);
 262                 so->texconst1 =
 263                         A6XX_TEX_CONST_1_WIDTH(u_minify(prsc->width0, lvl)) |
 264                         A6XX_TEX_CONST_1_HEIGHT(u_minify(prsc->height0, lvl));
 265                 so->texconst2 =
 266                         A6XX_TEX_CONST_2_PITCHALIGN(rsc->layout.pitchalign - 6) |
 267                         A6XX_TEX_CONST_2_PITCH(fd_resource_pitch(rsc, lvl));
 268                 so->offset = fd_resource_offset(rsc, lvl, cso->u.tex.first_layer);
 269                 so->ubwc_offset = fd_resource_ubwc_offset(rsc, lvl, cso->u.tex.first_layer);
 270                 so->ubwc_enabled = fd_resource_ubwc_enabled(rsc, lvl);
 271         }
 272
 273         so->texconst0 |= fd6_tex_const_0(prsc, lvl, cso->format,
 274                                 cso->swizzle_r, cso->swizzle_g,
 275                                 cso->swizzle_b, cso->swizzle_a);
 276
 277         so->texconst2 |= A6XX_TEX_CONST_2_TYPE(fd6_tex_type(cso->target));
 278
 279         switch (cso->target) {
 280         case PIPE_TEXTURE_RECT:
 281         case PIPE_TEXTURE_1D:
 282         case PIPE_TEXTURE_2D:
 283                 so->texconst3 =
 284                         A6XX_TEX_CONST_3_ARRAY_PITCH(rsc->layout.layer_size);
 285                 so->texconst5 =
 286                         A6XX_TEX_CONST_5_DEPTH(1);
 287                 break;
 288         case PIPE_TEXTURE_1D_ARRAY:
 289         case PIPE_TEXTURE_2D_ARRAY:
 290                 so->texconst3 =
 291                         A6XX_TEX_CONST_3_ARRAY_PITCH(rsc->layout.layer_size);
 292                 so->texconst5 =
 293                         A6XX_TEX_CONST_5_DEPTH(layers);
 294                 break;
 295         case PIPE_TEXTURE_CUBE:
 296         case PIPE_TEXTURE_CUBE_ARRAY:
 297                 so->texconst3 =
 298                         A6XX_TEX_CONST_3_ARRAY_PITCH(rsc->layout.layer_size);
 299                 so->texconst5 =
 300                         A6XX_TEX_CONST_5_DEPTH(layers / 6);
 301                 break;
 302         case PIPE_TEXTURE_3D:
 303                 so->texconst3 =
 304                         A6XX_TEX_CONST_3_MIN_LAYERSZ(
 305                                 fd_resource_slice(rsc, prsc->last_level)->size0) |
 306                         A6XX_TEX_CONST_3_ARRAY_PITCH(fd_resource_slice(rsc, lvl)->size0);
 307                 so->texconst5 =
 308                         A6XX_TEX_CONST_5_DEPTH(u_minify(prsc->depth0, lvl));
 309                 break;
 310         default:
 311                 break;
 312         }
 313
 314         if (so->ubwc_enabled) {
 315                 uint32_t block_width, block_height;
 316                 fdl6_get_ubwc_blockwidth(&rsc->layout, &block_width, &block_height);
 317
 318                 so->texconst3 |= A6XX_TEX_CONST_3_FLAG | A6XX_TEX_CONST_3_TILE_ALL;
 319                 so->texconst9 |= A6XX_TEX_CONST_9_FLAG_BUFFER_ARRAY_PITCH(rsc->layout.ubwc_layer_size >> 2);
 320                 so->texconst10 |=
 321                         A6XX_TEX_CONST_10_FLAG_BUFFER_PITCH(fdl_ubwc_pitch(&rsc->layout, lvl)) |
 322                         A6XX_TEX_CONST_10_FLAG_BUFFER_LOGW(util_logbase2_ceil(DIV_ROUND_UP(u_minify(prsc->width0, lvl), block_width))) |
 323                         A6XX_TEX_CONST_10_FLAG_BUFFER_LOGH(util_logbase2_ceil(DIV_ROUND_UP(u_minify(prsc->height0, lvl), block_height)));
 324         }
 325
 326         return &so->base;
 327 }
 328
 329 static void
 330 fd6_sampler_view_destroy(struct pipe_context *pctx,
 331                 struct pipe_sampler_view *_view)
 332 {
 333         struct fd6_context *fd6_ctx = fd6_context(fd_context(pctx));
 334         struct fd6_pipe_sampler_view *view = fd6_pipe_sampler_view(_view);
 335
 336         hash_table_foreach(fd6_ctx->tex_cache, entry) {
 337                 struct fd6_texture_state *state = entry->data;
 338
 339                 for (unsigned i = 0; i < ARRAY_SIZE(state->key.view); i++) {
 340                         if (view->seqno == state->key.view[i].seqno) {
 341                                 fd6_texture_state_destroy(entry->data);
 342                                 _mesa_hash_table_remove(fd6_ctx->tex_cache, entry);
 343                                 break;
 344                         }
 345                 }
 346         }
 347
 348         pipe_resource_reference(&view->base.texture, NULL);
 349
 350         free(view);
 351 }
 352
 353
 354 static uint32_t
 355 key_hash(const void *_key)
 356 {
 357         const struct fd6_texture_key *key = _key;
 358         return XXH32(key, sizeof(*key), 0);
 359 }
 360
 361 static bool
 362 key_equals(const void *_a, const void *_b)
 363 {
 364         const struct fd6_texture_key *a = _a;
 365         const struct fd6_texture_key *b = _b;
 366         return memcmp(a, b, sizeof(struct fd6_texture_key)) == 0;
 367 }
 368
 369 struct fd6_texture_state *
 370 fd6_texture_state(struct fd_context *ctx, enum pipe_shader_type type,
 371                 struct fd_texture_stateobj *tex)
 372 {
 373         struct fd6_context *fd6_ctx = fd6_context(ctx);
 374         struct fd6_texture_key key;
 375         bool needs_border = false;
 376
 377         memset(&key, 0, sizeof(key));
 378
 379         for (unsigned i = 0; i < tex->num_textures; i++) {
 380                 if (!tex->textures[i])
 381                         continue;
 382
 383                 struct fd6_pipe_sampler_view *view =
 384                         fd6_pipe_sampler_view(tex->textures[i]);
 385
 386                 key.view[i].rsc_seqno = fd_resource(view->base.texture)->seqno;
 387                 key.view[i].seqno = view->seqno;
 388         }
 389
 390         for (unsigned i = 0; i < tex->num_samplers; i++) {
 391                 if (!tex->samplers[i])
 392                         continue;
 393
 394                 struct fd6_sampler_stateobj *sampler =
 395                         fd6_sampler_stateobj(tex->samplers[i]);
 396
 397                 key.samp[i].seqno = sampler->seqno;
 398
 399                 needs_border |= sampler->needs_border;
 400         }
 401
 402         key.type = type;
 403         key.bcolor_offset = fd6_border_color_offset(ctx, type, tex);
 404
 405         uint32_t hash = key_hash(&key);
 406         struct hash_entry *entry =
 407                 _mesa_hash_table_search_pre_hashed(fd6_ctx->tex_cache, hash, &key);
 408
 409         if (entry) {
 410                 return entry->data;
 411         }
 412
 413         struct fd6_texture_state *state = CALLOC_STRUCT(fd6_texture_state);
 414
 415         state->key = key;
 416         state->stateobj = fd_ringbuffer_new_object(ctx->pipe, 0x1000);
 417         state->needs_border = needs_border;
 418
 419         fd6_emit_textures(ctx->pipe, state->stateobj, type, tex, key.bcolor_offset,
 420                         NULL, NULL);
 421
 422         /* NOTE: uses copy of key in state obj, because pointer passed by caller
 423          * is probably on the stack
 424          */
 425         _mesa_hash_table_insert_pre_hashed(fd6_ctx->tex_cache, hash,
 426                         &state->key, state);
 427
 428         return state;
 429 }
 430
 431 static void
 432 fd6_texture_state_destroy(struct fd6_texture_state *state)
 433 {
 434         fd_ringbuffer_del(state->stateobj);
 435         free(state);
 436 }
 437
 438 static void
 439 fd6_rebind_resource(struct fd_context *ctx, struct fd_resource *rsc)
 440 {
 441         if (!(rsc->dirty & FD_DIRTY_TEX))
 442                 return;
 443
 444         struct fd6_context *fd6_ctx = fd6_context(ctx);
 445
 446         hash_table_foreach (fd6_ctx->tex_cache, entry) {
 447                 struct fd6_texture_state *state = entry->data;
 448
 449                 for (unsigned i = 0; i < ARRAY_SIZE(state->key.view); i++) {
 450                         if (rsc->seqno == state->key.view[i].rsc_seqno) {
 451                                 fd6_texture_state_destroy(entry->data);
 452                                 _mesa_hash_table_remove(fd6_ctx->tex_cache, entry);
 453                         }
 454                 }
 455         }
 456 }
 457
 458 void
 459 fd6_texture_init(struct pipe_context *pctx)
 460 {
 461         struct fd_context *ctx = fd_context(pctx);
 462         struct fd6_context *fd6_ctx = fd6_context(ctx);
 463
 464         pctx->create_sampler_state = fd6_sampler_state_create;
 465         pctx->delete_sampler_state = fd6_sampler_state_delete;
 466         pctx->bind_sampler_states = fd6_sampler_states_bind;
 467
 468         pctx->create_sampler_view = fd6_sampler_view_create;
 469         pctx->sampler_view_destroy = fd6_sampler_view_destroy;
 470         pctx->set_sampler_views = fd_set_sampler_views;
 471
 472         ctx->rebind_resource = fd6_rebind_resource;
 473
 474         fd6_ctx->tex_cache = _mesa_hash_table_create(NULL, key_hash, key_equals);
 475 }
 476
 477 void
 478 fd6_texture_fini(struct pipe_context *pctx)
 479 {
 480         struct fd6_context *fd6_ctx = fd6_context(fd_context(pctx));
 481
 482         hash_table_foreach(fd6_ctx->tex_cache, entry) {
 483                 fd6_texture_state_destroy(entry->data);
 484         }
 485         ralloc_free(fd6_ctx->tex_cache);
 486 }