src/gallium/drivers/freedreno/freedreno_resource.c

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29 #include "util/u_format.h"
  30 #include "util/u_format_zs.h"
  31 #include "util/u_inlines.h"
  32 #include "util/u_transfer.h"
  33 #include "util/u_string.h"
  34 #include "util/u_surface.h"
  35
  36 #include "freedreno_resource.h"
  37 #include "freedreno_screen.h"
  38 #include "freedreno_surface.h"
  39 #include "freedreno_context.h"
  40 #include "freedreno_query_hw.h"
  41 #include "freedreno_util.h"
  42
  43 #include <errno.h>
  44
  45
  46 static bool
  47 pending(struct fd_resource *rsc, enum fd_resource_status status)
  48 {
  49         return (rsc->status & status) ||
  50                 (rsc->stencil && (rsc->stencil->status & status));
  51 }
  52
  53 static void
  54 fd_invalidate_resource(struct fd_context *ctx, struct pipe_resource *prsc)
  55 {
  56         int i;
  57
  58         /* Go through the entire state and see if the resource is bound
  59          * anywhere. If it is, mark the relevant state as dirty. This is called on
  60          * realloc_bo.
  61          */
  62
  63         /* Constbufs */
  64         for (i = 1; i < PIPE_MAX_CONSTANT_BUFFERS && !(ctx->dirty & FD_DIRTY_CONSTBUF); i++) {
  65                 if (ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer == prsc)
  66                         ctx->dirty |= FD_DIRTY_CONSTBUF;
  67                 if (ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer == prsc)
  68                         ctx->dirty |= FD_DIRTY_CONSTBUF;
  69         }
  70
  71         /* VBOs */
  72         for (i = 0; i < ctx->vtx.vertexbuf.count && !(ctx->dirty & FD_DIRTY_VTXBUF); i++) {
  73                 if (ctx->vtx.vertexbuf.vb[i].buffer == prsc)
  74                         ctx->dirty |= FD_DIRTY_VTXBUF;
  75         }
  76
  77         /* Index buffer */
  78         if (ctx->indexbuf.buffer == prsc)
  79                 ctx->dirty |= FD_DIRTY_INDEXBUF;
  80
  81         /* Textures */
  82         for (i = 0; i < ctx->verttex.num_textures && !(ctx->dirty & FD_DIRTY_VERTTEX); i++) {
  83                 if (ctx->verttex.textures[i] && (ctx->verttex.textures[i]->texture == prsc))
  84                         ctx->dirty |= FD_DIRTY_VERTTEX;
  85         }
  86         for (i = 0; i < ctx->fragtex.num_textures && !(ctx->dirty & FD_DIRTY_FRAGTEX); i++) {
  87                 if (ctx->fragtex.textures[i] && (ctx->fragtex.textures[i]->texture == prsc))
  88                         ctx->dirty |= FD_DIRTY_FRAGTEX;
  89         }
  90 }
  91
  92 static void
  93 realloc_bo(struct fd_resource *rsc, uint32_t size)
  94 {
  95         struct fd_screen *screen = fd_screen(rsc->base.b.screen);
  96         uint32_t flags = DRM_FREEDRENO_GEM_CACHE_WCOMBINE |
  97                         DRM_FREEDRENO_GEM_TYPE_KMEM; /* TODO */
  98
  99         /* if we start using things other than write-combine,
 100          * be sure to check for PIPE_RESOURCE_FLAG_MAP_COHERENT
 101          */
 102
 103         if (rsc->bo)
 104                 fd_bo_del(rsc->bo);
 105
 106         rsc->bo = fd_bo_new(screen->dev, size, flags);
 107         rsc->timestamp = 0;
 108         rsc->status = 0;
 109         rsc->pending_ctx = NULL;
 110         list_delinit(&rsc->list);
 111         util_range_set_empty(&rsc->valid_buffer_range);
 112 }
 113
 114 /* Currently this is only used for flushing Z32_S8 texture transfers, but
 115  * eventually it should handle everything.
 116  */
 117 static void
 118 fd_resource_flush(struct fd_transfer *trans, const struct pipe_box *box)
 119 {
 120         struct fd_resource *rsc = fd_resource(trans->base.resource);
 121         struct fd_resource_slice *slice = fd_resource_slice(rsc, trans->base.level);
 122         struct fd_resource_slice *sslice = fd_resource_slice(rsc->stencil, trans->base.level);
 123         enum pipe_format format = trans->base.resource->format;
 124
 125         float *depth = fd_bo_map(rsc->bo) + slice->offset +
 126                 (trans->base.box.y + box->y) * slice->pitch * 4 + (trans->base.box.x + box->x) * 4;
 127         uint8_t *stencil = fd_bo_map(rsc->stencil->bo) + sslice->offset +
 128                 (trans->base.box.y + box->y) * sslice->pitch + trans->base.box.x + box->x;
 129
 130         assert(format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT ||
 131                    format == PIPE_FORMAT_X32_S8X24_UINT);
 132
 133         if (format != PIPE_FORMAT_X32_S8X24_UINT)
 134                 util_format_z32_float_s8x24_uint_unpack_z_float(
 135                                 depth, slice->pitch * 4,
 136                                 trans->staging, trans->base.stride,
 137                                 box->width, box->height);
 138
 139         util_format_z32_float_s8x24_uint_unpack_s_8uint(
 140                         stencil, sslice->pitch,
 141                         trans->staging, trans->base.stride,
 142                         box->width, box->height);
 143 }
 144
 145 static void fd_resource_transfer_flush_region(struct pipe_context *pctx,
 146                 struct pipe_transfer *ptrans,
 147                 const struct pipe_box *box)
 148 {
 149         struct fd_resource *rsc = fd_resource(ptrans->resource);
 150         struct fd_transfer *trans = fd_transfer(ptrans);
 151
 152         if (ptrans->resource->target == PIPE_BUFFER)
 153                 util_range_add(&rsc->valid_buffer_range,
 154                                            ptrans->box.x + box->x,
 155                                            ptrans->box.x + box->x + box->width);
 156
 157         if (trans->staging)
 158                 fd_resource_flush(trans, box);
 159 }
 160
 161 static void
 162 fd_resource_transfer_unmap(struct pipe_context *pctx,
 163                 struct pipe_transfer *ptrans)
 164 {
 165         struct fd_context *ctx = fd_context(pctx);
 166         struct fd_resource *rsc = fd_resource(ptrans->resource);
 167         struct fd_transfer *trans = fd_transfer(ptrans);
 168
 169         if (trans->staging && !(ptrans->usage & PIPE_TRANSFER_FLUSH_EXPLICIT)) {
 170                 struct pipe_box box;
 171                 u_box_2d(0, 0, ptrans->box.width, ptrans->box.height, &box);
 172                 fd_resource_flush(trans, &box);
 173         }
 174
 175         if (!(ptrans->usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
 176                 fd_bo_cpu_fini(rsc->bo);
 177                 if (rsc->stencil)
 178                         fd_bo_cpu_fini(rsc->stencil->bo);
 179         }
 180
 181         util_range_add(&rsc->valid_buffer_range,
 182                                    ptrans->box.x,
 183                                    ptrans->box.x + ptrans->box.width);
 184
 185         pipe_resource_reference(&ptrans->resource, NULL);
 186         util_slab_free(&ctx->transfer_pool, ptrans);
 187
 188         if (trans->staging)
 189                 free(trans->staging);
 190 }
 191
 192 static void *
 193 fd_resource_transfer_map(struct pipe_context *pctx,
 194                 struct pipe_resource *prsc,
 195                 unsigned level, unsigned usage,
 196                 const struct pipe_box *box,
 197                 struct pipe_transfer **pptrans)
 198 {
 199         struct fd_context *ctx = fd_context(pctx);
 200         struct fd_resource *rsc = fd_resource(prsc);
 201         struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
 202         struct fd_transfer *trans;
 203         struct pipe_transfer *ptrans;
 204         enum pipe_format format = prsc->format;
 205         uint32_t op = 0;
 206         uint32_t offset;
 207         char *buf;
 208         int ret = 0;
 209
 210         DBG("prsc=%p, level=%u, usage=%x, box=%dx%d+%d,%d", prsc, level, usage,
 211                 box->width, box->height, box->x, box->y);
 212
 213         ptrans = util_slab_alloc(&ctx->transfer_pool);
 214         if (!ptrans)
 215                 return NULL;
 216
 217         /* util_slab_alloc() doesn't zero: */
 218         trans = fd_transfer(ptrans);
 219         memset(trans, 0, sizeof(*trans));
 220
 221         pipe_resource_reference(&ptrans->resource, prsc);
 222         ptrans->level = level;
 223         ptrans->usage = usage;
 224         ptrans->box = *box;
 225         ptrans->stride = slice->pitch * rsc->cpp;
 226         ptrans->layer_stride = slice->size0;
 227
 228         if (usage & PIPE_TRANSFER_READ)
 229                 op |= DRM_FREEDRENO_PREP_READ;
 230
 231         if (usage & PIPE_TRANSFER_WRITE)
 232                 op |= DRM_FREEDRENO_PREP_WRITE;
 233
 234         if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) {
 235                 realloc_bo(rsc, fd_bo_size(rsc->bo));
 236                 if (rsc->stencil)
 237                         realloc_bo(rsc->stencil, fd_bo_size(rsc->stencil->bo));
 238                 fd_invalidate_resource(ctx, prsc);
 239         } else if ((usage & PIPE_TRANSFER_WRITE) &&
 240                            prsc->target == PIPE_BUFFER &&
 241                            !util_ranges_intersect(&rsc->valid_buffer_range,
 242                                                                           box->x, box->x + box->width)) {
 243                 /* We are trying to write to a previously uninitialized range. No need
 244                  * to wait.
 245                  */
 246         } else if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
 247                 /* If the GPU is writing to the resource, or if it is reading from the
 248                  * resource and we're trying to write to it, flush the renders.
 249                  */
 250                 if (((ptrans->usage & PIPE_TRANSFER_WRITE) &&
 251                                         pending(rsc, FD_PENDING_READ | FD_PENDING_WRITE)) ||
 252                                 pending(rsc, FD_PENDING_WRITE))
 253                         fd_context_render(pctx);
 254
 255                 /* The GPU keeps track of how the various bo's are being used, and
 256                  * will wait if necessary for the proper operation to have
 257                  * completed.
 258                  */
 259                 ret = fd_bo_cpu_prep(rsc->bo, ctx->screen->pipe, op);
 260                 if (ret)
 261                         goto fail;
 262         }
 263
 264         buf = fd_bo_map(rsc->bo);
 265         if (!buf) {
 266                 fd_resource_transfer_unmap(pctx, ptrans);
 267                 return NULL;
 268         }
 269
 270         if (rsc->layer_first) {
 271                 offset = slice->offset +
 272                         box->y / util_format_get_blockheight(format) * ptrans->stride +
 273                         box->x / util_format_get_blockwidth(format) * rsc->cpp +
 274                         box->z * rsc->layer_size;
 275         } else {
 276                 offset = slice->offset +
 277                         box->y / util_format_get_blockheight(format) * ptrans->stride +
 278                         box->x / util_format_get_blockwidth(format) * rsc->cpp +
 279                         box->z * slice->size0;
 280         }
 281
 282         if (prsc->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT ||
 283                 prsc->format == PIPE_FORMAT_X32_S8X24_UINT) {
 284                 trans->base.stride = trans->base.box.width * rsc->cpp * 2;
 285                 trans->staging = malloc(trans->base.stride * trans->base.box.height);
 286                 if (!trans->staging)
 287                         goto fail;
 288
 289                 /* if we're not discarding the whole range (or resource), we must copy
 290                  * the real data in.
 291                  */
 292                 if (!(usage & (PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE |
 293                                            PIPE_TRANSFER_DISCARD_RANGE))) {
 294                         struct fd_resource_slice *sslice =
 295                                 fd_resource_slice(rsc->stencil, level);
 296                         void *sbuf = fd_bo_map(rsc->stencil->bo);
 297                         if (!sbuf)
 298                                 goto fail;
 299
 300                         float *depth = (float *)(buf + slice->offset +
 301                                 box->y * slice->pitch * 4 + box->x * 4);
 302                         uint8_t *stencil = sbuf + sslice->offset +
 303                                 box->y * sslice->pitch + box->x;
 304
 305                         if (format != PIPE_FORMAT_X32_S8X24_UINT)
 306                                 util_format_z32_float_s8x24_uint_pack_z_float(
 307                                                 trans->staging, trans->base.stride,
 308                                                 depth, slice->pitch * 4,
 309                                                 box->width, box->height);
 310
 311                         util_format_z32_float_s8x24_uint_pack_s_8uint(
 312                                         trans->staging, trans->base.stride,
 313                                         stencil, sslice->pitch,
 314                                         box->width, box->height);
 315                 }
 316
 317                 buf = trans->staging;
 318                 offset = 0;
 319         }
 320
 321         *pptrans = ptrans;
 322
 323         return buf + offset;
 324
 325 fail:
 326         fd_resource_transfer_unmap(pctx, ptrans);
 327         return NULL;
 328 }
 329
 330 static void
 331 fd_resource_destroy(struct pipe_screen *pscreen,
 332                 struct pipe_resource *prsc)
 333 {
 334         struct fd_resource *rsc = fd_resource(prsc);
 335         if (rsc->bo)
 336                 fd_bo_del(rsc->bo);
 337         list_delinit(&rsc->list);
 338         util_range_destroy(&rsc->valid_buffer_range);
 339         FREE(rsc);
 340 }
 341
 342 static boolean
 343 fd_resource_get_handle(struct pipe_screen *pscreen,
 344                 struct pipe_resource *prsc,
 345                 struct winsys_handle *handle)
 346 {
 347         struct fd_resource *rsc = fd_resource(prsc);
 348
 349         return fd_screen_bo_get_handle(pscreen, rsc->bo,
 350                         rsc->slices[0].pitch * rsc->cpp, handle);
 351 }
 352
 353
 354 static const struct u_resource_vtbl fd_resource_vtbl = {
 355                 .resource_get_handle      = fd_resource_get_handle,
 356                 .resource_destroy         = fd_resource_destroy,
 357                 .transfer_map             = fd_resource_transfer_map,
 358                 .transfer_flush_region    = fd_resource_transfer_flush_region,
 359                 .transfer_unmap           = fd_resource_transfer_unmap,
 360                 .transfer_inline_write    = u_default_transfer_inline_write,
 361 };
 362
 363 static uint32_t
 364 setup_slices(struct fd_resource *rsc, uint32_t alignment)
 365 {
 366         struct pipe_resource *prsc = &rsc->base.b;
 367         uint32_t level, size = 0;
 368         uint32_t width = prsc->width0;
 369         uint32_t height = prsc->height0;
 370         uint32_t depth = prsc->depth0;
 371         /* in layer_first layout, the level (slice) contains just one
 372          * layer (since in fact the layer contains the slices)
 373          */
 374         uint32_t layers_in_level = rsc->layer_first ? 1 : prsc->array_size;
 375
 376         for (level = 0; level <= prsc->last_level; level++) {
 377                 struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
 378
 379                 slice->pitch = width = align(width, 32);
 380                 slice->offset = size;
 381                 /* 1d array and 2d array textures must all have the same layer size
 382                  * for each miplevel on a3xx. 3d textures can have different layer
 383                  * sizes for high levels, but the hw auto-sizer is buggy (or at least
 384                  * different than what this code does), so as soon as the layer size
 385                  * range gets into range, we stop reducing it.
 386                  */
 387                 if (prsc->target == PIPE_TEXTURE_3D && (
 388                                         level == 1 ||
 389                                         (level > 1 && rsc->slices[level - 1].size0 > 0xf000)))
 390                         slice->size0 = align(slice->pitch * height * rsc->cpp, alignment);
 391                 else if (level == 0 || rsc->layer_first || alignment == 1)
 392                         slice->size0 = align(slice->pitch * height * rsc->cpp, alignment);
 393                 else
 394                         slice->size0 = rsc->slices[level - 1].size0;
 395
 396                 size += slice->size0 * depth * layers_in_level;
 397
 398                 width = u_minify(width, 1);
 399                 height = u_minify(height, 1);
 400                 depth = u_minify(depth, 1);
 401         }
 402
 403         return size;
 404 }
 405
 406 static uint32_t
 407 slice_alignment(struct pipe_screen *pscreen, const struct pipe_resource *tmpl)
 408 {
 409         /* on a3xx, 2d array and 3d textures seem to want their
 410          * layers aligned to page boundaries:
 411          */
 412         switch (tmpl->target) {
 413         case PIPE_TEXTURE_3D:
 414         case PIPE_TEXTURE_1D_ARRAY:
 415         case PIPE_TEXTURE_2D_ARRAY:
 416                 return 4096;
 417         default:
 418                 return 1;
 419         }
 420 }
 421
 422 /**
 423  * Create a new texture object, using the given template info.
 424  */
 425 static struct pipe_resource *
 426 fd_resource_create(struct pipe_screen *pscreen,
 427                 const struct pipe_resource *tmpl)
 428 {
 429         struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
 430         struct pipe_resource *prsc = &rsc->base.b;
 431         uint32_t size;
 432
 433         DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
 434                         "nr_samples=%u, usage=%u, bind=%x, flags=%x",
 435                         tmpl->target, util_format_name(tmpl->format),
 436                         tmpl->width0, tmpl->height0, tmpl->depth0,
 437                         tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
 438                         tmpl->usage, tmpl->bind, tmpl->flags);
 439
 440         if (!rsc)
 441                 return NULL;
 442
 443         *prsc = *tmpl;
 444
 445         pipe_reference_init(&prsc->reference, 1);
 446         list_inithead(&rsc->list);
 447         prsc->screen = pscreen;
 448
 449         util_range_init(&rsc->valid_buffer_range);
 450
 451         rsc->base.vtbl = &fd_resource_vtbl;
 452         if (tmpl->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT)
 453                 rsc->cpp = util_format_get_blocksize(PIPE_FORMAT_Z32_FLOAT);
 454         else
 455                 rsc->cpp = util_format_get_blocksize(tmpl->format);
 456
 457         assert(rsc->cpp);
 458
 459         if (is_a4xx(fd_screen(pscreen))) {
 460                 switch (tmpl->target) {
 461                 case PIPE_TEXTURE_3D:
 462                         rsc->layer_first = false;
 463                         break;
 464                 default:
 465                         rsc->layer_first = true;
 466                         break;
 467                 }
 468         }
 469
 470         size = setup_slices(rsc, slice_alignment(pscreen, tmpl));
 471
 472         if (rsc->layer_first) {
 473                 rsc->layer_size = align(size, 4096);
 474                 size = rsc->layer_size * prsc->array_size;
 475         }
 476
 477         realloc_bo(rsc, size);
 478         if (!rsc->bo)
 479                 goto fail;
 480
 481         /* There is no native Z32F_S8 sampling or rendering format, so this must
 482          * be emulated via two separate textures. The depth texture still keeps
 483          * its Z32F_S8 format though, and we also keep a reference to a separate
 484          * S8 texture.
 485          */
 486         if (tmpl->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
 487                 struct pipe_resource stencil = *tmpl;
 488                 stencil.format = PIPE_FORMAT_S8_UINT;
 489                 rsc->stencil = fd_resource(fd_resource_create(pscreen, &stencil));
 490                 if (!rsc->stencil)
 491                         goto fail;
 492         }
 493
 494         return prsc;
 495 fail:
 496         fd_resource_destroy(pscreen, prsc);
 497         return NULL;
 498 }
 499
 500 /**
 501  * Create a texture from a winsys_handle. The handle is often created in
 502  * another process by first creating a pipe texture and then calling
 503  * resource_get_handle.
 504  */
 505 static struct pipe_resource *
 506 fd_resource_from_handle(struct pipe_screen *pscreen,
 507                 const struct pipe_resource *tmpl,
 508                 struct winsys_handle *handle)
 509 {
 510         struct fd_resource *rsc = CALLOC_STRUCT(fd_resource);
 511         struct fd_resource_slice *slice = &rsc->slices[0];
 512         struct pipe_resource *prsc = &rsc->base.b;
 513
 514         DBG("target=%d, format=%s, %ux%ux%u, array_size=%u, last_level=%u, "
 515                         "nr_samples=%u, usage=%u, bind=%x, flags=%x",
 516                         tmpl->target, util_format_name(tmpl->format),
 517                         tmpl->width0, tmpl->height0, tmpl->depth0,
 518                         tmpl->array_size, tmpl->last_level, tmpl->nr_samples,
 519                         tmpl->usage, tmpl->bind, tmpl->flags);
 520
 521         if (!rsc)
 522                 return NULL;
 523
 524         *prsc = *tmpl;
 525
 526         pipe_reference_init(&prsc->reference, 1);
 527         list_inithead(&rsc->list);
 528         prsc->screen = pscreen;
 529
 530         util_range_init(&rsc->valid_buffer_range);
 531
 532         rsc->bo = fd_screen_bo_from_handle(pscreen, handle, &slice->pitch);
 533         if (!rsc->bo)
 534                 goto fail;
 535
 536         rsc->base.vtbl = &fd_resource_vtbl;
 537         rsc->cpp = util_format_get_blocksize(tmpl->format);
 538         slice->pitch /= rsc->cpp;
 539
 540         assert(rsc->cpp);
 541
 542         return prsc;
 543
 544 fail:
 545         fd_resource_destroy(pscreen, prsc);
 546         return NULL;
 547 }
 548
 549 static void fd_blitter_pipe_begin(struct fd_context *ctx);
 550 static void fd_blitter_pipe_end(struct fd_context *ctx);
 551
 552 /**
 553  * _copy_region using pipe (3d engine)
 554  */
 555 static bool
 556 fd_blitter_pipe_copy_region(struct fd_context *ctx,
 557                 struct pipe_resource *dst,
 558                 unsigned dst_level,
 559                 unsigned dstx, unsigned dsty, unsigned dstz,
 560                 struct pipe_resource *src,
 561                 unsigned src_level,
 562                 const struct pipe_box *src_box)
 563 {
 564         /* not until we allow rendertargets to be buffers */
 565         if (dst->target == PIPE_BUFFER || src->target == PIPE_BUFFER)
 566                 return false;
 567
 568         if (!util_blitter_is_copy_supported(ctx->blitter, dst, src))
 569                 return false;
 570
 571         fd_blitter_pipe_begin(ctx);
 572         util_blitter_copy_texture(ctx->blitter,
 573                         dst, dst_level, dstx, dsty, dstz,
 574                         src, src_level, src_box);
 575         fd_blitter_pipe_end(ctx);
 576
 577         return true;
 578 }
 579
 580 /**
 581  * Copy a block of pixels from one resource to another.
 582  * The resource must be of the same format.
 583  * Resources with nr_samples > 1 are not allowed.
 584  */
 585 static void
 586 fd_resource_copy_region(struct pipe_context *pctx,
 587                 struct pipe_resource *dst,
 588                 unsigned dst_level,
 589                 unsigned dstx, unsigned dsty, unsigned dstz,
 590                 struct pipe_resource *src,
 591                 unsigned src_level,
 592                 const struct pipe_box *src_box)
 593 {
 594         struct fd_context *ctx = fd_context(pctx);
 595
 596         /* TODO if we have 2d core, or other DMA engine that could be used
 597          * for simple copies and reasonably easily synchronized with the 3d
 598          * core, this is where we'd plug it in..
 599          */
 600
 601         /* try blit on 3d pipe: */
 602         if (fd_blitter_pipe_copy_region(ctx,
 603                         dst, dst_level, dstx, dsty, dstz,
 604                         src, src_level, src_box))
 605                 return;
 606
 607         /* else fallback to pure sw: */
 608         util_resource_copy_region(pctx,
 609                         dst, dst_level, dstx, dsty, dstz,
 610                         src, src_level, src_box);
 611 }
 612
 613 /**
 614  * Optimal hardware path for blitting pixels.
 615  * Scaling, format conversion, up- and downsampling (resolve) are allowed.
 616  */
 617 static void
 618 fd_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
 619 {
 620         struct fd_context *ctx = fd_context(pctx);
 621         struct pipe_blit_info info = *blit_info;
 622
 623         if (info.src.resource->nr_samples > 1 &&
 624                         info.dst.resource->nr_samples <= 1 &&
 625                         !util_format_is_depth_or_stencil(info.src.resource->format) &&
 626                         !util_format_is_pure_integer(info.src.resource->format)) {
 627                 DBG("color resolve unimplemented");
 628                 return;
 629         }
 630
 631         if (util_try_blit_via_copy_region(pctx, &info)) {
 632                 return; /* done */
 633         }
 634
 635         if (info.mask & PIPE_MASK_S) {
 636                 DBG("cannot blit stencil, skipping");
 637                 info.mask &= ~PIPE_MASK_S;
 638         }
 639
 640         if (!util_blitter_is_blit_supported(ctx->blitter, &info)) {
 641                 DBG("blit unsupported %s -> %s",
 642                                 util_format_short_name(info.src.resource->format),
 643                                 util_format_short_name(info.dst.resource->format));
 644                 return;
 645         }
 646
 647         fd_blitter_pipe_begin(ctx);
 648         util_blitter_blit(ctx->blitter, &info);
 649         fd_blitter_pipe_end(ctx);
 650 }
 651
 652 static void
 653 fd_blitter_pipe_begin(struct fd_context *ctx)
 654 {
 655         util_blitter_save_vertex_buffer_slot(ctx->blitter, ctx->vtx.vertexbuf.vb);
 656         util_blitter_save_vertex_elements(ctx->blitter, ctx->vtx.vtx);
 657         util_blitter_save_vertex_shader(ctx->blitter, ctx->prog.vp);
 658         util_blitter_save_so_targets(ctx->blitter, ctx->streamout.num_targets,
 659                         ctx->streamout.targets);
 660         util_blitter_save_rasterizer(ctx->blitter, ctx->rasterizer);
 661         util_blitter_save_viewport(ctx->blitter, &ctx->viewport);
 662         util_blitter_save_scissor(ctx->blitter, &ctx->scissor);
 663         util_blitter_save_fragment_shader(ctx->blitter, ctx->prog.fp);
 664         util_blitter_save_blend(ctx->blitter, ctx->blend);
 665         util_blitter_save_depth_stencil_alpha(ctx->blitter, ctx->zsa);
 666         util_blitter_save_stencil_ref(ctx->blitter, &ctx->stencil_ref);
 667         util_blitter_save_sample_mask(ctx->blitter, ctx->sample_mask);
 668         util_blitter_save_framebuffer(ctx->blitter, &ctx->framebuffer);
 669         util_blitter_save_fragment_sampler_states(ctx->blitter,
 670                         ctx->fragtex.num_samplers,
 671                         (void **)ctx->fragtex.samplers);
 672         util_blitter_save_fragment_sampler_views(ctx->blitter,
 673                         ctx->fragtex.num_textures, ctx->fragtex.textures);
 674
 675         fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_BLIT);
 676 }
 677
 678 static void
 679 fd_blitter_pipe_end(struct fd_context *ctx)
 680 {
 681         fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_NULL);
 682 }
 683
 684 static void
 685 fd_flush_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
 686 {
 687         struct fd_resource *rsc = fd_resource(prsc);
 688
 689         if (pending(rsc, FD_PENDING_WRITE | FD_PENDING_READ))
 690                 fd_context_render(pctx);
 691 }
 692
 693 void
 694 fd_resource_screen_init(struct pipe_screen *pscreen)
 695 {
 696         pscreen->resource_create = fd_resource_create;
 697         pscreen->resource_from_handle = fd_resource_from_handle;
 698         pscreen->resource_get_handle = u_resource_get_handle_vtbl;
 699         pscreen->resource_destroy = u_resource_destroy_vtbl;
 700 }
 701
 702 void
 703 fd_resource_context_init(struct pipe_context *pctx)
 704 {
 705         pctx->transfer_map = u_transfer_map_vtbl;
 706         pctx->transfer_flush_region = u_transfer_flush_region_vtbl;
 707         pctx->transfer_unmap = u_transfer_unmap_vtbl;
 708         pctx->transfer_inline_write = u_transfer_inline_write_vtbl;
 709         pctx->create_surface = fd_create_surface;
 710         pctx->surface_destroy = fd_surface_destroy;
 711         pctx->resource_copy_region = fd_resource_copy_region;
 712         pctx->blit = fd_blit;
 713         pctx->flush_resource = fd_flush_resource;
 714 }