src/gallium/drivers/radeonsi/si_blit.c

   1 /*
   2  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include "si_pipe.h"
  25 #include "util/u_format.h"
  26 #include "util/u_surface.h"
  27
  28 enum si_blitter_op /* bitmask */
  29 {
  30         SI_SAVE_TEXTURES      = 1,
  31         SI_SAVE_FRAMEBUFFER   = 2,
  32         SI_DISABLE_RENDER_COND = 4,
  33
  34         SI_CLEAR         = 0,
  35
  36         SI_CLEAR_SURFACE = SI_SAVE_FRAMEBUFFER,
  37
  38         SI_COPY          = SI_SAVE_FRAMEBUFFER | SI_SAVE_TEXTURES |
  39                            SI_DISABLE_RENDER_COND,
  40
  41         SI_BLIT          = SI_SAVE_FRAMEBUFFER | SI_SAVE_TEXTURES,
  42
  43         SI_DECOMPRESS    = SI_SAVE_FRAMEBUFFER | SI_DISABLE_RENDER_COND,
  44
  45         SI_COLOR_RESOLVE = SI_SAVE_FRAMEBUFFER
  46 };
  47
  48 static void si_blitter_begin(struct pipe_context *ctx, enum si_blitter_op op)
  49 {
  50         struct si_context *sctx = (struct si_context *)ctx;
  51
  52         r600_suspend_nontimer_queries(&sctx->b);
  53
  54         util_blitter_save_blend(sctx->blitter, sctx->queued.named.blend);
  55         util_blitter_save_depth_stencil_alpha(sctx->blitter, sctx->queued.named.dsa);
  56         util_blitter_save_stencil_ref(sctx->blitter, &sctx->stencil_ref);
  57         util_blitter_save_rasterizer(sctx->blitter, sctx->queued.named.rasterizer);
  58         util_blitter_save_fragment_shader(sctx->blitter, sctx->ps_shader);
  59         util_blitter_save_geometry_shader(sctx->blitter, sctx->gs_shader);
  60         util_blitter_save_tessctrl_shader(sctx->blitter, sctx->tcs_shader);
  61         util_blitter_save_tesseval_shader(sctx->blitter, sctx->tes_shader);
  62         util_blitter_save_vertex_shader(sctx->blitter, sctx->vs_shader);
  63         util_blitter_save_vertex_elements(sctx->blitter, sctx->vertex_elements);
  64         if (sctx->queued.named.sample_mask) {
  65                 util_blitter_save_sample_mask(sctx->blitter,
  66                                               sctx->queued.named.sample_mask->sample_mask);
  67         }
  68         util_blitter_save_viewport(sctx->blitter, &sctx->viewports.states[0]);
  69         util_blitter_save_scissor(sctx->blitter, &sctx->scissors.states[0]);
  70         util_blitter_save_vertex_buffer_slot(sctx->blitter, sctx->vertex_buffer);
  71         util_blitter_save_so_targets(sctx->blitter, sctx->b.streamout.num_targets,
  72                                      (struct pipe_stream_output_target**)sctx->b.streamout.targets);
  73
  74         if (op & SI_SAVE_FRAMEBUFFER)
  75                 util_blitter_save_framebuffer(sctx->blitter, &sctx->framebuffer.state);
  76
  77         if (op & SI_SAVE_TEXTURES) {
  78                 util_blitter_save_fragment_sampler_states(
  79                         sctx->blitter, 2,
  80                         sctx->samplers[PIPE_SHADER_FRAGMENT].states.saved_states);
  81
  82                 util_blitter_save_fragment_sampler_views(sctx->blitter, 2,
  83                         sctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
  84         }
  85
  86         if ((op & SI_DISABLE_RENDER_COND) && sctx->b.current_render_cond) {
  87                 util_blitter_save_render_condition(sctx->blitter,
  88                                                    sctx->b.current_render_cond,
  89                                                    sctx->b.current_render_cond_cond,
  90                                                    sctx->b.current_render_cond_mode);
  91         }
  92 }
  93
  94 static void si_blitter_end(struct pipe_context *ctx)
  95 {
  96         struct si_context *sctx = (struct si_context *)ctx;
  97         r600_resume_nontimer_queries(&sctx->b);
  98 }
  99
 100 static unsigned u_max_sample(struct pipe_resource *r)
 101 {
 102         return r->nr_samples ? r->nr_samples - 1 : 0;
 103 }
 104
 105 static void si_blit_decompress_depth(struct pipe_context *ctx,
 106                                      struct r600_texture *texture,
 107                                      struct r600_texture *staging,
 108                                      unsigned first_level, unsigned last_level,
 109                                      unsigned first_layer, unsigned last_layer,
 110                                      unsigned first_sample, unsigned last_sample)
 111 {
 112         struct si_context *sctx = (struct si_context *)ctx;
 113         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 114         float depth = 1.0f;
 115         const struct util_format_description *desc;
 116         struct r600_texture *flushed_depth_texture = staging ?
 117                         staging : texture->flushed_depth_texture;
 118
 119         if (!staging && !texture->dirty_level_mask)
 120                 return;
 121
 122         max_sample = u_max_sample(&texture->resource.b.b);
 123
 124         desc = util_format_description(flushed_depth_texture->resource.b.b.format);
 125
 126         if (util_format_has_depth(desc))
 127                 sctx->dbcb_depth_copy_enabled = true;
 128         if (util_format_has_stencil(desc))
 129                 sctx->dbcb_stencil_copy_enabled = true;
 130
 131         assert(sctx->dbcb_depth_copy_enabled || sctx->dbcb_stencil_copy_enabled);
 132
 133         for (level = first_level; level <= last_level; level++) {
 134                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 135                         continue;
 136
 137                 /* The smaller the mipmap level, the less layers there are
 138                  * as far as 3D textures are concerned. */
 139                 max_layer = util_max_layer(&texture->resource.b.b, level);
 140                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 141
 142                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 143                         for (sample = first_sample; sample <= last_sample; sample++) {
 144                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 145
 146                                 sctx->dbcb_copy_sample = sample;
 147                                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 148
 149                                 surf_tmpl.format = texture->resource.b.b.format;
 150                                 surf_tmpl.u.tex.level = level;
 151                                 surf_tmpl.u.tex.first_layer = layer;
 152                                 surf_tmpl.u.tex.last_layer = layer;
 153
 154                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 155
 156                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 157                                 cbsurf = ctx->create_surface(ctx,
 158                                                 (struct pipe_resource*)flushed_depth_texture, &surf_tmpl);
 159
 160                                 si_blitter_begin(ctx, SI_DECOMPRESS);
 161                                 util_blitter_custom_depth_stencil(sctx->blitter, zsurf, cbsurf, 1 << sample,
 162                                                                   sctx->custom_dsa_flush, depth);
 163                                 si_blitter_end(ctx);
 164
 165                                 pipe_surface_reference(&zsurf, NULL);
 166                                 pipe_surface_reference(&cbsurf, NULL);
 167                         }
 168                 }
 169
 170                 /* The texture will always be dirty if some layers aren't flushed.
 171                  * I don't think this case can occur though. */
 172                 if (!staging &&
 173                     first_layer == 0 && last_layer == max_layer &&
 174                     first_sample == 0 && last_sample == max_sample) {
 175                         texture->dirty_level_mask &= ~(1 << level);
 176                 }
 177         }
 178
 179         sctx->dbcb_depth_copy_enabled = false;
 180         sctx->dbcb_stencil_copy_enabled = false;
 181         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 182 }
 183
 184 static void si_blit_decompress_depth_in_place(struct si_context *sctx,
 185                                               struct r600_texture *texture,
 186                                               unsigned first_level, unsigned last_level,
 187                                               unsigned first_layer, unsigned last_layer)
 188 {
 189         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 190         unsigned layer, max_layer, checked_last_layer, level;
 191
 192         sctx->db_inplace_flush_enabled = true;
 193         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 194
 195         surf_tmpl.format = texture->resource.b.b.format;
 196
 197         for (level = first_level; level <= last_level; level++) {
 198                 if (!(texture->dirty_level_mask & (1 << level)))
 199                         continue;
 200
 201                 surf_tmpl.u.tex.level = level;
 202
 203                 /* The smaller the mipmap level, the less layers there are
 204                  * as far as 3D textures are concerned. */
 205                 max_layer = util_max_layer(&texture->resource.b.b, level);
 206                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 207
 208                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 209                         surf_tmpl.u.tex.first_layer = layer;
 210                         surf_tmpl.u.tex.last_layer = layer;
 211
 212                         zsurf = sctx->b.b.create_surface(&sctx->b.b, &texture->resource.b.b, &surf_tmpl);
 213
 214                         si_blitter_begin(&sctx->b.b, SI_DECOMPRESS);
 215                         util_blitter_custom_depth_stencil(sctx->blitter, zsurf, NULL, ~0,
 216                                                           sctx->custom_dsa_flush,
 217                                                           1.0f);
 218                         si_blitter_end(&sctx->b.b);
 219
 220                         pipe_surface_reference(&zsurf, NULL);
 221                 }
 222
 223                 /* The texture will always be dirty if some layers aren't flushed.
 224                  * I don't think this case occurs often though. */
 225                 if (first_layer == 0 && last_layer == max_layer) {
 226                         texture->dirty_level_mask &= ~(1 << level);
 227                 }
 228         }
 229
 230         sctx->db_inplace_flush_enabled = false;
 231         si_mark_atom_dirty(sctx, &sctx->db_render_state);
 232 }
 233
 234 void si_flush_depth_textures(struct si_context *sctx,
 235                              struct si_textures_info *textures)
 236 {
 237         unsigned i;
 238         unsigned mask = textures->depth_texture_mask;
 239
 240         while (mask) {
 241                 struct pipe_sampler_view *view;
 242                 struct r600_texture *tex;
 243
 244                 i = u_bit_scan(&mask);
 245
 246                 view = textures->views.views[i];
 247                 assert(view);
 248
 249                 tex = (struct r600_texture *)view->texture;
 250                 assert(tex->is_depth && !tex->is_flushing_texture);
 251
 252                 si_blit_decompress_depth_in_place(sctx, tex,
 253                                                   view->u.tex.first_level, view->u.tex.last_level,
 254                                                   0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 255         }
 256 }
 257
 258 static void si_blit_decompress_color(struct pipe_context *ctx,
 259                 struct r600_texture *rtex,
 260                 unsigned first_level, unsigned last_level,
 261                 unsigned first_layer, unsigned last_layer)
 262 {
 263         struct si_context *sctx = (struct si_context *)ctx;
 264         unsigned layer, level, checked_last_layer, max_layer;
 265
 266         if (!rtex->dirty_level_mask)
 267                 return;
 268
 269         for (level = first_level; level <= last_level; level++) {
 270                 if (!(rtex->dirty_level_mask & (1 << level)))
 271                         continue;
 272
 273                 /* The smaller the mipmap level, the less layers there are
 274                  * as far as 3D textures are concerned. */
 275                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 276                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 277
 278                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 279                         struct pipe_surface *cbsurf, surf_tmpl;
 280
 281                         surf_tmpl.format = rtex->resource.b.b.format;
 282                         surf_tmpl.u.tex.level = level;
 283                         surf_tmpl.u.tex.first_layer = layer;
 284                         surf_tmpl.u.tex.last_layer = layer;
 285                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 286
 287                         si_blitter_begin(ctx, SI_DECOMPRESS);
 288                         util_blitter_custom_color(sctx->blitter, cbsurf,
 289                                 rtex->fmask.size ? sctx->custom_blend_decompress :
 290                                                    sctx->custom_blend_fastclear);
 291                         si_blitter_end(ctx);
 292
 293                         pipe_surface_reference(&cbsurf, NULL);
 294                 }
 295
 296                 /* The texture will always be dirty if some layers aren't flushed.
 297                  * I don't think this case occurs often though. */
 298                 if (first_layer == 0 && last_layer == max_layer) {
 299                         rtex->dirty_level_mask &= ~(1 << level);
 300                 }
 301         }
 302 }
 303
 304 void si_decompress_color_textures(struct si_context *sctx,
 305                                   struct si_textures_info *textures)
 306 {
 307         unsigned i;
 308         unsigned mask = textures->compressed_colortex_mask;
 309
 310         while (mask) {
 311                 struct pipe_sampler_view *view;
 312                 struct r600_texture *tex;
 313
 314                 i = u_bit_scan(&mask);
 315
 316                 view = textures->views.views[i];
 317                 assert(view);
 318
 319                 tex = (struct r600_texture *)view->texture;
 320                 assert(tex->cmask.size || tex->fmask.size);
 321
 322                 si_blit_decompress_color(&sctx->b.b, tex,
 323                                          view->u.tex.first_level, view->u.tex.last_level,
 324                                          0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 325         }
 326 }
 327
 328 static void si_clear(struct pipe_context *ctx, unsigned buffers,
 329                      const union pipe_color_union *color,
 330                      double depth, unsigned stencil)
 331 {
 332         struct si_context *sctx = (struct si_context *)ctx;
 333         struct pipe_framebuffer_state *fb = &sctx->framebuffer.state;
 334         struct pipe_surface *zsbuf = fb->zsbuf;
 335         struct r600_texture *zstex =
 336                 zsbuf ? (struct r600_texture*)zsbuf->texture : NULL;
 337
 338         if (buffers & PIPE_CLEAR_COLOR) {
 339                 evergreen_do_fast_color_clear(&sctx->b, fb, &sctx->framebuffer.atom,
 340                                               &buffers, color);
 341                 if (!buffers)
 342                         return; /* all buffers have been fast cleared */
 343         }
 344
 345         if (buffers & PIPE_CLEAR_COLOR) {
 346                 int i;
 347
 348                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 349                 for (i = 0; i < fb->nr_cbufs; i++) {
 350                         struct r600_texture *tex;
 351
 352                         /* If not clearing this buffer, skip. */
 353                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 354                                 continue;
 355
 356                         if (!fb->cbufs[i])
 357                                 continue;
 358
 359                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 360                         if (tex->fmask.size == 0)
 361                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 362                 }
 363         }
 364
 365         if (buffers & PIPE_CLEAR_DEPTH &&
 366             zstex && zstex->htile_buffer &&
 367             zsbuf->u.tex.level == 0 &&
 368             zsbuf->u.tex.first_layer == 0 &&
 369             zsbuf->u.tex.last_layer == util_max_layer(&zstex->resource.b.b, 0)) {
 370                 /* Need to disable EXPCLEAR temporarily if clearing
 371                  * to a new value. */
 372                 if (zstex->depth_cleared && zstex->depth_clear_value != depth) {
 373                         sctx->db_depth_disable_expclear = true;
 374                 }
 375
 376                 zstex->depth_clear_value = depth;
 377                 si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_DEPTH_CLEAR */
 378                 sctx->db_depth_clear = true;
 379                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 380         }
 381
 382         si_blitter_begin(ctx, SI_CLEAR);
 383         util_blitter_clear(sctx->blitter, fb->width, fb->height,
 384                            util_framebuffer_get_num_layers(fb),
 385                            buffers, color, depth, stencil);
 386         si_blitter_end(ctx);
 387
 388         if (sctx->db_depth_clear) {
 389                 sctx->db_depth_clear = false;
 390                 sctx->db_depth_disable_expclear = false;
 391                 zstex->depth_cleared = true;
 392                 si_mark_atom_dirty(sctx, &sctx->db_render_state);
 393         }
 394 }
 395
 396 static void si_clear_render_target(struct pipe_context *ctx,
 397                                    struct pipe_surface *dst,
 398                                    const union pipe_color_union *color,
 399                                    unsigned dstx, unsigned dsty,
 400                                    unsigned width, unsigned height)
 401 {
 402         struct si_context *sctx = (struct si_context *)ctx;
 403
 404         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 405         util_blitter_clear_render_target(sctx->blitter, dst, color,
 406                                          dstx, dsty, width, height);
 407         si_blitter_end(ctx);
 408 }
 409
 410 static void si_clear_depth_stencil(struct pipe_context *ctx,
 411                                    struct pipe_surface *dst,
 412                                    unsigned clear_flags,
 413                                    double depth,
 414                                    unsigned stencil,
 415                                    unsigned dstx, unsigned dsty,
 416                                    unsigned width, unsigned height)
 417 {
 418         struct si_context *sctx = (struct si_context *)ctx;
 419
 420         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 421         util_blitter_clear_depth_stencil(sctx->blitter, dst, clear_flags, depth, stencil,
 422                                          dstx, dsty, width, height);
 423         si_blitter_end(ctx);
 424 }
 425
 426 /* Helper for decompressing a portion of a color or depth resource before
 427  * blitting if any decompression is needed.
 428  * The driver doesn't decompress resources automatically while u_blitter is
 429  * rendering. */
 430 static void si_decompress_subresource(struct pipe_context *ctx,
 431                                       struct pipe_resource *tex,
 432                                       unsigned level,
 433                                       unsigned first_layer, unsigned last_layer)
 434 {
 435         struct si_context *sctx = (struct si_context *)ctx;
 436         struct r600_texture *rtex = (struct r600_texture*)tex;
 437
 438         if (rtex->is_depth && !rtex->is_flushing_texture) {
 439                 si_blit_decompress_depth_in_place(sctx, rtex,
 440                                                   level, level,
 441                                                   first_layer, last_layer);
 442         } else if (rtex->fmask.size || rtex->cmask.size) {
 443                 si_blit_decompress_color(ctx, rtex, level, level,
 444                                          first_layer, last_layer);
 445         }
 446 }
 447
 448 struct texture_orig_info {
 449         unsigned format;
 450         unsigned width0;
 451         unsigned height0;
 452         unsigned npix_x;
 453         unsigned npix_y;
 454         unsigned npix0_x;
 455         unsigned npix0_y;
 456 };
 457
 458 void si_resource_copy_region(struct pipe_context *ctx,
 459                              struct pipe_resource *dst,
 460                              unsigned dst_level,
 461                              unsigned dstx, unsigned dsty, unsigned dstz,
 462                              struct pipe_resource *src,
 463                              unsigned src_level,
 464                              const struct pipe_box *src_box)
 465 {
 466         struct si_context *sctx = (struct si_context *)ctx;
 467         struct pipe_surface *dst_view, dst_templ;
 468         struct pipe_sampler_view src_templ, *src_view;
 469         unsigned dst_width, dst_height, src_width0, src_height0;
 470         unsigned src_force_level = 0;
 471         struct pipe_box sbox, dstbox;
 472
 473         /* Handle buffers first. */
 474         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 475                 si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width, false);
 476                 return;
 477         }
 478
 479         assert(u_max_sample(dst) == u_max_sample(src));
 480
 481         /* The driver doesn't decompress resources automatically while
 482          * u_blitter is rendering. */
 483         si_decompress_subresource(ctx, src, src_level,
 484                                   src_box->z, src_box->z + src_box->depth - 1);
 485
 486         dst_width = u_minify(dst->width0, dst_level);
 487         dst_height = u_minify(dst->height0, dst_level);
 488         src_width0 = src->width0;
 489         src_height0 = src->height0;
 490
 491         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 492         util_blitter_default_src_texture(&src_templ, src, src_level);
 493
 494         if (util_format_is_compressed(src->format) &&
 495             util_format_is_compressed(dst->format)) {
 496                 unsigned blocksize = util_format_get_blocksize(src->format);
 497
 498                 if (blocksize == 8)
 499                         src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 500                 else
 501                         src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 502                 dst_templ.format = src_templ.format;
 503
 504                 dst_width = util_format_get_nblocksx(dst->format, dst_width);
 505                 dst_height = util_format_get_nblocksy(dst->format, dst_height);
 506                 src_width0 = util_format_get_nblocksx(src->format, src_width0);
 507                 src_height0 = util_format_get_nblocksy(src->format, src_height0);
 508
 509                 dstx = util_format_get_nblocksx(dst->format, dstx);
 510                 dsty = util_format_get_nblocksy(dst->format, dsty);
 511
 512                 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 513                 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
 514                 sbox.z = src_box->z;
 515                 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 516                 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
 517                 sbox.depth = src_box->depth;
 518                 src_box = &sbox;
 519
 520                 src_force_level = src_level;
 521         } else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src) ||
 522                    /* also *8_SNORM has precision issues, use UNORM instead */
 523                    util_format_is_snorm(src->format)) {
 524                 if (util_format_is_subsampled_422(src->format)) {
 525                         src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 526                         dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 527
 528                         dst_width = util_format_get_nblocksx(dst->format, dst_width);
 529                         src_width0 = util_format_get_nblocksx(src->format, src_width0);
 530
 531                         dstx = util_format_get_nblocksx(dst->format, dstx);
 532
 533                         sbox = *src_box;
 534                         sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 535                         sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 536                         src_box = &sbox;
 537                 } else {
 538                         unsigned blocksize = util_format_get_blocksize(src->format);
 539
 540                         switch (blocksize) {
 541                         case 1:
 542                                 dst_templ.format = PIPE_FORMAT_R8_UNORM;
 543                                 src_templ.format = PIPE_FORMAT_R8_UNORM;
 544                                 break;
 545                         case 2:
 546                                 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
 547                                 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
 548                                 break;
 549                         case 4:
 550                                 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 551                                 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 552                                 break;
 553                         case 8:
 554                                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 555                                 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 556                                 break;
 557                         case 16:
 558                                 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 559                                 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 560                                 break;
 561                         default:
 562                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 563                                         util_format_short_name(src->format), blocksize);
 564                                 assert(0);
 565                         }
 566                 }
 567         }
 568
 569         /* Initialize the surface. */
 570         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
 571                                               dst_width, dst_height);
 572
 573         /* Initialize the sampler view. */
 574         src_view = si_create_sampler_view_custom(ctx, src, &src_templ,
 575                                                  src_width0, src_height0,
 576                                                  src_force_level);
 577
 578         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 579                  abs(src_box->depth), &dstbox);
 580
 581         /* Copy. */
 582         si_blitter_begin(ctx, SI_COPY);
 583         util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
 584                                   src_view, src_box, src_width0, src_height0,
 585                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
 586                                   FALSE);
 587         si_blitter_end(ctx);
 588
 589         pipe_surface_reference(&dst_view, NULL);
 590         pipe_sampler_view_reference(&src_view, NULL);
 591 }
 592
 593 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 594 static enum pipe_format int_to_norm_format(enum pipe_format format)
 595 {
 596         switch (format) {
 597 #define REPLACE_FORMAT_SIGN(format,sign) \
 598         case PIPE_FORMAT_##format##_##sign##INT: \
 599                 return PIPE_FORMAT_##format##_##sign##NORM
 600 #define REPLACE_FORMAT(format) \
 601                 REPLACE_FORMAT_SIGN(format, U); \
 602                 REPLACE_FORMAT_SIGN(format, S)
 603
 604         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 605         REPLACE_FORMAT(R8);
 606         REPLACE_FORMAT(R8G8);
 607         REPLACE_FORMAT(R8G8B8X8);
 608         REPLACE_FORMAT(R8G8B8A8);
 609         REPLACE_FORMAT(A8);
 610         REPLACE_FORMAT(I8);
 611         REPLACE_FORMAT(L8);
 612         REPLACE_FORMAT(L8A8);
 613         REPLACE_FORMAT(R16);
 614         REPLACE_FORMAT(R16G16);
 615         REPLACE_FORMAT(R16G16B16X16);
 616         REPLACE_FORMAT(R16G16B16A16);
 617         REPLACE_FORMAT(A16);
 618         REPLACE_FORMAT(I16);
 619         REPLACE_FORMAT(L16);
 620         REPLACE_FORMAT(L16A16);
 621
 622 #undef REPLACE_FORMAT
 623 #undef REPLACE_FORMAT_SIGN
 624         default:
 625                 return format;
 626         }
 627 }
 628
 629 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 630                                      const struct pipe_blit_info *info)
 631 {
 632         struct si_context *sctx = (struct si_context*)ctx;
 633         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 634         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 635         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 636         enum pipe_format format = int_to_norm_format(info->dst.format);
 637         unsigned sample_mask = ~0;
 638
 639         if (info->src.resource->nr_samples > 1 &&
 640             info->dst.resource->nr_samples <= 1 &&
 641             util_max_layer(info->src.resource, 0) == 0 &&
 642             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 643             info->dst.format == info->src.format &&
 644             !util_format_is_pure_integer(format) &&
 645             !util_format_is_depth_or_stencil(format) &&
 646             !info->scissor_enable &&
 647             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 648             dst_width == info->src.resource->width0 &&
 649             dst_height == info->src.resource->height0 &&
 650             info->dst.box.x == 0 &&
 651             info->dst.box.y == 0 &&
 652             info->dst.box.width == dst_width &&
 653             info->dst.box.height == dst_height &&
 654             info->dst.box.depth == 1 &&
 655             info->src.box.x == 0 &&
 656             info->src.box.y == 0 &&
 657             info->src.box.width == dst_width &&
 658             info->src.box.height == dst_height &&
 659             info->src.box.depth == 1 &&
 660             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 661             !(dst->surface.flags & RADEON_SURF_SCANOUT) &&
 662             (!dst->cmask.size || !dst->dirty_level_mask) /* dst cannot be fast-cleared */) {
 663                 si_blitter_begin(ctx, SI_COLOR_RESOLVE |
 664                                  (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 665                 util_blitter_custom_resolve_color(sctx->blitter,
 666                                                   info->dst.resource, info->dst.level,
 667                                                   info->dst.box.z,
 668                                                   info->src.resource, info->src.box.z,
 669                                                   sample_mask, sctx->custom_blend_resolve,
 670                                                   format);
 671                 si_blitter_end(ctx);
 672                 return true;
 673         }
 674         return false;
 675 }
 676
 677 static void si_blit(struct pipe_context *ctx,
 678                     const struct pipe_blit_info *info)
 679 {
 680         struct si_context *sctx = (struct si_context*)ctx;
 681
 682         if (do_hardware_msaa_resolve(ctx, info)) {
 683                 return;
 684         }
 685
 686         assert(util_blitter_is_blit_supported(sctx->blitter, info));
 687
 688         /* The driver doesn't decompress resources automatically while
 689          * u_blitter is rendering. */
 690         si_decompress_subresource(ctx, info->src.resource, info->src.level,
 691                                   info->src.box.z,
 692                                   info->src.box.z + info->src.box.depth - 1);
 693
 694         if (sctx->screen->b.debug_flags & DBG_FORCE_DMA &&
 695             util_try_blit_via_copy_region(ctx, info))
 696                 return;
 697
 698         si_blitter_begin(ctx, SI_BLIT |
 699                          (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 700         util_blitter_blit(sctx->blitter, info);
 701         si_blitter_end(ctx);
 702 }
 703
 704 static void si_flush_resource(struct pipe_context *ctx,
 705                               struct pipe_resource *res)
 706 {
 707         struct r600_texture *rtex = (struct r600_texture*)res;
 708
 709         assert(res->target != PIPE_BUFFER);
 710
 711         if (!rtex->is_depth && rtex->cmask.size) {
 712                 si_blit_decompress_color(ctx, rtex, 0, res->last_level,
 713                                          0, util_max_layer(res, 0));
 714         }
 715 }
 716
 717 void si_init_blit_functions(struct si_context *sctx)
 718 {
 719         sctx->b.b.clear = si_clear;
 720         sctx->b.b.clear_render_target = si_clear_render_target;
 721         sctx->b.b.clear_depth_stencil = si_clear_depth_stencil;
 722         sctx->b.b.resource_copy_region = si_resource_copy_region;
 723         sctx->b.b.blit = si_blit;
 724         sctx->b.b.flush_resource = si_flush_resource;
 725         sctx->b.blit_decompress_depth = si_blit_decompress_depth;
 726 }