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_blitter.h"
  26 #include "util/u_format.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_vertex_shader(sctx->blitter, sctx->vs_shader);
  61         util_blitter_save_vertex_elements(sctx->blitter, sctx->vertex_elements);
  62         if (sctx->queued.named.viewport) {
  63                 util_blitter_save_viewport(sctx->blitter, &sctx->queued.named.viewport->viewport);
  64         }
  65         util_blitter_save_vertex_buffer_slot(sctx->blitter, sctx->vertex_buffer);
  66         util_blitter_save_so_targets(sctx->blitter, sctx->b.streamout.num_targets,
  67                                      (struct pipe_stream_output_target**)sctx->b.streamout.targets);
  68
  69         if (op & SI_SAVE_FRAMEBUFFER)
  70                 util_blitter_save_framebuffer(sctx->blitter, &sctx->framebuffer.state);
  71
  72         if (op & SI_SAVE_TEXTURES) {
  73                 util_blitter_save_fragment_sampler_states(
  74                         sctx->blitter, sctx->samplers[PIPE_SHADER_FRAGMENT].n_samplers,
  75                         (void**)sctx->samplers[PIPE_SHADER_FRAGMENT].samplers);
  76
  77                 util_blitter_save_fragment_sampler_views(sctx->blitter,
  78                         util_last_bit(sctx->samplers[PIPE_SHADER_FRAGMENT].views.desc.enabled_mask &
  79                                       ((1 << NUM_TEX_UNITS) - 1)),
  80                         sctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
  81         }
  82
  83         if ((op & SI_DISABLE_RENDER_COND) && sctx->b.current_render_cond) {
  84                 util_blitter_save_render_condition(sctx->blitter,
  85                                                    sctx->b.current_render_cond,
  86                                                    sctx->b.current_render_cond_cond,
  87                                                    sctx->b.current_render_cond_mode);
  88         }
  89 }
  90
  91 static void si_blitter_end(struct pipe_context *ctx)
  92 {
  93         struct si_context *sctx = (struct si_context *)ctx;
  94         r600_resume_nontimer_queries(&sctx->b);
  95 }
  96
  97 static unsigned u_max_sample(struct pipe_resource *r)
  98 {
  99         return r->nr_samples ? r->nr_samples - 1 : 0;
 100 }
 101
 102 static void si_blit_decompress_depth(struct pipe_context *ctx,
 103                                      struct r600_texture *texture,
 104                                      struct r600_texture *staging,
 105                                      unsigned first_level, unsigned last_level,
 106                                      unsigned first_layer, unsigned last_layer,
 107                                      unsigned first_sample, unsigned last_sample)
 108 {
 109         struct si_context *sctx = (struct si_context *)ctx;
 110         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 111         float depth = 1.0f;
 112         const struct util_format_description *desc;
 113         void **custom_dsa;
 114         struct r600_texture *flushed_depth_texture = staging ?
 115                         staging : texture->flushed_depth_texture;
 116
 117         if (!staging && !texture->dirty_level_mask)
 118                 return;
 119
 120         max_sample = u_max_sample(&texture->resource.b.b);
 121
 122         desc = util_format_description(flushed_depth_texture->resource.b.b.format);
 123         switch (util_format_has_depth(desc) | util_format_has_stencil(desc) << 1) {
 124         default:
 125                 assert(!"No depth or stencil to uncompress");
 126                 return;
 127         case 3:
 128                 custom_dsa = sctx->custom_dsa_flush_depth_stencil;
 129                 break;
 130         case 2:
 131                 custom_dsa = sctx->custom_dsa_flush_stencil;
 132                 break;
 133         case 1:
 134                 custom_dsa = sctx->custom_dsa_flush_depth;
 135                 break;
 136         }
 137
 138         for (level = first_level; level <= last_level; level++) {
 139                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 140                         continue;
 141
 142                 /* The smaller the mipmap level, the less layers there are
 143                  * as far as 3D textures are concerned. */
 144                 max_layer = util_max_layer(&texture->resource.b.b, level);
 145                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 146
 147                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 148                         for (sample = first_sample; sample <= last_sample; sample++) {
 149                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 150
 151                                 surf_tmpl.format = texture->resource.b.b.format;
 152                                 surf_tmpl.u.tex.level = level;
 153                                 surf_tmpl.u.tex.first_layer = layer;
 154                                 surf_tmpl.u.tex.last_layer = layer;
 155
 156                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 157
 158                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 159                                 cbsurf = ctx->create_surface(ctx,
 160                                                 (struct pipe_resource*)flushed_depth_texture, &surf_tmpl);
 161
 162                                 si_blitter_begin(ctx, SI_DECOMPRESS);
 163                                 util_blitter_custom_depth_stencil(sctx->blitter, zsurf, cbsurf, 1 << sample,
 164                                                                   custom_dsa[sample], depth);
 165                                 si_blitter_end(ctx);
 166
 167                                 pipe_surface_reference(&zsurf, NULL);
 168                                 pipe_surface_reference(&cbsurf, NULL);
 169                         }
 170                 }
 171
 172                 /* The texture will always be dirty if some layers aren't flushed.
 173                  * I don't think this case can occur though. */
 174                 if (!staging &&
 175                     first_layer == 0 && last_layer == max_layer &&
 176                     first_sample == 0 && last_sample == max_sample) {
 177                         texture->dirty_level_mask &= ~(1 << level);
 178                 }
 179         }
 180 }
 181
 182 static void si_blit_decompress_depth_in_place(struct si_context *sctx,
 183                                               struct r600_texture *texture,
 184                                               unsigned first_level, unsigned last_level,
 185                                               unsigned first_layer, unsigned last_layer)
 186 {
 187         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 188         unsigned layer, max_layer, checked_last_layer, level;
 189
 190         surf_tmpl.format = texture->resource.b.b.format;
 191
 192         for (level = first_level; level <= last_level; level++) {
 193                 if (!(texture->dirty_level_mask & (1 << level)))
 194                         continue;
 195
 196                 surf_tmpl.u.tex.level = level;
 197
 198                 /* The smaller the mipmap level, the less layers there are
 199                  * as far as 3D textures are concerned. */
 200                 max_layer = util_max_layer(&texture->resource.b.b, level);
 201                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 202
 203                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 204                         surf_tmpl.u.tex.first_layer = layer;
 205                         surf_tmpl.u.tex.last_layer = layer;
 206
 207                         zsurf = sctx->b.b.create_surface(&sctx->b.b, &texture->resource.b.b, &surf_tmpl);
 208
 209                         si_blitter_begin(&sctx->b.b, SI_DECOMPRESS);
 210                         util_blitter_custom_depth_stencil(sctx->blitter, zsurf, NULL, ~0,
 211                                                           sctx->custom_dsa_flush_inplace,
 212                                                           1.0f);
 213                         si_blitter_end(&sctx->b.b);
 214
 215                         pipe_surface_reference(&zsurf, NULL);
 216                 }
 217
 218                 /* The texture will always be dirty if some layers aren't flushed.
 219                  * I don't think this case occurs often though. */
 220                 if (first_layer == 0 && last_layer == max_layer) {
 221                         texture->dirty_level_mask &= ~(1 << level);
 222                 }
 223         }
 224 }
 225
 226 void si_flush_depth_textures(struct si_context *sctx,
 227                              struct si_textures_info *textures)
 228 {
 229         unsigned i;
 230
 231         for (i = 0; i < textures->n_views; ++i) {
 232                 struct pipe_sampler_view *view;
 233                 struct r600_texture *tex;
 234
 235                 view = textures->views.views[i];
 236                 if (!view) continue;
 237
 238                 tex = (struct r600_texture *)view->texture;
 239                 if (!tex->is_depth || tex->is_flushing_texture)
 240                         continue;
 241
 242                 si_blit_decompress_depth_in_place(sctx, tex,
 243                                                   view->u.tex.first_level, view->u.tex.last_level,
 244                                                   0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 245         }
 246 }
 247
 248 static void si_blit_decompress_color(struct pipe_context *ctx,
 249                 struct r600_texture *rtex,
 250                 unsigned first_level, unsigned last_level,
 251                 unsigned first_layer, unsigned last_layer)
 252 {
 253         struct si_context *sctx = (struct si_context *)ctx;
 254         unsigned layer, level, checked_last_layer, max_layer;
 255
 256         if (!rtex->dirty_level_mask)
 257                 return;
 258
 259         for (level = first_level; level <= last_level; level++) {
 260                 if (!(rtex->dirty_level_mask & (1 << level)))
 261                         continue;
 262
 263                 /* The smaller the mipmap level, the less layers there are
 264                  * as far as 3D textures are concerned. */
 265                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 266                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 267
 268                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 269                         struct pipe_surface *cbsurf, surf_tmpl;
 270
 271                         surf_tmpl.format = rtex->resource.b.b.format;
 272                         surf_tmpl.u.tex.level = level;
 273                         surf_tmpl.u.tex.first_layer = layer;
 274                         surf_tmpl.u.tex.last_layer = layer;
 275                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 276
 277                         si_blitter_begin(ctx, SI_DECOMPRESS);
 278                         util_blitter_custom_color(sctx->blitter, cbsurf,
 279                                 rtex->fmask.size ? sctx->custom_blend_decompress :
 280                                                    sctx->custom_blend_fastclear);
 281                         si_blitter_end(ctx);
 282
 283                         pipe_surface_reference(&cbsurf, NULL);
 284                 }
 285
 286                 /* The texture will always be dirty if some layers aren't flushed.
 287                  * I don't think this case occurs often though. */
 288                 if (first_layer == 0 && last_layer == max_layer) {
 289                         rtex->dirty_level_mask &= ~(1 << level);
 290                 }
 291         }
 292 }
 293
 294 void si_decompress_color_textures(struct si_context *sctx,
 295                                   struct si_textures_info *textures)
 296 {
 297         unsigned i;
 298         unsigned mask = textures->compressed_colortex_mask;
 299
 300         while (mask) {
 301                 struct pipe_sampler_view *view;
 302                 struct r600_texture *tex;
 303
 304                 i = u_bit_scan(&mask);
 305
 306                 view = textures->views.views[i];
 307                 assert(view);
 308
 309                 tex = (struct r600_texture *)view->texture;
 310                 assert(tex->cmask.size || tex->fmask.size);
 311
 312                 si_blit_decompress_color(&sctx->b.b, tex,
 313                                          view->u.tex.first_level, view->u.tex.last_level,
 314                                          0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 315         }
 316 }
 317
 318 static void si_clear(struct pipe_context *ctx, unsigned buffers,
 319                      const union pipe_color_union *color,
 320                      double depth, unsigned stencil)
 321 {
 322         struct si_context *sctx = (struct si_context *)ctx;
 323         struct pipe_framebuffer_state *fb = &sctx->framebuffer.state;
 324
 325         if (buffers & PIPE_CLEAR_COLOR) {
 326                 evergreen_do_fast_color_clear(&sctx->b, fb, &sctx->framebuffer.atom,
 327                                               &buffers, color);
 328         }
 329
 330         if (buffers & PIPE_CLEAR_COLOR) {
 331                 int i;
 332
 333                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 334                 for (i = 0; i < fb->nr_cbufs; i++) {
 335                         struct r600_texture *tex;
 336
 337                         /* If not clearing this buffer, skip. */
 338                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 339                                 continue;
 340
 341                         if (!fb->cbufs[i])
 342                                 continue;
 343
 344                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 345                         if (tex->fmask.size == 0)
 346                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 347                 }
 348         }
 349
 350         si_blitter_begin(ctx, SI_CLEAR);
 351         util_blitter_clear(sctx->blitter, fb->width, fb->height,
 352                            util_framebuffer_get_num_layers(fb),
 353                            buffers, color, depth, stencil);
 354         si_blitter_end(ctx);
 355 }
 356
 357 static void si_clear_render_target(struct pipe_context *ctx,
 358                                    struct pipe_surface *dst,
 359                                    const union pipe_color_union *color,
 360                                    unsigned dstx, unsigned dsty,
 361                                    unsigned width, unsigned height)
 362 {
 363         struct si_context *sctx = (struct si_context *)ctx;
 364
 365         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 366         util_blitter_clear_render_target(sctx->blitter, dst, color,
 367                                          dstx, dsty, width, height);
 368         si_blitter_end(ctx);
 369 }
 370
 371 static void si_clear_depth_stencil(struct pipe_context *ctx,
 372                                    struct pipe_surface *dst,
 373                                    unsigned clear_flags,
 374                                    double depth,
 375                                    unsigned stencil,
 376                                    unsigned dstx, unsigned dsty,
 377                                    unsigned width, unsigned height)
 378 {
 379         struct si_context *sctx = (struct si_context *)ctx;
 380
 381         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 382         util_blitter_clear_depth_stencil(sctx->blitter, dst, clear_flags, depth, stencil,
 383                                          dstx, dsty, width, height);
 384         si_blitter_end(ctx);
 385 }
 386
 387 /* Helper for decompressing a portion of a color or depth resource before
 388  * blitting if any decompression is needed.
 389  * The driver doesn't decompress resources automatically while u_blitter is
 390  * rendering. */
 391 static void si_decompress_subresource(struct pipe_context *ctx,
 392                                       struct pipe_resource *tex,
 393                                       unsigned level,
 394                                       unsigned first_layer, unsigned last_layer)
 395 {
 396         struct si_context *sctx = (struct si_context *)ctx;
 397         struct r600_texture *rtex = (struct r600_texture*)tex;
 398
 399         if (rtex->is_depth && !rtex->is_flushing_texture) {
 400                 si_blit_decompress_depth_in_place(sctx, rtex,
 401                                                   level, level,
 402                                                   first_layer, last_layer);
 403         } else if (rtex->fmask.size || rtex->cmask.size) {
 404                 si_blit_decompress_color(ctx, rtex, level, level,
 405                                          first_layer, last_layer);
 406         }
 407 }
 408
 409 struct texture_orig_info {
 410         unsigned format;
 411         unsigned width0;
 412         unsigned height0;
 413         unsigned npix_x;
 414         unsigned npix_y;
 415         unsigned npix0_x;
 416         unsigned npix0_y;
 417 };
 418
 419 static void si_compressed_to_blittable(struct pipe_resource *tex,
 420                                        unsigned level,
 421                                        struct texture_orig_info *orig)
 422 {
 423         struct r600_texture *rtex = (struct r600_texture*)tex;
 424         unsigned pixsize = util_format_get_blocksize(rtex->resource.b.b.format);
 425         int new_format;
 426         int new_height, new_width;
 427
 428         orig->format = tex->format;
 429         orig->width0 = tex->width0;
 430         orig->height0 = tex->height0;
 431         orig->npix0_x = rtex->surface.level[0].npix_x;
 432         orig->npix0_y = rtex->surface.level[0].npix_y;
 433         orig->npix_x = rtex->surface.level[level].npix_x;
 434         orig->npix_y = rtex->surface.level[level].npix_y;
 435
 436         if (pixsize == 8)
 437                 new_format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 438         else
 439                 new_format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 440
 441         new_width = util_format_get_nblocksx(tex->format, orig->width0);
 442         new_height = util_format_get_nblocksy(tex->format, orig->height0);
 443
 444         tex->width0 = new_width;
 445         tex->height0 = new_height;
 446         tex->format = new_format;
 447         rtex->surface.level[0].npix_x = util_format_get_nblocksx(orig->format, orig->npix0_x);
 448         rtex->surface.level[0].npix_y = util_format_get_nblocksy(orig->format, orig->npix0_y);
 449         rtex->surface.level[level].npix_x = util_format_get_nblocksx(orig->format, orig->npix_x);
 450         rtex->surface.level[level].npix_y = util_format_get_nblocksy(orig->format, orig->npix_y);
 451
 452         /* By dividing the dimensions by 4, we effectively decrement
 453          * last_level by 2, therefore the last 2 mipmap levels disappear and
 454          * aren't blittable. Note that the last 3 mipmap levels (4x4, 2x2,
 455          * 1x1) have equal slice sizes, which is an important assumption
 456          * for this to work.
 457          *
 458          * In order to make the last 2 mipmap levels blittable, we have to
 459          * add the slice size of the last mipmap level to the texture
 460          * address, so that even though the hw thinks it reads last_level-2,
 461          * it will actually read last_level-1, and if we add the slice size*2,
 462          * it will read last_level. That's how this workaround works.
 463          */
 464         if (level > rtex->resource.b.b.last_level-2)
 465                 rtex->mipmap_shift = level - (rtex->resource.b.b.last_level-2);
 466 }
 467
 468 static void si_change_format(struct pipe_resource *tex,
 469                              unsigned level,
 470                              struct texture_orig_info *orig,
 471                              enum pipe_format format)
 472 {
 473         struct r600_texture *rtex = (struct r600_texture*)tex;
 474
 475         orig->format = tex->format;
 476         orig->width0 = tex->width0;
 477         orig->height0 = tex->height0;
 478         orig->npix0_x = rtex->surface.level[0].npix_x;
 479         orig->npix0_y = rtex->surface.level[0].npix_y;
 480         orig->npix_x = rtex->surface.level[level].npix_x;
 481         orig->npix_y = rtex->surface.level[level].npix_y;
 482
 483         tex->format = format;
 484 }
 485
 486 static void si_reset_blittable_to_orig(struct pipe_resource *tex,
 487                                        unsigned level,
 488                                        struct texture_orig_info *orig)
 489 {
 490         struct r600_texture *rtex = (struct r600_texture*)tex;
 491
 492         tex->format = orig->format;
 493         tex->width0 = orig->width0;
 494         tex->height0 = orig->height0;
 495         rtex->surface.level[0].npix_x = orig->npix0_x;
 496         rtex->surface.level[0].npix_y = orig->npix0_y;
 497         rtex->surface.level[level].npix_x = orig->npix_x;
 498         rtex->surface.level[level].npix_y = orig->npix_y;
 499         rtex->mipmap_shift = 0;
 500 }
 501
 502 static void si_resource_copy_region(struct pipe_context *ctx,
 503                                     struct pipe_resource *dst,
 504                                     unsigned dst_level,
 505                                     unsigned dstx, unsigned dsty, unsigned dstz,
 506                                     struct pipe_resource *src,
 507                                     unsigned src_level,
 508                                     const struct pipe_box *src_box)
 509 {
 510         struct si_context *sctx = (struct si_context *)ctx;
 511         struct r600_texture *rdst = (struct r600_texture*)dst;
 512         struct pipe_surface *dst_view, dst_templ;
 513         struct pipe_sampler_view src_templ, *src_view;
 514         struct texture_orig_info orig_info[2];
 515         struct pipe_box sbox, dstbox;
 516         boolean restore_orig[2];
 517
 518         /* Fallback for buffers. */
 519         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 520                 si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width);
 521                 return;
 522         }
 523
 524         memset(orig_info, 0, sizeof(orig_info));
 525
 526         /* The driver doesn't decompress resources automatically while
 527          * u_blitter is rendering. */
 528         si_decompress_subresource(ctx, src, src_level,
 529                                   src_box->z, src_box->z + src_box->depth - 1);
 530
 531         restore_orig[0] = restore_orig[1] = FALSE;
 532
 533         if (util_format_is_compressed(src->format) &&
 534             util_format_is_compressed(dst->format)) {
 535                 si_compressed_to_blittable(src, src_level, &orig_info[0]);
 536                 restore_orig[0] = TRUE;
 537                 sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
 538                 sbox.y = util_format_get_nblocksy(orig_info[0].format, src_box->y);
 539                 sbox.z = src_box->z;
 540                 sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
 541                 sbox.height = util_format_get_nblocksy(orig_info[0].format, src_box->height);
 542                 sbox.depth = src_box->depth;
 543                 src_box = &sbox;
 544
 545                 si_compressed_to_blittable(dst, dst_level, &orig_info[1]);
 546                 restore_orig[1] = TRUE;
 547                 /* translate the dst box as well */
 548                 dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
 549                 dsty = util_format_get_nblocksy(orig_info[1].format, dsty);
 550         } else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src)) {
 551                 if (util_format_is_subsampled_422(src->format)) {
 552                         /* XXX untested */
 553                         si_change_format(src, src_level, &orig_info[0],
 554                                          PIPE_FORMAT_R8G8B8A8_UINT);
 555                         si_change_format(dst, dst_level, &orig_info[1],
 556                                          PIPE_FORMAT_R8G8B8A8_UINT);
 557
 558                         sbox = *src_box;
 559                         sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
 560                         sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
 561                         src_box = &sbox;
 562                         dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
 563
 564                         restore_orig[0] = TRUE;
 565                         restore_orig[1] = TRUE;
 566                 } else {
 567                         unsigned blocksize = util_format_get_blocksize(src->format);
 568
 569                         switch (blocksize) {
 570                         case 1:
 571                                 si_change_format(src, src_level, &orig_info[0],
 572                                                 PIPE_FORMAT_R8_UNORM);
 573                                 si_change_format(dst, dst_level, &orig_info[1],
 574                                                 PIPE_FORMAT_R8_UNORM);
 575                                 break;
 576                         case 2:
 577                                 si_change_format(src, src_level, &orig_info[0],
 578                                                 PIPE_FORMAT_R8G8_UNORM);
 579                                 si_change_format(dst, dst_level, &orig_info[1],
 580                                                 PIPE_FORMAT_R8G8_UNORM);
 581                                 break;
 582                         case 4:
 583                                 si_change_format(src, src_level, &orig_info[0],
 584                                                 PIPE_FORMAT_R8G8B8A8_UNORM);
 585                                 si_change_format(dst, dst_level, &orig_info[1],
 586                                                 PIPE_FORMAT_R8G8B8A8_UNORM);
 587                                 break;
 588                         case 8:
 589                                 si_change_format(src, src_level, &orig_info[0],
 590                                                 PIPE_FORMAT_R16G16B16A16_UINT);
 591                                 si_change_format(dst, dst_level, &orig_info[1],
 592                                                 PIPE_FORMAT_R16G16B16A16_UINT);
 593                                 break;
 594                         case 16:
 595                                 si_change_format(src, src_level, &orig_info[0],
 596                                                 PIPE_FORMAT_R32G32B32A32_UINT);
 597                                 si_change_format(dst, dst_level, &orig_info[1],
 598                                                 PIPE_FORMAT_R32G32B32A32_UINT);
 599                                 break;
 600                         default:
 601                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 602                                         util_format_short_name(src->format), blocksize);
 603                                 assert(0);
 604                         }
 605                         restore_orig[0] = TRUE;
 606                         restore_orig[1] = TRUE;
 607                 }
 608         }
 609
 610         /* Initialize the surface. */
 611         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 612         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
 613                                               rdst->surface.level[dst_level].npix_x,
 614                                               rdst->surface.level[dst_level].npix_y);
 615
 616         /* Initialize the sampler view. */
 617         util_blitter_default_src_texture(&src_templ, src, src_level);
 618         src_view = ctx->create_sampler_view(ctx, src, &src_templ);
 619
 620         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 621                  abs(src_box->depth), &dstbox);
 622
 623         /* Copy. */
 624         si_blitter_begin(ctx, SI_COPY);
 625         util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
 626                                   src_view, src_box, src->width0, src->height0,
 627                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
 628         si_blitter_end(ctx);
 629
 630         pipe_surface_reference(&dst_view, NULL);
 631         pipe_sampler_view_reference(&src_view, NULL);
 632
 633         if (restore_orig[0])
 634                 si_reset_blittable_to_orig(src, src_level, &orig_info[0]);
 635
 636         if (restore_orig[1])
 637                 si_reset_blittable_to_orig(dst, dst_level, &orig_info[1]);
 638 }
 639
 640 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 641 static enum pipe_format int_to_norm_format(enum pipe_format format)
 642 {
 643         switch (format) {
 644 #define REPLACE_FORMAT_SIGN(format,sign) \
 645         case PIPE_FORMAT_##format##_##sign##INT: \
 646                 return PIPE_FORMAT_##format##_##sign##NORM
 647 #define REPLACE_FORMAT(format) \
 648                 REPLACE_FORMAT_SIGN(format, U); \
 649                 REPLACE_FORMAT_SIGN(format, S)
 650
 651         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 652         REPLACE_FORMAT(R8);
 653         REPLACE_FORMAT(R8G8);
 654         REPLACE_FORMAT(R8G8B8X8);
 655         REPLACE_FORMAT(R8G8B8A8);
 656         REPLACE_FORMAT(A8);
 657         REPLACE_FORMAT(I8);
 658         REPLACE_FORMAT(L8);
 659         REPLACE_FORMAT(L8A8);
 660         REPLACE_FORMAT(R16);
 661         REPLACE_FORMAT(R16G16);
 662         REPLACE_FORMAT(R16G16B16X16);
 663         REPLACE_FORMAT(R16G16B16A16);
 664         REPLACE_FORMAT(A16);
 665         REPLACE_FORMAT(I16);
 666         REPLACE_FORMAT(L16);
 667         REPLACE_FORMAT(L16A16);
 668
 669 #undef REPLACE_FORMAT
 670 #undef REPLACE_FORMAT_SIGN
 671         default:
 672                 return format;
 673         }
 674 }
 675
 676 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 677                                      const struct pipe_blit_info *info)
 678 {
 679         struct si_context *sctx = (struct si_context*)ctx;
 680         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 681         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 682         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 683         enum pipe_format format = int_to_norm_format(info->dst.format);
 684         unsigned sample_mask = ~0;
 685
 686         if (info->src.resource->nr_samples > 1 &&
 687             info->dst.resource->nr_samples <= 1 &&
 688             util_max_layer(info->src.resource, 0) == 0 &&
 689             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 690             info->dst.format == info->src.format &&
 691             !util_format_is_pure_integer(format) &&
 692             !util_format_is_depth_or_stencil(format) &&
 693             !info->scissor_enable &&
 694             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 695             dst_width == info->src.resource->width0 &&
 696             dst_height == info->src.resource->height0 &&
 697             info->dst.box.x == 0 &&
 698             info->dst.box.y == 0 &&
 699             info->dst.box.width == dst_width &&
 700             info->dst.box.height == dst_height &&
 701             info->dst.box.depth == 1 &&
 702             info->src.box.x == 0 &&
 703             info->src.box.y == 0 &&
 704             info->src.box.width == dst_width &&
 705             info->src.box.height == dst_height &&
 706             info->src.box.depth == 1 &&
 707             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 708             !(dst->surface.flags & RADEON_SURF_SCANOUT) &&
 709             (!dst->cmask.size || !dst->dirty_level_mask) /* dst cannot be fast-cleared */) {
 710                 si_blitter_begin(ctx, SI_COLOR_RESOLVE |
 711                                  (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 712                 util_blitter_custom_resolve_color(sctx->blitter,
 713                                                   info->dst.resource, info->dst.level,
 714                                                   info->dst.box.z,
 715                                                   info->src.resource, info->src.box.z,
 716                                                   sample_mask, sctx->custom_blend_resolve,
 717                                                   format);
 718                 si_blitter_end(ctx);
 719                 return true;
 720         }
 721         return false;
 722 }
 723
 724 static void si_blit(struct pipe_context *ctx,
 725                     const struct pipe_blit_info *info)
 726 {
 727         struct si_context *sctx = (struct si_context*)ctx;
 728
 729         if (do_hardware_msaa_resolve(ctx, info)) {
 730                 return;
 731         }
 732
 733         assert(util_blitter_is_blit_supported(sctx->blitter, info));
 734
 735         /* The driver doesn't decompress resources automatically while
 736          * u_blitter is rendering. */
 737         si_decompress_subresource(ctx, info->src.resource, info->src.level,
 738                                   info->src.box.z,
 739                                   info->src.box.z + info->src.box.depth - 1);
 740
 741         si_blitter_begin(ctx, SI_BLIT |
 742                          (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
 743         util_blitter_blit(sctx->blitter, info);
 744         si_blitter_end(ctx);
 745 }
 746
 747 static void si_flush_resource(struct pipe_context *ctx,
 748                               struct pipe_resource *res)
 749 {
 750         struct r600_texture *rtex = (struct r600_texture*)res;
 751
 752         assert(res->target != PIPE_BUFFER);
 753
 754         if (!rtex->is_depth && rtex->cmask.size) {
 755                 si_blit_decompress_color(ctx, rtex, 0, res->last_level,
 756                                          0, res->array_size - 1);
 757         }
 758 }
 759
 760 void si_init_blit_functions(struct si_context *sctx)
 761 {
 762         sctx->b.b.clear = si_clear;
 763         sctx->b.b.clear_render_target = si_clear_render_target;
 764         sctx->b.b.clear_depth_stencil = si_clear_depth_stencil;
 765         sctx->b.b.resource_copy_region = si_resource_copy_region;
 766         sctx->b.b.blit = si_blit;
 767         sctx->b.b.flush_resource = si_flush_resource;
 768         sctx->b.blit_decompress_depth = si_blit_decompress_depth;
 769 }