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                            SI_DISABLE_RENDER_COND,
  43
  44         SI_DECOMPRESS    = SI_SAVE_FRAMEBUFFER | SI_DISABLE_RENDER_COND,
  45
  46         SI_COLOR_RESOLVE = SI_SAVE_FRAMEBUFFER | SI_DISABLE_RENDER_COND
  47 };
  48
  49 static void si_blitter_begin(struct pipe_context *ctx, enum si_blitter_op op)
  50 {
  51         struct si_context *sctx = (struct si_context *)ctx;
  52
  53         r600_suspend_nontimer_queries(&sctx->b);
  54
  55         util_blitter_save_blend(sctx->blitter, sctx->queued.named.blend);
  56         util_blitter_save_depth_stencil_alpha(sctx->blitter, sctx->queued.named.dsa);
  57         util_blitter_save_stencil_ref(sctx->blitter, &sctx->stencil_ref);
  58         util_blitter_save_rasterizer(sctx->blitter, sctx->queued.named.rasterizer);
  59         util_blitter_save_fragment_shader(sctx->blitter, sctx->ps_shader);
  60         util_blitter_save_geometry_shader(sctx->blitter, sctx->gs_shader);
  61         util_blitter_save_vertex_shader(sctx->blitter, sctx->vs_shader);
  62         util_blitter_save_vertex_elements(sctx->blitter, sctx->vertex_elements);
  63         if (sctx->queued.named.viewport) {
  64                 util_blitter_save_viewport(sctx->blitter, &sctx->queued.named.viewport->viewport);
  65         }
  66         util_blitter_save_vertex_buffer_slot(sctx->blitter, sctx->vertex_buffer);
  67         util_blitter_save_so_targets(sctx->blitter, sctx->b.streamout.num_targets,
  68                                      (struct pipe_stream_output_target**)sctx->b.streamout.targets);
  69
  70         if (op & SI_SAVE_FRAMEBUFFER)
  71                 util_blitter_save_framebuffer(sctx->blitter, &sctx->framebuffer);
  72
  73         if (op & SI_SAVE_TEXTURES) {
  74                 util_blitter_save_fragment_sampler_states(
  75                         sctx->blitter, sctx->samplers[PIPE_SHADER_FRAGMENT].n_samplers,
  76                         (void**)sctx->samplers[PIPE_SHADER_FRAGMENT].samplers);
  77
  78                 util_blitter_save_fragment_sampler_views(sctx->blitter,
  79                         util_last_bit(sctx->samplers[PIPE_SHADER_FRAGMENT].views.desc.enabled_mask &
  80                                       ((1 << NUM_TEX_UNITS) - 1)),
  81                         sctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
  82         }
  83
  84         if ((op & SI_DISABLE_RENDER_COND) && sctx->b.current_render_cond) {
  85                 util_blitter_save_render_condition(sctx->blitter,
  86                                                    sctx->b.current_render_cond,
  87                                                    sctx->b.current_render_cond_cond,
  88                                                    sctx->b.current_render_cond_mode);
  89         }
  90 }
  91
  92 static void si_blitter_end(struct pipe_context *ctx)
  93 {
  94         struct si_context *sctx = (struct si_context *)ctx;
  95         r600_resume_nontimer_queries(&sctx->b);
  96 }
  97
  98 static unsigned u_max_sample(struct pipe_resource *r)
  99 {
 100         return r->nr_samples ? r->nr_samples - 1 : 0;
 101 }
 102
 103 static void si_blit_decompress_depth(struct pipe_context *ctx,
 104                                      struct r600_texture *texture,
 105                                      struct r600_texture *staging,
 106                                      unsigned first_level, unsigned last_level,
 107                                      unsigned first_layer, unsigned last_layer,
 108                                      unsigned first_sample, unsigned last_sample)
 109 {
 110         struct si_context *sctx = (struct si_context *)ctx;
 111         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 112         float depth = 1.0f;
 113         const struct util_format_description *desc;
 114         void **custom_dsa;
 115         struct r600_texture *flushed_depth_texture = staging ?
 116                         staging : texture->flushed_depth_texture;
 117
 118         if (!staging && !texture->dirty_level_mask)
 119                 return;
 120
 121         max_sample = u_max_sample(&texture->resource.b.b);
 122
 123         desc = util_format_description(flushed_depth_texture->resource.b.b.format);
 124         switch (util_format_has_depth(desc) | util_format_has_stencil(desc) << 1) {
 125         default:
 126                 assert(!"No depth or stencil to uncompress");
 127                 return;
 128         case 3:
 129                 custom_dsa = sctx->custom_dsa_flush_depth_stencil;
 130                 break;
 131         case 2:
 132                 custom_dsa = sctx->custom_dsa_flush_stencil;
 133                 break;
 134         case 1:
 135                 custom_dsa = sctx->custom_dsa_flush_depth;
 136                 break;
 137         }
 138
 139         for (level = first_level; level <= last_level; level++) {
 140                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 141                         continue;
 142
 143                 /* The smaller the mipmap level, the less layers there are
 144                  * as far as 3D textures are concerned. */
 145                 max_layer = util_max_layer(&texture->resource.b.b, level);
 146                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 147
 148                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 149                         for (sample = first_sample; sample <= last_sample; sample++) {
 150                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 151
 152                                 surf_tmpl.format = texture->resource.b.b.format;
 153                                 surf_tmpl.u.tex.level = level;
 154                                 surf_tmpl.u.tex.first_layer = layer;
 155                                 surf_tmpl.u.tex.last_layer = layer;
 156
 157                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 158
 159                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 160                                 cbsurf = ctx->create_surface(ctx,
 161                                                 (struct pipe_resource*)flushed_depth_texture, &surf_tmpl);
 162
 163                                 si_blitter_begin(ctx, SI_DECOMPRESS);
 164                                 util_blitter_custom_depth_stencil(sctx->blitter, zsurf, cbsurf, 1 << sample,
 165                                                                   custom_dsa[sample], depth);
 166                                 si_blitter_end(ctx);
 167
 168                                 pipe_surface_reference(&zsurf, NULL);
 169                                 pipe_surface_reference(&cbsurf, NULL);
 170                         }
 171                 }
 172
 173                 /* The texture will always be dirty if some layers aren't flushed.
 174                  * I don't think this case can occur though. */
 175                 if (!staging &&
 176                     first_layer == 0 && last_layer == max_layer &&
 177                     first_sample == 0 && last_sample == max_sample) {
 178                         texture->dirty_level_mask &= ~(1 << level);
 179                 }
 180         }
 181 }
 182
 183 static void si_blit_decompress_depth_in_place(struct si_context *sctx,
 184                                               struct r600_texture *texture,
 185                                               unsigned first_level, unsigned last_level,
 186                                               unsigned first_layer, unsigned last_layer)
 187 {
 188         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 189         unsigned layer, max_layer, checked_last_layer, level;
 190
 191         surf_tmpl.format = texture->resource.b.b.format;
 192
 193         for (level = first_level; level <= last_level; level++) {
 194                 if (!(texture->dirty_level_mask & (1 << level)))
 195                         continue;
 196
 197                 surf_tmpl.u.tex.level = level;
 198
 199                 /* The smaller the mipmap level, the less layers there are
 200                  * as far as 3D textures are concerned. */
 201                 max_layer = util_max_layer(&texture->resource.b.b, level);
 202                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 203
 204                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 205                         surf_tmpl.u.tex.first_layer = layer;
 206                         surf_tmpl.u.tex.last_layer = layer;
 207
 208                         zsurf = sctx->b.b.create_surface(&sctx->b.b, &texture->resource.b.b, &surf_tmpl);
 209
 210                         si_blitter_begin(&sctx->b.b, SI_DECOMPRESS);
 211                         util_blitter_custom_depth_stencil(sctx->blitter, zsurf, NULL, ~0,
 212                                                           sctx->custom_dsa_flush_inplace,
 213                                                           1.0f);
 214                         si_blitter_end(&sctx->b.b);
 215
 216                         pipe_surface_reference(&zsurf, NULL);
 217                 }
 218
 219                 /* The texture will always be dirty if some layers aren't flushed.
 220                  * I don't think this case occurs often though. */
 221                 if (first_layer == 0 && last_layer == max_layer) {
 222                         texture->dirty_level_mask &= ~(1 << level);
 223                 }
 224         }
 225 }
 226
 227 void si_flush_depth_textures(struct si_context *sctx,
 228                              struct si_textures_info *textures)
 229 {
 230         unsigned i;
 231
 232         for (i = 0; i < textures->n_views; ++i) {
 233                 struct pipe_sampler_view *view;
 234                 struct r600_texture *tex;
 235
 236                 view = textures->views.views[i];
 237                 if (!view) continue;
 238
 239                 tex = (struct r600_texture *)view->texture;
 240                 if (!tex->is_depth || tex->is_flushing_texture)
 241                         continue;
 242
 243                 si_blit_decompress_depth_in_place(sctx, tex,
 244                                                   view->u.tex.first_level, view->u.tex.last_level,
 245                                                   0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 246         }
 247 }
 248
 249 static void si_blit_decompress_color(struct pipe_context *ctx,
 250                 struct r600_texture *rtex,
 251                 unsigned first_level, unsigned last_level,
 252                 unsigned first_layer, unsigned last_layer)
 253 {
 254         struct si_context *sctx = (struct si_context *)ctx;
 255         unsigned layer, level, checked_last_layer, max_layer;
 256
 257         if (!rtex->dirty_level_mask)
 258                 return;
 259
 260         for (level = first_level; level <= last_level; level++) {
 261                 if (!(rtex->dirty_level_mask & (1 << level)))
 262                         continue;
 263
 264                 /* The smaller the mipmap level, the less layers there are
 265                  * as far as 3D textures are concerned. */
 266                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 267                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 268
 269                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 270                         struct pipe_surface *cbsurf, surf_tmpl;
 271
 272                         surf_tmpl.format = rtex->resource.b.b.format;
 273                         surf_tmpl.u.tex.level = level;
 274                         surf_tmpl.u.tex.first_layer = layer;
 275                         surf_tmpl.u.tex.last_layer = layer;
 276                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 277
 278                         si_blitter_begin(ctx, SI_DECOMPRESS);
 279                         util_blitter_custom_color(sctx->blitter, cbsurf,
 280                                                   sctx->custom_blend_decompress);
 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;
 324
 325         si_blitter_begin(ctx, SI_CLEAR);
 326         util_blitter_clear(sctx->blitter, fb->width, fb->height,
 327                            util_framebuffer_get_num_layers(fb),
 328                            buffers, color, depth, stencil);
 329         si_blitter_end(ctx);
 330 }
 331
 332 static void si_clear_render_target(struct pipe_context *ctx,
 333                                    struct pipe_surface *dst,
 334                                    const union pipe_color_union *color,
 335                                    unsigned dstx, unsigned dsty,
 336                                    unsigned width, unsigned height)
 337 {
 338         struct si_context *sctx = (struct si_context *)ctx;
 339
 340         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 341         util_blitter_clear_render_target(sctx->blitter, dst, color,
 342                                          dstx, dsty, width, height);
 343         si_blitter_end(ctx);
 344 }
 345
 346 static void si_clear_depth_stencil(struct pipe_context *ctx,
 347                                    struct pipe_surface *dst,
 348                                    unsigned clear_flags,
 349                                    double depth,
 350                                    unsigned stencil,
 351                                    unsigned dstx, unsigned dsty,
 352                                    unsigned width, unsigned height)
 353 {
 354         struct si_context *sctx = (struct si_context *)ctx;
 355
 356         si_blitter_begin(ctx, SI_CLEAR_SURFACE);
 357         util_blitter_clear_depth_stencil(sctx->blitter, dst, clear_flags, depth, stencil,
 358                                          dstx, dsty, width, height);
 359         si_blitter_end(ctx);
 360 }
 361
 362 /* Helper for decompressing a portion of a color or depth resource before
 363  * blitting if any decompression is needed.
 364  * The driver doesn't decompress resources automatically while u_blitter is
 365  * rendering. */
 366 static void si_decompress_subresource(struct pipe_context *ctx,
 367                                       struct pipe_resource *tex,
 368                                       unsigned level,
 369                                       unsigned first_layer, unsigned last_layer)
 370 {
 371         struct si_context *sctx = (struct si_context *)ctx;
 372         struct r600_texture *rtex = (struct r600_texture*)tex;
 373
 374         if (rtex->is_depth && !rtex->is_flushing_texture) {
 375                 si_blit_decompress_depth_in_place(sctx, rtex,
 376                                                   level, level,
 377                                                   first_layer, last_layer);
 378         } else if (rtex->fmask.size || rtex->cmask.size) {
 379                 si_blit_decompress_color(ctx, rtex, level, level,
 380                                          first_layer, last_layer);
 381         }
 382 }
 383
 384 struct texture_orig_info {
 385         unsigned format;
 386         unsigned width0;
 387         unsigned height0;
 388         unsigned npix_x;
 389         unsigned npix_y;
 390         unsigned npix0_x;
 391         unsigned npix0_y;
 392 };
 393
 394 static void si_compressed_to_blittable(struct pipe_resource *tex,
 395                                        unsigned level,
 396                                        struct texture_orig_info *orig)
 397 {
 398         struct r600_texture *rtex = (struct r600_texture*)tex;
 399         unsigned pixsize = util_format_get_blocksize(rtex->resource.b.b.format);
 400         int new_format;
 401         int new_height, new_width;
 402
 403         orig->format = tex->format;
 404         orig->width0 = tex->width0;
 405         orig->height0 = tex->height0;
 406         orig->npix0_x = rtex->surface.level[0].npix_x;
 407         orig->npix0_y = rtex->surface.level[0].npix_y;
 408         orig->npix_x = rtex->surface.level[level].npix_x;
 409         orig->npix_y = rtex->surface.level[level].npix_y;
 410
 411         if (pixsize == 8)
 412                 new_format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 413         else
 414                 new_format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 415
 416         new_width = util_format_get_nblocksx(tex->format, orig->width0);
 417         new_height = util_format_get_nblocksy(tex->format, orig->height0);
 418
 419         tex->width0 = new_width;
 420         tex->height0 = new_height;
 421         tex->format = new_format;
 422         rtex->surface.level[0].npix_x = util_format_get_nblocksx(orig->format, orig->npix0_x);
 423         rtex->surface.level[0].npix_y = util_format_get_nblocksy(orig->format, orig->npix0_y);
 424         rtex->surface.level[level].npix_x = util_format_get_nblocksx(orig->format, orig->npix_x);
 425         rtex->surface.level[level].npix_y = util_format_get_nblocksy(orig->format, orig->npix_y);
 426
 427         /* By dividing the dimensions by 4, we effectively decrement
 428          * last_level by 2, therefore the last 2 mipmap levels disappear and
 429          * aren't blittable. Note that the last 3 mipmap levels (4x4, 2x2,
 430          * 1x1) have equal slice sizes, which is an important assumption
 431          * for this to work.
 432          *
 433          * In order to make the last 2 mipmap levels blittable, we have to
 434          * add the slice size of the last mipmap level to the texture
 435          * address, so that even though the hw thinks it reads last_level-2,
 436          * it will actually read last_level-1, and if we add the slice size*2,
 437          * it will read last_level. That's how this workaround works.
 438          */
 439         if (level > rtex->resource.b.b.last_level-2)
 440                 rtex->mipmap_shift = level - (rtex->resource.b.b.last_level-2);
 441 }
 442
 443 static void si_change_format(struct pipe_resource *tex,
 444                              unsigned level,
 445                              struct texture_orig_info *orig,
 446                              enum pipe_format format)
 447 {
 448         struct r600_texture *rtex = (struct r600_texture*)tex;
 449
 450         orig->format = tex->format;
 451         orig->width0 = tex->width0;
 452         orig->height0 = tex->height0;
 453         orig->npix0_x = rtex->surface.level[0].npix_x;
 454         orig->npix0_y = rtex->surface.level[0].npix_y;
 455         orig->npix_x = rtex->surface.level[level].npix_x;
 456         orig->npix_y = rtex->surface.level[level].npix_y;
 457
 458         tex->format = format;
 459 }
 460
 461 static void si_reset_blittable_to_orig(struct pipe_resource *tex,
 462                                        unsigned level,
 463                                        struct texture_orig_info *orig)
 464 {
 465         struct r600_texture *rtex = (struct r600_texture*)tex;
 466
 467         tex->format = orig->format;
 468         tex->width0 = orig->width0;
 469         tex->height0 = orig->height0;
 470         rtex->surface.level[0].npix_x = orig->npix0_x;
 471         rtex->surface.level[0].npix_y = orig->npix0_y;
 472         rtex->surface.level[level].npix_x = orig->npix_x;
 473         rtex->surface.level[level].npix_y = orig->npix_y;
 474         rtex->mipmap_shift = 0;
 475 }
 476
 477 static void si_resource_copy_region(struct pipe_context *ctx,
 478                                     struct pipe_resource *dst,
 479                                     unsigned dst_level,
 480                                     unsigned dstx, unsigned dsty, unsigned dstz,
 481                                     struct pipe_resource *src,
 482                                     unsigned src_level,
 483                                     const struct pipe_box *src_box)
 484 {
 485         struct si_context *sctx = (struct si_context *)ctx;
 486         struct pipe_surface *dst_view, dst_templ;
 487         struct pipe_sampler_view src_templ, *src_view;
 488         struct texture_orig_info orig_info[2];
 489         struct pipe_box sbox, dstbox;
 490         boolean restore_orig[2];
 491
 492         /* Fallback for buffers. */
 493         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 494                 si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width);
 495                 return;
 496         }
 497
 498         memset(orig_info, 0, sizeof(orig_info));
 499
 500         /* The driver doesn't decompress resources automatically while
 501          * u_blitter is rendering. */
 502         si_decompress_subresource(ctx, src, src_level,
 503                                   src_box->z, src_box->z + src_box->depth - 1);
 504
 505         restore_orig[0] = restore_orig[1] = FALSE;
 506
 507         if (util_format_is_compressed(src->format) &&
 508             util_format_is_compressed(dst->format)) {
 509                 si_compressed_to_blittable(src, src_level, &orig_info[0]);
 510                 restore_orig[0] = TRUE;
 511                 sbox.x = util_format_get_nblocksx(orig_info[0].format, src_box->x);
 512                 sbox.y = util_format_get_nblocksy(orig_info[0].format, src_box->y);
 513                 sbox.z = src_box->z;
 514                 sbox.width = util_format_get_nblocksx(orig_info[0].format, src_box->width);
 515                 sbox.height = util_format_get_nblocksy(orig_info[0].format, src_box->height);
 516                 sbox.depth = src_box->depth;
 517                 src_box = &sbox;
 518
 519                 si_compressed_to_blittable(dst, dst_level, &orig_info[1]);
 520                 restore_orig[1] = TRUE;
 521                 /* translate the dst box as well */
 522                 dstx = util_format_get_nblocksx(orig_info[1].format, dstx);
 523                 dsty = util_format_get_nblocksy(orig_info[1].format, dsty);
 524         } else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src)) {
 525                 unsigned blocksize = util_format_get_blocksize(src->format);
 526
 527                 switch (blocksize) {
 528                 case 1:
 529                         si_change_format(src, src_level, &orig_info[0],
 530                                          PIPE_FORMAT_R8_UNORM);
 531                         si_change_format(dst, dst_level, &orig_info[1],
 532                                          PIPE_FORMAT_R8_UNORM);
 533                         break;
 534                 case 2:
 535                         si_change_format(src, src_level, &orig_info[0],
 536                                          PIPE_FORMAT_R8G8_UNORM);
 537                         si_change_format(dst, dst_level, &orig_info[1],
 538                                          PIPE_FORMAT_R8G8_UNORM);
 539                         break;
 540                 case 4:
 541                         si_change_format(src, src_level, &orig_info[0],
 542                                          PIPE_FORMAT_R8G8B8A8_UNORM);
 543                         si_change_format(dst, dst_level, &orig_info[1],
 544                                          PIPE_FORMAT_R8G8B8A8_UNORM);
 545                         break;
 546                 case 8:
 547                         si_change_format(src, src_level, &orig_info[0],
 548                                          PIPE_FORMAT_R16G16B16A16_UINT);
 549                         si_change_format(dst, dst_level, &orig_info[1],
 550                                          PIPE_FORMAT_R16G16B16A16_UINT);
 551                         break;
 552                 case 16:
 553                         si_change_format(src, src_level, &orig_info[0],
 554                                          PIPE_FORMAT_R32G32B32A32_UINT);
 555                         si_change_format(dst, dst_level, &orig_info[1],
 556                                          PIPE_FORMAT_R32G32B32A32_UINT);
 557                         break;
 558                 default:
 559                         fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 560                                 util_format_short_name(src->format), blocksize);
 561                         assert(0);
 562                 }
 563                 restore_orig[0] = TRUE;
 564                 restore_orig[1] = TRUE;
 565         }
 566
 567         /* Initialize the surface. */
 568         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 569         dst_view = ctx->create_surface(ctx, dst, &dst_templ);
 570
 571         /* Initialize the sampler view. */
 572         util_blitter_default_src_texture(&src_templ, src, src_level);
 573         src_view = ctx->create_sampler_view(ctx, src, &src_templ);
 574
 575         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 576                  abs(src_box->depth), &dstbox);
 577
 578         /* Copy. */
 579         si_blitter_begin(ctx, SI_COPY);
 580         util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
 581                                   src_view, src_box, src->width0, src->height0,
 582                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
 583         si_blitter_end(ctx);
 584
 585         pipe_surface_reference(&dst_view, NULL);
 586         pipe_sampler_view_reference(&src_view, NULL);
 587
 588         if (restore_orig[0])
 589                 si_reset_blittable_to_orig(src, src_level, &orig_info[0]);
 590
 591         if (restore_orig[1])
 592                 si_reset_blittable_to_orig(dst, dst_level, &orig_info[1]);
 593 }
 594
 595 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 596 static enum pipe_format int_to_norm_format(enum pipe_format format)
 597 {
 598         switch (format) {
 599 #define REPLACE_FORMAT_SIGN(format,sign) \
 600         case PIPE_FORMAT_##format##_##sign##INT: \
 601                 return PIPE_FORMAT_##format##_##sign##NORM
 602 #define REPLACE_FORMAT(format) \
 603                 REPLACE_FORMAT_SIGN(format, U); \
 604                 REPLACE_FORMAT_SIGN(format, S)
 605
 606         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 607         REPLACE_FORMAT(R8);
 608         REPLACE_FORMAT(R8G8);
 609         REPLACE_FORMAT(R8G8B8X8);
 610         REPLACE_FORMAT(R8G8B8A8);
 611         REPLACE_FORMAT(A8);
 612         REPLACE_FORMAT(I8);
 613         REPLACE_FORMAT(L8);
 614         REPLACE_FORMAT(L8A8);
 615         REPLACE_FORMAT(R16);
 616         REPLACE_FORMAT(R16G16);
 617         REPLACE_FORMAT(R16G16B16X16);
 618         REPLACE_FORMAT(R16G16B16A16);
 619         REPLACE_FORMAT(A16);
 620         REPLACE_FORMAT(I16);
 621         REPLACE_FORMAT(L16);
 622         REPLACE_FORMAT(L16A16);
 623
 624 #undef REPLACE_FORMAT
 625 #undef REPLACE_FORMAT_SIGN
 626         default:
 627                 return format;
 628         }
 629 }
 630
 631 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 632                                      const struct pipe_blit_info *info)
 633 {
 634         struct si_context *sctx = (struct si_context*)ctx;
 635         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 636         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 637         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 638         enum pipe_format format = int_to_norm_format(info->dst.format);
 639         unsigned sample_mask = ~0;
 640
 641         if (info->src.resource->nr_samples > 1 &&
 642             info->dst.resource->nr_samples <= 1 &&
 643             util_max_layer(info->src.resource, 0) == 0 &&
 644             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 645             info->dst.format == info->src.format &&
 646             !util_format_is_pure_integer(format) &&
 647             !util_format_is_depth_or_stencil(format) &&
 648             !info->scissor_enable &&
 649             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 650             dst_width == info->src.resource->width0 &&
 651             dst_height == info->src.resource->height0 &&
 652             info->dst.box.x == 0 &&
 653             info->dst.box.y == 0 &&
 654             info->dst.box.width == dst_width &&
 655             info->dst.box.height == dst_height &&
 656             info->dst.box.depth == 1 &&
 657             info->src.box.x == 0 &&
 658             info->src.box.y == 0 &&
 659             info->src.box.width == dst_width &&
 660             info->src.box.height == dst_height &&
 661             info->src.box.depth == 1 &&
 662             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 663             !(dst->surface.flags & RADEON_SURF_SCANOUT)) {
 664                 si_blitter_begin(ctx, SI_COLOR_RESOLVE);
 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         si_blitter_begin(ctx, SI_BLIT);
 695         util_blitter_blit(sctx->blitter, info);
 696         si_blitter_end(ctx);
 697 }
 698
 699 static void si_flush_resource(struct pipe_context *ctx,
 700                               struct pipe_resource *resource)
 701 {
 702 }
 703
 704 void si_init_blit_functions(struct si_context *sctx)
 705 {
 706         sctx->b.b.clear = si_clear;
 707         sctx->b.b.clear_render_target = si_clear_render_target;
 708         sctx->b.b.clear_depth_stencil = si_clear_depth_stencil;
 709         sctx->b.b.resource_copy_region = si_resource_copy_region;
 710         sctx->b.b.blit = si_blit;
 711         sctx->b.b.flush_resource = si_flush_resource;
 712         sctx->b.blit_decompress_depth = si_blit_decompress_depth;
 713 }