src/gallium/drivers/r600/r600_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 #include "r600_pipe.h"
  24 #include "util/u_surface.h"
  25 #include "util/u_blitter.h"
  26 #include "util/u_format.h"
  27 #include "evergreend.h"
  28
  29 enum r600_blitter_op /* bitmask */
  30 {
  31         R600_SAVE_FRAGMENT_STATE = 1,
  32         R600_SAVE_TEXTURES       = 2,
  33         R600_SAVE_FRAMEBUFFER    = 4,
  34         R600_DISABLE_RENDER_COND = 8,
  35
  36         R600_CLEAR         = R600_SAVE_FRAGMENT_STATE,
  37
  38         R600_CLEAR_SURFACE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER,
  39
  40         R600_COPY_BUFFER   = R600_DISABLE_RENDER_COND,
  41
  42         R600_COPY_TEXTURE  = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
  43                              R600_DISABLE_RENDER_COND,
  44
  45         R600_BLIT          = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES,
  46
  47         R600_DECOMPRESS    = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND,
  48
  49         R600_COLOR_RESOLVE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER
  50 };
  51
  52 static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
  53 {
  54         struct r600_context *rctx = (struct r600_context *)ctx;
  55
  56         r600_suspend_nontimer_queries(&rctx->b);
  57
  58         util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer_state.vb);
  59         util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_fetch_shader.cso);
  60         util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
  61         util_blitter_save_geometry_shader(rctx->blitter, rctx->gs_shader);
  62         util_blitter_save_so_targets(rctx->blitter, rctx->b.streamout.num_targets,
  63                                      (struct pipe_stream_output_target**)rctx->b.streamout.targets);
  64         util_blitter_save_rasterizer(rctx->blitter, rctx->rasterizer_state.cso);
  65
  66         if (op & R600_SAVE_FRAGMENT_STATE) {
  67                 util_blitter_save_viewport(rctx->blitter, &rctx->viewport[0].state);
  68                 util_blitter_save_scissor(rctx->blitter, &rctx->scissor[0].scissor);
  69                 util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
  70                 util_blitter_save_blend(rctx->blitter, rctx->blend_state.cso);
  71                 util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->dsa_state.cso);
  72                 util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref.pipe_state);
  73                 util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask);
  74         }
  75
  76         if (op & R600_SAVE_FRAMEBUFFER)
  77                 util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer.state);
  78
  79         if (op & R600_SAVE_TEXTURES) {
  80                 util_blitter_save_fragment_sampler_states(
  81                         rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].states.enabled_mask),
  82                         (void**)rctx->samplers[PIPE_SHADER_FRAGMENT].states.states);
  83
  84                 util_blitter_save_fragment_sampler_views(
  85                         rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.enabled_mask),
  86                         (struct pipe_sampler_view**)rctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
  87         }
  88
  89         if ((op & R600_DISABLE_RENDER_COND) && rctx->b.current_render_cond) {
  90            util_blitter_save_render_condition(rctx->blitter,
  91                                               rctx->b.current_render_cond,
  92                                               rctx->b.current_render_cond_cond,
  93                                               rctx->b.current_render_cond_mode);
  94         }
  95 }
  96
  97 static void r600_blitter_end(struct pipe_context *ctx)
  98 {
  99         struct r600_context *rctx = (struct r600_context *)ctx;
 100         r600_resume_nontimer_queries(&rctx->b);
 101 }
 102
 103 static unsigned u_max_sample(struct pipe_resource *r)
 104 {
 105         return r->nr_samples ? r->nr_samples - 1 : 0;
 106 }
 107
 108 static void r600_blit_decompress_depth(struct pipe_context *ctx,
 109                                        struct r600_texture *texture,
 110                                        struct r600_texture *staging,
 111                                        unsigned first_level, unsigned last_level,
 112                                        unsigned first_layer, unsigned last_layer,
 113                                        unsigned first_sample, unsigned last_sample)
 114 {
 115         struct r600_context *rctx = (struct r600_context *)ctx;
 116         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 117         struct r600_texture *flushed_depth_texture = staging ?
 118                         staging : texture->flushed_depth_texture;
 119         const struct util_format_description *desc =
 120                 util_format_description(texture->resource.b.b.format);
 121         float depth;
 122
 123         if (!staging && !texture->dirty_level_mask)
 124                 return;
 125
 126         max_sample = u_max_sample(&texture->resource.b.b);
 127
 128         /* XXX Decompressing MSAA depth textures is broken on R6xx.
 129          * There is also a hardlock if CMASK and FMASK are not present.
 130          * Just skip this until we find out how to fix it. */
 131         if (rctx->b.chip_class == R600 && max_sample > 0) {
 132                 texture->dirty_level_mask = 0;
 133                 return;
 134         }
 135
 136         if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
 137             rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
 138                 depth = 0.0f;
 139         else
 140                 depth = 1.0f;
 141
 142         /* Enable decompression in DB_RENDER_CONTROL */
 143         rctx->db_misc_state.flush_depthstencil_through_cb = true;
 144         rctx->db_misc_state.copy_depth = util_format_has_depth(desc);
 145         rctx->db_misc_state.copy_stencil = util_format_has_stencil(desc);
 146         rctx->db_misc_state.copy_sample = first_sample;
 147         rctx->db_misc_state.atom.dirty = true;
 148
 149         for (level = first_level; level <= last_level; level++) {
 150                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 151                         continue;
 152
 153                 /* The smaller the mipmap level, the less layers there are
 154                  * as far as 3D textures are concerned. */
 155                 max_layer = util_max_layer(&texture->resource.b.b, level);
 156                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 157
 158                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 159                         for (sample = first_sample; sample <= last_sample; sample++) {
 160                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 161
 162                                 if (sample != rctx->db_misc_state.copy_sample) {
 163                                         rctx->db_misc_state.copy_sample = sample;
 164                                         rctx->db_misc_state.atom.dirty = true;
 165                                 }
 166
 167                                 surf_tmpl.format = texture->resource.b.b.format;
 168                                 surf_tmpl.u.tex.level = level;
 169                                 surf_tmpl.u.tex.first_layer = layer;
 170                                 surf_tmpl.u.tex.last_layer = layer;
 171
 172                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 173
 174                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 175                                 cbsurf = ctx->create_surface(ctx,
 176                                                 &flushed_depth_texture->resource.b.b, &surf_tmpl);
 177
 178                                 r600_blitter_begin(ctx, R600_DECOMPRESS);
 179                                 util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, 1 << sample,
 180                                                                   rctx->custom_dsa_flush, depth);
 181                                 r600_blitter_end(ctx);
 182
 183                                 pipe_surface_reference(&zsurf, NULL);
 184                                 pipe_surface_reference(&cbsurf, NULL);
 185                         }
 186                 }
 187
 188                 /* The texture will always be dirty if some layers or samples aren't flushed.
 189                  * I don't think this case occurs often though. */
 190                 if (!staging &&
 191                     first_layer == 0 && last_layer == max_layer &&
 192                     first_sample == 0 && last_sample == max_sample) {
 193                         texture->dirty_level_mask &= ~(1 << level);
 194                 }
 195         }
 196
 197         /* reenable compression in DB_RENDER_CONTROL */
 198         rctx->db_misc_state.flush_depthstencil_through_cb = false;
 199         rctx->db_misc_state.atom.dirty = true;
 200 }
 201
 202 static void r600_blit_decompress_depth_in_place(struct r600_context *rctx,
 203                                                 struct r600_texture *texture,
 204                                                 unsigned first_level, unsigned last_level,
 205                                                 unsigned first_layer, unsigned last_layer)
 206 {
 207         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 208         unsigned layer, max_layer, checked_last_layer, level;
 209
 210         /* Enable decompression in DB_RENDER_CONTROL */
 211         rctx->db_misc_state.flush_depthstencil_in_place = true;
 212         rctx->db_misc_state.atom.dirty = true;
 213
 214         surf_tmpl.format = texture->resource.b.b.format;
 215
 216         for (level = first_level; level <= last_level; level++) {
 217                 if (!(texture->dirty_level_mask & (1 << level)))
 218                         continue;
 219
 220                 surf_tmpl.u.tex.level = level;
 221
 222                 /* The smaller the mipmap level, the less layers there are
 223                  * as far as 3D textures are concerned. */
 224                 max_layer = util_max_layer(&texture->resource.b.b, level);
 225                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 226
 227                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 228                         surf_tmpl.u.tex.first_layer = layer;
 229                         surf_tmpl.u.tex.last_layer = layer;
 230
 231                         zsurf = rctx->b.b.create_surface(&rctx->b.b, &texture->resource.b.b, &surf_tmpl);
 232
 233                         r600_blitter_begin(&rctx->b.b, R600_DECOMPRESS);
 234                         util_blitter_custom_depth_stencil(rctx->blitter, zsurf, NULL, ~0,
 235                                                           rctx->custom_dsa_flush, 1.0f);
 236                         r600_blitter_end(&rctx->b.b);
 237
 238                         pipe_surface_reference(&zsurf, NULL);
 239                 }
 240
 241                 /* The texture will always be dirty if some layers or samples aren't flushed.
 242                  * I don't think this case occurs often though. */
 243                 if (first_layer == 0 && last_layer == max_layer) {
 244                         texture->dirty_level_mask &= ~(1 << level);
 245                 }
 246         }
 247
 248         /* Disable decompression in DB_RENDER_CONTROL */
 249         rctx->db_misc_state.flush_depthstencil_in_place = false;
 250         rctx->db_misc_state.atom.dirty = true;
 251 }
 252
 253 void r600_decompress_depth_textures(struct r600_context *rctx,
 254                                struct r600_samplerview_state *textures)
 255 {
 256         unsigned i;
 257         unsigned depth_texture_mask = textures->compressed_depthtex_mask;
 258
 259         while (depth_texture_mask) {
 260                 struct pipe_sampler_view *view;
 261                 struct r600_texture *tex;
 262
 263                 i = u_bit_scan(&depth_texture_mask);
 264
 265                 view = &textures->views[i]->base;
 266                 assert(view);
 267
 268                 tex = (struct r600_texture *)view->texture;
 269                 assert(tex->is_depth && !tex->is_flushing_texture);
 270
 271                 if (rctx->b.chip_class >= EVERGREEN ||
 272                     r600_can_read_depth(tex)) {
 273                         r600_blit_decompress_depth_in_place(rctx, tex,
 274                                                    view->u.tex.first_level, view->u.tex.last_level,
 275                                                    0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 276                 } else {
 277                         r600_blit_decompress_depth(&rctx->b.b, tex, NULL,
 278                                                    view->u.tex.first_level, view->u.tex.last_level,
 279                                                    0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level),
 280                                                    0, u_max_sample(&tex->resource.b.b));
 281                 }
 282         }
 283 }
 284
 285 static void r600_blit_decompress_color(struct pipe_context *ctx,
 286                 struct r600_texture *rtex,
 287                 unsigned first_level, unsigned last_level,
 288                 unsigned first_layer, unsigned last_layer)
 289 {
 290         struct r600_context *rctx = (struct r600_context *)ctx;
 291         unsigned layer, level, checked_last_layer, max_layer;
 292
 293         if (!rtex->dirty_level_mask)
 294                 return;
 295
 296         for (level = first_level; level <= last_level; level++) {
 297                 if (!(rtex->dirty_level_mask & (1 << level)))
 298                         continue;
 299
 300                 /* The smaller the mipmap level, the less layers there are
 301                  * as far as 3D textures are concerned. */
 302                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 303                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 304
 305                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 306                         struct pipe_surface *cbsurf, surf_tmpl;
 307
 308                         surf_tmpl.format = rtex->resource.b.b.format;
 309                         surf_tmpl.u.tex.level = level;
 310                         surf_tmpl.u.tex.first_layer = layer;
 311                         surf_tmpl.u.tex.last_layer = layer;
 312                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 313
 314                         r600_blitter_begin(ctx, R600_DECOMPRESS);
 315                         util_blitter_custom_color(rctx->blitter, cbsurf,
 316                                 rtex->fmask.size ? rctx->custom_blend_decompress : rctx->custom_blend_fastclear);
 317                         r600_blitter_end(ctx);
 318
 319                         pipe_surface_reference(&cbsurf, NULL);
 320                 }
 321
 322                 /* The texture will always be dirty if some layers aren't flushed.
 323                  * I don't think this case occurs often though. */
 324                 if (first_layer == 0 && last_layer == max_layer) {
 325                         rtex->dirty_level_mask &= ~(1 << level);
 326                 }
 327         }
 328 }
 329
 330 void r600_decompress_color_textures(struct r600_context *rctx,
 331                                     struct r600_samplerview_state *textures)
 332 {
 333         unsigned i;
 334         unsigned mask = textures->compressed_colortex_mask;
 335
 336         while (mask) {
 337                 struct pipe_sampler_view *view;
 338                 struct r600_texture *tex;
 339
 340                 i = u_bit_scan(&mask);
 341
 342                 view = &textures->views[i]->base;
 343                 assert(view);
 344
 345                 tex = (struct r600_texture *)view->texture;
 346                 assert(tex->cmask.size);
 347
 348                 r600_blit_decompress_color(&rctx->b.b, tex,
 349                                            view->u.tex.first_level, view->u.tex.last_level,
 350                                            0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 351         }
 352 }
 353
 354 /* Helper for decompressing a portion of a color or depth resource before
 355  * blitting if any decompression is needed.
 356  * The driver doesn't decompress resources automatically while u_blitter is
 357  * rendering. */
 358 static bool r600_decompress_subresource(struct pipe_context *ctx,
 359                                         struct pipe_resource *tex,
 360                                         unsigned level,
 361                                         unsigned first_layer, unsigned last_layer)
 362 {
 363         struct r600_context *rctx = (struct r600_context *)ctx;
 364         struct r600_texture *rtex = (struct r600_texture*)tex;
 365
 366         if (rtex->is_depth && !rtex->is_flushing_texture) {
 367                 if (rctx->b.chip_class >= EVERGREEN ||
 368                     r600_can_read_depth(rtex)) {
 369                         r600_blit_decompress_depth_in_place(rctx, rtex,
 370                                                    level, level,
 371                                                    first_layer, last_layer);
 372                 } else {
 373                         if (!r600_init_flushed_depth_texture(ctx, tex, NULL))
 374                                 return false; /* error */
 375
 376                         r600_blit_decompress_depth(ctx, rtex, NULL,
 377                                                    level, level,
 378                                                    first_layer, last_layer,
 379                                                    0, u_max_sample(tex));
 380                 }
 381         } else if (rtex->cmask.size) {
 382                 r600_blit_decompress_color(ctx, rtex, level, level,
 383                                            first_layer, last_layer);
 384         }
 385         return true;
 386 }
 387
 388 static void r600_clear(struct pipe_context *ctx, unsigned buffers,
 389                        const union pipe_color_union *color,
 390                        double depth, unsigned stencil)
 391 {
 392         struct r600_context *rctx = (struct r600_context *)ctx;
 393         struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
 394
 395         if (buffers & PIPE_CLEAR_COLOR && rctx->b.chip_class >= EVERGREEN) {
 396                 evergreen_do_fast_color_clear(&rctx->b, fb, &rctx->framebuffer.atom,
 397                                               &buffers, color);
 398         }
 399
 400         if (buffers & PIPE_CLEAR_COLOR) {
 401                 int i;
 402
 403                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 404                 for (i = 0; i < fb->nr_cbufs; i++) {
 405                         struct r600_texture *tex;
 406
 407                         /* If not clearing this buffer, skip. */
 408                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 409                                 continue;
 410
 411                         if (!fb->cbufs[i])
 412                                 continue;
 413
 414                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 415                         if (tex->fmask.size == 0)
 416                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 417                 }
 418         }
 419
 420         /* if hyperz enabled just clear hyperz */
 421         if (fb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
 422                 struct r600_texture *rtex;
 423                 unsigned level = fb->zsbuf->u.tex.level;
 424
 425                 rtex = (struct r600_texture*)fb->zsbuf->texture;
 426
 427                 /* We can't use hyperz fast clear if each slice of a texture
 428                  * array are clear to different value. To simplify code just
 429                  * disable fast clear for texture array.
 430                  */
 431                 /* Only use htile for first level */
 432                 if (rtex->htile_buffer && !level && rtex->surface.array_size == 1) {
 433                         if (rtex->depth_clear_value != depth) {
 434                                 rtex->depth_clear_value = depth;
 435                                 rctx->db_state.atom.dirty = true;
 436                         }
 437                         rctx->db_misc_state.htile_clear = true;
 438                         rctx->db_misc_state.atom.dirty = true;
 439                 }
 440         }
 441
 442         r600_blitter_begin(ctx, R600_CLEAR);
 443         util_blitter_clear(rctx->blitter, fb->width, fb->height, 1,
 444                            buffers, color, depth, stencil);
 445         r600_blitter_end(ctx);
 446
 447         /* disable fast clear */
 448         if (rctx->db_misc_state.htile_clear) {
 449                 rctx->db_misc_state.htile_clear = false;
 450                 rctx->db_misc_state.atom.dirty = true;
 451         }
 452 }
 453
 454 static void r600_clear_render_target(struct pipe_context *ctx,
 455                                      struct pipe_surface *dst,
 456                                      const union pipe_color_union *color,
 457                                      unsigned dstx, unsigned dsty,
 458                                      unsigned width, unsigned height)
 459 {
 460         struct r600_context *rctx = (struct r600_context *)ctx;
 461
 462         r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
 463         util_blitter_clear_render_target(rctx->blitter, dst, color,
 464                                          dstx, dsty, width, height);
 465         r600_blitter_end(ctx);
 466 }
 467
 468 static void r600_clear_depth_stencil(struct pipe_context *ctx,
 469                                      struct pipe_surface *dst,
 470                                      unsigned clear_flags,
 471                                      double depth,
 472                                      unsigned stencil,
 473                                      unsigned dstx, unsigned dsty,
 474                                      unsigned width, unsigned height)
 475 {
 476         struct r600_context *rctx = (struct r600_context *)ctx;
 477
 478         r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
 479         util_blitter_clear_depth_stencil(rctx->blitter, dst, clear_flags, depth, stencil,
 480                                          dstx, dsty, width, height);
 481         r600_blitter_end(ctx);
 482 }
 483
 484 static void r600_copy_buffer(struct pipe_context *ctx, struct pipe_resource *dst, unsigned dstx,
 485                              struct pipe_resource *src, const struct pipe_box *src_box)
 486 {
 487         struct r600_context *rctx = (struct r600_context*)ctx;
 488
 489         if (rctx->screen->b.has_cp_dma) {
 490                 r600_cp_dma_copy_buffer(rctx, dst, dstx, src, src_box->x, src_box->width);
 491         }
 492         else if (rctx->screen->b.has_streamout &&
 493                  /* Require 4-byte alignment. */
 494                  dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
 495
 496                 r600_blitter_begin(ctx, R600_COPY_BUFFER);
 497                 util_blitter_copy_buffer(rctx->blitter, dst, dstx, src, src_box->x, src_box->width);
 498                 r600_blitter_end(ctx);
 499         } else {
 500                 util_resource_copy_region(ctx, dst, 0, dstx, 0, 0, src, 0, src_box);
 501         }
 502
 503         /* The index buffer (VGT) doesn't seem to see the result of the copying.
 504          * Can we somehow flush the index buffer cache? Starting a new IB seems
 505          * to do the trick. */
 506         if (rctx->b.chip_class <= R700)
 507                 rctx->b.rings.gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
 508 }
 509
 510 /**
 511  * Global buffers are not really resources, they are are actually offsets
 512  * into a single global resource (r600_screen::global_pool).  The means
 513  * they don't have their own cs_buf handle, so they cannot be passed
 514  * to r600_copy_buffer() and must be handled separately.
 515  *
 516  * XXX: It should be possible to implement this function using
 517  * r600_copy_buffer() by passing the memory_pool resource as both src
 518  * and dst and updating dstx and src_box to point to the correct offsets.
 519  * This would likely perform better than the current implementation.
 520  */
 521 static void r600_copy_global_buffer(struct pipe_context *ctx,
 522                                     struct pipe_resource *dst, unsigned
 523                                     dstx, struct pipe_resource *src,
 524                                     const struct pipe_box *src_box)
 525 {
 526         struct pipe_box dst_box; struct pipe_transfer *src_pxfer,
 527         *dst_pxfer;
 528
 529         u_box_1d(dstx, src_box->width, &dst_box);
 530         void *src_ptr = ctx->transfer_map(ctx, src, 0, PIPE_TRANSFER_READ,
 531                                           src_box, &src_pxfer);
 532         void *dst_ptr = ctx->transfer_map(ctx, dst, 0, PIPE_TRANSFER_WRITE,
 533                                           &dst_box, &dst_pxfer);
 534         memcpy(dst_ptr, src_ptr, src_box->width);
 535
 536         ctx->transfer_unmap(ctx, src_pxfer);
 537         ctx->transfer_unmap(ctx, dst_pxfer);
 538 }
 539
 540 static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
 541                               unsigned offset, unsigned size, unsigned value)
 542 {
 543         struct r600_context *rctx = (struct r600_context*)ctx;
 544
 545         if (rctx->screen->b.has_cp_dma &&
 546             rctx->b.chip_class >= EVERGREEN &&
 547             offset % 4 == 0 && size % 4 == 0) {
 548                 evergreen_cp_dma_clear_buffer(rctx, dst, offset, size, value);
 549         } else if (rctx->screen->b.has_streamout && offset % 4 == 0 && size % 4 == 0) {
 550                 union pipe_color_union clear_value;
 551                 clear_value.ui[0] = value;
 552
 553                 r600_blitter_begin(ctx, R600_DISABLE_RENDER_COND);
 554                 util_blitter_clear_buffer(rctx->blitter, dst, offset, size,
 555                                           1, &clear_value);
 556                 r600_blitter_end(ctx);
 557         } else {
 558                 uint32_t *map = r600_buffer_map_sync_with_rings(&rctx->b, r600_resource(dst),
 559                                                                  PIPE_TRANSFER_WRITE);
 560                 size /= 4;
 561                 for (unsigned i = 0; i < size; i++)
 562                         *map++ = value;
 563         }
 564 }
 565
 566 static bool util_format_is_subsampled_2x1_32bpp(enum pipe_format format)
 567 {
 568         const struct util_format_description *desc = util_format_description(format);
 569
 570         return desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED &&
 571                desc->block.width == 2 &&
 572                desc->block.height == 1 &&
 573                desc->block.bits == 32;
 574 }
 575
 576 static void r600_resource_copy_region(struct pipe_context *ctx,
 577                                       struct pipe_resource *dst,
 578                                       unsigned dst_level,
 579                                       unsigned dstx, unsigned dsty, unsigned dstz,
 580                                       struct pipe_resource *src,
 581                                       unsigned src_level,
 582                                       const struct pipe_box *src_box)
 583 {
 584         struct r600_context *rctx = (struct r600_context *)ctx;
 585         struct pipe_surface *dst_view, dst_templ;
 586         struct pipe_sampler_view src_templ, *src_view;
 587         unsigned dst_width, dst_height, src_width0, src_height0, src_widthFL, src_heightFL;
 588         unsigned src_force_level = 0;
 589         struct pipe_box sbox, dstbox;
 590
 591         /* Handle buffers first. */
 592         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 593                 if ((src->bind & PIPE_BIND_GLOBAL) ||
 594                                         (dst->bind & PIPE_BIND_GLOBAL)) {
 595                         r600_copy_global_buffer(ctx, dst, dstx, src, src_box);
 596                 } else {
 597                         r600_copy_buffer(ctx, dst, dstx, src, src_box);
 598                 }
 599                 return;
 600         }
 601
 602         assert(u_max_sample(dst) == u_max_sample(src));
 603
 604         /* The driver doesn't decompress resources automatically while
 605          * u_blitter is rendering. */
 606         if (!r600_decompress_subresource(ctx, src, src_level,
 607                                          src_box->z, src_box->z + src_box->depth - 1)) {
 608                 return; /* error */
 609         }
 610
 611         dst_width = u_minify(dst->width0, dst_level);
 612         dst_height = u_minify(dst->height0, dst_level);
 613         src_width0 = src->width0;
 614         src_height0 = src->height0;
 615         src_widthFL = u_minify(src->width0, src_level);
 616         src_heightFL = u_minify(src->height0, src_level);
 617
 618         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 619         util_blitter_default_src_texture(&src_templ, src, src_level);
 620
 621         if (util_format_is_compressed(src->format)) {
 622                 unsigned blocksize = util_format_get_blocksize(src->format);
 623
 624                 if (blocksize == 8)
 625                         src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 626                 else
 627                         src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 628                 dst_templ.format = src_templ.format;
 629
 630                 dst_width = util_format_get_nblocksx(dst->format, dst_width);
 631                 dst_height = util_format_get_nblocksy(dst->format, dst_height);
 632                 src_width0 = util_format_get_nblocksx(src->format, src_width0);
 633                 src_height0 = util_format_get_nblocksy(src->format, src_height0);
 634                 src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
 635                 src_heightFL = util_format_get_nblocksy(src->format, src_heightFL);
 636
 637                 dstx = util_format_get_nblocksx(dst->format, dstx);
 638                 dsty = util_format_get_nblocksy(dst->format, dsty);
 639
 640                 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 641                 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
 642                 sbox.z = src_box->z;
 643                 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 644                 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
 645                 sbox.depth = src_box->depth;
 646                 src_box = &sbox;
 647
 648                 src_force_level = src_level;
 649         } else if (!util_blitter_is_copy_supported(rctx->blitter, dst, src)) {
 650                 if (util_format_is_subsampled_2x1_32bpp(src->format)) {
 651
 652                         src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 653                         dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 654
 655                         dst_width = util_format_get_nblocksx(dst->format, dst_width);
 656                         src_width0 = util_format_get_nblocksx(src->format, src_width0);
 657                         src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
 658
 659                         dstx = util_format_get_nblocksx(dst->format, dstx);
 660
 661                         sbox = *src_box;
 662                         sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 663                         sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 664                         src_box = &sbox;
 665                 } else {
 666                         unsigned blocksize = util_format_get_blocksize(src->format);
 667
 668                         switch (blocksize) {
 669                         case 1:
 670                                 dst_templ.format = PIPE_FORMAT_R8_UNORM;
 671                                 src_templ.format = PIPE_FORMAT_R8_UNORM;
 672                                 break;
 673                         case 2:
 674                                 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
 675                                 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
 676                                 break;
 677                         case 4:
 678                                 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 679                                 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 680                                 break;
 681                         case 8:
 682                                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 683                                 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 684                                 break;
 685                         case 16:
 686                                 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 687                                 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 688                                 break;
 689                         default:
 690                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 691                                         util_format_short_name(src->format), blocksize);
 692                                 assert(0);
 693                         }
 694                 }
 695         }
 696
 697         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ, dst_width, dst_height);
 698
 699         if (rctx->b.chip_class >= EVERGREEN) {
 700                 src_view = evergreen_create_sampler_view_custom(ctx, src, &src_templ,
 701                                                                 src_width0, src_height0,
 702                                                                 src_force_level);
 703         } else {
 704                 src_view = r600_create_sampler_view_custom(ctx, src, &src_templ,
 705                                                            src_widthFL, src_heightFL);
 706         }
 707
 708         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 709                  abs(src_box->depth), &dstbox);
 710
 711         /* Copy. */
 712         r600_blitter_begin(ctx, R600_COPY_TEXTURE);
 713         util_blitter_blit_generic(rctx->blitter, dst_view, &dstbox,
 714                                   src_view, src_box, src_width0, src_height0,
 715                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
 716         r600_blitter_end(ctx);
 717
 718         pipe_surface_reference(&dst_view, NULL);
 719         pipe_sampler_view_reference(&src_view, NULL);
 720 }
 721
 722 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 723 static enum pipe_format int_to_norm_format(enum pipe_format format)
 724 {
 725         switch (format) {
 726 #define REPLACE_FORMAT_SIGN(format,sign) \
 727         case PIPE_FORMAT_##format##_##sign##INT: \
 728                 return PIPE_FORMAT_##format##_##sign##NORM
 729 #define REPLACE_FORMAT(format) \
 730                 REPLACE_FORMAT_SIGN(format, U); \
 731                 REPLACE_FORMAT_SIGN(format, S)
 732
 733         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 734         REPLACE_FORMAT(R8);
 735         REPLACE_FORMAT(R8G8);
 736         REPLACE_FORMAT(R8G8B8X8);
 737         REPLACE_FORMAT(R8G8B8A8);
 738         REPLACE_FORMAT(A8);
 739         REPLACE_FORMAT(I8);
 740         REPLACE_FORMAT(L8);
 741         REPLACE_FORMAT(L8A8);
 742         REPLACE_FORMAT(R16);
 743         REPLACE_FORMAT(R16G16);
 744         REPLACE_FORMAT(R16G16B16X16);
 745         REPLACE_FORMAT(R16G16B16A16);
 746         REPLACE_FORMAT(A16);
 747         REPLACE_FORMAT(I16);
 748         REPLACE_FORMAT(L16);
 749         REPLACE_FORMAT(L16A16);
 750
 751 #undef REPLACE_FORMAT
 752 #undef REPLACE_FORMAT_SIGN
 753         default:
 754                 return format;
 755         }
 756 }
 757
 758 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 759                                      const struct pipe_blit_info *info)
 760 {
 761         struct r600_context *rctx = (struct r600_context*)ctx;
 762         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 763         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 764         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 765         enum pipe_format format = int_to_norm_format(info->dst.format);
 766         unsigned sample_mask =
 767                 rctx->b.chip_class == CAYMAN ? ~0 :
 768                 ((1ull << MAX2(1, info->src.resource->nr_samples)) - 1);
 769
 770         if (info->src.resource->nr_samples > 1 &&
 771             info->dst.resource->nr_samples <= 1 &&
 772             util_max_layer(info->src.resource, 0) == 0 &&
 773             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 774             info->dst.format == info->src.format &&
 775             !util_format_is_pure_integer(format) &&
 776             !util_format_is_depth_or_stencil(format) &&
 777             !info->scissor_enable &&
 778             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 779             dst_width == info->src.resource->width0 &&
 780             dst_height == info->src.resource->height0 &&
 781             info->dst.box.x == 0 &&
 782             info->dst.box.y == 0 &&
 783             info->dst.box.width == dst_width &&
 784             info->dst.box.height == dst_height &&
 785             info->dst.box.depth == 1 &&
 786             info->src.box.x == 0 &&
 787             info->src.box.y == 0 &&
 788             info->src.box.width == dst_width &&
 789             info->src.box.height == dst_height &&
 790             info->src.box.depth == 1 &&
 791             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 792             (!dst->cmask.size || !dst->dirty_level_mask) /* dst cannot be fast-cleared */) {
 793                 r600_blitter_begin(ctx, R600_COLOR_RESOLVE |
 794                                    (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
 795                 util_blitter_custom_resolve_color(rctx->blitter,
 796                                                   info->dst.resource, info->dst.level,
 797                                                   info->dst.box.z,
 798                                                   info->src.resource, info->src.box.z,
 799                                                   sample_mask, rctx->custom_blend_resolve,
 800                                                   format);
 801                 r600_blitter_end(ctx);
 802                 return true;
 803         }
 804         return false;
 805 }
 806
 807 static void r600_blit(struct pipe_context *ctx,
 808                       const struct pipe_blit_info *info)
 809 {
 810         struct r600_context *rctx = (struct r600_context*)ctx;
 811
 812         if (do_hardware_msaa_resolve(ctx, info)) {
 813                 return;
 814         }
 815
 816         assert(util_blitter_is_blit_supported(rctx->blitter, info));
 817
 818         /* The driver doesn't decompress resources automatically while
 819          * u_blitter is rendering. */
 820         if (!r600_decompress_subresource(ctx, info->src.resource, info->src.level,
 821                                          info->src.box.z,
 822                                          info->src.box.z + info->src.box.depth - 1)) {
 823                 return; /* error */
 824         }
 825
 826         r600_blitter_begin(ctx, R600_BLIT |
 827                            (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
 828         util_blitter_blit(rctx->blitter, info);
 829         r600_blitter_end(ctx);
 830 }
 831
 832 static void r600_flush_resource(struct pipe_context *ctx,
 833                                 struct pipe_resource *res)
 834 {
 835         struct r600_texture *rtex = (struct r600_texture*)res;
 836
 837         assert(res->target != PIPE_BUFFER);
 838
 839         if (!rtex->is_depth && rtex->cmask.size) {
 840                 r600_blit_decompress_color(ctx, rtex, 0, res->last_level,
 841                                            0, res->array_size - 1);
 842         }
 843 }
 844
 845 void r600_init_blit_functions(struct r600_context *rctx)
 846 {
 847         rctx->b.b.clear = r600_clear;
 848         rctx->b.b.clear_render_target = r600_clear_render_target;
 849         rctx->b.b.clear_depth_stencil = r600_clear_depth_stencil;
 850         rctx->b.b.resource_copy_region = r600_resource_copy_region;
 851         rctx->b.b.blit = r600_blit;
 852         rctx->b.b.flush_resource = r600_flush_resource;
 853         rctx->b.clear_buffer = r600_clear_buffer;
 854         rctx->b.blit_decompress_depth = r600_blit_decompress_depth;
 855 }