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_format.h"
  26 #include "evergreend.h"
  27
  28 enum r600_blitter_op /* bitmask */
  29 {
  30         R600_SAVE_FRAGMENT_STATE = 1,
  31         R600_SAVE_TEXTURES       = 2,
  32         R600_SAVE_FRAMEBUFFER    = 4,
  33         R600_DISABLE_RENDER_COND = 8,
  34
  35         R600_CLEAR         = R600_SAVE_FRAGMENT_STATE,
  36
  37         R600_CLEAR_SURFACE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER,
  38
  39         R600_COPY_BUFFER   = R600_DISABLE_RENDER_COND,
  40
  41         R600_COPY_TEXTURE  = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
  42                              R600_DISABLE_RENDER_COND,
  43
  44         R600_BLIT          = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES,
  45
  46         R600_DECOMPRESS    = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND,
  47
  48         R600_COLOR_RESOLVE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER
  49 };
  50
  51 static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
  52 {
  53         struct r600_context *rctx = (struct r600_context *)ctx;
  54
  55         r600_suspend_nontimer_queries(&rctx->b);
  56
  57         util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer_state.vb);
  58         util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_fetch_shader.cso);
  59         util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
  60         util_blitter_save_geometry_shader(rctx->blitter, rctx->gs_shader);
  61         util_blitter_save_so_targets(rctx->blitter, rctx->b.streamout.num_targets,
  62                                      (struct pipe_stream_output_target**)rctx->b.streamout.targets);
  63         util_blitter_save_rasterizer(rctx->blitter, rctx->rasterizer_state.cso);
  64
  65         if (op & R600_SAVE_FRAGMENT_STATE) {
  66                 util_blitter_save_viewport(rctx->blitter, &rctx->viewport[0].state);
  67                 util_blitter_save_scissor(rctx->blitter, &rctx->scissor[0].scissor);
  68                 util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
  69                 util_blitter_save_blend(rctx->blitter, rctx->blend_state.cso);
  70                 util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->dsa_state.cso);
  71                 util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref.pipe_state);
  72                 util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask);
  73         }
  74
  75         if (op & R600_SAVE_FRAMEBUFFER)
  76                 util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer.state);
  77
  78         if (op & R600_SAVE_TEXTURES) {
  79                 util_blitter_save_fragment_sampler_states(
  80                         rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].states.enabled_mask),
  81                         (void**)rctx->samplers[PIPE_SHADER_FRAGMENT].states.states);
  82
  83                 util_blitter_save_fragment_sampler_views(
  84                         rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.enabled_mask),
  85                         (struct pipe_sampler_view**)rctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
  86         }
  87
  88         if ((op & R600_DISABLE_RENDER_COND) && rctx->b.current_render_cond) {
  89            util_blitter_save_render_condition(rctx->blitter,
  90                                               rctx->b.current_render_cond,
  91                                               rctx->b.current_render_cond_cond,
  92                                               rctx->b.current_render_cond_mode);
  93         }
  94 }
  95
  96 static void r600_blitter_end(struct pipe_context *ctx)
  97 {
  98         struct r600_context *rctx = (struct r600_context *)ctx;
  99         r600_resume_nontimer_queries(&rctx->b);
 100 }
 101
 102 static unsigned u_max_sample(struct pipe_resource *r)
 103 {
 104         return r->nr_samples ? r->nr_samples - 1 : 0;
 105 }
 106
 107 static void r600_blit_decompress_depth(struct pipe_context *ctx,
 108                                        struct r600_texture *texture,
 109                                        struct r600_texture *staging,
 110                                        unsigned first_level, unsigned last_level,
 111                                        unsigned first_layer, unsigned last_layer,
 112                                        unsigned first_sample, unsigned last_sample)
 113 {
 114         struct r600_context *rctx = (struct r600_context *)ctx;
 115         unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 116         struct r600_texture *flushed_depth_texture = staging ?
 117                         staging : texture->flushed_depth_texture;
 118         const struct util_format_description *desc =
 119                 util_format_description(texture->resource.b.b.format);
 120         float depth;
 121
 122         if (!staging && !texture->dirty_level_mask)
 123                 return;
 124
 125         max_sample = u_max_sample(&texture->resource.b.b);
 126
 127         /* XXX Decompressing MSAA depth textures is broken on R6xx.
 128          * There is also a hardlock if CMASK and FMASK are not present.
 129          * Just skip this until we find out how to fix it. */
 130         if (rctx->b.chip_class == R600 && max_sample > 0) {
 131                 texture->dirty_level_mask = 0;
 132                 return;
 133         }
 134
 135         if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
 136             rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
 137                 depth = 0.0f;
 138         else
 139                 depth = 1.0f;
 140
 141         /* Enable decompression in DB_RENDER_CONTROL */
 142         rctx->db_misc_state.flush_depthstencil_through_cb = true;
 143         rctx->db_misc_state.copy_depth = util_format_has_depth(desc);
 144         rctx->db_misc_state.copy_stencil = util_format_has_stencil(desc);
 145         rctx->db_misc_state.copy_sample = first_sample;
 146         rctx->db_misc_state.atom.dirty = true;
 147
 148         for (level = first_level; level <= last_level; level++) {
 149                 if (!staging && !(texture->dirty_level_mask & (1 << level)))
 150                         continue;
 151
 152                 /* The smaller the mipmap level, the less layers there are
 153                  * as far as 3D textures are concerned. */
 154                 max_layer = util_max_layer(&texture->resource.b.b, level);
 155                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 156
 157                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 158                         for (sample = first_sample; sample <= last_sample; sample++) {
 159                                 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 160
 161                                 if (sample != rctx->db_misc_state.copy_sample) {
 162                                         rctx->db_misc_state.copy_sample = sample;
 163                                         rctx->db_misc_state.atom.dirty = true;
 164                                 }
 165
 166                                 surf_tmpl.format = texture->resource.b.b.format;
 167                                 surf_tmpl.u.tex.level = level;
 168                                 surf_tmpl.u.tex.first_layer = layer;
 169                                 surf_tmpl.u.tex.last_layer = layer;
 170
 171                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
 172
 173                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
 174                                 cbsurf = ctx->create_surface(ctx,
 175                                                 &flushed_depth_texture->resource.b.b, &surf_tmpl);
 176
 177                                 r600_blitter_begin(ctx, R600_DECOMPRESS);
 178                                 util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, 1 << sample,
 179                                                                   rctx->custom_dsa_flush, depth);
 180                                 r600_blitter_end(ctx);
 181
 182                                 pipe_surface_reference(&zsurf, NULL);
 183                                 pipe_surface_reference(&cbsurf, NULL);
 184                         }
 185                 }
 186
 187                 /* The texture will always be dirty if some layers or samples aren't flushed.
 188                  * I don't think this case occurs often though. */
 189                 if (!staging &&
 190                     first_layer == 0 && last_layer == max_layer &&
 191                     first_sample == 0 && last_sample == max_sample) {
 192                         texture->dirty_level_mask &= ~(1 << level);
 193                 }
 194         }
 195
 196         /* reenable compression in DB_RENDER_CONTROL */
 197         rctx->db_misc_state.flush_depthstencil_through_cb = false;
 198         rctx->db_misc_state.atom.dirty = true;
 199 }
 200
 201 static void r600_blit_decompress_depth_in_place(struct r600_context *rctx,
 202                                                 struct r600_texture *texture,
 203                                                 unsigned first_level, unsigned last_level,
 204                                                 unsigned first_layer, unsigned last_layer)
 205 {
 206         struct pipe_surface *zsurf, surf_tmpl = {{0}};
 207         unsigned layer, max_layer, checked_last_layer, level;
 208
 209         /* Enable decompression in DB_RENDER_CONTROL */
 210         rctx->db_misc_state.flush_depthstencil_in_place = true;
 211         rctx->db_misc_state.atom.dirty = true;
 212
 213         surf_tmpl.format = texture->resource.b.b.format;
 214
 215         for (level = first_level; level <= last_level; level++) {
 216                 if (!(texture->dirty_level_mask & (1 << level)))
 217                         continue;
 218
 219                 surf_tmpl.u.tex.level = level;
 220
 221                 /* The smaller the mipmap level, the less layers there are
 222                  * as far as 3D textures are concerned. */
 223                 max_layer = util_max_layer(&texture->resource.b.b, level);
 224                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 225
 226                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 227                         surf_tmpl.u.tex.first_layer = layer;
 228                         surf_tmpl.u.tex.last_layer = layer;
 229
 230                         zsurf = rctx->b.b.create_surface(&rctx->b.b, &texture->resource.b.b, &surf_tmpl);
 231
 232                         r600_blitter_begin(&rctx->b.b, R600_DECOMPRESS);
 233                         util_blitter_custom_depth_stencil(rctx->blitter, zsurf, NULL, ~0,
 234                                                           rctx->custom_dsa_flush, 1.0f);
 235                         r600_blitter_end(&rctx->b.b);
 236
 237                         pipe_surface_reference(&zsurf, NULL);
 238                 }
 239
 240                 /* The texture will always be dirty if some layers or samples aren't flushed.
 241                  * I don't think this case occurs often though. */
 242                 if (first_layer == 0 && last_layer == max_layer) {
 243                         texture->dirty_level_mask &= ~(1 << level);
 244                 }
 245         }
 246
 247         /* Disable decompression in DB_RENDER_CONTROL */
 248         rctx->db_misc_state.flush_depthstencil_in_place = false;
 249         rctx->db_misc_state.atom.dirty = true;
 250 }
 251
 252 void r600_decompress_depth_textures(struct r600_context *rctx,
 253                                struct r600_samplerview_state *textures)
 254 {
 255         unsigned i;
 256         unsigned depth_texture_mask = textures->compressed_depthtex_mask;
 257
 258         while (depth_texture_mask) {
 259                 struct pipe_sampler_view *view;
 260                 struct r600_texture *tex;
 261
 262                 i = u_bit_scan(&depth_texture_mask);
 263
 264                 view = &textures->views[i]->base;
 265                 assert(view);
 266
 267                 tex = (struct r600_texture *)view->texture;
 268                 assert(tex->is_depth && !tex->is_flushing_texture);
 269
 270                 if (rctx->b.chip_class >= EVERGREEN ||
 271                     r600_can_read_depth(tex)) {
 272                         r600_blit_decompress_depth_in_place(rctx, tex,
 273                                                    view->u.tex.first_level, view->u.tex.last_level,
 274                                                    0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 275                 } else {
 276                         r600_blit_decompress_depth(&rctx->b.b, tex, NULL,
 277                                                    view->u.tex.first_level, view->u.tex.last_level,
 278                                                    0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level),
 279                                                    0, u_max_sample(&tex->resource.b.b));
 280                 }
 281         }
 282 }
 283
 284 static void r600_blit_decompress_color(struct pipe_context *ctx,
 285                 struct r600_texture *rtex,
 286                 unsigned first_level, unsigned last_level,
 287                 unsigned first_layer, unsigned last_layer)
 288 {
 289         struct r600_context *rctx = (struct r600_context *)ctx;
 290         unsigned layer, level, checked_last_layer, max_layer;
 291
 292         if (!rtex->dirty_level_mask)
 293                 return;
 294
 295         for (level = first_level; level <= last_level; level++) {
 296                 if (!(rtex->dirty_level_mask & (1 << level)))
 297                         continue;
 298
 299                 /* The smaller the mipmap level, the less layers there are
 300                  * as far as 3D textures are concerned. */
 301                 max_layer = util_max_layer(&rtex->resource.b.b, level);
 302                 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
 303
 304                 for (layer = first_layer; layer <= checked_last_layer; layer++) {
 305                         struct pipe_surface *cbsurf, surf_tmpl;
 306
 307                         surf_tmpl.format = rtex->resource.b.b.format;
 308                         surf_tmpl.u.tex.level = level;
 309                         surf_tmpl.u.tex.first_layer = layer;
 310                         surf_tmpl.u.tex.last_layer = layer;
 311                         cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
 312
 313                         r600_blitter_begin(ctx, R600_DECOMPRESS);
 314                         util_blitter_custom_color(rctx->blitter, cbsurf,
 315                                 rtex->fmask.size ? rctx->custom_blend_decompress : rctx->custom_blend_fastclear);
 316                         r600_blitter_end(ctx);
 317
 318                         pipe_surface_reference(&cbsurf, NULL);
 319                 }
 320
 321                 /* The texture will always be dirty if some layers aren't flushed.
 322                  * I don't think this case occurs often though. */
 323                 if (first_layer == 0 && last_layer == max_layer) {
 324                         rtex->dirty_level_mask &= ~(1 << level);
 325                 }
 326         }
 327 }
 328
 329 void r600_decompress_color_textures(struct r600_context *rctx,
 330                                     struct r600_samplerview_state *textures)
 331 {
 332         unsigned i;
 333         unsigned mask = textures->compressed_colortex_mask;
 334
 335         while (mask) {
 336                 struct pipe_sampler_view *view;
 337                 struct r600_texture *tex;
 338
 339                 i = u_bit_scan(&mask);
 340
 341                 view = &textures->views[i]->base;
 342                 assert(view);
 343
 344                 tex = (struct r600_texture *)view->texture;
 345                 assert(tex->cmask.size);
 346
 347                 r600_blit_decompress_color(&rctx->b.b, tex,
 348                                            view->u.tex.first_level, view->u.tex.last_level,
 349                                            0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
 350         }
 351 }
 352
 353 /* Helper for decompressing a portion of a color or depth resource before
 354  * blitting if any decompression is needed.
 355  * The driver doesn't decompress resources automatically while u_blitter is
 356  * rendering. */
 357 static bool r600_decompress_subresource(struct pipe_context *ctx,
 358                                         struct pipe_resource *tex,
 359                                         unsigned level,
 360                                         unsigned first_layer, unsigned last_layer)
 361 {
 362         struct r600_context *rctx = (struct r600_context *)ctx;
 363         struct r600_texture *rtex = (struct r600_texture*)tex;
 364
 365         if (rtex->is_depth && !rtex->is_flushing_texture) {
 366                 if (rctx->b.chip_class >= EVERGREEN ||
 367                     r600_can_read_depth(rtex)) {
 368                         r600_blit_decompress_depth_in_place(rctx, rtex,
 369                                                    level, level,
 370                                                    first_layer, last_layer);
 371                 } else {
 372                         if (!r600_init_flushed_depth_texture(ctx, tex, NULL))
 373                                 return false; /* error */
 374
 375                         r600_blit_decompress_depth(ctx, rtex, NULL,
 376                                                    level, level,
 377                                                    first_layer, last_layer,
 378                                                    0, u_max_sample(tex));
 379                 }
 380         } else if (rtex->cmask.size) {
 381                 r600_blit_decompress_color(ctx, rtex, level, level,
 382                                            first_layer, last_layer);
 383         }
 384         return true;
 385 }
 386
 387 static void r600_clear(struct pipe_context *ctx, unsigned buffers,
 388                        const union pipe_color_union *color,
 389                        double depth, unsigned stencil)
 390 {
 391         struct r600_context *rctx = (struct r600_context *)ctx;
 392         struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
 393
 394         if (buffers & PIPE_CLEAR_COLOR && rctx->b.chip_class >= EVERGREEN) {
 395                 evergreen_do_fast_color_clear(&rctx->b, fb, &rctx->framebuffer.atom,
 396                                               &buffers, color);
 397         }
 398
 399         if (buffers & PIPE_CLEAR_COLOR) {
 400                 int i;
 401
 402                 /* These buffers cannot use fast clear, make sure to disable expansion. */
 403                 for (i = 0; i < fb->nr_cbufs; i++) {
 404                         struct r600_texture *tex;
 405
 406                         /* If not clearing this buffer, skip. */
 407                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 408                                 continue;
 409
 410                         if (!fb->cbufs[i])
 411                                 continue;
 412
 413                         tex = (struct r600_texture *)fb->cbufs[i]->texture;
 414                         if (tex->fmask.size == 0)
 415                                 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
 416                 }
 417         }
 418
 419         /* if hyperz enabled just clear hyperz */
 420         if (fb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
 421                 struct r600_texture *rtex;
 422                 unsigned level = fb->zsbuf->u.tex.level;
 423
 424                 rtex = (struct r600_texture*)fb->zsbuf->texture;
 425
 426                 /* We can't use hyperz fast clear if each slice of a texture
 427                  * array are clear to different value. To simplify code just
 428                  * disable fast clear for texture array.
 429                  */
 430                 /* Only use htile for first level */
 431                 if (rtex->htile_buffer && !level &&
 432                    fb->zsbuf->u.tex.first_layer == 0 &&
 433                    fb->zsbuf->u.tex.last_layer == util_max_layer(&rtex->resource.b.b, level)) {
 434                         if (rtex->depth_clear_value != depth) {
 435                                 rtex->depth_clear_value = depth;
 436                                 rctx->db_state.atom.dirty = true;
 437                         }
 438                         rctx->db_misc_state.htile_clear = true;
 439                         rctx->db_misc_state.atom.dirty = true;
 440                 }
 441         }
 442
 443         r600_blitter_begin(ctx, R600_CLEAR);
 444         util_blitter_clear(rctx->blitter, fb->width, fb->height, 1,
 445                            buffers, color, depth, stencil);
 446         r600_blitter_end(ctx);
 447
 448         /* disable fast clear */
 449         if (rctx->db_misc_state.htile_clear) {
 450                 rctx->db_misc_state.htile_clear = false;
 451                 rctx->db_misc_state.atom.dirty = true;
 452         }
 453 }
 454
 455 static void r600_clear_render_target(struct pipe_context *ctx,
 456                                      struct pipe_surface *dst,
 457                                      const union pipe_color_union *color,
 458                                      unsigned dstx, unsigned dsty,
 459                                      unsigned width, unsigned height)
 460 {
 461         struct r600_context *rctx = (struct r600_context *)ctx;
 462
 463         r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
 464         util_blitter_clear_render_target(rctx->blitter, dst, color,
 465                                          dstx, dsty, width, height);
 466         r600_blitter_end(ctx);
 467 }
 468
 469 static void r600_clear_depth_stencil(struct pipe_context *ctx,
 470                                      struct pipe_surface *dst,
 471                                      unsigned clear_flags,
 472                                      double depth,
 473                                      unsigned stencil,
 474                                      unsigned dstx, unsigned dsty,
 475                                      unsigned width, unsigned height)
 476 {
 477         struct r600_context *rctx = (struct r600_context *)ctx;
 478
 479         r600_blitter_begin(ctx, R600_CLEAR_SURFACE);
 480         util_blitter_clear_depth_stencil(rctx->blitter, dst, clear_flags, depth, stencil,
 481                                          dstx, dsty, width, height);
 482         r600_blitter_end(ctx);
 483 }
 484
 485 static void r600_copy_buffer(struct pipe_context *ctx, struct pipe_resource *dst, unsigned dstx,
 486                              struct pipe_resource *src, const struct pipe_box *src_box)
 487 {
 488         struct r600_context *rctx = (struct r600_context*)ctx;
 489
 490         if (rctx->screen->b.has_cp_dma) {
 491                 r600_cp_dma_copy_buffer(rctx, dst, dstx, src, src_box->x, src_box->width);
 492         }
 493         else if (rctx->screen->b.has_streamout &&
 494                  /* Require 4-byte alignment. */
 495                  dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
 496
 497                 r600_blitter_begin(ctx, R600_COPY_BUFFER);
 498                 util_blitter_copy_buffer(rctx->blitter, dst, dstx, src, src_box->x, src_box->width);
 499                 r600_blitter_end(ctx);
 500         } else {
 501                 util_resource_copy_region(ctx, dst, 0, dstx, 0, 0, src, 0, src_box);
 502         }
 503
 504         /* The index buffer (VGT) doesn't seem to see the result of the copying.
 505          * Can we somehow flush the index buffer cache? Starting a new IB seems
 506          * to do the trick. */
 507         if (rctx->b.chip_class <= R700)
 508                 rctx->b.rings.gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
 509 }
 510
 511 /**
 512  * Global buffers are not really resources, they are are actually offsets
 513  * into a single global resource (r600_screen::global_pool).  The means
 514  * they don't have their own cs_buf handle, so they cannot be passed
 515  * to r600_copy_buffer() and must be handled separately.
 516  *
 517  * XXX: It should be possible to implement this function using
 518  * r600_copy_buffer() by passing the memory_pool resource as both src
 519  * and dst and updating dstx and src_box to point to the correct offsets.
 520  * This would likely perform better than the current implementation.
 521  */
 522 static void r600_copy_global_buffer(struct pipe_context *ctx,
 523                                     struct pipe_resource *dst, unsigned
 524                                     dstx, struct pipe_resource *src,
 525                                     const struct pipe_box *src_box)
 526 {
 527         struct pipe_box dst_box; struct pipe_transfer *src_pxfer,
 528         *dst_pxfer;
 529
 530         u_box_1d(dstx, src_box->width, &dst_box);
 531         void *src_ptr = ctx->transfer_map(ctx, src, 0, PIPE_TRANSFER_READ,
 532                                           src_box, &src_pxfer);
 533         void *dst_ptr = ctx->transfer_map(ctx, dst, 0, PIPE_TRANSFER_WRITE,
 534                                           &dst_box, &dst_pxfer);
 535         memcpy(dst_ptr, src_ptr, src_box->width);
 536
 537         ctx->transfer_unmap(ctx, src_pxfer);
 538         ctx->transfer_unmap(ctx, dst_pxfer);
 539 }
 540
 541 static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
 542                               unsigned offset, unsigned size, unsigned value)
 543 {
 544         struct r600_context *rctx = (struct r600_context*)ctx;
 545
 546         if (rctx->screen->b.has_cp_dma &&
 547             rctx->b.chip_class >= EVERGREEN &&
 548             offset % 4 == 0 && size % 4 == 0) {
 549                 evergreen_cp_dma_clear_buffer(rctx, dst, offset, size, value);
 550         } else if (rctx->screen->b.has_streamout && offset % 4 == 0 && size % 4 == 0) {
 551                 union pipe_color_union clear_value;
 552                 clear_value.ui[0] = value;
 553
 554                 r600_blitter_begin(ctx, R600_DISABLE_RENDER_COND);
 555                 util_blitter_clear_buffer(rctx->blitter, dst, offset, size,
 556                                           1, &clear_value);
 557                 r600_blitter_end(ctx);
 558         } else {
 559                 uint32_t *map = r600_buffer_map_sync_with_rings(&rctx->b, r600_resource(dst),
 560                                                                  PIPE_TRANSFER_WRITE);
 561                 size /= 4;
 562                 for (unsigned i = 0; i < size; i++)
 563                         *map++ = value;
 564         }
 565 }
 566
 567 static void r600_resource_copy_region(struct pipe_context *ctx,
 568                                       struct pipe_resource *dst,
 569                                       unsigned dst_level,
 570                                       unsigned dstx, unsigned dsty, unsigned dstz,
 571                                       struct pipe_resource *src,
 572                                       unsigned src_level,
 573                                       const struct pipe_box *src_box)
 574 {
 575         struct r600_context *rctx = (struct r600_context *)ctx;
 576         struct pipe_surface *dst_view, dst_templ;
 577         struct pipe_sampler_view src_templ, *src_view;
 578         unsigned dst_width, dst_height, src_width0, src_height0, src_widthFL, src_heightFL;
 579         unsigned src_force_level = 0;
 580         struct pipe_box sbox, dstbox;
 581
 582         /* Handle buffers first. */
 583         if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 584                 if ((src->bind & PIPE_BIND_GLOBAL) ||
 585                                         (dst->bind & PIPE_BIND_GLOBAL)) {
 586                         r600_copy_global_buffer(ctx, dst, dstx, src, src_box);
 587                 } else {
 588                         r600_copy_buffer(ctx, dst, dstx, src, src_box);
 589                 }
 590                 return;
 591         }
 592
 593         assert(u_max_sample(dst) == u_max_sample(src));
 594
 595         /* The driver doesn't decompress resources automatically while
 596          * u_blitter is rendering. */
 597         if (!r600_decompress_subresource(ctx, src, src_level,
 598                                          src_box->z, src_box->z + src_box->depth - 1)) {
 599                 return; /* error */
 600         }
 601
 602         dst_width = u_minify(dst->width0, dst_level);
 603         dst_height = u_minify(dst->height0, dst_level);
 604         src_width0 = src->width0;
 605         src_height0 = src->height0;
 606         src_widthFL = u_minify(src->width0, src_level);
 607         src_heightFL = u_minify(src->height0, src_level);
 608
 609         util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 610         util_blitter_default_src_texture(&src_templ, src, src_level);
 611
 612         if (util_format_is_compressed(src->format)) {
 613                 unsigned blocksize = util_format_get_blocksize(src->format);
 614
 615                 if (blocksize == 8)
 616                         src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
 617                 else
 618                         src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
 619                 dst_templ.format = src_templ.format;
 620
 621                 dst_width = util_format_get_nblocksx(dst->format, dst_width);
 622                 dst_height = util_format_get_nblocksy(dst->format, dst_height);
 623                 src_width0 = util_format_get_nblocksx(src->format, src_width0);
 624                 src_height0 = util_format_get_nblocksy(src->format, src_height0);
 625                 src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
 626                 src_heightFL = util_format_get_nblocksy(src->format, src_heightFL);
 627
 628                 dstx = util_format_get_nblocksx(dst->format, dstx);
 629                 dsty = util_format_get_nblocksy(dst->format, dsty);
 630
 631                 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 632                 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
 633                 sbox.z = src_box->z;
 634                 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 635                 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
 636                 sbox.depth = src_box->depth;
 637                 src_box = &sbox;
 638
 639                 src_force_level = src_level;
 640         } else if (!util_blitter_is_copy_supported(rctx->blitter, dst, src)) {
 641                 if (util_format_is_subsampled_422(src->format)) {
 642
 643                         src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 644                         dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
 645
 646                         dst_width = util_format_get_nblocksx(dst->format, dst_width);
 647                         src_width0 = util_format_get_nblocksx(src->format, src_width0);
 648                         src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
 649
 650                         dstx = util_format_get_nblocksx(dst->format, dstx);
 651
 652                         sbox = *src_box;
 653                         sbox.x = util_format_get_nblocksx(src->format, src_box->x);
 654                         sbox.width = util_format_get_nblocksx(src->format, src_box->width);
 655                         src_box = &sbox;
 656                 } else {
 657                         unsigned blocksize = util_format_get_blocksize(src->format);
 658
 659                         switch (blocksize) {
 660                         case 1:
 661                                 dst_templ.format = PIPE_FORMAT_R8_UNORM;
 662                                 src_templ.format = PIPE_FORMAT_R8_UNORM;
 663                                 break;
 664                         case 2:
 665                                 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
 666                                 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
 667                                 break;
 668                         case 4:
 669                                 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 670                                 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 671                                 break;
 672                         case 8:
 673                                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 674                                 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
 675                                 break;
 676                         case 16:
 677                                 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 678                                 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
 679                                 break;
 680                         default:
 681                                 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
 682                                         util_format_short_name(src->format), blocksize);
 683                                 assert(0);
 684                         }
 685                 }
 686         }
 687
 688         dst_view = r600_create_surface_custom(ctx, dst, &dst_templ, dst_width, dst_height);
 689
 690         if (rctx->b.chip_class >= EVERGREEN) {
 691                 src_view = evergreen_create_sampler_view_custom(ctx, src, &src_templ,
 692                                                                 src_width0, src_height0,
 693                                                                 src_force_level);
 694         } else {
 695                 src_view = r600_create_sampler_view_custom(ctx, src, &src_templ,
 696                                                            src_widthFL, src_heightFL);
 697         }
 698
 699         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 700                  abs(src_box->depth), &dstbox);
 701
 702         /* Copy. */
 703         r600_blitter_begin(ctx, R600_COPY_TEXTURE);
 704         util_blitter_blit_generic(rctx->blitter, dst_view, &dstbox,
 705                                   src_view, src_box, src_width0, src_height0,
 706                                   PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL);
 707         r600_blitter_end(ctx);
 708
 709         pipe_surface_reference(&dst_view, NULL);
 710         pipe_sampler_view_reference(&src_view, NULL);
 711 }
 712
 713 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
 714 static enum pipe_format int_to_norm_format(enum pipe_format format)
 715 {
 716         switch (format) {
 717 #define REPLACE_FORMAT_SIGN(format,sign) \
 718         case PIPE_FORMAT_##format##_##sign##INT: \
 719                 return PIPE_FORMAT_##format##_##sign##NORM
 720 #define REPLACE_FORMAT(format) \
 721                 REPLACE_FORMAT_SIGN(format, U); \
 722                 REPLACE_FORMAT_SIGN(format, S)
 723
 724         REPLACE_FORMAT_SIGN(B10G10R10A2, U);
 725         REPLACE_FORMAT(R8);
 726         REPLACE_FORMAT(R8G8);
 727         REPLACE_FORMAT(R8G8B8X8);
 728         REPLACE_FORMAT(R8G8B8A8);
 729         REPLACE_FORMAT(A8);
 730         REPLACE_FORMAT(I8);
 731         REPLACE_FORMAT(L8);
 732         REPLACE_FORMAT(L8A8);
 733         REPLACE_FORMAT(R16);
 734         REPLACE_FORMAT(R16G16);
 735         REPLACE_FORMAT(R16G16B16X16);
 736         REPLACE_FORMAT(R16G16B16A16);
 737         REPLACE_FORMAT(A16);
 738         REPLACE_FORMAT(I16);
 739         REPLACE_FORMAT(L16);
 740         REPLACE_FORMAT(L16A16);
 741
 742 #undef REPLACE_FORMAT
 743 #undef REPLACE_FORMAT_SIGN
 744         default:
 745                 return format;
 746         }
 747 }
 748
 749 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
 750                                      const struct pipe_blit_info *info)
 751 {
 752         struct r600_context *rctx = (struct r600_context*)ctx;
 753         struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
 754         unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 755         unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 756         enum pipe_format format = int_to_norm_format(info->dst.format);
 757         unsigned sample_mask =
 758                 rctx->b.chip_class == CAYMAN ? ~0 :
 759                 ((1ull << MAX2(1, info->src.resource->nr_samples)) - 1);
 760
 761         if (info->src.resource->nr_samples > 1 &&
 762             info->dst.resource->nr_samples <= 1 &&
 763             util_max_layer(info->src.resource, 0) == 0 &&
 764             util_max_layer(info->dst.resource, info->dst.level) == 0 &&
 765             info->dst.format == info->src.format &&
 766             !util_format_is_pure_integer(format) &&
 767             !util_format_is_depth_or_stencil(format) &&
 768             !info->scissor_enable &&
 769             (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
 770             dst_width == info->src.resource->width0 &&
 771             dst_height == info->src.resource->height0 &&
 772             info->dst.box.x == 0 &&
 773             info->dst.box.y == 0 &&
 774             info->dst.box.width == dst_width &&
 775             info->dst.box.height == dst_height &&
 776             info->dst.box.depth == 1 &&
 777             info->src.box.x == 0 &&
 778             info->src.box.y == 0 &&
 779             info->src.box.width == dst_width &&
 780             info->src.box.height == dst_height &&
 781             info->src.box.depth == 1 &&
 782             dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
 783             (!dst->cmask.size || !dst->dirty_level_mask) /* dst cannot be fast-cleared */) {
 784                 r600_blitter_begin(ctx, R600_COLOR_RESOLVE |
 785                                    (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
 786                 util_blitter_custom_resolve_color(rctx->blitter,
 787                                                   info->dst.resource, info->dst.level,
 788                                                   info->dst.box.z,
 789                                                   info->src.resource, info->src.box.z,
 790                                                   sample_mask, rctx->custom_blend_resolve,
 791                                                   format);
 792                 r600_blitter_end(ctx);
 793                 return true;
 794         }
 795         return false;
 796 }
 797
 798 static void r600_blit(struct pipe_context *ctx,
 799                       const struct pipe_blit_info *info)
 800 {
 801         struct r600_context *rctx = (struct r600_context*)ctx;
 802
 803         if (do_hardware_msaa_resolve(ctx, info)) {
 804                 return;
 805         }
 806
 807         assert(util_blitter_is_blit_supported(rctx->blitter, info));
 808
 809         /* The driver doesn't decompress resources automatically while
 810          * u_blitter is rendering. */
 811         if (!r600_decompress_subresource(ctx, info->src.resource, info->src.level,
 812                                          info->src.box.z,
 813                                          info->src.box.z + info->src.box.depth - 1)) {
 814                 return; /* error */
 815         }
 816
 817         r600_blitter_begin(ctx, R600_BLIT |
 818                            (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
 819         util_blitter_blit(rctx->blitter, info);
 820         r600_blitter_end(ctx);
 821 }
 822
 823 static void r600_flush_resource(struct pipe_context *ctx,
 824                                 struct pipe_resource *res)
 825 {
 826         struct r600_texture *rtex = (struct r600_texture*)res;
 827
 828         assert(res->target != PIPE_BUFFER);
 829
 830         if (!rtex->is_depth && rtex->cmask.size) {
 831                 r600_blit_decompress_color(ctx, rtex, 0, res->last_level,
 832                                            0, util_max_layer(res, 0));
 833         }
 834 }
 835
 836 void r600_init_blit_functions(struct r600_context *rctx)
 837 {
 838         rctx->b.b.clear = r600_clear;
 839         rctx->b.b.clear_render_target = r600_clear_render_target;
 840         rctx->b.b.clear_depth_stencil = r600_clear_depth_stencil;
 841         rctx->b.b.resource_copy_region = r600_resource_copy_region;
 842         rctx->b.b.blit = r600_blit;
 843         rctx->b.b.flush_resource = r600_flush_resource;
 844         rctx->b.clear_buffer = r600_clear_buffer;
 845         rctx->b.blit_decompress_depth = r600_blit_decompress_depth;
 846 }