src/gallium/drivers/r300/r300_blit.c

   1 /*
   2  * Copyright 2009 Marek Olšák <maraeo@gmail.com>
   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 "r300_context.h"
  24 #include "r300_emit.h"
  25 #include "r300_texture.h"
  26 #include "r300_reg.h"
  27
  28 #include "util/u_format.h"
  29 #include "util/u_pack_color.h"
  30 #include "util/u_surface.h"
  31
  32 enum r300_blitter_op /* bitmask */
  33 {
  34     R300_STOP_QUERY         = 1,
  35     R300_SAVE_TEXTURES      = 2,
  36     R300_SAVE_FRAMEBUFFER   = 4,
  37     R300_IGNORE_RENDER_COND = 8,
  38
  39     R300_CLEAR         = R300_STOP_QUERY,
  40
  41     R300_CLEAR_SURFACE = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER,
  42
  43     R300_COPY          = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
  44                          R300_SAVE_TEXTURES | R300_IGNORE_RENDER_COND,
  45
  46     R300_BLIT          = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
  47                          R300_SAVE_TEXTURES | R300_IGNORE_RENDER_COND,
  48
  49     R300_DECOMPRESS    = R300_STOP_QUERY | R300_IGNORE_RENDER_COND,
  50 };
  51
  52 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
  53 {
  54     if ((op & R300_STOP_QUERY) && r300->query_current) {
  55         r300->blitter_saved_query = r300->query_current;
  56         r300_stop_query(r300);
  57     }
  58
  59     /* Yeah we have to save all those states to ensure the blitter operation
  60      * is really transparent. The states will be restored by the blitter once
  61      * copying is done. */
  62     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
  63     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
  64     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
  65     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
  66     util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
  67     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
  68     util_blitter_save_viewport(r300->blitter, &r300->viewport);
  69     util_blitter_save_scissor(r300->blitter, r300->scissor_state.state);
  70     util_blitter_save_sample_mask(r300->blitter, *(unsigned*)r300->sample_mask.state);
  71     util_blitter_save_vertex_buffer_slot(r300->blitter, r300->vertex_buffer);
  72     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
  73
  74     if (op & R300_SAVE_FRAMEBUFFER) {
  75         util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
  76     }
  77
  78     if (op & R300_SAVE_TEXTURES) {
  79         struct r300_textures_state* state =
  80             (struct r300_textures_state*)r300->textures_state.state;
  81
  82         util_blitter_save_fragment_sampler_states(
  83             r300->blitter, state->sampler_state_count,
  84             (void**)state->sampler_states);
  85
  86         util_blitter_save_fragment_sampler_views(
  87             r300->blitter, state->sampler_view_count,
  88             (struct pipe_sampler_view**)state->sampler_views);
  89     }
  90
  91     if (op & R300_IGNORE_RENDER_COND) {
  92         /* Save the flag. */
  93         r300->blitter_saved_skip_rendering = r300->skip_rendering+1;
  94         r300->skip_rendering = FALSE;
  95     } else {
  96         r300->blitter_saved_skip_rendering = 0;
  97     }
  98 }
  99
 100 static void r300_blitter_end(struct r300_context *r300)
 101 {
 102     if (r300->blitter_saved_query) {
 103         r300_resume_query(r300, r300->blitter_saved_query);
 104         r300->blitter_saved_query = NULL;
 105     }
 106
 107     if (r300->blitter_saved_skip_rendering) {
 108         /* Restore the flag. */
 109         r300->skip_rendering = r300->blitter_saved_skip_rendering-1;
 110     }
 111 }
 112
 113 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
 114                                           const float* rgba)
 115 {
 116     union util_color uc;
 117     util_pack_color(rgba, format, &uc);
 118
 119     if (util_format_get_blocksizebits(format) == 32)
 120         return uc.ui;
 121     else
 122         return uc.us | (uc.us << 16);
 123 }
 124
 125 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
 126                                        unsigned clear_buffers)
 127 {
 128     struct pipe_framebuffer_state *fb =
 129         (struct pipe_framebuffer_state*)r300->fb_state.state;
 130
 131     /* Only color clear allowed, and only one colorbuffer. */
 132     if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
 133         return FALSE;
 134
 135     return r300_surface(fb->cbufs[0])->cbzb_allowed;
 136 }
 137
 138 static boolean r300_fast_zclear_allowed(struct r300_context *r300,
 139                                         unsigned clear_buffers)
 140 {
 141     struct pipe_framebuffer_state *fb =
 142         (struct pipe_framebuffer_state*)r300->fb_state.state;
 143
 144     return r300_resource(fb->zsbuf->texture)->tex.zmask_dwords[fb->zsbuf->u.tex.level] != 0;
 145 }
 146
 147 static boolean r300_hiz_clear_allowed(struct r300_context *r300)
 148 {
 149     struct pipe_framebuffer_state *fb =
 150         (struct pipe_framebuffer_state*)r300->fb_state.state;
 151
 152     return r300_resource(fb->zsbuf->texture)->tex.hiz_dwords[fb->zsbuf->u.tex.level] != 0;
 153 }
 154
 155 static uint32_t r300_depth_clear_value(enum pipe_format format,
 156                                        double depth, unsigned stencil)
 157 {
 158     switch (format) {
 159         case PIPE_FORMAT_Z16_UNORM:
 160         case PIPE_FORMAT_X8Z24_UNORM:
 161             return util_pack_z(format, depth);
 162
 163         case PIPE_FORMAT_S8_UINT_Z24_UNORM:
 164             return util_pack_z_stencil(format, depth, stencil);
 165
 166         default:
 167             assert(0);
 168             return 0;
 169     }
 170 }
 171
 172 static uint32_t r300_hiz_clear_value(double depth)
 173 {
 174     uint32_t r = (uint32_t)(CLAMP(depth, 0, 1) * 255.5);
 175     assert(r <= 255);
 176     return r | (r << 8) | (r << 16) | (r << 24);
 177 }
 178
 179 DEBUG_GET_ONCE_BOOL_OPTION(hyperz, "RADEON_HYPERZ", FALSE)
 180
 181 /* Clear currently bound buffers. */
 182 static void r300_clear(struct pipe_context* pipe,
 183                        unsigned buffers,
 184                        const union pipe_color_union *color,
 185                        double depth,
 186                        unsigned stencil)
 187 {
 188     /* My notes about Zbuffer compression:
 189      *
 190      * 1) The zbuffer must be micro-tiled and whole microtiles must be
 191      *    written if compression is enabled. If microtiling is disabled,
 192      *    it locks up.
 193      *
 194      * 2) There is ZMASK RAM which contains a compressed zbuffer.
 195      *    Each dword of the Z Mask contains compression information
 196      *    for 16 4x4 pixel tiles, that is 2 bits for each tile.
 197      *    On chips with 2 Z pipes, every other dword maps to a different
 198      *    pipe. On newer chipsets, there is a new compression mode
 199      *    with 8x8 pixel tiles per 2 bits.
 200      *
 201      * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
 202      *    it should look in the ZMASK RAM first before fetching from a real
 203      *    zbuffer.
 204      *
 205      * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
 206      *    during zbuffer reads instead of the value that is actually stored
 207      *    in the zbuffer memory. A pixel is in a cleared state when its ZMASK
 208      *    is equal to 0. Therefore, if you clear ZMASK with zeros, you may
 209      *    leave the zbuffer memory uninitialized, but then you must enable
 210      *    compression, so that the ZMASK RAM is actually used.
 211      *
 212      * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
 213      *    during zbuffer updates. A special decompressing operation should be
 214      *    used to fully decompress a zbuffer, which basically just stores all
 215      *    compressed tiles in ZMASK to the zbuffer memory.
 216      *
 217      * 6) For a 16-bit zbuffer, compression causes a hung with one or
 218      *    two samples and should not be used.
 219      *
 220      * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
 221      *    to avoid needless decompression.
 222      *
 223      * 8) Fastfill must not be used if reading of compressed Z data is disabled
 224      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
 225      *    i.e. it cannot be used to compress the zbuffer.
 226      *
 227      * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
 228      *
 229      * - Marek
 230      */
 231
 232     struct r300_context* r300 = r300_context(pipe);
 233     struct pipe_framebuffer_state *fb =
 234         (struct pipe_framebuffer_state*)r300->fb_state.state;
 235     struct r300_hyperz_state *hyperz =
 236         (struct r300_hyperz_state*)r300->hyperz_state.state;
 237     uint32_t width = fb->width;
 238     uint32_t height = fb->height;
 239     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
 240
 241     /* Use fast Z clear.
 242      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
 243     if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
 244         boolean zmask_clear, hiz_clear;
 245
 246         /* If both depth and stencil are present, they must be cleared together. */
 247         if (fb->zsbuf->texture->format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
 248             (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL) {
 249             zmask_clear = FALSE;
 250             hiz_clear = FALSE;
 251         } else {
 252             zmask_clear = r300_fast_zclear_allowed(r300, buffers);
 253             hiz_clear = r300_hiz_clear_allowed(r300);
 254         }
 255
 256         /* If we need Hyper-Z. */
 257         if (zmask_clear || hiz_clear) {
 258             /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
 259             if (!r300->hyperz_enabled &&
 260                 (r300->screen->caps.is_r500 || debug_get_option_hyperz())) {
 261                 r300->hyperz_enabled =
 262                     r300->rws->cs_request_feature(r300->cs,
 263                                                 RADEON_FID_R300_HYPERZ_ACCESS,
 264                                                 TRUE);
 265                 if (r300->hyperz_enabled) {
 266                    /* Need to emit HyperZ buffer regs for the first time. */
 267                    r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 268                 }
 269             }
 270
 271             /* Setup Hyper-Z clears. */
 272             if (r300->hyperz_enabled) {
 273                 if (zmask_clear) {
 274                     hyperz_dcv = hyperz->zb_depthclearvalue =
 275                         r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
 276
 277                     r300_mark_atom_dirty(r300, &r300->zmask_clear);
 278                     buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 279                 }
 280
 281                 if (hiz_clear) {
 282                     r300->hiz_clear_value = r300_hiz_clear_value(depth);
 283                     r300_mark_atom_dirty(r300, &r300->hiz_clear);
 284                 }
 285                 r300->num_z_clears++;
 286             }
 287         }
 288     }
 289
 290     /* Enable CBZB clear. */
 291     if (r300_cbzb_clear_allowed(r300, buffers)) {
 292         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 293
 294         hyperz->zb_depthclearvalue =
 295                 r300_depth_clear_cb_value(surf->base.format, color->f);
 296
 297         width = surf->cbzb_width;
 298         height = surf->cbzb_height;
 299
 300         r300->cbzb_clear = TRUE;
 301         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 302     }
 303
 304     /* Clear. */
 305     if (buffers) {
 306         enum pipe_format cformat = fb->nr_cbufs ? fb->cbufs[0]->format : PIPE_FORMAT_NONE;
 307         /* Clear using the blitter. */
 308         r300_blitter_begin(r300, R300_CLEAR);
 309         util_blitter_clear(r300->blitter,
 310                            width,
 311                            height,
 312                            fb->nr_cbufs,
 313                            buffers, cformat, color, depth, stencil);
 314         r300_blitter_end(r300);
 315     } else if (r300->zmask_clear.dirty || r300->hiz_clear.dirty) {
 316         /* Just clear zmask and hiz now, this does not use the standard draw
 317          * procedure. */
 318         /* Calculate zmask_clear and hiz_clear atom sizes. */
 319         unsigned dwords =
 320             (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
 321             (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
 322             r300_get_num_cs_end_dwords(r300);
 323
 324         /* Reserve CS space. */
 325         if (dwords > (RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
 326             r300_flush(&r300->context, RADEON_FLUSH_ASYNC, NULL);
 327         }
 328
 329         /* Emit clear packets. */
 330         if (r300->zmask_clear.dirty) {
 331             r300_emit_zmask_clear(r300, r300->zmask_clear.size,
 332                                   r300->zmask_clear.state);
 333             r300->zmask_clear.dirty = FALSE;
 334         }
 335         if (r300->hiz_clear.dirty) {
 336             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
 337                                 r300->hiz_clear.state);
 338             r300->hiz_clear.dirty = FALSE;
 339         }
 340     } else {
 341         assert(0);
 342     }
 343
 344     /* Disable CBZB clear. */
 345     if (r300->cbzb_clear) {
 346         r300->cbzb_clear = FALSE;
 347         hyperz->zb_depthclearvalue = hyperz_dcv;
 348         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 349     }
 350
 351     /* Enable fastfill and/or hiz.
 352      *
 353      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
 354      * looks if zmask/hiz is in use and programs hardware accordingly. */
 355     if (r300->zmask_in_use || r300->hiz_in_use) {
 356         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 357     }
 358 }
 359
 360 /* Clear a region of a color surface to a constant value. */
 361 static void r300_clear_render_target(struct pipe_context *pipe,
 362                                      struct pipe_surface *dst,
 363                                      const union pipe_color_union *color,
 364                                      unsigned dstx, unsigned dsty,
 365                                      unsigned width, unsigned height)
 366 {
 367     struct r300_context *r300 = r300_context(pipe);
 368
 369     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 370     util_blitter_clear_render_target(r300->blitter, dst, color,
 371                                      dstx, dsty, width, height);
 372     r300_blitter_end(r300);
 373 }
 374
 375 /* Clear a region of a depth stencil surface. */
 376 static void r300_clear_depth_stencil(struct pipe_context *pipe,
 377                                      struct pipe_surface *dst,
 378                                      unsigned clear_flags,
 379                                      double depth,
 380                                      unsigned stencil,
 381                                      unsigned dstx, unsigned dsty,
 382                                      unsigned width, unsigned height)
 383 {
 384     struct r300_context *r300 = r300_context(pipe);
 385     struct pipe_framebuffer_state *fb =
 386         (struct pipe_framebuffer_state*)r300->fb_state.state;
 387
 388     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 389         if (fb->zsbuf->texture == dst->texture) {
 390             r300_decompress_zmask(r300);
 391         }
 392     }
 393
 394     /* XXX Do not decompress ZMask of the currently-set zbuffer. */
 395     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 396     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
 397                                      dstx, dsty, width, height);
 398     r300_blitter_end(r300);
 399 }
 400
 401 void r300_decompress_zmask(struct r300_context *r300)
 402 {
 403     struct pipe_framebuffer_state *fb =
 404         (struct pipe_framebuffer_state*)r300->fb_state.state;
 405
 406     if (!r300->zmask_in_use || r300->locked_zbuffer)
 407         return;
 408
 409     r300->zmask_decompress = TRUE;
 410     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 411
 412     r300_blitter_begin(r300, R300_DECOMPRESS);
 413     util_blitter_custom_clear_depth(r300->blitter, fb->width, fb->height, 0,
 414                                     r300->dsa_decompress_zmask);
 415     r300_blitter_end(r300);
 416
 417     r300->zmask_decompress = FALSE;
 418     r300->zmask_in_use = FALSE;
 419     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 420 }
 421
 422 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
 423 {
 424     struct pipe_framebuffer_state fb;
 425
 426     memset(&fb, 0, sizeof(fb));
 427     fb.width = r300->locked_zbuffer->width;
 428     fb.height = r300->locked_zbuffer->height;
 429     fb.zsbuf = r300->locked_zbuffer;
 430
 431     r300->context.set_framebuffer_state(&r300->context, &fb);
 432     r300_decompress_zmask(r300);
 433 }
 434
 435 void r300_decompress_zmask_locked(struct r300_context *r300)
 436 {
 437     struct pipe_framebuffer_state saved_fb;
 438
 439     memset(&saved_fb, 0, sizeof(saved_fb));
 440     util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
 441     r300_decompress_zmask_locked_unsafe(r300);
 442     r300->context.set_framebuffer_state(&r300->context, &saved_fb);
 443     util_unreference_framebuffer_state(&saved_fb);
 444
 445     pipe_surface_reference(&r300->locked_zbuffer, NULL);
 446 }
 447
 448 bool r300_is_blit_supported(enum pipe_format format)
 449 {
 450     const struct util_format_description *desc =
 451         util_format_description(format);
 452
 453     return desc->layout == UTIL_FORMAT_LAYOUT_PLAIN ||
 454            desc->layout == UTIL_FORMAT_LAYOUT_S3TC ||
 455            desc->layout == UTIL_FORMAT_LAYOUT_RGTC;
 456 }
 457
 458 /* Copy a block of pixels from one surface to another. */
 459 static void r300_resource_copy_region(struct pipe_context *pipe,
 460                                       struct pipe_resource *dst,
 461                                       unsigned dst_level,
 462                                       unsigned dstx, unsigned dsty, unsigned dstz,
 463                                       struct pipe_resource *src,
 464                                       unsigned src_level,
 465                                       const struct pipe_box *src_box)
 466 {
 467     struct pipe_screen *screen = pipe->screen;
 468     struct r300_context *r300 = r300_context(pipe);
 469     struct pipe_framebuffer_state *fb =
 470         (struct pipe_framebuffer_state*)r300->fb_state.state;
 471     unsigned src_width0 = r300_resource(src)->tex.width0;
 472     unsigned src_height0 = r300_resource(src)->tex.height0;
 473     unsigned dst_width0 = r300_resource(dst)->tex.width0;
 474     unsigned dst_height0 = r300_resource(dst)->tex.height0;
 475     unsigned layout;
 476     struct pipe_box box, dstbox;
 477     struct pipe_sampler_view src_templ, *src_view;
 478     struct pipe_surface dst_templ, *dst_view;
 479
 480     /* Fallback for buffers. */
 481     if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) ||
 482         !r300_is_blit_supported(dst->format)) {
 483         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 484                                   src, src_level, src_box);
 485         return;
 486     }
 487
 488     /* Can't read MSAA textures. */
 489     if (src->nr_samples > 1 || dst->nr_samples > 1) {
 490         return;
 491     }
 492
 493     /* The code below changes the texture format so that the copy can be done
 494      * on hardware. E.g. depth-stencil surfaces are copied as RGBA
 495      * colorbuffers. */
 496
 497     util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 498     util_blitter_default_src_texture(&src_templ, src, src_level);
 499
 500     layout = util_format_description(dst_templ.format)->layout;
 501
 502     /* Handle non-renderable plain formats. */
 503     if (layout == UTIL_FORMAT_LAYOUT_PLAIN &&
 504         (!screen->is_format_supported(screen, src_templ.format, src->target,
 505                                       src->nr_samples,
 506                                       PIPE_BIND_SAMPLER_VIEW) ||
 507          !screen->is_format_supported(screen, dst_templ.format, dst->target,
 508                                       dst->nr_samples,
 509                                       PIPE_BIND_RENDER_TARGET))) {
 510         switch (util_format_get_blocksize(dst_templ.format)) {
 511             case 1:
 512                 dst_templ.format = PIPE_FORMAT_I8_UNORM;
 513                 break;
 514             case 2:
 515                 dst_templ.format = PIPE_FORMAT_B4G4R4A4_UNORM;
 516                 break;
 517             case 4:
 518                 dst_templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 519                 break;
 520             case 8:
 521                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UNORM;
 522                 break;
 523             default:
 524                 debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
 525                              "r300: copy_region: Software fallback doesn't work for tiled textures.\n",
 526                              util_format_short_name(dst_templ.format));
 527         }
 528         src_templ.format = dst_templ.format;
 529     }
 530
 531     /* Handle compressed formats. */
 532     if (layout == UTIL_FORMAT_LAYOUT_S3TC ||
 533         layout == UTIL_FORMAT_LAYOUT_RGTC) {
 534         assert(src_templ.format == dst_templ.format);
 535
 536         box = *src_box;
 537         src_box = &box;
 538
 539         dst_width0 = align(dst_width0, 4);
 540         dst_height0 = align(dst_height0, 4);
 541         src_width0 = align(src_width0, 4);
 542         src_height0 = align(src_height0, 4);
 543         box.width = align(box.width, 4);
 544         box.height = align(box.height, 4);
 545
 546         switch (util_format_get_blocksize(dst_templ.format)) {
 547         case 8:
 548             /* one 4x4 pixel block has 8 bytes.
 549              * we set 1 pixel = 4 bytes ===> 1 block corrensponds to 2 pixels. */
 550             dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 551             dst_width0 = dst_width0 / 2;
 552             src_width0 = src_width0 / 2;
 553             dstx /= 2;
 554             box.x /= 2;
 555             box.width /= 2;
 556             break;
 557         case 16:
 558             /* one 4x4 pixel block has 16 bytes.
 559              * we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
 560             dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 561             break;
 562         }
 563         src_templ.format = dst_templ.format;
 564
 565         dst_height0 = dst_height0 / 4;
 566         src_height0 = src_height0 / 4;
 567         dsty /= 4;
 568         box.y /= 4;
 569         box.height /= 4;
 570     }
 571
 572     /* Fallback for textures. */
 573     if (!screen->is_format_supported(screen, dst_templ.format,
 574                                      dst->target, dst->nr_samples,
 575                                      PIPE_BIND_RENDER_TARGET) ||
 576         !screen->is_format_supported(screen, src_templ.format,
 577                                      src->target, src->nr_samples,
 578                                      PIPE_BIND_SAMPLER_VIEW)) {
 579         assert(0 && "this shouldn't happen, update r300_is_blit_supported");
 580         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 581                                   src, src_level, src_box);
 582         return;
 583     }
 584
 585     /* Decompress ZMASK. */
 586     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 587         if (fb->zsbuf->texture == src ||
 588             fb->zsbuf->texture == dst) {
 589             r300_decompress_zmask(r300);
 590         }
 591     }
 592
 593     dst_view = r300_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
 594     src_view = r300_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
 595
 596     u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 597              abs(src_box->depth), &dstbox);
 598
 599     r300_blitter_begin(r300, R300_COPY);
 600     util_blitter_blit_generic(r300->blitter, dst_view, &dstbox,
 601                               src_view, src_box, src_width0, src_height0,
 602                               PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
 603                               FALSE);
 604     r300_blitter_end(r300);
 605
 606     pipe_surface_reference(&dst_view, NULL);
 607     pipe_sampler_view_reference(&src_view, NULL);
 608 }
 609
 610 static boolean r300_is_simple_msaa_resolve(const struct pipe_blit_info *info)
 611 {
 612     unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 613     unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 614
 615     return info->dst.resource->format == info->src.resource->format &&
 616            info->dst.resource->format == info->dst.format &&
 617            info->src.resource->format == info->src.format &&
 618            !info->scissor_enable &&
 619            info->mask == PIPE_MASK_RGBA &&
 620            dst_width == info->src.resource->width0 &&
 621            dst_height == info->src.resource->height0 &&
 622            info->dst.box.x == 0 &&
 623            info->dst.box.y == 0 &&
 624            info->dst.box.width == dst_width &&
 625            info->dst.box.height == dst_height &&
 626            info->src.box.x == 0 &&
 627            info->src.box.y == 0 &&
 628            info->src.box.width == dst_width &&
 629            info->src.box.height == dst_height &&
 630            (r300_resource(info->dst.resource)->tex.microtile != RADEON_LAYOUT_LINEAR ||
 631             r300_resource(info->dst.resource)->tex.macrotile[info->dst.level] != RADEON_LAYOUT_LINEAR);
 632 }
 633
 634 static void r300_simple_msaa_resolve(struct pipe_context *pipe,
 635                                      struct pipe_resource *dst,
 636                                      unsigned dst_level,
 637                                      unsigned dst_layer,
 638                                      struct pipe_resource *src,
 639                                      enum pipe_format format)
 640 {
 641     struct r300_context *r300 = r300_context(pipe);
 642     struct r300_surface *srcsurf, *dstsurf;
 643     struct pipe_surface surf_tmpl;
 644     struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
 645
 646     memset(&surf_tmpl, 0, sizeof(surf_tmpl));
 647     surf_tmpl.format = format;
 648     srcsurf = r300_surface(pipe->create_surface(pipe, src, &surf_tmpl));
 649
 650     surf_tmpl.format = format;
 651     surf_tmpl.u.tex.level = dst_level;
 652     surf_tmpl.u.tex.first_layer =
 653     surf_tmpl.u.tex.last_layer = dst_layer;
 654     dstsurf = r300_surface(pipe->create_surface(pipe, dst, &surf_tmpl));
 655
 656     /* COLORPITCH should contain the tiling info of the resolve buffer.
 657      * The tiling of the AA buffer isn't programmable anyway. */
 658     srcsurf->pitch &= ~(R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
 659     srcsurf->pitch |= dstsurf->pitch & (R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
 660
 661     /* Enable AA resolve. */
 662     aa->dest = dstsurf;
 663     r300->aa_state.size = 8;
 664     r300_mark_atom_dirty(r300, &r300->aa_state);
 665
 666     /* Resolve the surface. */
 667     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 668     util_blitter_custom_color(r300->blitter, &srcsurf->base, NULL);
 669     r300_blitter_end(r300);
 670
 671     /* Disable AA resolve. */
 672     aa->dest = NULL;
 673     r300->aa_state.size = 4;
 674     r300_mark_atom_dirty(r300, &r300->aa_state);
 675
 676     pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
 677     pipe_surface_reference((struct pipe_surface**)&dstsurf, NULL);
 678 }
 679
 680 static void r300_msaa_resolve(struct pipe_context *pipe,
 681                               const struct pipe_blit_info *info)
 682 {
 683     struct r300_context *r300 = r300_context(pipe);
 684     struct pipe_screen *screen = pipe->screen;
 685     struct pipe_resource *tmp, templ;
 686     struct pipe_blit_info blit;
 687
 688     assert(info->src.level == 0);
 689     assert(info->src.box.z == 0);
 690     assert(info->src.box.depth == 1);
 691     assert(info->dst.box.depth == 1);
 692
 693     if (r300_is_simple_msaa_resolve(info)) {
 694         r300_simple_msaa_resolve(pipe, info->dst.resource, info->dst.level,
 695                                  info->dst.box.z, info->src.resource,
 696                                  info->src.format);
 697         return;
 698     }
 699
 700     /* resolve into a temporary texture, then blit */
 701     memset(&templ, 0, sizeof(templ));
 702     templ.target = PIPE_TEXTURE_2D;
 703     templ.format = info->src.resource->format;
 704     templ.width0 = info->src.resource->width0;
 705     templ.height0 = info->src.resource->height0;
 706     templ.depth0 = 1;
 707     templ.array_size = 1;
 708     templ.usage = PIPE_USAGE_STATIC;
 709     templ.flags = R300_RESOURCE_FORCE_MICROTILING;
 710
 711     tmp = screen->resource_create(screen, &templ);
 712
 713     /* resolve */
 714     r300_simple_msaa_resolve(pipe, tmp, 0, 0, info->src.resource,
 715                              info->src.format);
 716
 717     /* blit */
 718     blit = *info;
 719     blit.src.resource = tmp;
 720     blit.src.box.z = 0;
 721
 722     r300_blitter_begin(r300, R300_BLIT);
 723     util_blitter_blit(r300->blitter, &blit);
 724     r300_blitter_end(r300);
 725
 726     pipe_resource_reference(&tmp, NULL);
 727 }
 728
 729 static void r300_blit(struct pipe_context *pipe,
 730                       const struct pipe_blit_info *blit)
 731 {
 732     struct r300_context *r300 = r300_context(pipe);
 733     struct pipe_framebuffer_state *fb =
 734         (struct pipe_framebuffer_state*)r300->fb_state.state;
 735     struct pipe_blit_info info = *blit;
 736
 737     /* MSAA resolve. */
 738     if (info.src.resource->nr_samples > 1 &&
 739         info.dst.resource->nr_samples <= 1 &&
 740         !util_format_is_depth_or_stencil(info.src.resource->format)) {
 741         r300_msaa_resolve(pipe, &info);
 742         return;
 743     }
 744
 745     /* Can't read MSAA textures. */
 746     if (info.src.resource->nr_samples > 1) {
 747         return;
 748     }
 749
 750     /* Blit a combined depth-stencil resource as color.
 751      * S8Z24 is the only supported stencil format. */
 752     if ((info.mask & PIPE_MASK_S) &&
 753         info.src.format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
 754         info.dst.format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
 755         if (info.dst.resource->nr_samples > 1) {
 756             /* Cannot do that with MSAA buffers. */
 757             info.mask &= ~PIPE_MASK_S;
 758             if (!(info.mask & PIPE_MASK_Z)) {
 759                 return;
 760             }
 761         } else {
 762             /* Single-sample buffer. */
 763             info.src.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 764             info.dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 765             if (info.mask & PIPE_MASK_Z) {
 766                 info.mask = PIPE_MASK_RGBA; /* depth+stencil */
 767             } else {
 768                 info.mask = PIPE_MASK_B; /* stencil only */
 769             }
 770         }
 771     }
 772
 773     /* Decompress ZMASK. */
 774     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 775         if (fb->zsbuf->texture == info.src.resource ||
 776             fb->zsbuf->texture == info.dst.resource) {
 777             r300_decompress_zmask(r300);
 778         }
 779     }
 780
 781     r300_blitter_begin(r300, R300_BLIT);
 782     util_blitter_blit(r300->blitter, &info);
 783     r300_blitter_end(r300);
 784 }
 785
 786 void r300_init_blit_functions(struct r300_context *r300)
 787 {
 788     r300->context.clear = r300_clear;
 789     r300->context.clear_render_target = r300_clear_render_target;
 790     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
 791     r300->context.resource_copy_region = r300_resource_copy_region;
 792     r300->context.blit = r300_blit;
 793 }