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_half.h"
  30 #include "util/u_pack_color.h"
  31 #include "util/u_surface.h"
  32
  33 enum r300_blitter_op /* bitmask */
  34 {
  35     R300_STOP_QUERY         = 1,
  36     R300_SAVE_TEXTURES      = 2,
  37     R300_SAVE_FRAMEBUFFER   = 4,
  38     R300_IGNORE_RENDER_COND = 8,
  39
  40     R300_CLEAR         = R300_STOP_QUERY,
  41
  42     R300_CLEAR_SURFACE = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER,
  43
  44     R300_COPY          = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
  45                          R300_SAVE_TEXTURES | R300_IGNORE_RENDER_COND,
  46
  47     R300_BLIT          = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
  48                          R300_SAVE_TEXTURES | R300_IGNORE_RENDER_COND,
  49
  50     R300_DECOMPRESS    = R300_STOP_QUERY | R300_IGNORE_RENDER_COND,
  51 };
  52
  53 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
  54 {
  55     if ((op & R300_STOP_QUERY) && r300->query_current) {
  56         r300->blitter_saved_query = r300->query_current;
  57         r300_stop_query(r300);
  58     }
  59
  60     /* Yeah we have to save all those states to ensure the blitter operation
  61      * is really transparent. The states will be restored by the blitter once
  62      * copying is done. */
  63     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
  64     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
  65     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
  66     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
  67     util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
  68     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
  69     util_blitter_save_viewport(r300->blitter, &r300->viewport);
  70     util_blitter_save_scissor(r300->blitter, r300->scissor_state.state);
  71     util_blitter_save_sample_mask(r300->blitter, *(unsigned*)r300->sample_mask.state);
  72     util_blitter_save_vertex_buffer_slot(r300->blitter, r300->vertex_buffer);
  73     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
  74
  75     if (op & R300_SAVE_FRAMEBUFFER) {
  76         util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
  77     }
  78
  79     if (op & R300_SAVE_TEXTURES) {
  80         struct r300_textures_state* state =
  81             (struct r300_textures_state*)r300->textures_state.state;
  82
  83         util_blitter_save_fragment_sampler_states(
  84             r300->blitter, state->sampler_state_count,
  85             (void**)state->sampler_states);
  86
  87         util_blitter_save_fragment_sampler_views(
  88             r300->blitter, state->sampler_view_count,
  89             (struct pipe_sampler_view**)state->sampler_views);
  90     }
  91
  92     if (op & R300_IGNORE_RENDER_COND) {
  93         /* Save the flag. */
  94         r300->blitter_saved_skip_rendering = r300->skip_rendering+1;
  95         r300->skip_rendering = FALSE;
  96     } else {
  97         r300->blitter_saved_skip_rendering = 0;
  98     }
  99 }
 100
 101 static void r300_blitter_end(struct r300_context *r300)
 102 {
 103     if (r300->blitter_saved_query) {
 104         r300_resume_query(r300, r300->blitter_saved_query);
 105         r300->blitter_saved_query = NULL;
 106     }
 107
 108     if (r300->blitter_saved_skip_rendering) {
 109         /* Restore the flag. */
 110         r300->skip_rendering = r300->blitter_saved_skip_rendering-1;
 111     }
 112 }
 113
 114 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
 115                                           const float* rgba)
 116 {
 117     union util_color uc;
 118     util_pack_color(rgba, format, &uc);
 119
 120     if (util_format_get_blocksizebits(format) == 32)
 121         return uc.ui;
 122     else
 123         return uc.us | (uc.us << 16);
 124 }
 125
 126 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
 127                                        unsigned clear_buffers)
 128 {
 129     struct pipe_framebuffer_state *fb =
 130         (struct pipe_framebuffer_state*)r300->fb_state.state;
 131
 132     /* Only color clear allowed, and only one colorbuffer. */
 133     if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
 134         return FALSE;
 135
 136     return r300_surface(fb->cbufs[0])->cbzb_allowed;
 137 }
 138
 139 static boolean r300_fast_zclear_allowed(struct r300_context *r300,
 140                                         unsigned clear_buffers)
 141 {
 142     struct pipe_framebuffer_state *fb =
 143         (struct pipe_framebuffer_state*)r300->fb_state.state;
 144
 145     return r300_resource(fb->zsbuf->texture)->tex.zmask_dwords[fb->zsbuf->u.tex.level] != 0;
 146 }
 147
 148 static boolean r300_hiz_clear_allowed(struct r300_context *r300)
 149 {
 150     struct pipe_framebuffer_state *fb =
 151         (struct pipe_framebuffer_state*)r300->fb_state.state;
 152
 153     return r300_resource(fb->zsbuf->texture)->tex.hiz_dwords[fb->zsbuf->u.tex.level] != 0;
 154 }
 155
 156 static uint32_t r300_depth_clear_value(enum pipe_format format,
 157                                        double depth, unsigned stencil)
 158 {
 159     switch (format) {
 160         case PIPE_FORMAT_Z16_UNORM:
 161         case PIPE_FORMAT_X8Z24_UNORM:
 162             return util_pack_z(format, depth);
 163
 164         case PIPE_FORMAT_S8_UINT_Z24_UNORM:
 165             return util_pack_z_stencil(format, depth, stencil);
 166
 167         default:
 168             assert(0);
 169             return 0;
 170     }
 171 }
 172
 173 static uint32_t r300_hiz_clear_value(double depth)
 174 {
 175     uint32_t r = (uint32_t)(CLAMP(depth, 0, 1) * 255.5);
 176     assert(r <= 255);
 177     return r | (r << 8) | (r << 16) | (r << 24);
 178 }
 179
 180 static void r300_set_clear_color(struct r300_context *r300,
 181                                  const union pipe_color_union *color)
 182 {
 183     struct pipe_framebuffer_state *fb =
 184         (struct pipe_framebuffer_state*)r300->fb_state.state;
 185     union util_color uc;
 186
 187     memset(&uc, 0, sizeof(uc));
 188     util_pack_color(color->f, fb->cbufs[0]->format, &uc);
 189
 190     if (fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
 191         fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16X16_FLOAT) {
 192         /* (0,1,2,3) maps to (B,G,R,A) */
 193         r300->color_clear_value_gb = uc.h[0] | ((uint32_t)uc.h[1] << 16);
 194         r300->color_clear_value_ar = uc.h[2] | ((uint32_t)uc.h[3] << 16);
 195     } else {
 196         r300->color_clear_value = uc.ui;
 197     }
 198 }
 199
 200 DEBUG_GET_ONCE_BOOL_OPTION(hyperz, "RADEON_HYPERZ", FALSE)
 201
 202 /* Clear currently bound buffers. */
 203 static void r300_clear(struct pipe_context* pipe,
 204                        unsigned buffers,
 205                        const union pipe_color_union *color,
 206                        double depth,
 207                        unsigned stencil)
 208 {
 209     /* My notes about Zbuffer compression:
 210      *
 211      * 1) The zbuffer must be micro-tiled and whole microtiles must be
 212      *    written if compression is enabled. If microtiling is disabled,
 213      *    it locks up.
 214      *
 215      * 2) There is ZMASK RAM which contains a compressed zbuffer.
 216      *    Each dword of the Z Mask contains compression information
 217      *    for 16 4x4 pixel tiles, that is 2 bits for each tile.
 218      *    On chips with 2 Z pipes, every other dword maps to a different
 219      *    pipe. On newer chipsets, there is a new compression mode
 220      *    with 8x8 pixel tiles per 2 bits.
 221      *
 222      * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
 223      *    it should look in the ZMASK RAM first before fetching from a real
 224      *    zbuffer.
 225      *
 226      * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
 227      *    during zbuffer reads instead of the value that is actually stored
 228      *    in the zbuffer memory. A pixel is in a cleared state when its ZMASK
 229      *    is equal to 0. Therefore, if you clear ZMASK with zeros, you may
 230      *    leave the zbuffer memory uninitialized, but then you must enable
 231      *    compression, so that the ZMASK RAM is actually used.
 232      *
 233      * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
 234      *    during zbuffer updates. A special decompressing operation should be
 235      *    used to fully decompress a zbuffer, which basically just stores all
 236      *    compressed tiles in ZMASK to the zbuffer memory.
 237      *
 238      * 6) For a 16-bit zbuffer, compression causes a hung with one or
 239      *    two samples and should not be used.
 240      *
 241      * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
 242      *    to avoid needless decompression.
 243      *
 244      * 8) Fastfill must not be used if reading of compressed Z data is disabled
 245      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
 246      *    i.e. it cannot be used to compress the zbuffer.
 247      *
 248      * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
 249      *
 250      * - Marek
 251      */
 252
 253     struct r300_context* r300 = r300_context(pipe);
 254     struct pipe_framebuffer_state *fb =
 255         (struct pipe_framebuffer_state*)r300->fb_state.state;
 256     struct r300_hyperz_state *hyperz =
 257         (struct r300_hyperz_state*)r300->hyperz_state.state;
 258     uint32_t width = fb->width;
 259     uint32_t height = fb->height;
 260     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
 261
 262     /* Use fast Z clear.
 263      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
 264     if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
 265         boolean zmask_clear, hiz_clear;
 266
 267         /* If both depth and stencil are present, they must be cleared together. */
 268         if (fb->zsbuf->texture->format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
 269             (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL) {
 270             zmask_clear = FALSE;
 271             hiz_clear = FALSE;
 272         } else {
 273             zmask_clear = r300_fast_zclear_allowed(r300, buffers);
 274             hiz_clear = r300_hiz_clear_allowed(r300);
 275         }
 276
 277         /* If we need Hyper-Z. */
 278         if (zmask_clear || hiz_clear) {
 279             /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
 280             if (!r300->hyperz_enabled &&
 281                 (r300->screen->caps.is_r500 || debug_get_option_hyperz())) {
 282                 r300->hyperz_enabled =
 283                     r300->rws->cs_request_feature(r300->cs,
 284                                                 RADEON_FID_R300_HYPERZ_ACCESS,
 285                                                 TRUE);
 286                 if (r300->hyperz_enabled) {
 287                    /* Need to emit HyperZ buffer regs for the first time. */
 288                    r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 289                 }
 290             }
 291
 292             /* Setup Hyper-Z clears. */
 293             if (r300->hyperz_enabled) {
 294                 if (zmask_clear) {
 295                     hyperz_dcv = hyperz->zb_depthclearvalue =
 296                         r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
 297
 298                     r300_mark_atom_dirty(r300, &r300->zmask_clear);
 299                     r300_mark_atom_dirty(r300, &r300->gpu_flush);
 300                     buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 301                 }
 302
 303                 if (hiz_clear) {
 304                     r300->hiz_clear_value = r300_hiz_clear_value(depth);
 305                     r300_mark_atom_dirty(r300, &r300->hiz_clear);
 306                     r300_mark_atom_dirty(r300, &r300->gpu_flush);
 307                 }
 308                 r300->num_z_clears++;
 309             }
 310         }
 311     }
 312
 313     /* Use fast color clear for an AA colorbuffer.
 314      * The CMASK is shared between all colorbuffers, so we use it
 315      * if there is only one colorbuffer bound. */
 316     if ((buffers & PIPE_CLEAR_COLOR) && fb->nr_cbufs == 1 &&
 317         r300_resource(fb->cbufs[0]->texture)->tex.cmask_dwords) {
 318         /* Try to obtain the access to the CMASK if we don't have one. */
 319         if (!r300->cmask_access) {
 320             r300->cmask_access =
 321                 r300->rws->cs_request_feature(r300->cs,
 322                                               RADEON_FID_R300_CMASK_ACCESS,
 323                                               TRUE);
 324         }
 325
 326         /* Setup the clear. */
 327         if (r300->cmask_access) {
 328             /* Pair the resource with the CMASK to avoid other resources
 329              * accessing it. */
 330             if (!r300->screen->cmask_resource) {
 331                 pipe_mutex_lock(r300->screen->cmask_mutex);
 332                 /* Double checking (first unlocked, then locked). */
 333                 if (!r300->screen->cmask_resource) {
 334                     /* Don't reference this, so that the texture can be
 335                      * destroyed while set in cmask_resource.
 336                      * Then in texture_destroy, we set cmask_resource to NULL. */
 337                     r300->screen->cmask_resource = fb->cbufs[0]->texture;
 338                 }
 339                 pipe_mutex_unlock(r300->screen->cmask_mutex);
 340             }
 341
 342             if (r300->screen->cmask_resource == fb->cbufs[0]->texture) {
 343                 r300_set_clear_color(r300, color);
 344                 r300_mark_atom_dirty(r300, &r300->cmask_clear);
 345                 r300_mark_atom_dirty(r300, &r300->gpu_flush);
 346                 buffers &= ~PIPE_CLEAR_COLOR;
 347             }
 348         }
 349     }
 350     /* Enable CBZB clear. */
 351     else if (r300_cbzb_clear_allowed(r300, buffers)) {
 352         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 353
 354         hyperz->zb_depthclearvalue =
 355                 r300_depth_clear_cb_value(surf->base.format, color->f);
 356
 357         width = surf->cbzb_width;
 358         height = surf->cbzb_height;
 359
 360         r300->cbzb_clear = TRUE;
 361         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 362     }
 363
 364     /* Clear. */
 365     if (buffers) {
 366         /* Clear using the blitter. */
 367         r300_blitter_begin(r300, R300_CLEAR);
 368         util_blitter_clear(r300->blitter,
 369                            width,
 370                            height,
 371                            buffers, color, depth, stencil);
 372         r300_blitter_end(r300);
 373     } else if (r300->zmask_clear.dirty ||
 374                r300->hiz_clear.dirty ||
 375                r300->cmask_clear.dirty) {
 376         /* Just clear zmask and hiz now, this does not use the standard draw
 377          * procedure. */
 378         /* Calculate zmask_clear and hiz_clear atom sizes. */
 379         unsigned dwords =
 380             r300->gpu_flush.size +
 381             (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
 382             (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
 383             (r300->cmask_clear.dirty ? r300->cmask_clear.size : 0) +
 384             r300_get_num_cs_end_dwords(r300);
 385
 386         /* Reserve CS space. */
 387         if (dwords > (RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
 388             r300_flush(&r300->context, RADEON_FLUSH_ASYNC, NULL);
 389         }
 390
 391         /* Emit clear packets. */
 392         r300_emit_gpu_flush(r300, r300->gpu_flush.size, r300->gpu_flush.state);
 393         r300->gpu_flush.dirty = FALSE;
 394
 395         if (r300->zmask_clear.dirty) {
 396             r300_emit_zmask_clear(r300, r300->zmask_clear.size,
 397                                   r300->zmask_clear.state);
 398             r300->zmask_clear.dirty = FALSE;
 399         }
 400         if (r300->hiz_clear.dirty) {
 401             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
 402                                 r300->hiz_clear.state);
 403             r300->hiz_clear.dirty = FALSE;
 404         }
 405         if (r300->cmask_clear.dirty) {
 406             r300_emit_cmask_clear(r300, r300->cmask_clear.size,
 407                                   r300->cmask_clear.state);
 408             r300->cmask_clear.dirty = FALSE;
 409         }
 410     } else {
 411         assert(0);
 412     }
 413
 414     /* Disable CBZB clear. */
 415     if (r300->cbzb_clear) {
 416         r300->cbzb_clear = FALSE;
 417         hyperz->zb_depthclearvalue = hyperz_dcv;
 418         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 419     }
 420
 421     /* Enable fastfill and/or hiz.
 422      *
 423      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
 424      * looks if zmask/hiz is in use and programs hardware accordingly. */
 425     if (r300->zmask_in_use || r300->hiz_in_use) {
 426         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 427     }
 428 }
 429
 430 /* Clear a region of a color surface to a constant value. */
 431 static void r300_clear_render_target(struct pipe_context *pipe,
 432                                      struct pipe_surface *dst,
 433                                      const union pipe_color_union *color,
 434                                      unsigned dstx, unsigned dsty,
 435                                      unsigned width, unsigned height)
 436 {
 437     struct r300_context *r300 = r300_context(pipe);
 438
 439     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 440     util_blitter_clear_render_target(r300->blitter, dst, color,
 441                                      dstx, dsty, width, height);
 442     r300_blitter_end(r300);
 443 }
 444
 445 /* Clear a region of a depth stencil surface. */
 446 static void r300_clear_depth_stencil(struct pipe_context *pipe,
 447                                      struct pipe_surface *dst,
 448                                      unsigned clear_flags,
 449                                      double depth,
 450                                      unsigned stencil,
 451                                      unsigned dstx, unsigned dsty,
 452                                      unsigned width, unsigned height)
 453 {
 454     struct r300_context *r300 = r300_context(pipe);
 455     struct pipe_framebuffer_state *fb =
 456         (struct pipe_framebuffer_state*)r300->fb_state.state;
 457
 458     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 459         if (fb->zsbuf->texture == dst->texture) {
 460             r300_decompress_zmask(r300);
 461         }
 462     }
 463
 464     /* XXX Do not decompress ZMask of the currently-set zbuffer. */
 465     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 466     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
 467                                      dstx, dsty, width, height);
 468     r300_blitter_end(r300);
 469 }
 470
 471 void r300_decompress_zmask(struct r300_context *r300)
 472 {
 473     struct pipe_framebuffer_state *fb =
 474         (struct pipe_framebuffer_state*)r300->fb_state.state;
 475
 476     if (!r300->zmask_in_use || r300->locked_zbuffer)
 477         return;
 478
 479     r300->zmask_decompress = TRUE;
 480     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 481
 482     r300_blitter_begin(r300, R300_DECOMPRESS);
 483     util_blitter_custom_clear_depth(r300->blitter, fb->width, fb->height, 0,
 484                                     r300->dsa_decompress_zmask);
 485     r300_blitter_end(r300);
 486
 487     r300->zmask_decompress = FALSE;
 488     r300->zmask_in_use = FALSE;
 489     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 490 }
 491
 492 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
 493 {
 494     struct pipe_framebuffer_state fb;
 495
 496     memset(&fb, 0, sizeof(fb));
 497     fb.width = r300->locked_zbuffer->width;
 498     fb.height = r300->locked_zbuffer->height;
 499     fb.zsbuf = r300->locked_zbuffer;
 500
 501     r300->context.set_framebuffer_state(&r300->context, &fb);
 502     r300_decompress_zmask(r300);
 503 }
 504
 505 void r300_decompress_zmask_locked(struct r300_context *r300)
 506 {
 507     struct pipe_framebuffer_state saved_fb;
 508
 509     memset(&saved_fb, 0, sizeof(saved_fb));
 510     util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
 511     r300_decompress_zmask_locked_unsafe(r300);
 512     r300->context.set_framebuffer_state(&r300->context, &saved_fb);
 513     util_unreference_framebuffer_state(&saved_fb);
 514
 515     pipe_surface_reference(&r300->locked_zbuffer, NULL);
 516 }
 517
 518 bool r300_is_blit_supported(enum pipe_format format)
 519 {
 520     const struct util_format_description *desc =
 521         util_format_description(format);
 522
 523     return desc->layout == UTIL_FORMAT_LAYOUT_PLAIN ||
 524            desc->layout == UTIL_FORMAT_LAYOUT_S3TC ||
 525            desc->layout == UTIL_FORMAT_LAYOUT_RGTC;
 526 }
 527
 528 /* Copy a block of pixels from one surface to another. */
 529 static void r300_resource_copy_region(struct pipe_context *pipe,
 530                                       struct pipe_resource *dst,
 531                                       unsigned dst_level,
 532                                       unsigned dstx, unsigned dsty, unsigned dstz,
 533                                       struct pipe_resource *src,
 534                                       unsigned src_level,
 535                                       const struct pipe_box *src_box)
 536 {
 537     struct pipe_screen *screen = pipe->screen;
 538     struct r300_context *r300 = r300_context(pipe);
 539     struct pipe_framebuffer_state *fb =
 540         (struct pipe_framebuffer_state*)r300->fb_state.state;
 541     unsigned src_width0 = r300_resource(src)->tex.width0;
 542     unsigned src_height0 = r300_resource(src)->tex.height0;
 543     unsigned dst_width0 = r300_resource(dst)->tex.width0;
 544     unsigned dst_height0 = r300_resource(dst)->tex.height0;
 545     unsigned layout;
 546     struct pipe_box box, dstbox;
 547     struct pipe_sampler_view src_templ, *src_view;
 548     struct pipe_surface dst_templ, *dst_view;
 549
 550     /* Fallback for buffers. */
 551     if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) ||
 552         !r300_is_blit_supported(dst->format)) {
 553         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 554                                   src, src_level, src_box);
 555         return;
 556     }
 557
 558     /* Can't read MSAA textures. */
 559     if (src->nr_samples > 1 || dst->nr_samples > 1) {
 560         return;
 561     }
 562
 563     /* The code below changes the texture format so that the copy can be done
 564      * on hardware. E.g. depth-stencil surfaces are copied as RGBA
 565      * colorbuffers. */
 566
 567     util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
 568     util_blitter_default_src_texture(&src_templ, src, src_level);
 569
 570     layout = util_format_description(dst_templ.format)->layout;
 571
 572     /* Handle non-renderable plain formats. */
 573     if (layout == UTIL_FORMAT_LAYOUT_PLAIN &&
 574         (!screen->is_format_supported(screen, src_templ.format, src->target,
 575                                       src->nr_samples,
 576                                       PIPE_BIND_SAMPLER_VIEW) ||
 577          !screen->is_format_supported(screen, dst_templ.format, dst->target,
 578                                       dst->nr_samples,
 579                                       PIPE_BIND_RENDER_TARGET))) {
 580         switch (util_format_get_blocksize(dst_templ.format)) {
 581             case 1:
 582                 dst_templ.format = PIPE_FORMAT_I8_UNORM;
 583                 break;
 584             case 2:
 585                 dst_templ.format = PIPE_FORMAT_B4G4R4A4_UNORM;
 586                 break;
 587             case 4:
 588                 dst_templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 589                 break;
 590             case 8:
 591                 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UNORM;
 592                 break;
 593             default:
 594                 debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
 595                              "r300: copy_region: Software fallback doesn't work for tiled textures.\n",
 596                              util_format_short_name(dst_templ.format));
 597         }
 598         src_templ.format = dst_templ.format;
 599     }
 600
 601     /* Handle compressed formats. */
 602     if (layout == UTIL_FORMAT_LAYOUT_S3TC ||
 603         layout == UTIL_FORMAT_LAYOUT_RGTC) {
 604         assert(src_templ.format == dst_templ.format);
 605
 606         box = *src_box;
 607         src_box = &box;
 608
 609         dst_width0 = align(dst_width0, 4);
 610         dst_height0 = align(dst_height0, 4);
 611         src_width0 = align(src_width0, 4);
 612         src_height0 = align(src_height0, 4);
 613         box.width = align(box.width, 4);
 614         box.height = align(box.height, 4);
 615
 616         switch (util_format_get_blocksize(dst_templ.format)) {
 617         case 8:
 618             /* one 4x4 pixel block has 8 bytes.
 619              * we set 1 pixel = 4 bytes ===> 1 block corrensponds to 2 pixels. */
 620             dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 621             dst_width0 = dst_width0 / 2;
 622             src_width0 = src_width0 / 2;
 623             dstx /= 2;
 624             box.x /= 2;
 625             box.width /= 2;
 626             break;
 627         case 16:
 628             /* one 4x4 pixel block has 16 bytes.
 629              * we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
 630             dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
 631             break;
 632         }
 633         src_templ.format = dst_templ.format;
 634
 635         dst_height0 = dst_height0 / 4;
 636         src_height0 = src_height0 / 4;
 637         dsty /= 4;
 638         box.y /= 4;
 639         box.height /= 4;
 640     }
 641
 642     /* Fallback for textures. */
 643     if (!screen->is_format_supported(screen, dst_templ.format,
 644                                      dst->target, dst->nr_samples,
 645                                      PIPE_BIND_RENDER_TARGET) ||
 646         !screen->is_format_supported(screen, src_templ.format,
 647                                      src->target, src->nr_samples,
 648                                      PIPE_BIND_SAMPLER_VIEW)) {
 649         assert(0 && "this shouldn't happen, update r300_is_blit_supported");
 650         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 651                                   src, src_level, src_box);
 652         return;
 653     }
 654
 655     /* Decompress ZMASK. */
 656     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 657         if (fb->zsbuf->texture == src ||
 658             fb->zsbuf->texture == dst) {
 659             r300_decompress_zmask(r300);
 660         }
 661     }
 662
 663     dst_view = r300_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
 664     src_view = r300_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
 665
 666     u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
 667              abs(src_box->depth), &dstbox);
 668
 669     r300_blitter_begin(r300, R300_COPY);
 670     util_blitter_blit_generic(r300->blitter, dst_view, &dstbox,
 671                               src_view, src_box, src_width0, src_height0,
 672                               PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
 673                               FALSE);
 674     r300_blitter_end(r300);
 675
 676     pipe_surface_reference(&dst_view, NULL);
 677     pipe_sampler_view_reference(&src_view, NULL);
 678 }
 679
 680 static boolean r300_is_simple_msaa_resolve(const struct pipe_blit_info *info)
 681 {
 682     unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
 683     unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
 684
 685     return info->dst.resource->format == info->src.resource->format &&
 686            info->dst.resource->format == info->dst.format &&
 687            info->src.resource->format == info->src.format &&
 688            !info->scissor_enable &&
 689            info->mask == PIPE_MASK_RGBA &&
 690            dst_width == info->src.resource->width0 &&
 691            dst_height == info->src.resource->height0 &&
 692            info->dst.box.x == 0 &&
 693            info->dst.box.y == 0 &&
 694            info->dst.box.width == dst_width &&
 695            info->dst.box.height == dst_height &&
 696            info->src.box.x == 0 &&
 697            info->src.box.y == 0 &&
 698            info->src.box.width == dst_width &&
 699            info->src.box.height == dst_height &&
 700            (r300_resource(info->dst.resource)->tex.microtile != RADEON_LAYOUT_LINEAR ||
 701             r300_resource(info->dst.resource)->tex.macrotile[info->dst.level] != RADEON_LAYOUT_LINEAR);
 702 }
 703
 704 static void r300_simple_msaa_resolve(struct pipe_context *pipe,
 705                                      struct pipe_resource *dst,
 706                                      unsigned dst_level,
 707                                      unsigned dst_layer,
 708                                      struct pipe_resource *src,
 709                                      enum pipe_format format)
 710 {
 711     struct r300_context *r300 = r300_context(pipe);
 712     struct r300_surface *srcsurf, *dstsurf;
 713     struct pipe_surface surf_tmpl;
 714     struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
 715
 716     memset(&surf_tmpl, 0, sizeof(surf_tmpl));
 717     surf_tmpl.format = format;
 718     srcsurf = r300_surface(pipe->create_surface(pipe, src, &surf_tmpl));
 719
 720     surf_tmpl.format = format;
 721     surf_tmpl.u.tex.level = dst_level;
 722     surf_tmpl.u.tex.first_layer =
 723     surf_tmpl.u.tex.last_layer = dst_layer;
 724     dstsurf = r300_surface(pipe->create_surface(pipe, dst, &surf_tmpl));
 725
 726     /* COLORPITCH should contain the tiling info of the resolve buffer.
 727      * The tiling of the AA buffer isn't programmable anyway. */
 728     srcsurf->pitch &= ~(R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
 729     srcsurf->pitch |= dstsurf->pitch & (R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
 730
 731     /* Enable AA resolve. */
 732     aa->dest = dstsurf;
 733     r300->aa_state.size = 8;
 734     r300_mark_atom_dirty(r300, &r300->aa_state);
 735
 736     /* Resolve the surface. */
 737     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 738     util_blitter_custom_color(r300->blitter, &srcsurf->base, NULL);
 739     r300_blitter_end(r300);
 740
 741     /* Disable AA resolve. */
 742     aa->dest = NULL;
 743     r300->aa_state.size = 4;
 744     r300_mark_atom_dirty(r300, &r300->aa_state);
 745
 746     pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
 747     pipe_surface_reference((struct pipe_surface**)&dstsurf, NULL);
 748 }
 749
 750 static void r300_msaa_resolve(struct pipe_context *pipe,
 751                               const struct pipe_blit_info *info)
 752 {
 753     struct r300_context *r300 = r300_context(pipe);
 754     struct pipe_screen *screen = pipe->screen;
 755     struct pipe_resource *tmp, templ;
 756     struct pipe_blit_info blit;
 757
 758     assert(info->src.level == 0);
 759     assert(info->src.box.z == 0);
 760     assert(info->src.box.depth == 1);
 761     assert(info->dst.box.depth == 1);
 762
 763     if (r300_is_simple_msaa_resolve(info)) {
 764         r300_simple_msaa_resolve(pipe, info->dst.resource, info->dst.level,
 765                                  info->dst.box.z, info->src.resource,
 766                                  info->src.format);
 767         return;
 768     }
 769
 770     /* resolve into a temporary texture, then blit */
 771     memset(&templ, 0, sizeof(templ));
 772     templ.target = PIPE_TEXTURE_2D;
 773     templ.format = info->src.resource->format;
 774     templ.width0 = info->src.resource->width0;
 775     templ.height0 = info->src.resource->height0;
 776     templ.depth0 = 1;
 777     templ.array_size = 1;
 778     templ.usage = PIPE_USAGE_STATIC;
 779     templ.flags = R300_RESOURCE_FORCE_MICROTILING;
 780
 781     tmp = screen->resource_create(screen, &templ);
 782
 783     /* resolve */
 784     r300_simple_msaa_resolve(pipe, tmp, 0, 0, info->src.resource,
 785                              info->src.format);
 786
 787     /* blit */
 788     blit = *info;
 789     blit.src.resource = tmp;
 790     blit.src.box.z = 0;
 791
 792     r300_blitter_begin(r300, R300_BLIT);
 793     util_blitter_blit(r300->blitter, &blit);
 794     r300_blitter_end(r300);
 795
 796     pipe_resource_reference(&tmp, NULL);
 797 }
 798
 799 static void r300_blit(struct pipe_context *pipe,
 800                       const struct pipe_blit_info *blit)
 801 {
 802     struct r300_context *r300 = r300_context(pipe);
 803     struct pipe_framebuffer_state *fb =
 804         (struct pipe_framebuffer_state*)r300->fb_state.state;
 805     struct pipe_blit_info info = *blit;
 806
 807     /* MSAA resolve. */
 808     if (info.src.resource->nr_samples > 1 &&
 809         info.dst.resource->nr_samples <= 1 &&
 810         !util_format_is_depth_or_stencil(info.src.resource->format)) {
 811         r300_msaa_resolve(pipe, &info);
 812         return;
 813     }
 814
 815     /* Can't read MSAA textures. */
 816     if (info.src.resource->nr_samples > 1) {
 817         return;
 818     }
 819
 820     /* Blit a combined depth-stencil resource as color.
 821      * S8Z24 is the only supported stencil format. */
 822     if ((info.mask & PIPE_MASK_S) &&
 823         info.src.format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
 824         info.dst.format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
 825         if (info.dst.resource->nr_samples > 1) {
 826             /* Cannot do that with MSAA buffers. */
 827             info.mask &= ~PIPE_MASK_S;
 828             if (!(info.mask & PIPE_MASK_Z)) {
 829                 return;
 830             }
 831         } else {
 832             /* Single-sample buffer. */
 833             info.src.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 834             info.dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 835             if (info.mask & PIPE_MASK_Z) {
 836                 info.mask = PIPE_MASK_RGBA; /* depth+stencil */
 837             } else {
 838                 info.mask = PIPE_MASK_B; /* stencil only */
 839             }
 840         }
 841     }
 842
 843     /* Decompress ZMASK. */
 844     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 845         if (fb->zsbuf->texture == info.src.resource ||
 846             fb->zsbuf->texture == info.dst.resource) {
 847             r300_decompress_zmask(r300);
 848         }
 849     }
 850
 851     r300_blitter_begin(r300, R300_BLIT);
 852     util_blitter_blit(r300->blitter, &info);
 853     r300_blitter_end(r300);
 854 }
 855
 856 void r300_init_blit_functions(struct r300_context *r300)
 857 {
 858     r300->context.clear = r300_clear;
 859     r300->context.clear_render_target = r300_clear_render_target;
 860     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
 861     r300->context.resource_copy_region = r300_resource_copy_region;
 862     r300->context.blit = r300_blit;
 863 }