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