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_winsys.h"
  27
  28 #include "util/u_format.h"
  29 #include "util/u_pack_color.h"
  30
  31 enum r300_blitter_op /* bitmask */
  32 {
  33     R300_CLEAR         = 1,
  34     R300_CLEAR_SURFACE = 2,
  35     R300_COPY          = 4
  36 };
  37
  38 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
  39 {
  40     if (r300->query_current) {
  41         r300->blitter_saved_query = r300->query_current;
  42         r300_stop_query(r300);
  43     }
  44
  45     /* Yeah we have to save all those states to ensure the blitter operation
  46      * is really transparent. The states will be restored by the blitter once
  47      * copying is done. */
  48     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
  49     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
  50     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
  51     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
  52     util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
  53     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
  54     util_blitter_save_viewport(r300->blitter, &r300->viewport);
  55     util_blitter_save_clip(r300->blitter, (struct pipe_clip_state*)r300->clip_state.state);
  56     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
  57     util_blitter_save_vertex_buffers(r300->blitter, r300->vbuf_mgr->nr_vertex_buffers,
  58                                      r300->vbuf_mgr->vertex_buffer);
  59
  60     if (op & (R300_CLEAR_SURFACE | R300_COPY)) {
  61         util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
  62     }
  63
  64     if (op & R300_COPY) {
  65         struct r300_textures_state* state =
  66             (struct r300_textures_state*)r300->textures_state.state;
  67
  68         util_blitter_save_fragment_sampler_states(
  69             r300->blitter, state->sampler_state_count,
  70             (void**)state->sampler_states);
  71
  72         util_blitter_save_fragment_sampler_views(
  73             r300->blitter, state->sampler_view_count,
  74             (struct pipe_sampler_view**)state->sampler_views);
  75     }
  76 }
  77
  78 static void r300_blitter_end(struct r300_context *r300)
  79 {
  80     if (r300->blitter_saved_query) {
  81         r300_resume_query(r300, r300->blitter_saved_query);
  82         r300->blitter_saved_query = NULL;
  83     }
  84 }
  85
  86 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
  87                                           const float* rgba)
  88 {
  89     union util_color uc;
  90     util_pack_color(rgba, format, &uc);
  91
  92     if (util_format_get_blocksizebits(format) == 32)
  93         return uc.ui;
  94     else
  95         return uc.us | (uc.us << 16);
  96 }
  97
  98 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
  99                                        unsigned clear_buffers)
 100 {
 101     struct pipe_framebuffer_state *fb =
 102         (struct pipe_framebuffer_state*)r300->fb_state.state;
 103
 104     /* Only color clear allowed, and only one colorbuffer. */
 105     if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
 106         return FALSE;
 107
 108     return r300_surface(fb->cbufs[0])->cbzb_allowed;
 109 }
 110
 111 static boolean r300_fast_zclear_allowed(struct r300_context *r300)
 112 {
 113     struct pipe_framebuffer_state *fb =
 114         (struct pipe_framebuffer_state*)r300->fb_state.state;
 115
 116     return r300_resource(fb->zsbuf->texture)->tex.zmask_dwords[fb->zsbuf->u.tex.level];
 117 }
 118
 119 static boolean r300_hiz_clear_allowed(struct r300_context *r300)
 120 {
 121     struct pipe_framebuffer_state *fb =
 122         (struct pipe_framebuffer_state*)r300->fb_state.state;
 123
 124     return r300_resource(fb->zsbuf->texture)->tex.hiz_dwords[fb->zsbuf->u.tex.level];
 125 }
 126
 127 static uint32_t r300_depth_clear_value(enum pipe_format format,
 128                                        double depth, unsigned stencil)
 129 {
 130     switch (format) {
 131         case PIPE_FORMAT_Z16_UNORM:
 132         case PIPE_FORMAT_X8Z24_UNORM:
 133             return util_pack_z(format, depth);
 134
 135         case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
 136             return util_pack_z_stencil(format, depth, stencil);
 137
 138         default:
 139             assert(0);
 140             return 0;
 141     }
 142 }
 143
 144 /* Clear currently bound buffers. */
 145 static void r300_clear(struct pipe_context* pipe,
 146                        unsigned buffers,
 147                        const float* rgba,
 148                        double depth,
 149                        unsigned stencil)
 150 {
 151     /* My notes about Zbuffer compression:
 152      *
 153      * 1) The zbuffer must be micro-tiled and whole microtiles must be
 154      *    written if compression is enabled. If microtiling is disabled,
 155      *    it locks up.
 156      *
 157      * 2) There is ZMASK RAM which contains a compressed zbuffer.
 158      *    Each dword of the Z Mask contains compression information
 159      *    for 16 4x4 pixel tiles, that is 2 bits for each tile.
 160      *    On chips with 2 Z pipes, every other dword maps to a different
 161      *    pipe. On newer chipsets, there is a new compression mode
 162      *    with 8x8 pixel tiles per 2 bits.
 163      *
 164      * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
 165      *    it should look in the ZMASK RAM first before fetching from a real
 166      *    zbuffer.
 167      *
 168      * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
 169      *    during zbuffer reads instead of the value that is actually stored
 170      *    in the zbuffer memory. A pixel is in a cleared state when its ZMASK
 171      *    is equal to 0. Therefore, if you clear ZMASK with zeros, you may
 172      *    leave the zbuffer memory uninitialized, but then you must enable
 173      *    compression, so that the ZMASK RAM is actually used.
 174      *
 175      * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
 176      *    during zbuffer updates. A special decompressing operation should be
 177      *    used to fully decompress a zbuffer, which basically just stores all
 178      *    compressed tiles in ZMASK to the zbuffer memory.
 179      *
 180      * 6) For a 16-bit zbuffer, compression causes a hung with one or
 181      *    two samples and should not be used.
 182      *
 183      * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
 184      *    to avoid needless decompression.
 185      *
 186      * 8) Fastfill must not be used if reading of compressed Z data is disabled
 187      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
 188      *    i.e. it cannot be used to compress the zbuffer.
 189      *
 190      * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
 191      *
 192      * - Marek
 193      */
 194
 195     struct r300_context* r300 = r300_context(pipe);
 196     struct pipe_framebuffer_state *fb =
 197         (struct pipe_framebuffer_state*)r300->fb_state.state;
 198     struct r300_hyperz_state *hyperz =
 199         (struct r300_hyperz_state*)r300->hyperz_state.state;
 200     uint32_t width = fb->width;
 201     uint32_t height = fb->height;
 202     boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
 203     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
 204
 205     /* Enable fast Z clear.
 206      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
 207     if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && can_hyperz) {
 208         if (r300_fast_zclear_allowed(r300)) {
 209             hyperz_dcv = hyperz->zb_depthclearvalue =
 210                 r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
 211
 212             r300_mark_atom_dirty(r300, &r300->zmask_clear);
 213             buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 214         }
 215
 216         if (r300_hiz_clear_allowed(r300)) {
 217             r300_mark_atom_dirty(r300, &r300->hiz_clear);
 218         }
 219     }
 220
 221     /* Enable CBZB clear. */
 222     if (r300_cbzb_clear_allowed(r300, buffers)) {
 223         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 224
 225         hyperz->zb_depthclearvalue =
 226                 r300_depth_clear_cb_value(surf->base.format, rgba);
 227
 228         width = surf->cbzb_width;
 229         height = surf->cbzb_height;
 230
 231         r300->cbzb_clear = TRUE;
 232         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 233     }
 234
 235     /* Clear. */
 236     if (buffers) {
 237         /* Clear using the blitter. */
 238         r300_blitter_begin(r300, R300_CLEAR);
 239         util_blitter_clear(r300->blitter,
 240                            width,
 241                            height,
 242                            fb->nr_cbufs,
 243                            buffers, rgba, depth, stencil);
 244         r300_blitter_end(r300);
 245     } else if (r300->zmask_clear.dirty || r300->hiz_clear.dirty) {
 246         /* Just clear zmask and hiz now, this does not use the standard draw
 247          * procedure. */
 248         unsigned dwords;
 249
 250         /* Calculate zmask_clear and hiz_clear atom sizes. */
 251         r300_update_hyperz_state(r300);
 252         dwords = (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
 253                  (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
 254                  r300_get_num_cs_end_dwords(r300);
 255
 256         /* Reserve CS space. */
 257         if (dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
 258             r300->context.flush(&r300->context, 0, NULL);
 259         }
 260
 261         /* Emit clear packets. */
 262         if (r300->zmask_clear.dirty) {
 263             r300_emit_zmask_clear(r300, r300->zmask_clear.size,
 264                                   r300->zmask_clear.state);
 265             r300->zmask_clear.dirty = FALSE;
 266         }
 267         if (r300->hiz_clear.dirty) {
 268             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
 269                                 r300->hiz_clear.state);
 270             r300->hiz_clear.dirty = FALSE;
 271         }
 272     } else {
 273         assert(0);
 274     }
 275
 276     /* Disable CBZB clear. */
 277     if (r300->cbzb_clear) {
 278         r300->cbzb_clear = FALSE;
 279         hyperz->zb_depthclearvalue = hyperz_dcv;
 280         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 281     }
 282
 283     /* Enable fastfill and/or hiz.
 284      *
 285      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
 286      * looks if zmask/hiz is in use and programs hardware accordingly. */
 287     if (r300->zmask_in_use || r300->hiz_in_use) {
 288         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 289     }
 290 }
 291
 292 /* Clear a region of a color surface to a constant value. */
 293 static void r300_clear_render_target(struct pipe_context *pipe,
 294                                      struct pipe_surface *dst,
 295                                      const float *rgba,
 296                                      unsigned dstx, unsigned dsty,
 297                                      unsigned width, unsigned height)
 298 {
 299     struct r300_context *r300 = r300_context(pipe);
 300
 301     r300->hyperz_locked = TRUE;
 302     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 303
 304     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 305     util_blitter_clear_render_target(r300->blitter, dst, rgba,
 306                                      dstx, dsty, width, height);
 307     r300_blitter_end(r300);
 308
 309     r300->hyperz_locked = FALSE;
 310     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 311 }
 312
 313 /* Clear a region of a depth stencil surface. */
 314 static void r300_clear_depth_stencil(struct pipe_context *pipe,
 315                                      struct pipe_surface *dst,
 316                                      unsigned clear_flags,
 317                                      double depth,
 318                                      unsigned stencil,
 319                                      unsigned dstx, unsigned dsty,
 320                                      unsigned width, unsigned height)
 321 {
 322     struct r300_context *r300 = r300_context(pipe);
 323     struct pipe_framebuffer_state *fb =
 324         (struct pipe_framebuffer_state*)r300->fb_state.state;
 325
 326     if (r300->zmask_in_use && !r300->hyperz_locked) {
 327         if (fb->zsbuf->texture == dst->texture) {
 328             r300_decompress_zmask(r300);
 329         } else {
 330             r300->hyperz_locked = TRUE;
 331             r300_mark_atom_dirty(r300, &r300->hyperz_state);
 332         }
 333     }
 334
 335     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 336     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
 337                                      dstx, dsty, width, height);
 338     r300_blitter_end(r300);
 339
 340     if (r300->hyperz_locked) {
 341         r300->hyperz_locked = FALSE;
 342         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 343     }
 344 }
 345
 346 void r300_decompress_zmask(struct r300_context *r300)
 347 {
 348     struct pipe_framebuffer_state *fb =
 349         (struct pipe_framebuffer_state*)r300->fb_state.state;
 350
 351     if (!r300->zmask_in_use || r300->hyperz_locked)
 352         return;
 353
 354     r300->zmask_decompress = TRUE;
 355     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 356
 357     r300_blitter_begin(r300, R300_CLEAR);
 358     util_blitter_clear_depth_custom(r300->blitter, fb->width, fb->height, 0,
 359                                     r300->dsa_decompress_zmask);
 360     r300_blitter_end(r300);
 361
 362     r300->zmask_decompress = FALSE;
 363     r300->zmask_in_use = FALSE;
 364     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 365 }
 366
 367 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
 368 {
 369     struct pipe_framebuffer_state fb = {0};
 370     fb.width = r300->locked_zbuffer->width;
 371     fb.height = r300->locked_zbuffer->height;
 372     fb.nr_cbufs = 0;
 373     fb.zsbuf = r300->locked_zbuffer;
 374
 375     r300->context.set_framebuffer_state(&r300->context, &fb);
 376     r300_decompress_zmask(r300);
 377 }
 378
 379 void r300_decompress_zmask_locked(struct r300_context *r300)
 380 {
 381     struct pipe_framebuffer_state saved_fb = {0};
 382
 383     util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
 384     r300_decompress_zmask_locked_unsafe(r300);
 385     r300->context.set_framebuffer_state(&r300->context, &saved_fb);
 386     util_unreference_framebuffer_state(&saved_fb);
 387 }
 388
 389 /* Copy a block of pixels from one surface to another using HW. */
 390 static void r300_hw_copy_region(struct pipe_context* pipe,
 391                                 struct pipe_resource *dst,
 392                                 unsigned dst_level,
 393                                 unsigned dstx, unsigned dsty, unsigned dstz,
 394                                 struct pipe_resource *src,
 395                                 unsigned src_level,
 396                                 const struct pipe_box *src_box)
 397 {
 398     struct r300_context* r300 = r300_context(pipe);
 399
 400     r300_blitter_begin(r300, R300_COPY);
 401     util_blitter_copy_region(r300->blitter, dst, dst_level, dstx, dsty, dstz,
 402                              src, src_level, src_box, TRUE);
 403     r300_blitter_end(r300);
 404 }
 405
 406 /* Copy a block of pixels from one surface to another. */
 407 static void r300_resource_copy_region(struct pipe_context *pipe,
 408                                       struct pipe_resource *dst,
 409                                       unsigned dst_level,
 410                                       unsigned dstx, unsigned dsty, unsigned dstz,
 411                                       struct pipe_resource *src,
 412                                       unsigned src_level,
 413                                       const struct pipe_box *src_box)
 414 {
 415     struct r300_context *r300 = r300_context(pipe);
 416     struct pipe_framebuffer_state *fb =
 417         (struct pipe_framebuffer_state*)r300->fb_state.state;
 418     struct pipe_resource old_src = *src;
 419     struct pipe_resource old_dst = *dst;
 420     struct pipe_resource new_src = old_src;
 421     struct pipe_resource new_dst = old_dst;
 422     const struct util_format_description *desc =
 423             util_format_description(dst->format);
 424     struct pipe_box box;
 425
 426     if (r300->zmask_in_use && !r300->hyperz_locked) {
 427         if (fb->zsbuf->texture == src ||
 428             fb->zsbuf->texture == dst) {
 429             r300_decompress_zmask(r300);
 430         } else {
 431             r300->hyperz_locked = TRUE;
 432             r300_mark_atom_dirty(r300, &r300->hyperz_state);
 433         }
 434     }
 435
 436     /* Handle non-renderable plain formats. */
 437     if (desc->layout == UTIL_FORMAT_LAYOUT_PLAIN &&
 438         (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB ||
 439          !pipe->screen->is_format_supported(pipe->screen,
 440                                             src->format, src->target,
 441                                             src->nr_samples,
 442                                             PIPE_BIND_SAMPLER_VIEW, 0) ||
 443          !pipe->screen->is_format_supported(pipe->screen,
 444                                             dst->format, dst->target,
 445                                             dst->nr_samples,
 446                                             PIPE_BIND_RENDER_TARGET, 0))) {
 447         switch (util_format_get_blocksize(old_dst.format)) {
 448             case 1:
 449                 new_dst.format = PIPE_FORMAT_I8_UNORM;
 450                 break;
 451             case 2:
 452                 new_dst.format = PIPE_FORMAT_B4G4R4A4_UNORM;
 453                 break;
 454             case 4:
 455                 new_dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 456                 break;
 457             case 8:
 458                 new_dst.format = PIPE_FORMAT_R16G16B16A16_UNORM;
 459                 break;
 460             default:
 461                 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
 462                              "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
 463                              util_format_short_name(dst->format));
 464         }
 465         new_src.format = new_dst.format;
 466     }
 467
 468     /* Handle compressed formats. */
 469     if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
 470         switch (util_format_get_blocksize(old_dst.format)) {
 471         case 8:
 472             /* 1 pixel = 4 bits,
 473              * we set 1 pixel = 2 bytes ===> 4 times larger pixels. */
 474             new_dst.format = PIPE_FORMAT_B4G4R4A4_UNORM;
 475             break;
 476         case 16:
 477             /* 1 pixel = 8 bits,
 478              * we set 1 pixel = 4 bytes ===> 4 times larger pixels. */
 479             new_dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 480             break;
 481         }
 482
 483         /* Since the pixels are 4 times larger, we must decrease
 484          * the image size and the coordinates 4 times. */
 485         new_src.format = new_dst.format;
 486         new_dst.height0 = (new_dst.height0 + 3) / 4;
 487         new_src.height0 = (new_src.height0 + 3) / 4;
 488         dsty /= 4;
 489         box = *src_box;
 490         box.y /= 4;
 491         box.height = (box.height + 3) / 4;
 492         src_box = &box;
 493     }
 494
 495     if (old_src.format != new_src.format)
 496         r300_resource_set_properties(pipe->screen, src, 0, &new_src);
 497     if (old_dst.format != new_dst.format)
 498         r300_resource_set_properties(pipe->screen, dst, 0, &new_dst);
 499
 500     r300_hw_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 501                         src, src_level, src_box);
 502
 503     if (old_src.format != new_src.format)
 504         r300_resource_set_properties(pipe->screen, src, 0, &old_src);
 505     if (old_dst.format != new_dst.format)
 506         r300_resource_set_properties(pipe->screen, dst, 0, &old_dst);
 507
 508     if (r300->hyperz_locked) {
 509         r300->hyperz_locked = FALSE;
 510         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 511     }
 512 }
 513
 514 void r300_init_blit_functions(struct r300_context *r300)
 515 {
 516     r300->context.clear = r300_clear;
 517     r300->context.clear_render_target = r300_clear_render_target;
 518     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
 519     r300->context.resource_copy_region = r300_resource_copy_region;
 520 }