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_hyperz.h"
  26 #include "r300_texture.h"
  27 #include "r300_winsys.h"
  28
  29 #include "util/u_format.h"
  30 #include "util/u_pack_color.h"
  31
  32 enum r300_blitter_op /* bitmask */
  33 {
  34     R300_CLEAR         = 1,
  35     R300_CLEAR_SURFACE = 2,
  36     R300_COPY          = 4
  37 };
  38
  39 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
  40 {
  41     if (r300->query_current) {
  42         r300->blitter_saved_query = r300->query_current;
  43         r300_stop_query(r300);
  44     }
  45
  46     /* Yeah we have to save all those states to ensure the blitter operation
  47      * is really transparent. The states will be restored by the blitter once
  48      * copying is done. */
  49     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
  50     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
  51     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
  52     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
  53     util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
  54     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
  55     util_blitter_save_viewport(r300->blitter, &r300->viewport);
  56     util_blitter_save_clip(r300->blitter, (struct pipe_clip_state*)r300->clip_state.state);
  57     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
  58     util_blitter_save_vertex_buffers(r300->blitter, r300->vertex_buffer_count,
  59                                      r300->vertex_buffer);
  60
  61     if (op & (R300_CLEAR_SURFACE | R300_COPY)) {
  62         util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
  63     }
  64
  65     if (op & R300_COPY) {
  66         struct r300_textures_state* state =
  67             (struct r300_textures_state*)r300->textures_state.state;
  68
  69         util_blitter_save_fragment_sampler_states(
  70             r300->blitter, state->sampler_state_count,
  71             (void**)state->sampler_states);
  72
  73         util_blitter_save_fragment_sampler_views(
  74             r300->blitter, state->sampler_view_count,
  75             (struct pipe_sampler_view**)state->sampler_views);
  76     }
  77 }
  78
  79 static void r300_blitter_end(struct r300_context *r300)
  80 {
  81     if (r300->blitter_saved_query) {
  82         r300_resume_query(r300, r300->blitter_saved_query);
  83         r300->blitter_saved_query = NULL;
  84     }
  85 }
  86
  87 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
  88                                           const float* rgba)
  89 {
  90     union util_color uc;
  91     util_pack_color(rgba, format, &uc);
  92
  93     if (util_format_get_blocksizebits(format) == 32)
  94         return uc.ui;
  95     else
  96         return uc.us | (uc.us << 16);
  97 }
  98
  99 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
 100                                        unsigned clear_buffers)
 101 {
 102     struct pipe_framebuffer_state *fb =
 103         (struct pipe_framebuffer_state*)r300->fb_state.state;
 104
 105     /* Only color clear allowed, and only one colorbuffer. */
 106     if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
 107         return FALSE;
 108
 109     return r300_surface(fb->cbufs[0])->cbzb_allowed;
 110 }
 111
 112 static boolean r300_fast_zclear_allowed(struct r300_context *r300)
 113 {
 114     struct pipe_framebuffer_state *fb =
 115         (struct pipe_framebuffer_state*)r300->fb_state.state;
 116
 117     return r300_texture(fb->zsbuf->texture)->desc.zmask_dwords[fb->zsbuf->u.tex.level];
 118 }
 119
 120 static uint32_t r300_depth_clear_value(enum pipe_format format,
 121                                        double depth, unsigned stencil)
 122 {
 123     switch (format) {
 124         case PIPE_FORMAT_Z16_UNORM:
 125         case PIPE_FORMAT_X8Z24_UNORM:
 126             return util_pack_z(format, depth);
 127
 128         case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
 129             return util_pack_z_stencil(format, depth, stencil);
 130
 131         default:
 132             assert(0);
 133             return 0;
 134     }
 135 }
 136
 137 /* Clear currently bound buffers. */
 138 static void r300_clear(struct pipe_context* pipe,
 139                        unsigned buffers,
 140                        const float* rgba,
 141                        double depth,
 142                        unsigned stencil)
 143 {
 144     /* My notes about Zbuffer compression:
 145      *
 146      * 1) The zbuffer must be micro-tiled and whole microtiles must be
 147      *    written if compression is enabled. If microtiling is disabled,
 148      *    it locks up.
 149      *
 150      * 2) There is ZMASK RAM which contains a compressed zbuffer.
 151      *    Each dword of the Z Mask contains compression information
 152      *    for 16 4x4 pixel tiles, that is 2 bits for each tile.
 153      *    On chips with 2 Z pipes, every other dword maps to a different
 154      *    pipe. On newer chipsets, there is a new compression mode
 155      *    with 8x8 pixel tiles per 2 bits.
 156      *
 157      * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
 158      *    it should look in the ZMASK RAM first before fetching from a real
 159      *    zbuffer.
 160      *
 161      * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
 162      *    during zbuffer reads instead of the value that is actually stored
 163      *    in the zbuffer memory. A pixel is in a cleared state when its ZMASK
 164      *    is equal to 0. Therefore, if you clear ZMASK with zeros, you may
 165      *    leave the zbuffer memory uninitialized, but then you must enable
 166      *    compression, so that the ZMASK RAM is actually used.
 167      *
 168      * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
 169      *    during zbuffer updates. A special decompressing operation should be
 170      *    used to fully decompress a zbuffer, which basically just stores all
 171      *    compressed tiles in ZMASK to the zbuffer memory.
 172      *
 173      * 6) For a 16-bit zbuffer, compression causes a hung with one or
 174      *    two samples and should not be used.
 175      *
 176      * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
 177      *    to avoid needless decompression.
 178      *
 179      * 8) Fastfill must not be used if reading of compressed Z data is disabled
 180      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
 181      *    i.e. it cannot be used to compress the zbuffer.
 182      *
 183      * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
 184      *
 185      * - Marek
 186      */
 187
 188     struct r300_context* r300 = r300_context(pipe);
 189     struct pipe_framebuffer_state *fb =
 190         (struct pipe_framebuffer_state*)r300->fb_state.state;
 191     struct r300_hyperz_state *hyperz =
 192         (struct r300_hyperz_state*)r300->hyperz_state.state;
 193     struct r300_texture *zstex =
 194             fb->zsbuf ? r300_texture(fb->zsbuf->texture) : NULL;
 195     uint32_t width = fb->width;
 196     uint32_t height = fb->height;
 197     boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
 198     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
 199
 200     /* Enable fast Z clear.
 201      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
 202     if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && can_hyperz) {
 203         hyperz_dcv = hyperz->zb_depthclearvalue =
 204             r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
 205
 206         if (r300_fast_zclear_allowed(r300)) {
 207             r300_mark_atom_dirty(r300, &r300->zmask_clear);
 208             buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 209         }
 210
 211         if (zstex->hiz_mem[fb->zsbuf->u.tex.level])
 212             r300_mark_atom_dirty(r300, &r300->hiz_clear);
 213
 214         /* XXX Change this to r300_mark_atom_dirty(r300, &r300->hyperz_state);
 215          * once hiz offset is constant. */
 216         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 217     }
 218
 219     /* Enable CBZB clear. */
 220     if (r300_cbzb_clear_allowed(r300, buffers)) {
 221         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 222
 223         hyperz->zb_depthclearvalue =
 224                 r300_depth_clear_cb_value(surf->base.format, rgba);
 225
 226         width = surf->cbzb_width;
 227         height = surf->cbzb_height;
 228
 229         r300->cbzb_clear = TRUE;
 230         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 231     }
 232
 233     /* Clear. */
 234     if (buffers) {
 235         /* Clear using the blitter. */
 236         r300_blitter_begin(r300, R300_CLEAR);
 237         util_blitter_clear(r300->blitter,
 238                            width,
 239                            height,
 240                            fb->nr_cbufs,
 241                            buffers, rgba, depth, stencil);
 242         r300_blitter_end(r300);
 243     } else if (r300->zmask_clear.dirty) {
 244         /* Just clear zmask and hiz now, this does not use the standard draw
 245          * procedure. */
 246         unsigned dwords;
 247
 248         /* Calculate zmask_clear and hiz_clear atom sizes. */
 249         r300_update_hyperz_state(r300);
 250         dwords = r300->zmask_clear.size +
 251                  (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
 252                  r300_get_num_cs_end_dwords(r300);
 253
 254         /* Reserve CS space. */
 255         if (dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
 256             r300->context.flush(&r300->context, 0, NULL);
 257         }
 258
 259         /* Emit clear packets. */
 260         r300_emit_zmask_clear(r300, r300->zmask_clear.size,
 261                               r300->zmask_clear.state);
 262         r300->zmask_clear.dirty = FALSE;
 263         if (r300->hiz_clear.dirty) {
 264             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
 265                                 r300->hiz_clear.state);
 266             r300->hiz_clear.dirty = FALSE;
 267         }
 268     } else {
 269         assert(0);
 270     }
 271
 272     /* Disable CBZB clear. */
 273     if (r300->cbzb_clear) {
 274         r300->cbzb_clear = FALSE;
 275         hyperz->zb_depthclearvalue = hyperz_dcv;
 276         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 277     }
 278
 279     /* Enable fastfill and/or hiz.
 280      *
 281      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
 282      * looks if zmask/hiz is in use and enables fastfill accordingly. */
 283     if (r300->zmask_in_use ||
 284         (zstex && zstex->hiz_in_use[fb->zsbuf->u.tex.level])) {
 285         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 286     }
 287 }
 288
 289 /* Clear a region of a color surface to a constant value. */
 290 static void r300_clear_render_target(struct pipe_context *pipe,
 291                                      struct pipe_surface *dst,
 292                                      const float *rgba,
 293                                      unsigned dstx, unsigned dsty,
 294                                      unsigned width, unsigned height)
 295 {
 296     struct r300_context *r300 = r300_context(pipe);
 297
 298     r300->zmask_locked = TRUE;
 299     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 300
 301     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 302     util_blitter_clear_render_target(r300->blitter, dst, rgba,
 303                                      dstx, dsty, width, height);
 304     r300_blitter_end(r300);
 305
 306     r300->zmask_locked = FALSE;
 307     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 308 }
 309
 310 /* Clear a region of a depth stencil surface. */
 311 static void r300_clear_depth_stencil(struct pipe_context *pipe,
 312                                      struct pipe_surface *dst,
 313                                      unsigned clear_flags,
 314                                      double depth,
 315                                      unsigned stencil,
 316                                      unsigned dstx, unsigned dsty,
 317                                      unsigned width, unsigned height)
 318 {
 319     struct r300_context *r300 = r300_context(pipe);
 320     struct pipe_framebuffer_state *fb =
 321         (struct pipe_framebuffer_state*)r300->fb_state.state;
 322
 323     if (r300->zmask_in_use && !r300->zmask_locked) {
 324         if (fb->zsbuf->texture == dst->texture) {
 325             r300_decompress_zmask(r300);
 326         } else {
 327             r300->zmask_locked = TRUE;
 328             r300_mark_atom_dirty(r300, &r300->hyperz_state);
 329         }
 330     }
 331
 332     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 333     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
 334                                      dstx, dsty, width, height);
 335     r300_blitter_end(r300);
 336
 337     if (r300->zmask_locked) {
 338         r300->zmask_locked = FALSE;
 339         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 340     }
 341 }
 342
 343 void r300_decompress_zmask(struct r300_context *r300)
 344 {
 345     struct pipe_framebuffer_state *fb =
 346         (struct pipe_framebuffer_state*)r300->fb_state.state;
 347
 348     if (!r300->zmask_in_use || r300->zmask_locked)
 349         return;
 350
 351     r300->zmask_decompress = TRUE;
 352     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 353
 354     r300_blitter_begin(r300, R300_CLEAR);
 355     util_blitter_clear_depth_custom(r300->blitter, fb->width, fb->height, 0,
 356                                     r300->dsa_decompress_zmask);
 357     r300_blitter_end(r300);
 358
 359     r300->zmask_decompress = FALSE;
 360     r300->zmask_in_use = FALSE;
 361     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 362 }
 363
 364 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
 365 {
 366     struct pipe_framebuffer_state fb = {0};
 367     fb.width = r300->locked_zbuffer->width;
 368     fb.height = r300->locked_zbuffer->height;
 369     fb.nr_cbufs = 0;
 370     fb.zsbuf = r300->locked_zbuffer;
 371
 372     r300->context.set_framebuffer_state(&r300->context, &fb);
 373     r300_decompress_zmask(r300);
 374 }
 375
 376 void r300_decompress_zmask_locked(struct r300_context *r300)
 377 {
 378     struct pipe_framebuffer_state saved_fb = {0};
 379
 380     util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
 381     r300_decompress_zmask_locked_unsafe(r300);
 382     r300->context.set_framebuffer_state(&r300->context, &saved_fb);
 383     util_unreference_framebuffer_state(&saved_fb);
 384 }
 385
 386 /* Copy a block of pixels from one surface to another using HW. */
 387 static void r300_hw_copy_region(struct pipe_context* pipe,
 388                                 struct pipe_resource *dst,
 389                                 unsigned dst_level,
 390                                 unsigned dstx, unsigned dsty, unsigned dstz,
 391                                 struct pipe_resource *src,
 392                                 unsigned src_level,
 393                                 const struct pipe_box *src_box)
 394 {
 395     struct r300_context* r300 = r300_context(pipe);
 396
 397     r300_blitter_begin(r300, R300_COPY);
 398     util_blitter_copy_region(r300->blitter, dst, dst_level, dstx, dsty, dstz,
 399                              src, src_level, src_box, TRUE);
 400     r300_blitter_end(r300);
 401 }
 402
 403 /* Copy a block of pixels from one surface to another. */
 404 static void r300_resource_copy_region(struct pipe_context *pipe,
 405                                       struct pipe_resource *dst,
 406                                       unsigned dst_level,
 407                                       unsigned dstx, unsigned dsty, unsigned dstz,
 408                                       struct pipe_resource *src,
 409                                       unsigned src_level,
 410                                       const struct pipe_box *src_box)
 411 {
 412     struct r300_context *r300 = r300_context(pipe);
 413     struct pipe_framebuffer_state *fb =
 414         (struct pipe_framebuffer_state*)r300->fb_state.state;
 415     enum pipe_format old_format = dst->format;
 416     enum pipe_format new_format = old_format;
 417     const struct util_format_description *desc = util_format_description(old_format);
 418
 419     if (r300->zmask_in_use && !r300->zmask_locked) {
 420         if (fb->zsbuf->texture == src ||
 421             fb->zsbuf->texture == dst) {
 422             r300_decompress_zmask(r300);
 423         } else {
 424             r300->zmask_locked = TRUE;
 425             r300_mark_atom_dirty(r300, &r300->hyperz_state);
 426         }
 427     }
 428
 429     if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB ||
 430         (!pipe->screen->is_format_supported(pipe->screen,
 431                                             old_format, src->target,
 432                                             src->nr_samples,
 433                                             PIPE_BIND_RENDER_TARGET |
 434                                             PIPE_BIND_SAMPLER_VIEW, 0) &&
 435          desc->layout == UTIL_FORMAT_LAYOUT_PLAIN)) {
 436         switch (util_format_get_blocksize(old_format)) {
 437             case 1:
 438                 new_format = PIPE_FORMAT_I8_UNORM;
 439                 break;
 440             case 2:
 441                 new_format = PIPE_FORMAT_B4G4R4A4_UNORM;
 442                 break;
 443             case 4:
 444                 new_format = PIPE_FORMAT_B8G8R8A8_UNORM;
 445                 break;
 446             case 8:
 447                 new_format = PIPE_FORMAT_R16G16B16A16_UNORM;
 448                 break;
 449             default:
 450                 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
 451                              "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
 452                              util_format_short_name(old_format));
 453         }
 454     }
 455
 456     if (old_format != new_format) {
 457         r300_texture_reinterpret_format(pipe->screen,
 458                                         dst, new_format);
 459         r300_texture_reinterpret_format(pipe->screen,
 460                                         src, new_format);
 461     }
 462
 463     r300_hw_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 464                         src, src_level, src_box);
 465
 466     if (old_format != new_format) {
 467         r300_texture_reinterpret_format(pipe->screen,
 468                                         dst, old_format);
 469         r300_texture_reinterpret_format(pipe->screen,
 470                                         src, old_format);
 471     }
 472
 473     if (r300->zmask_locked) {
 474         r300->zmask_locked = FALSE;
 475         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 476     }
 477 }
 478
 479 void r300_init_blit_functions(struct r300_context *r300)
 480 {
 481     r300->context.clear = r300_clear;
 482     r300->context.clear_render_target = r300_clear_render_target;
 483     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
 484     r300->context.resource_copy_region = r300_resource_copy_region;
 485 }