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
  27 #include "util/u_format.h"
  28 #include "util/u_pack_color.h"
  29 #include "util/u_surface.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] != 0;
 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] != 0;
 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 static uint32_t r300_hiz_clear_value(double depth)
 145 {
 146     uint32_t r = (uint32_t)(CLAMP(depth, 0, 1) * 255.5);
 147     assert(r <= 255);
 148     return r | (r << 8) | (r << 16) | (r << 24);
 149 }
 150
 151 /* Clear currently bound buffers. */
 152 static void r300_clear(struct pipe_context* pipe,
 153                        unsigned buffers,
 154                        const float* rgba,
 155                        double depth,
 156                        unsigned stencil)
 157 {
 158     /* My notes about Zbuffer compression:
 159      *
 160      * 1) The zbuffer must be micro-tiled and whole microtiles must be
 161      *    written if compression is enabled. If microtiling is disabled,
 162      *    it locks up.
 163      *
 164      * 2) There is ZMASK RAM which contains a compressed zbuffer.
 165      *    Each dword of the Z Mask contains compression information
 166      *    for 16 4x4 pixel tiles, that is 2 bits for each tile.
 167      *    On chips with 2 Z pipes, every other dword maps to a different
 168      *    pipe. On newer chipsets, there is a new compression mode
 169      *    with 8x8 pixel tiles per 2 bits.
 170      *
 171      * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
 172      *    it should look in the ZMASK RAM first before fetching from a real
 173      *    zbuffer.
 174      *
 175      * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
 176      *    during zbuffer reads instead of the value that is actually stored
 177      *    in the zbuffer memory. A pixel is in a cleared state when its ZMASK
 178      *    is equal to 0. Therefore, if you clear ZMASK with zeros, you may
 179      *    leave the zbuffer memory uninitialized, but then you must enable
 180      *    compression, so that the ZMASK RAM is actually used.
 181      *
 182      * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
 183      *    during zbuffer updates. A special decompressing operation should be
 184      *    used to fully decompress a zbuffer, which basically just stores all
 185      *    compressed tiles in ZMASK to the zbuffer memory.
 186      *
 187      * 6) For a 16-bit zbuffer, compression causes a hung with one or
 188      *    two samples and should not be used.
 189      *
 190      * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
 191      *    to avoid needless decompression.
 192      *
 193      * 8) Fastfill must not be used if reading of compressed Z data is disabled
 194      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
 195      *    i.e. it cannot be used to compress the zbuffer.
 196      *
 197      * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
 198      *
 199      * - Marek
 200      */
 201
 202     struct r300_context* r300 = r300_context(pipe);
 203     struct pipe_framebuffer_state *fb =
 204         (struct pipe_framebuffer_state*)r300->fb_state.state;
 205     struct r300_hyperz_state *hyperz =
 206         (struct r300_hyperz_state*)r300->hyperz_state.state;
 207     uint32_t width = fb->width;
 208     uint32_t height = fb->height;
 209     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
 210
 211     /* Enable fast Z clear.
 212      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
 213     if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
 214         boolean zmask_clear, hiz_clear;
 215
 216         zmask_clear = r300_fast_zclear_allowed(r300);
 217         hiz_clear = r300_hiz_clear_allowed(r300);
 218
 219         /* If we need Hyper-Z. */
 220         if (zmask_clear || hiz_clear) {
 221             r300->num_z_clears++;
 222
 223             /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
 224             if (!r300->hyperz_enabled) {
 225                 r300->hyperz_enabled =
 226                     r300->rws->cs_request_feature(r300->cs,
 227                                                 RADEON_FID_HYPERZ_RAM_ACCESS,
 228                                                 TRUE);
 229                 if (r300->hyperz_enabled) {
 230                    /* Need to emit HyperZ buffer regs for the first time. */
 231                    r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 232                 }
 233             }
 234
 235             /* Setup Hyper-Z clears. */
 236             if (r300->hyperz_enabled) {
 237                 if (zmask_clear) {
 238                     hyperz_dcv = hyperz->zb_depthclearvalue =
 239                         r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
 240
 241                     r300_mark_atom_dirty(r300, &r300->zmask_clear);
 242                     buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 243                 }
 244
 245                 if (hiz_clear) {
 246                     r300->hiz_clear_value = r300_hiz_clear_value(depth);
 247                     r300_mark_atom_dirty(r300, &r300->hiz_clear);
 248                 }
 249             }
 250         }
 251     }
 252
 253     /* Enable CBZB clear. */
 254     if (r300_cbzb_clear_allowed(r300, buffers)) {
 255         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 256
 257         hyperz->zb_depthclearvalue =
 258                 r300_depth_clear_cb_value(surf->base.format, rgba);
 259
 260         width = surf->cbzb_width;
 261         height = surf->cbzb_height;
 262
 263         r300->cbzb_clear = TRUE;
 264         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 265     }
 266
 267     /* Clear. */
 268     if (buffers) {
 269         /* Clear using the blitter. */
 270         r300_blitter_begin(r300, R300_CLEAR);
 271         util_blitter_clear(r300->blitter,
 272                            width,
 273                            height,
 274                            fb->nr_cbufs,
 275                            buffers, rgba, depth, stencil);
 276         r300_blitter_end(r300);
 277     } else if (r300->zmask_clear.dirty || r300->hiz_clear.dirty) {
 278         /* Just clear zmask and hiz now, this does not use the standard draw
 279          * procedure. */
 280         /* Calculate zmask_clear and hiz_clear atom sizes. */
 281         unsigned dwords =
 282             (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
 283             (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
 284             r300_get_num_cs_end_dwords(r300);
 285
 286         /* Reserve CS space. */
 287         if (dwords > (RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
 288             r300_flush(&r300->context, RADEON_FLUSH_ASYNC, NULL);
 289         }
 290
 291         /* Emit clear packets. */
 292         if (r300->zmask_clear.dirty) {
 293             r300_emit_zmask_clear(r300, r300->zmask_clear.size,
 294                                   r300->zmask_clear.state);
 295             r300->zmask_clear.dirty = FALSE;
 296         }
 297         if (r300->hiz_clear.dirty) {
 298             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
 299                                 r300->hiz_clear.state);
 300             r300->hiz_clear.dirty = FALSE;
 301         }
 302     } else {
 303         assert(0);
 304     }
 305
 306     /* Disable CBZB clear. */
 307     if (r300->cbzb_clear) {
 308         r300->cbzb_clear = FALSE;
 309         hyperz->zb_depthclearvalue = hyperz_dcv;
 310         r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
 311     }
 312
 313     /* Enable fastfill and/or hiz.
 314      *
 315      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
 316      * looks if zmask/hiz is in use and programs hardware accordingly. */
 317     if (r300->zmask_in_use || r300->hiz_in_use) {
 318         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 319     }
 320 }
 321
 322 /* Clear a region of a color surface to a constant value. */
 323 static void r300_clear_render_target(struct pipe_context *pipe,
 324                                      struct pipe_surface *dst,
 325                                      const float *rgba,
 326                                      unsigned dstx, unsigned dsty,
 327                                      unsigned width, unsigned height)
 328 {
 329     struct r300_context *r300 = r300_context(pipe);
 330
 331     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 332     util_blitter_clear_render_target(r300->blitter, dst, rgba,
 333                                      dstx, dsty, width, height);
 334     r300_blitter_end(r300);
 335 }
 336
 337 /* Clear a region of a depth stencil surface. */
 338 static void r300_clear_depth_stencil(struct pipe_context *pipe,
 339                                      struct pipe_surface *dst,
 340                                      unsigned clear_flags,
 341                                      double depth,
 342                                      unsigned stencil,
 343                                      unsigned dstx, unsigned dsty,
 344                                      unsigned width, unsigned height)
 345 {
 346     struct r300_context *r300 = r300_context(pipe);
 347     struct pipe_framebuffer_state *fb =
 348         (struct pipe_framebuffer_state*)r300->fb_state.state;
 349
 350     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 351         if (fb->zsbuf->texture == dst->texture) {
 352             r300_decompress_zmask(r300);
 353         }
 354     }
 355
 356     /* XXX Do not decompress ZMask of the currently-set zbuffer. */
 357     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 358     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
 359                                      dstx, dsty, width, height);
 360     r300_blitter_end(r300);
 361 }
 362
 363 void r300_decompress_zmask(struct r300_context *r300)
 364 {
 365     struct pipe_framebuffer_state *fb =
 366         (struct pipe_framebuffer_state*)r300->fb_state.state;
 367
 368     if (!r300->zmask_in_use || r300->locked_zbuffer)
 369         return;
 370
 371     r300->zmask_decompress = TRUE;
 372     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 373
 374     r300_blitter_begin(r300, R300_CLEAR);
 375     util_blitter_clear_depth_custom(r300->blitter, fb->width, fb->height, 0,
 376                                     r300->dsa_decompress_zmask);
 377     r300_blitter_end(r300);
 378
 379     r300->zmask_decompress = FALSE;
 380     r300->zmask_in_use = FALSE;
 381     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 382 }
 383
 384 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
 385 {
 386     struct pipe_framebuffer_state fb = {0};
 387     fb.width = r300->locked_zbuffer->width;
 388     fb.height = r300->locked_zbuffer->height;
 389     fb.nr_cbufs = 0;
 390     fb.zsbuf = r300->locked_zbuffer;
 391
 392     r300->context.set_framebuffer_state(&r300->context, &fb);
 393     r300_decompress_zmask(r300);
 394 }
 395
 396 void r300_decompress_zmask_locked(struct r300_context *r300)
 397 {
 398     struct pipe_framebuffer_state saved_fb = {0};
 399
 400     util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
 401     r300_decompress_zmask_locked_unsafe(r300);
 402     r300->context.set_framebuffer_state(&r300->context, &saved_fb);
 403     util_unreference_framebuffer_state(&saved_fb);
 404
 405     pipe_surface_reference(&r300->locked_zbuffer, NULL);
 406 }
 407
 408 /* Copy a block of pixels from one surface to another using HW. */
 409 static void r300_hw_copy_region(struct pipe_context* pipe,
 410                                 struct pipe_resource *dst,
 411                                 unsigned dst_level,
 412                                 unsigned dstx, unsigned dsty, unsigned dstz,
 413                                 struct pipe_resource *src,
 414                                 unsigned src_level,
 415                                 const struct pipe_box *src_box)
 416 {
 417     struct r300_context* r300 = r300_context(pipe);
 418
 419     r300_blitter_begin(r300, R300_COPY);
 420     util_blitter_copy_region(r300->blitter, dst, dst_level, dstx, dsty, dstz,
 421                              src, src_level, src_box, TRUE);
 422     r300_blitter_end(r300);
 423 }
 424
 425 /* Copy a block of pixels from one surface to another. */
 426 static void r300_resource_copy_region(struct pipe_context *pipe,
 427                                       struct pipe_resource *dst,
 428                                       unsigned dst_level,
 429                                       unsigned dstx, unsigned dsty, unsigned dstz,
 430                                       struct pipe_resource *src,
 431                                       unsigned src_level,
 432                                       const struct pipe_box *src_box)
 433 {
 434     struct r300_context *r300 = r300_context(pipe);
 435     struct pipe_framebuffer_state *fb =
 436         (struct pipe_framebuffer_state*)r300->fb_state.state;
 437     struct pipe_resource old_src = *src;
 438     struct pipe_resource old_dst = *dst;
 439     struct pipe_resource new_src = old_src;
 440     struct pipe_resource new_dst = old_dst;
 441     const struct util_format_description *desc =
 442             util_format_description(dst->format);
 443     struct pipe_box box;
 444
 445     /* Fallback for buffers. */
 446     if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
 447         util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 448                                   src, src_level, src_box);
 449         return;
 450     }
 451
 452     if (r300->zmask_in_use && !r300->locked_zbuffer) {
 453         if (fb->zsbuf->texture == src ||
 454             fb->zsbuf->texture == dst) {
 455             r300_decompress_zmask(r300);
 456         }
 457     }
 458
 459     /* Handle non-renderable plain formats. */
 460     if (desc->layout == UTIL_FORMAT_LAYOUT_PLAIN &&
 461         (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB ||
 462          !pipe->screen->is_format_supported(pipe->screen,
 463                                             src->format, src->target,
 464                                             src->nr_samples,
 465                                             PIPE_BIND_SAMPLER_VIEW) ||
 466          !pipe->screen->is_format_supported(pipe->screen,
 467                                             dst->format, dst->target,
 468                                             dst->nr_samples,
 469                                             PIPE_BIND_RENDER_TARGET))) {
 470         switch (util_format_get_blocksize(old_dst.format)) {
 471             case 1:
 472                 new_dst.format = PIPE_FORMAT_I8_UNORM;
 473                 break;
 474             case 2:
 475                 new_dst.format = PIPE_FORMAT_B4G4R4A4_UNORM;
 476                 break;
 477             case 4:
 478                 new_dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 479                 break;
 480             case 8:
 481                 new_dst.format = PIPE_FORMAT_R16G16B16A16_UNORM;
 482                 break;
 483             default:
 484                 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
 485                              "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
 486                              util_format_short_name(dst->format));
 487         }
 488         new_src.format = new_dst.format;
 489     }
 490
 491     /* Handle compressed formats. */
 492     if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC ||
 493         desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
 494         switch (util_format_get_blocksize(old_dst.format)) {
 495         case 8:
 496             /* 1 pixel = 4 bits,
 497              * we set 1 pixel = 2 bytes ===> 4 times larger pixels. */
 498             new_dst.format = PIPE_FORMAT_B4G4R4A4_UNORM;
 499             break;
 500         case 16:
 501             /* 1 pixel = 8 bits,
 502              * we set 1 pixel = 4 bytes ===> 4 times larger pixels. */
 503             new_dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 504             break;
 505         }
 506
 507         /* Since the pixels are 4 times larger, we must decrease
 508          * the image size and the coordinates 4 times. */
 509         new_src.format = new_dst.format;
 510         new_dst.height0 = (new_dst.height0 + 3) / 4;
 511         new_src.height0 = (new_src.height0 + 3) / 4;
 512         dsty /= 4;
 513         box = *src_box;
 514         box.y /= 4;
 515         box.height = (box.height + 3) / 4;
 516         src_box = &box;
 517     }
 518
 519     if (old_src.format != new_src.format)
 520         r300_resource_set_properties(pipe->screen, src, 0, &new_src);
 521     if (old_dst.format != new_dst.format)
 522         r300_resource_set_properties(pipe->screen, dst, 0, &new_dst);
 523
 524     r300_hw_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 525                         src, src_level, src_box);
 526
 527     if (old_src.format != new_src.format)
 528         r300_resource_set_properties(pipe->screen, src, 0, &old_src);
 529     if (old_dst.format != new_dst.format)
 530         r300_resource_set_properties(pipe->screen, dst, 0, &old_dst);
 531 }
 532
 533 void r300_init_blit_functions(struct r300_context *r300)
 534 {
 535     r300->context.clear = r300_clear;
 536     r300->context.clear_render_target = r300_clear_render_target;
 537     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
 538     r300->context.resource_copy_region = r300_resource_copy_region;
 539 }