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     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 uint32_t r300_depth_clear_value(enum pipe_format format,
 112                                        double depth, unsigned stencil)
 113 {
 114     switch (format) {
 115         case PIPE_FORMAT_Z16_UNORM:
 116         case PIPE_FORMAT_X8Z24_UNORM:
 117             return util_pack_z(format, depth);
 118
 119         case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
 120             return util_pack_z_stencil(format, depth, stencil);
 121
 122         default:
 123             assert(0);
 124             return 0;
 125     }
 126 }
 127
 128 /* Clear currently bound buffers. */
 129 static void r300_clear(struct pipe_context* pipe,
 130                        unsigned buffers,
 131                        const float* rgba,
 132                        double depth,
 133                        unsigned stencil)
 134 {
 135     /* My notes about fastfill:
 136      *
 137      * 1) Only the zbuffer is cleared.
 138      *
 139      * 2) The zbuffer must be micro-tiled and whole microtiles must be
 140      *    written. If microtiling is disabled, it locks up.
 141      *
 142      * 3) There is Z Mask RAM which contains a compressed zbuffer and
 143      *    it interacts with fastfill. We should figure out how to use it
 144      *    to get more performance.
 145      *    This is what we know about the Z Mask:
 146      *
 147      *       Each dword of the Z Mask contains compression information
 148      *       for 16 4x4 pixel blocks, that is 2 bits for each block.
 149      *       On chips with 2 Z pipes, every other dword maps to a different
 150      *       pipe.
 151      *
 152      * 4) ZB_DEPTHCLEARVALUE is used to clear the zbuffer and the Z Mask must
 153      *    be equal to 0. (clear the Z Mask RAM with zeros)
 154      *
 155      * 5) For 16-bit zbuffer, compression causes a hung with one or
 156      *    two samples and should not be used.
 157      *
 158      * 6) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
 159      *    to avoid needless decompression.
 160      *
 161      * 7) Fastfill must not be used if reading of compressed Z data is disabled
 162      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
 163      *    i.e. it cannot be used to compress the zbuffer.
 164      *
 165      * 8) ZB_CB_CLEAR does not interact with fastfill in any way.
 166      *
 167      * - Marek
 168      */
 169
 170     struct r300_context* r300 = r300_context(pipe);
 171     struct pipe_framebuffer_state *fb =
 172         (struct pipe_framebuffer_state*)r300->fb_state.state;
 173     struct r300_hyperz_state *hyperz =
 174         (struct r300_hyperz_state*)r300->hyperz_state.state;
 175     struct r300_texture *zstex =
 176             fb->zsbuf ? r300_texture(fb->zsbuf->texture) : NULL;
 177     uint32_t width = fb->width;
 178     uint32_t height = fb->height;
 179     boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
 180     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
 181
 182     /* Decompress zbuffers that are bound as textures. If we didn't flush here,
 183      * it would happen inside the blitter when updating derived state,
 184      * causing a blitter operation to be called from inside the blitter,
 185      * which would overwrite saved states and they would never get restored. */
 186     r300_flush_depth_textures(r300);
 187
 188     /* Enable fast Z clear.
 189      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
 190     if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && can_hyperz) {
 191         hyperz_dcv = hyperz->zb_depthclearvalue =
 192             r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
 193
 194         r300_mark_fb_state_dirty(r300, R300_CHANGED_ZCLEAR_FLAG);
 195         if (zstex->zmask_mem[fb->zsbuf->u.tex.level]) {
 196             r300_mark_atom_dirty(r300, &r300->zmask_clear);
 197             buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 198         }
 199         if (zstex->hiz_mem[fb->zsbuf->u.tex.level])
 200             r300_mark_atom_dirty(r300, &r300->hiz_clear);
 201     }
 202
 203     /* Enable CBZB clear. */
 204     if (r300_cbzb_clear_allowed(r300, buffers)) {
 205         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 206
 207         hyperz->zb_depthclearvalue =
 208                 r300_depth_clear_cb_value(surf->base.format, rgba);
 209
 210         width = surf->cbzb_width;
 211         height = surf->cbzb_height;
 212
 213         r300->cbzb_clear = TRUE;
 214         r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
 215     }
 216
 217     /* Clear. */
 218     if (buffers) {
 219         /* Clear using the blitter. */
 220         r300_blitter_begin(r300, R300_CLEAR);
 221         util_blitter_clear(r300->blitter,
 222                            width,
 223                            height,
 224                            fb->nr_cbufs,
 225                            buffers, rgba, depth, stencil);
 226         r300_blitter_end(r300);
 227     } else if (r300->zmask_clear.dirty) {
 228         /* Just clear zmask and hiz now, this does not use a standard draw
 229          * procedure. */
 230         unsigned dwords;
 231
 232         /* Calculate zmask_clear and hiz_clear atom sizes. */
 233         r300_update_hyperz_state(r300);
 234         dwords = r300->zmask_clear.size +
 235                  (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
 236                  r300_get_num_cs_end_dwords(r300);
 237
 238         /* Reserve CS space. */
 239         if (dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
 240             r300->context.flush(&r300->context, 0, NULL);
 241         }
 242
 243         /* Emit clear packets. */
 244         r300_emit_zmask_clear(r300, r300->zmask_clear.size,
 245                               r300->zmask_clear.state);
 246         r300->zmask_clear.dirty = FALSE;
 247         if (r300->hiz_clear.dirty) {
 248             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
 249                                 r300->hiz_clear.state);
 250             r300->hiz_clear.dirty = FALSE;
 251         }
 252     } else {
 253         assert(0);
 254     }
 255
 256     /* Disable CBZB clear. */
 257     if (r300->cbzb_clear) {
 258         r300->cbzb_clear = FALSE;
 259         hyperz->zb_depthclearvalue = hyperz_dcv;
 260         r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
 261     }
 262
 263     /* Enable fastfill and/or hiz.
 264      *
 265      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
 266      * looks if zmask/hiz is in use and enables fastfill accordingly. */
 267     if (zstex &&
 268         (zstex->zmask_in_use[fb->zsbuf->u.tex.level] ||
 269          zstex->hiz_in_use[fb->zsbuf->u.tex.level])) {
 270         r300_mark_atom_dirty(r300, &r300->hyperz_state);
 271     }
 272 }
 273
 274 /* Clear a region of a color surface to a constant value. */
 275 static void r300_clear_render_target(struct pipe_context *pipe,
 276                                      struct pipe_surface *dst,
 277                                      const float *rgba,
 278                                      unsigned dstx, unsigned dsty,
 279                                      unsigned width, unsigned height)
 280 {
 281     struct r300_context *r300 = r300_context(pipe);
 282
 283     /* Decompress zbuffers that are bound as textures. If we didn't flush here,
 284      * it would happen inside the blitter when updating derived state,
 285      * causing a blitter operation to be called from inside the blitter,
 286      * which would overwrite saved states and they would never get restored. */
 287     r300_flush_depth_textures(r300);
 288
 289     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 290     util_blitter_clear_render_target(r300->blitter, dst, rgba,
 291                                      dstx, dsty, width, height);
 292     r300_blitter_end(r300);
 293 }
 294
 295 /* Clear a region of a depth stencil surface. */
 296 static void r300_clear_depth_stencil(struct pipe_context *pipe,
 297                                      struct pipe_surface *dst,
 298                                      unsigned clear_flags,
 299                                      double depth,
 300                                      unsigned stencil,
 301                                      unsigned dstx, unsigned dsty,
 302                                      unsigned width, unsigned height)
 303 {
 304     struct r300_context *r300 = r300_context(pipe);
 305
 306     /* Decompress zbuffers that are bound as textures. If we didn't flush here,
 307      * it would happen inside the blitter when updating derived state,
 308      * causing a blitter operation to be called from inside the blitter,
 309      * which would overwrite saved states and they would never get restored. */
 310     r300_flush_depth_textures(r300);
 311
 312     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 313     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
 314                                      dstx, dsty, width, height);
 315     r300_blitter_end(r300);
 316 }
 317
 318 /* Flush a depth stencil buffer. */
 319 static void r300_flush_depth_stencil(struct pipe_context *pipe,
 320                                      struct pipe_resource *dst,
 321                                      unsigned level,
 322                                      unsigned layer)
 323 {
 324     struct r300_context *r300 = r300_context(pipe);
 325     struct pipe_surface *dstsurf, surf_tmpl;
 326     struct r300_texture *tex = r300_texture(dst);
 327
 328     if (!tex->zmask_mem[level])
 329         return;
 330     if (!tex->zmask_in_use[level])
 331         return;
 332
 333     surf_tmpl.format = dst->format;
 334     surf_tmpl.usage = PIPE_BIND_DEPTH_STENCIL;
 335     surf_tmpl.u.tex.level = level;
 336     surf_tmpl.u.tex.first_layer = layer;
 337     surf_tmpl.u.tex.last_layer = layer;
 338     dstsurf = pipe->create_surface(pipe, dst, &surf_tmpl);
 339
 340     r300->z_decomp_rd = TRUE;
 341
 342     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 343     util_blitter_flush_depth_stencil(r300->blitter, dstsurf);
 344     r300_blitter_end(r300);
 345     r300->z_decomp_rd = FALSE;
 346
 347     tex->zmask_in_use[level] = FALSE;
 348     pipe_surface_reference(&dstsurf, NULL);
 349 }
 350
 351 /* We can't use compressed zbuffers as samplers. */
 352 void r300_flush_depth_textures(struct r300_context *r300)
 353 {
 354     struct r300_textures_state *state =
 355         (struct r300_textures_state*)r300->textures_state.state;
 356     unsigned i, level;
 357     unsigned count = MIN2(state->sampler_view_count,
 358                           state->sampler_state_count);
 359
 360     if (r300->z_decomp_rd)
 361         return;
 362
 363     for (i = 0; i < count; i++)
 364         if (state->sampler_views[i] && state->sampler_states[i]) {
 365             struct pipe_resource *tex = state->sampler_views[i]->base.texture;
 366
 367             if (tex->target == PIPE_TEXTURE_3D ||
 368                 tex->target == PIPE_TEXTURE_CUBE)
 369                 continue;
 370
 371             /* Ignore non-depth textures.
 372              * Also ignore reinterpreted depth textures, e.g. resource_copy. */
 373             if (!util_format_is_depth_or_stencil(tex->format))
 374                 continue;
 375
 376             for (level = 0; level <= tex->last_level; level++)
 377                 if (r300_texture(tex)->zmask_in_use[level]) {
 378                     /* We don't handle 3D textures and cubemaps yet. */
 379                     r300_flush_depth_stencil(&r300->context, tex, level, 0);
 380                  }
 381         }
 382 }
 383
 384 /* Copy a block of pixels from one surface to another using HW. */
 385 static void r300_hw_copy_region(struct pipe_context* pipe,
 386                                 struct pipe_resource *dst,
 387                                 unsigned dst_level,
 388                                 unsigned dstx, unsigned dsty, unsigned dstz,
 389                                 struct pipe_resource *src,
 390                                 unsigned src_level,
 391                                 const struct pipe_box *src_box)
 392 {
 393     struct r300_context* r300 = r300_context(pipe);
 394
 395     r300_blitter_begin(r300, R300_COPY);
 396
 397     /* Do a 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     enum pipe_format old_format = dst->format;
 413     enum pipe_format new_format = old_format;
 414     boolean is_depth;
 415
 416     if (!pipe->screen->is_format_supported(pipe->screen,
 417                                            old_format, src->target,
 418                                            src->nr_samples,
 419                                            PIPE_BIND_RENDER_TARGET |
 420                                            PIPE_BIND_SAMPLER_VIEW, 0) &&
 421         util_format_is_plain(old_format)) {
 422         switch (util_format_get_blocksize(old_format)) {
 423             case 1:
 424                 new_format = PIPE_FORMAT_I8_UNORM;
 425                 break;
 426             case 2:
 427                 new_format = PIPE_FORMAT_B4G4R4A4_UNORM;
 428                 break;
 429             case 4:
 430                 new_format = PIPE_FORMAT_B8G8R8A8_UNORM;
 431                 break;
 432             case 8:
 433                 new_format = PIPE_FORMAT_R16G16B16A16_UNORM;
 434                 break;
 435             default:
 436                 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
 437                              "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
 438                              util_format_short_name(old_format));
 439         }
 440     }
 441
 442     is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
 443     if (is_depth) {
 444         r300_flush_depth_stencil(pipe, src, src_level, src_box->z);
 445     }
 446
 447     if (old_format != new_format) {
 448         r300_texture_reinterpret_format(pipe->screen,
 449                                         dst, new_format);
 450         r300_texture_reinterpret_format(pipe->screen,
 451                                         src, new_format);
 452     }
 453
 454     r300_hw_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
 455                         src, src_level, src_box);
 456
 457     if (old_format != new_format) {
 458         r300_texture_reinterpret_format(pipe->screen,
 459                                         dst, old_format);
 460         r300_texture_reinterpret_format(pipe->screen,
 461                                         src, old_format);
 462     }
 463 }
 464
 465 void r300_init_blit_functions(struct r300_context *r300)
 466 {
 467     r300->context.clear = r300_clear;
 468     r300->context.clear_render_target = r300_clear_render_target;
 469     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
 470     r300->context.resource_copy_region = r300_resource_copy_region;
 471 }