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_texture.h"
  25 #include "r300_winsys.h"
  26
  27 #include "util/u_format.h"
  28 #include "util/u_pack_color.h"
  29
  30 enum r300_blitter_op /* bitmask */
  31 {
  32     R300_CLEAR         = 1,
  33     R300_CLEAR_SURFACE = 2,
  34     R300_COPY          = 4
  35 };
  36
  37 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
  38 {
  39     if (r300->query_current) {
  40         r300->blitter_saved_query = r300->query_current;
  41         r300_stop_query(r300);
  42     }
  43
  44     /* Yeah we have to save all those states to ensure the blitter operation
  45      * is really transparent. The states will be restored by the blitter once
  46      * copying is done. */
  47     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
  48     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
  49     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
  50     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
  51     util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
  52     util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
  53     util_blitter_save_viewport(r300->blitter, &r300->viewport);
  54     util_blitter_save_clip(r300->blitter, (struct pipe_clip_state*)r300->clip_state.state);
  55     util_blitter_save_vertex_elements(r300->blitter, r300->velems);
  56     util_blitter_save_vertex_buffers(r300->blitter, r300->vertex_buffer_count,
  57                                      r300->vertex_buffer);
  58
  59     if (op & (R300_CLEAR_SURFACE | R300_COPY))
  60         util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
  61
  62     if (op & R300_COPY) {
  63         struct r300_textures_state* state =
  64             (struct r300_textures_state*)r300->textures_state.state;
  65
  66         util_blitter_save_fragment_sampler_states(
  67             r300->blitter, state->sampler_state_count,
  68             (void**)state->sampler_states);
  69
  70         util_blitter_save_fragment_sampler_views(
  71             r300->blitter, state->sampler_view_count,
  72             (struct pipe_sampler_view**)state->sampler_views);
  73     }
  74 }
  75
  76 static void r300_blitter_end(struct r300_context *r300)
  77 {
  78     if (r300->blitter_saved_query) {
  79         r300_resume_query(r300, r300->blitter_saved_query);
  80         r300->blitter_saved_query = NULL;
  81     }
  82 }
  83
  84 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
  85                                           const float* rgba)
  86 {
  87     union util_color uc;
  88     util_pack_color(rgba, format, &uc);
  89
  90     if (util_format_get_blocksizebits(format) == 32)
  91         return uc.ui;
  92     else
  93         return uc.us | (uc.us << 16);
  94 }
  95
  96 static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
  97                                        unsigned clear_buffers)
  98 {
  99     struct pipe_framebuffer_state *fb =
 100         (struct pipe_framebuffer_state*)r300->fb_state.state;
 101
 102     /* Only color clear allowed, and only one colorbuffer. */
 103     if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
 104         return FALSE;
 105
 106     return r300_surface(fb->cbufs[0])->cbzb_allowed;
 107 }
 108
 109 static uint32_t r300_depth_clear_value(enum pipe_format format,
 110                                        double depth, unsigned stencil)
 111 {
 112     switch (format) {
 113         case PIPE_FORMAT_Z16_UNORM:
 114         case PIPE_FORMAT_X8Z24_UNORM:
 115             return util_pack_z(format, depth);
 116
 117         case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
 118             return util_pack_z_stencil(format, depth, stencil);
 119
 120         default:
 121             assert(0);
 122             return 0;
 123     }
 124 }
 125
 126 /* Clear currently bound buffers. */
 127 static void r300_clear(struct pipe_context* pipe,
 128                        unsigned buffers,
 129                        const float* rgba,
 130                        double depth,
 131                        unsigned stencil)
 132 {
 133     /* My notes about fastfill:
 134      *
 135      * 1) Only the zbuffer is cleared.
 136      *
 137      * 2) The zbuffer must be micro-tiled and whole microtiles must be
 138      *    written. If microtiling is disabled, it locks up.
 139      *
 140      * 3) There is Z Mask RAM which contains a compressed zbuffer and
 141      *    it interacts with fastfill. We should figure out how to use it
 142      *    to get more performance.
 143      *    This is what we know about the Z Mask:
 144      *
 145      *       Each dword of the Z Mask contains compression information
 146      *       for 16 4x4 pixel blocks, that is 2 bits for each block.
 147      *       On chips with 2 Z pipes, every other dword maps to a different
 148      *       pipe.
 149      *
 150      * 4) ZB_DEPTHCLEARVALUE is used to clear the zbuffer and the Z Mask must
 151      *    be equal to 0. (clear the Z Mask RAM with zeros)
 152      *
 153      * 5) For 16-bit zbuffer, compression causes a hung with one or
 154      *    two samples and should not be used.
 155      *
 156      * 6) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
 157      *    to avoid needless decompression.
 158      *
 159      * 7) Fastfill must not be used if reading of compressed Z data is disabled
 160      *    and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
 161      *    i.e. it cannot be used to compress the zbuffer.
 162      *
 163      * 8) ZB_CB_CLEAR does not interact with fastfill in any way.
 164      *
 165      * - Marek
 166      */
 167
 168     struct r300_context* r300 = r300_context(pipe);
 169     struct pipe_framebuffer_state *fb =
 170         (struct pipe_framebuffer_state*)r300->fb_state.state;
 171     struct r300_hyperz_state *hyperz =
 172         (struct r300_hyperz_state*)r300->hyperz_state.state;
 173     uint32_t width = fb->width;
 174     uint32_t height = fb->height;
 175     boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
 176     uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
 177
 178     /* Enable fast Z clear.
 179      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
 180     if ((buffers & (PIPE_CLEAR_DEPTH|PIPE_CLEAR_STENCIL)) && has_hyperz) {
 181
 182         hyperz_dcv = hyperz->zb_depthclearvalue =
 183             r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
 184
 185         r300_mark_fb_state_dirty(r300, R300_CHANGED_ZCLEAR_FLAG);
 186         if (r300_texture(fb->zsbuf->texture)->zmask_mem[fb->zsbuf->level]) {
 187             r300->zmask_clear.dirty = TRUE;
 188             buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 189         }
 190         if (r300->hiz_enable)
 191             r300->hiz_clear.dirty = TRUE;
 192     }
 193
 194     /* Enable CBZB clear. */
 195     if (r300_cbzb_clear_allowed(r300, buffers)) {
 196         struct r300_surface *surf = r300_surface(fb->cbufs[0]);
 197
 198         hyperz->zb_depthclearvalue =
 199                 r300_depth_clear_cb_value(surf->base.format, rgba);
 200
 201         width = surf->cbzb_width;
 202         height = surf->cbzb_height;
 203
 204         r300->cbzb_clear = TRUE;
 205         r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
 206     }
 207
 208     /* Clear. */
 209     r300_blitter_begin(r300, R300_CLEAR);
 210     util_blitter_clear(r300->blitter,
 211                        width,
 212                        height,
 213                        fb->nr_cbufs,
 214                        buffers, rgba, depth, stencil);
 215     r300_blitter_end(r300);
 216
 217     /* Disable CBZB clear. */
 218     if (r300->cbzb_clear) {
 219         r300->cbzb_clear = FALSE;
 220         hyperz->zb_depthclearvalue = hyperz_dcv;
 221         r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
 222     }
 223
 224     /* Enable fastfill.
 225      *
 226      * If we cleared the zmask, it's dirty now. The Hyper-Z state update
 227      * looks for a dirty zmask and enables fastfill accordingly. */
 228     if (fb->zsbuf &&
 229         r300_texture(fb->zsbuf->texture)->dirty_zmask[fb->zsbuf->level]) {
 230         r300->hyperz_state.dirty = TRUE;
 231     }
 232
 233     /* XXX this flush "fixes" a hardlock in the cubestorm xscreensaver */
 234     if (r300->flush_counter == 0)
 235         pipe->flush(pipe, 0, NULL);
 236 }
 237
 238 /* Clear a region of a color surface to a constant value. */
 239 static void r300_clear_render_target(struct pipe_context *pipe,
 240                                      struct pipe_surface *dst,
 241                                      const float *rgba,
 242                                      unsigned dstx, unsigned dsty,
 243                                      unsigned width, unsigned height)
 244 {
 245     struct r300_context *r300 = r300_context(pipe);
 246
 247     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 248     util_blitter_clear_render_target(r300->blitter, dst, rgba,
 249                                      dstx, dsty, width, height);
 250     r300_blitter_end(r300);
 251 }
 252
 253 /* Clear a region of a depth stencil surface. */
 254 static void r300_clear_depth_stencil(struct pipe_context *pipe,
 255                                      struct pipe_surface *dst,
 256                                      unsigned clear_flags,
 257                                      double depth,
 258                                      unsigned stencil,
 259                                      unsigned dstx, unsigned dsty,
 260                                      unsigned width, unsigned height)
 261 {
 262     struct r300_context *r300 = r300_context(pipe);
 263
 264     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 265     util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
 266                                      dstx, dsty, width, height);
 267     r300_blitter_end(r300);
 268 }
 269
 270 /* Flush a depth stencil buffer. */
 271 void r300_flush_depth_stencil(struct pipe_context *pipe,
 272                               struct pipe_resource *dst,
 273                               struct pipe_subresource subdst,
 274                               unsigned zslice)
 275 {
 276     struct r300_context *r300 = r300_context(pipe);
 277     struct pipe_surface *dstsurf;
 278     struct r300_texture *tex = r300_texture(dst);
 279
 280     if (!tex->zmask_mem[subdst.level])
 281         return;
 282     if (!tex->dirty_zmask[subdst.level])
 283         return;
 284
 285     dstsurf = pipe->screen->get_tex_surface(pipe->screen, dst,
 286                                             subdst.face, subdst.level, zslice,
 287                                             PIPE_BIND_DEPTH_STENCIL);
 288     r300->z_decomp_rd = TRUE;
 289     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
 290     util_blitter_flush_depth_stencil(r300->blitter, dstsurf);
 291     r300_blitter_end(r300);
 292     r300->z_decomp_rd = FALSE;
 293
 294     tex->dirty_zmask[subdst.level] = FALSE;
 295 }
 296
 297 /* Copy a block of pixels from one surface to another using HW. */
 298 static void r300_hw_copy_region(struct pipe_context* pipe,
 299                                 struct pipe_resource *dst,
 300                                 struct pipe_subresource subdst,
 301                                 unsigned dstx, unsigned dsty, unsigned dstz,
 302                                 struct pipe_resource *src,
 303                                 struct pipe_subresource subsrc,
 304                                 unsigned srcx, unsigned srcy, unsigned srcz,
 305                                 unsigned width, unsigned height)
 306 {
 307     struct r300_context* r300 = r300_context(pipe);
 308
 309     r300_blitter_begin(r300, R300_COPY);
 310     util_blitter_copy_region(r300->blitter, dst, subdst, dstx, dsty, dstz,
 311                              src, subsrc, srcx, srcy, srcz, width, height,
 312                              TRUE);
 313     r300_blitter_end(r300);
 314 }
 315
 316 /* Copy a block of pixels from one surface to another. */
 317 static void r300_resource_copy_region(struct pipe_context *pipe,
 318                                       struct pipe_resource *dst,
 319                                       struct pipe_subresource subdst,
 320                                       unsigned dstx, unsigned dsty, unsigned dstz,
 321                                       struct pipe_resource *src,
 322                                       struct pipe_subresource subsrc,
 323                                       unsigned srcx, unsigned srcy, unsigned srcz,
 324                                       unsigned width, unsigned height)
 325 {
 326     enum pipe_format old_format = dst->format;
 327     enum pipe_format new_format = old_format;
 328     boolean is_depth;
 329     if (!pipe->screen->is_format_supported(pipe->screen,
 330                                            old_format, src->target,
 331                                            src->nr_samples,
 332                                            PIPE_BIND_RENDER_TARGET |
 333                                            PIPE_BIND_SAMPLER_VIEW, 0) &&
 334         util_format_is_plain(old_format)) {
 335         switch (util_format_get_blocksize(old_format)) {
 336             case 1:
 337                 new_format = PIPE_FORMAT_I8_UNORM;
 338                 break;
 339             case 2:
 340                 new_format = PIPE_FORMAT_B4G4R4A4_UNORM;
 341                 break;
 342             case 4:
 343                 new_format = PIPE_FORMAT_B8G8R8A8_UNORM;
 344                 break;
 345             case 8:
 346                 new_format = PIPE_FORMAT_R16G16B16A16_UNORM;
 347                 break;
 348             default:
 349                 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
 350                              "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
 351                              util_format_short_name(old_format));
 352         }
 353     }
 354
 355     is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
 356     if (is_depth) {
 357         r300_flush_depth_stencil(pipe, src, subsrc, srcz);
 358     }
 359     if (old_format != new_format) {
 360         dst->format = new_format;
 361         src->format = new_format;
 362
 363         r300_texture_reinterpret_format(pipe->screen,
 364                                         dst, new_format);
 365         r300_texture_reinterpret_format(pipe->screen,
 366                                         src, new_format);
 367     }
 368
 369     r300_hw_copy_region(pipe, dst, subdst, dstx, dsty, dstz,
 370                         src, subsrc, srcx, srcy, srcz, width, height);
 371
 372     if (old_format != new_format) {
 373         dst->format = old_format;
 374         src->format = old_format;
 375
 376         r300_texture_reinterpret_format(pipe->screen,
 377                                         dst, old_format);
 378         r300_texture_reinterpret_format(pipe->screen,
 379                                         src, old_format);
 380     }
 381 }
 382
 383 void r300_init_blit_functions(struct r300_context *r300)
 384 {
 385     r300->context.clear = r300_clear;
 386     r300->context.clear_render_target = r300_clear_render_target;
 387     r300->context.clear_depth_stencil = r300_clear_depth_stencil;
 388     r300->context.resource_copy_region = r300_resource_copy_region;
 389 }